Datasets:

Hendrik-a

/

arxiv

like 0

herein we study the inverse problem on inferring depth profile of near - surface residual stress in a weakly anisotropic medium by boundary measurement of rayleigh - wave dispersion if all other relevant material parameters of the elastic medium are known . our solution of this inverse problem is based on a recently developed algorithm by which each term of a high - frequency asymptotic formula for dispersion relations can be computed for rayleigh waves that propagate in various directions along the free surface of a vertically - inhomogeneous , prestressed , and weakly anisotropic half - space . as a prime example of possible applications we focus on a thick - plate sample of aa 7075-t651 aluminum alloy , which has one face treated by low plasticity burnishing ( lpb ) that induced a depth - dependent prestress at and immediately beneath the treated surface . we model the sample as a prestressed , weakly - textured orthorhombic aggregate of cubic crystallites and assume that by nondestructive and/or destructive measurements we have ascertained everything about the sample , including the lpb - induced prestress , before it is put into service . under the supposition that the prestress be partially relaxed but other material parameters remain unchanged after the sample undergoes a period of service , we examine the possibility of inferring the depth profile of the partially relaxed stress by boundary measurement of rayleigh - wave dispersion . * keywords * : inverse problem , acoustoelasticity , residual stress , rayleigh waves , dispersion relations , textured media , surface conditioning [ section ] [ thm]corollary [ thm]lemma [ thm]proposition [ thm]definition [ thm]remark

1610.06413

a common practice to provide lifetime enhancement against fatigue and stress - corrosion cracking of metallic parts ( e.g. , critical components of aircraft engines , welds in steel structures , etc . ) is to impart , through surface - conditioning treatments such as shot peening , sand blasting , laser peening , and low plasticity burnishing , a thin surface layer of compressive residual stress on the parts so treated . the protective compressive stress induced by surface conditioning , however , may relax as a result of thermomechanical loadings experienced by the treated part after it is put into service , thus compromising the very purpose of the surface - conditioning treatment . to ensure safety and good performance , a nondestructive technique should be developed so that retention of the protective compressive stress in the treated parts can be monitored in - situ from time to time , thereby providing a basis for deciding whether a treated part should be taken out of service for replacement or re - conditioning treatment . depending on the specific surface - conditioning technique and processing parameters , the layer of compressive residual stress typically starts from the surface and runs to a depth of about 0.3 mm to 1 or 2 mm . the residual stress @xmath0 thus created varies with depth from the surface . at the free surface the principal stress of @xmath0 which has the free surface as principal plane is zero . if another principal stress of @xmath0 is plotted against depth from the surface , the graph typically assumes the shape of a check mark with a long tail ( see fig . 2 in section 5 ) : the principal stress starts negative ( i.e. , compressive ) at the surface and goes through a quick dip , then after a blunt turn at a minimum value ( i.e. , maximum compressive stress ) increases monotonically until it becomes slightly tensile and reaches a maximum , and then decays in a long tail to approximately zero while remaining tensile . for life - prediction purposes , monitoring of not only the surface residual stress but also the profile and depth of penetration of the protective stress layer ( particularly the maximum compressive principal stresses and their locations ) are required , because they all strongly affect the fatigue life and corrosion - crack resistance of the treated part . the presence of stress in a body affects the velocities of elastic waves propagating in it . this phenomenon is called the acoustoelastic effect . there is ample experimental evidence ( see , for example , @xcite ) that the presence of a surface layer of inhomogeneous residual stress in an otherwise homogeneous medium will lead to the dispersion of rayleigh waves , the quantitative data of which can be ascertained by boundary measurements . in this paper we shall explore whether we could monitor the retention of the surface - treatment induced layer of protective compressive stress by measurements of rayleigh - wave dispersion . besides inhomogeneous stress , there are other material characteristics ( e.g. , surface roughness , inhomogeneity in crystallographic texture ) of a treated part that will lead to dispersion of rayleigh waves , often with effects comparable to or stronger than those due to initial stress ( see , e.g. , @xcite ) . should some such characteristics have also changed after the treated part is put in service , other measurements in addition to rayleigh - wave dispersion would be needed to infer the depth profile of the stress . as a first step towards the development of an ultrasonic technique for monitoring stress retention in surface - treated samples , here we will restrict our discussion to the following simple situation : except for the unknown depth - dependent residual stress , all other relevant material parameters are known . one scenario where this could happen is that we have ascertained all relevant material characteristics of the treated sample , including the residual stress imparted by surface - conditioning , before the sample is put into service . after a period of service , the protective residual stress may have suffered from partial or total relaxation , but all other material characteristics of the treated sample remain unchanged after its production . under the theoretical framework of linear elasticity with initial stress @xcite , man et al . @xcite recently presented a general procedure for obtaining a high - frequency asymptotic formula for the dispersion of the phase velocity of rayleigh waves propagating in a vertically - inhomogeneous , prestressed and anisotropic half - space . as a further development , the general procedure given in @xcite was adapted by tanuma et al . @xcite to the case where the incremental elasticity tensor @xmath1 can be written as the sum of an isotropic part @xmath2 and a perturbative part @xmath3 . under a cartesian coordinate system where the material medium occupies the half - space @xmath4 , the perturbative part @xmath5 , the initial stress @xmath6 , and the mass density @xmath7 were assumed to be smooth functions of @xmath8 . moreover , the following linearization assumption ( * ) was made : at the free surface @xmath9 of the material medium the perturbative part @xmath10 and the initial stress @xmath11 are sufficiently small as compared with @xmath2 that for all expressions and formulas which depend on @xmath10 and @xmath11 it suffices to keep only those terms linear in the components of these tensors . under this setting , specific formulas are derived @xcite with which the procedure presented in @xcite can be implemented to compute iteratively each term of a high - frequency asymptotic formula for dispersion relations that pertain to rayleigh waves with various propagation directions . thus for rayleigh waves of sufficiently high frequencies , dispersion curves can be generated by the method developed in @xcite when requisite data on material and stress are given . once we have that capability , the inverse problem of inferring stress retention from rayleigh - wave dispersion can be attacked by an iterative approach . the theory developed in @xcite is meant for applications that include as typical example ultrasonic measurement of stress in metal structural parts , where the perturbative part @xmath3 in the splitting @xmath12 of the incremental elasticity tensor is originated from the presence of crystallographic texture and of the prestress @xmath0 . moreover , the shifts in phase velocities of elastic waves caused by texture and initial stress ( with the latter bounded by the yield surface ) are typically within 2% of their values for the corresponding isotropic medium with @xmath13 , which suggests that linearization assumption ( * ) would be adequate . on the other hand , the theory developed in @xcite does not take into consideration the effects of surface roughness on rayleigh - wave dispersion . several empirical studies ( see e.g. , @xcite ) have shown that if rayleigh - wave dispersion is used for measurement of stress induced by shot - peening or laser - shock peening , the effect of surface roughness on the dispersion curves can not be ignored , for it can totally mask the dispersion due to inhomogeneous stress . surface conditioning by low plasticity burnishing ( lpb ) , however , is different , for lpb leaves a mirror - smooth finish on processed parts . for the dispersion of rayleigh waves which have frequencies suitable for interrogation of the compressive stress induced by lpb treatment , surface roughness is not an issue @xcite . in this paper we will study as prime example the possibility of using the high - frequency formula for rayleigh - wave dispersion developed in @xcite to infer retention of near - surface compressive stress in a thick - plate sample of an aa 7075-t651 aluminum alloy which was surface - treated by low plasticity burnishing ( lpb ) . the plan of this paper is as follows . in section 2 we present , within the context of linear elasticity with initial stress , the constitutive equation of a prestressed medium which is a polycrystalline aggregate of cubic crystallites that carries an orthorhombic texture . details on material parameters and texture coefficients specific to the aluminum sample , which serves as the prime example of our present study , are given in appendix a. in section 3 , after we briefly outline the procedure given in @xcite to arrive at a high - frequency asymptotic formula for rayleigh - wave dispersion , we present a theorem and its corollary which will be instrumental for reducing the inverse problem in question to solving systems of linear equations iteratively . section 4 is devoted to a statement of the inverse problem on monitoring of stress retention and its solution . in section 5 we apply the theory to a specific inverse problem pertaining to the aluminum sample . there after we describe how `` experimental '' data on rayleigh - wave dispersion are simulated over the frequency window from 4 mhz to 70 mhz , we use the third - order approximation of the dispersion relations to infer the depth profile of the residual stress . in our example the inferred and real " stress profiles match well for the range of depth from 0 to 0.7 mm . in section 6 we examine the scenario in which experimental conditions ( e.g. , diffraction errors ) rule out the use of data at frequencies lower than 7 mhz . we use the second - order approximation of and the simulated data on the dispersion relations over the frequency window from 7 mhz to 70 mhz to infer the stress profile . our example shows that the inferred and real " stress profiles match quite well for the range of depth from 0 to 0.5 mm . we end the paper with some closing remarks in section 7 . the thick - plate sample in question ( see @xcite for details on sample preparation ) is that of an aa 7075-t651 aluminum alloy , one face of which was surface - treated by low plasticity burnishing ( lpb ) . the lpb - treatment , in general , would introduce a depth - dependent residual stress , which is compressive at and near the treated surface . it would also induce changes in material properties ( e.g. , elastic and acoustoelastic constants ) which , in the engineering literature , are qualitatively referred to as the effects of `` cold work '' . in the theory that we adopt in this paper , such effects are described quantitatively as being caused by changes in surface and near - surface crystallographic texture . we model the lpb - treated aluminum sample as a prestressed and textured polycrystalline medium , which occupies the half - space @xmath4 under a spatial cartesian coordinate system @xmath14 with @xmath15 being the treated surface whereas the 1- and 2-axis are chosen arbitrarily . we assume that the prestress and all material properties of the polycrystalline medium be macroscopically homogeneous with respect to planar translations for a fixed @xmath8 , but they may vary with @xmath8 . in what follows , dependence of material tensors , prestress , and texture on @xmath8 will be suppressed except on occasions when we want to emphasize that dependence . let the lattice of a fixed single crystal be chosen as reference . the lattice orientation at a sampling point in the polycrystalline medium is specified by a rotation @xmath16 which brings the reference lattice to the lattice at the sampling point . the crystallographic texture @xcite of a material point in the plane defined by @xmath8 is characterized by an orientation distribution function ( odf ) @xmath17 defined on the rotation group so(3 ) . ) and ( [ d ] ) . ] following the convention adopted by roe @xcite , we endow so(3 ) with volume measure @xmath18 , where @xmath19 is the haar measure with @xmath20 . the odf @xmath21 can be expressed as an infinite series in terms of the wigner @xmath22-functions @xcite : @xmath23 following roe , in this paper we work with @xmath24 instead of @xmath25 , and we call them the texture coefficients . if the crystallites in the polycrystal have no preferred orientations , all the texture coefficients vanish and the odf reduces to @xmath26 we assume that elastic deformations superimposed on the given polycrystalline medium can be adequately described by the theory of linear elasticity with initial stress @xcite . the general constitutive equation in that theory can be written @xcite as @xmath27 ; \label{constitutive}\ ] ] here @xmath28 is the first piola - kirchhoff stress , @xmath29 the initial stress , @xmath30 the displacement gradient pertaining to the superimposed small elastic motion , and @xmath31 the corresponding infinitesimal strain , where the superscript @xmath32 denotes transposition ; @xmath1 is the incremental elasticity tensor which , when regarded as a fourth - order tensor on symmetric tensors , has its components @xmath33 satisfy the major and minor symmetries . motivated by hartig s law on the affine dependence of the young s modulus with strain an empirical finding supported by wholly independent , individual experiments from 1811 to the present ... for one solid after another , including all of the metals " ( @xcite , p. 155 ) , for the prestressed and textured polycrystalline medium we regard @xcite @xmath1 as a function of the odf @xmath21 and the initial stress , and we write @xmath34={\mbox{${\mathbb l}$}}(w , { \mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } ) [ \mbox{\boldmath $ e$ } ] = { \mbox{${\mathbb c}$}}(w)[\mbox{\boldmath $ e$}]+{\mbox{${\mathbb d}$}}(w)[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , \mbox{\boldmath $ e$}],\ ] ] where @xmath35 is the fourth - order elasticity tensor defined on symmetric tensors @xmath36 , and @xmath37 is the sixth - order acoustoelastic tensor defined on ordered pairs of @xmath0 and @xmath36 , and we replace the functions @xmath38 and @xmath39 , respectively , by their affine approximation : @xmath40 , \label{c - affine } \\ { \mbox{${\mathbb d}$}}&={\mbox{${\mathbb d}$}}(w)= { \mbox{${\mathbb d}$}}(w_{\rm iso})+ { \mbox{${\mathbb d}$}}'(w_{\rm iso})[w - w_{\rm iso } ] , \label{d - affine}\end{aligned}\ ] ] where @xmath41 and @xmath42 denote the frchet derivative of @xmath35 and @xmath37 at @xmath43 , respectively . note that the fourth - order tensor @xmath44 $ ] and the sixth - order tensor @xmath45 $ ] depend linearly on the texture coefficients . when the initial configuration is stress - free and the constituting crystallites have no preferred orientations , i.e. , @xmath46 and @xmath47 , the incremental elasticity tensor @xmath48 reduces to the elasticity tensor of classical isotropic elasticity given by @xmath49 = \lambda({\rm tr } \mbox{\boldmath $ e$}){\bf i}+2\mu \mbox{\boldmath $ e$},\ ] ] where @xmath50 and @xmath51 are the lam constants . from ( [ l1])([ciso ] ) , we observe that the incremental elasticity tensor can be expressed as @xmath52 where @xmath53 and @xmath54 = { \mbox{${\mathbb d}$}}(w_{\rm iso})[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , \cdot ] + \left({\mbox{${\mathbb c}$}}'(w_{\rm iso})[w - w_{\rm iso}]\right)[\cdot ] + \left({\mbox{${\mathbb d}$}}'(w_{\rm iso})[w - w_{\rm iso}]\right)[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , \cdot].\ ] ] the isotropic sixth - order tensor @xmath55 is given by the representation formula @xcite @xmath56 = \beta_{1}(\text{tr}\,{\mbox{\boldmath $ e$}})(\text{tr}\ , { \mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } \,){\bf i}+ \beta_2(\text{tr}\,{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } \,){\mbox{\boldmath $ e$}}+\beta_3\left((\text{tr}\,{\mbox{\boldmath $ e$}}){\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } + ( \text{tr}\,{\mbox{\boldmath $ e$}}{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } \,){\bf i}\right ) + \beta_4({\mbox{\boldmath $ e$}}{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } + { \mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } { \mbox{\boldmath $ e$}}),\ ] ] where @xmath57 ( @xmath58 ) are material constants . aluminum single crystals have cubic crystal symmetry specified by point group @xmath59 . as far as the effects of crystallographic texture on the even - order tensors @xmath60\right)[\cdot]$ ] and @xmath61\right)[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , \cdot]$ ] are concerned , we may use classical texture analysis and treat the aluminum crystallites as if their point group were @xmath62 , the proper point group in the same laue class as @xmath59 @xcite . the surface and near - surface ( up to a depth of 0.225 mm ) crystallographic texture of the sample at the lpb - treated face were measured by x - ray diffraction and serial sectioning . the texture was found to be essentially constant with depth and was orthorhombic with one of the 2-fold axes of rotational symmetry parallel to @xmath63 . in the present paper we simply assume that the sample have a homogeneous texture . let @xmath64 be a cartesian coordinate system which has its coordinate axes parallel to the three 2-fold axes of the orthorhombic texture and the @xmath65 axis agree with @xmath63 . let @xmath66 be the texture coefficients of the sample under the coordinate system @xmath64 and the choice that the reference crystal lattice has its three 4-fold axes parallel to the coordinate axes @xmath67 , @xmath68 , and @xmath65 . the fourth - order tensor @xmath44 $ ] and sixth - order tensor @xmath45 $ ] are then given @xcite respectively by @xmath69\right)[{\mbox{\boldmath $ e$ } } ] & = \alpha { \boldsymbol{\phi}}(w'_{400 } , w'_{420 } , w'_{440})[{\mbox{\boldmath $ e$ } } ] , \label{c ' } \\ \left({\mbox{${\mathbb d}$}}'(w_{\rm iso})[w - w_{\rm iso}]\right)[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , { \mbox{\boldmath $ e$ } } ] & = \sum_{j=1}^{4 } \tilde{b}_j { \boldsymbol{\psi}}^{(i)}(w'_{400 } , w'_{420 } , w'_{440})[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , { \mbox{\boldmath $ e$ } } ] \nonumber \\ & \hspace{0.5in}+ a { \boldsymbol{\theta}}(w'_{600 } , w'_{620 } , w'_{640 } , w'_{660})[{\mbox{$ { \mbox{\boldmath $ t$ } } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } , { \mbox{\boldmath $ e$ } } ] ; \label{d'}\end{aligned}\ ] ] here @xmath70 , @xmath71 , and @xmath72 @xmath73 are material constants ; @xmath74 is a fourth - order tensor and @xmath75 are sixth - order tensors defined in terms of the texture coefficients @xmath76 , @xmath77 , @xmath78 , and @xmath79 a sixth - order tensor defined in terms of @xmath80 , @xmath81 , @xmath82 and @xmath83 . the components of these tensors in any @xmath14 coordinate system which has the @xmath63-axis agree with the @xmath65 axis are given explicitly in appendix a. the constitutive equation in question , as defined by ( [ constitutive ] ) and ( [ l2])([d ] ) , has 12 material parameters , namely @xmath50 , @xmath51 , @xmath70 , @xmath57 @xmath84 , @xmath72 @xmath85 , and @xmath71 . the values of these material parameters and of the relevant texture coefficients @xmath66 that pertain to the aluminum sample in question are given in appendix a. in what follows we adopt the following basic assumptions @xcite on the initial stress @xmath86 , the incremental elasticity tensor @xmath87 , the perturbative part @xmath3 of @xmath1 , and the mass density @xmath88 : * @xmath89 and @xmath90 are smooth functions of the coordinate @xmath91 . * the initial stress @xmath0 is residual , i.e. , it satisfies the equation of equilibrium @xmath92 for @xmath93 and the components @xmath94 vanish at the surface @xmath9 . * at the free surface @xmath95 , the perturbative part @xmath3 of @xmath1 and the initial stress @xmath0 are sufficiently small as compared with the isotropic part @xmath96 of @xmath1 ( i.e. , @xmath97 , @xmath98 , where @xmath99 denotes the euclidean norm ) that for all expressions and formulas which depend on @xmath10 and @xmath11 it suffices to keep only those terms linear in the components of these tensors . note that by ( [ l2 ] ) @xmath100 ; hence by assumption ( a ) the perturbative part @xmath3 of @xmath1 also depends smoothly on @xmath8 . the objective of this paper is to study the inverse problem that pertains to using boundary measurement of rayleigh - wave dispersion to infer the near - surface depth profile of residual stress induced by surface - conditioning treatments such as low - plasticity burnishing if all other material parameters of the sample are known . our solution of this inverse problem is based on the algorithm developed in @xcite which , under the aforementioned conditions on @xmath0 , @xmath1 , @xmath3 and @xmath101 , can iteratively generate each term in a high - frequency asymptotic formula of the dispersion relation of rayleigh waves once the propagation direction and all the relevant material parameters including the residual stress depth - profile are specified . in this section we briefly outline the steps ( cf . @xcite for details ) to solve the direct problem and present a theorem which will be instrumental to solving the inverse problem . consider rayleigh waves that propagate with phase velocity @xmath102 , wave number @xmath103 , and propagation direction @xmath104 along the traction - free surface of the sample , which is modeled as a vertically - inhomogeneous half - space . let @xmath105 be the @xmath106 surface impedance matrix that expresses a linear relationship between the displacements at the free surface and the surface tractions needed to sustain them . in @xcite an algorithm is given by which each term @xmath107 of the asymptotic expansion @xmath108 can be computed iteratively once @xmath109 is determined . specifically @xmath110 is obtained by solving lyapunov - type equations ( 20 ) and ( 21 ) in @xcite , and @xmath111 @xmath112 are obtained by solving equations ( 23 ) to ( 25 ) in @xcite . note that @xmath113 is the surface impedance matrix of the comparative homogeneous elastic half - space which has its incremental elasticity tensor , mass density , and initial stress equal to @xmath114 , @xmath115 , and @xmath11 , i.e. , their values at the surface @xmath9 , respectively . let @xmath116 . from ( [ z - asym ] ) the truncated sum of the asymptotic expansion for the rayleigh - wave velocity @xmath117 up to the order @xmath118 , namely @xmath119 is obtained by applying the implicit function theorem to the approximate secular equation @xmath120=0 . \label{nth - approximatesecular}\end{aligned}\ ] ] note that @xmath121 satisfies @xmath122 . under assumption ( c ) , we are concerned only with the terms in @xmath123 up to those linear in @xmath11 and @xmath10 , which leads us to write @xmath124 here we use the notation @xmath125 to indicate that we are retaining terms up to those linear in @xmath10 and @xmath11 and that we are neglecting the higher order terms . @xmath126 is of zeroth order in @xmath11 and @xmath10 , whereas @xmath127 is of first order in @xmath11 and @xmath10 . note that @xmath128 is the surface impedance matrix pertaining to a homogeneous isotropic elastic half - space with constitutive equation @xmath129 $ ] and with density @xmath130 . the following mathematical theorem , which describes how @xmath131 affects @xmath132 @xmath133 in ( [ nth - v - esp ] ) , will prove to be instrumental when we study the inverse problem . for @xmath134 , @xmath132 in the @xmath135th - order term of ( [ nth - v - esp ] ) depends on @xmath11 and on the @xmath8-derivatives of @xmath131 at @xmath95 up to those of order @xmath135 ; in particular , @xmath132 is of first order in the @xmath135th - order @xmath8-derivative of @xmath131 at @xmath95 . a proof of this theorem is given in appendix b. in the inverse problem we will adopt a numerical setting where each component of @xmath0 is represented as a polynomial in @xmath8 of degree @xmath136 . therefore we will use only the first three equations of ( [ stress - polynomial ] ) in the present paper . on the other hand , when the initial stress @xmath137 is not residual ( see for example @xcite ) , we will have to keep the last three equations of ( [ stress - polynomial ] ) . ] @xmath138 here the coefficients @xmath139 are to be determined in the implementation for the inverse problem . from theorem 3.1 we immediately obtain for @xmath134 , @xmath132 in the @xmath135th - order term of the expansion ( [ nth - v - esp ] ) has the following dependency on the parameters @xmath140 : \(1 ) @xmath141 is a first - order function of its arguments . \(2 ) for @xmath142 , the function @xmath143 is of first - order in @xmath144 . tanuma et al . @xcite provide an algorithm for the computation of the functions @xmath132 @xmath145 if all relevant material parameters , texture coefficients , and the initial stress @xmath0 are available as functions of @xmath8 . conversely , corollary 3.2 allows us to infer the parameters @xmath144 @xmath133 from experimental data on rayleigh - wave dispersion in six propagation directions as follows . first the parameters @xmath146 are determined from the six values of @xmath147 for the different propagation directions by solving the six equations on @xmath147 , which are linear in the unknowns @xmath146 . second these values of @xmath146 are substituted into the function @xmath148 , which is linear in the unknowns @xmath149 . these unknowns are evaluated by solving the six equations on @xmath150 for the six different propagation directions . this iterative process is then repeated for @xmath151 to get @xmath152 from the six equations on @xmath132 , which are linear in @xmath152 . a special case where the initial stress @xmath0 is residual and @xmath153 is discussed in detail in section 4.2 . the inverse problem in question can be described as follows . suppose we have ascertained all relevant material characteristics of the treated sample , including the profile of the residual stress imparted by surface - conditioning , before the sample is put into service ( say , the sample at state 0 ) . after a service period of the sample , suppose only the residual stress in the sample ( say , at state 1 ) may have changed . can we infer the depth profile of the current residual stress from measurement data on the dispersion of rayleigh waves that propagate in various directions along the free surface of the sample ? recall that we work with two cartesian coordinate systems ; see section 2 . we choose and fix a spatial cartesian coordinate system @xmath14 such that the material medium occupies the half - space @xmath4 and @xmath15 is the surface treated by low plasticity burnishing ( lpb ) . the fixed 1- and 2-axis are chosen arbitrarily . the @xmath64 system is the material coordinate system which has its coordinate axes parallel to the three 2-fold axes of the orthorhombic texture of the sample and the @xmath65 axis agrees with the @xmath63 axis . we shall consider rayleigh waves that propagate in the 2-direction . let @xmath155 be the angle of rotation about the @xmath156-axis that will bring the @xmath157-axis to the @xmath158-axis . different propagation directions in the sample are obtained by rotating the material half - space about the 3-axis , i.e. , by varying @xmath155 . henceforth we call @xmath155 the propagation direction of the rayleigh wave ( relative to the @xmath158-direction of the material half - space ) . since we assume that the depth - dependent initial stress @xmath131 be residual , i.e. , it satisfies the equation @xmath159 for @xmath93 and the boundary condition of zero traction at @xmath9 , it is of the form @xmath160 { \mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ { 12}(x_3 ) & { \mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ { 22}(x_3 ) & 0 \\ [ 1.4ex ] 0 & 0 & 0 \end{array } \right ) \label{prestress}\ ] ] under the @xmath14 coordinate system . let @xmath161 and @xmath162 be the principal directions of the stress that are perpendicular to the 3-axis , and @xmath163 and @xmath164 be the corresponding principal stresses . let @xmath165 be the angle between @xmath162 and the @xmath158-axis . then @xmath166 is the angle of rotation about the 3-axis that will bring the direction of the 2-axis to @xmath162 ; see fig . [ coordinate ] . it follows that @xmath167 in ( [ prestress ] ) can be written as @xmath168 where @xmath169 let @xmath170 be the density of the aluminum alloy in question when it is stress free . the presence of vertically - inhomogeneous residual stress @xmath131 will change the density of the material point from @xmath170 to @xmath90 , which is related to @xmath170 and @xmath131 by the formula @xmath171.\ ] ] in this paper we take @xmath172 kg / m@xmath173 , which is the ( nominal ) density of aa7075 alloy as computed from those of its alloying elements and their concentrations ( @xcite , pp . 214 ) . now we are ready to make an assertion ( * * ) that serves as an affirmative answer to the question raised at the beginning of this section on the inverse problem : ( * * ) suppose all the relevant material parameters and texture coefficients , except for the residual stress @xmath131 , are known functions of @xmath8 . let @xmath174 , @xmath175 , and @xmath176 be modeled as polynomials of degree @xmath177 in @xmath8 as in ( [ stress - polynomial ] ) . if the parameters @xmath178 in the dispersion relation @xmath179 can be evaluated unambiguously from experimental data on rayleigh - wave dispersion for three different propagation directions @xmath155 , then @xmath180 , @xmath181 , @xmath182 , and all the parameters @xmath183 for @xmath184 in ( [ stress - polynomial ] ) can be determined . for definiteness , for the rest of this section we will show how the polynomial model functions @xmath185 , @xmath186 , and @xmath187 can be determined for the case @xmath153 . the dispersion relations in question are then @xmath188 where @xmath189 , @xmath190 , and @xmath103 denotes the wave number . in ( [ asymptotic ] ) , @xmath194 is the zeroth - order term . as shown by corollary 8.2 of @xcite , it is given by the formula @xmath195 here @xmath115 is the density of the material at the free surface @xmath9 . formulas that express the parameters @xmath196 , @xmath197 , @xmath198 , @xmath199 and @xmath200 ( the phase velocity of rayleigh waves in the isotropic base material ) in terms of the material parameters and texture coefficients are given in @xcite . let @xmath201 from ( [ rotation ] ) and ( [ unkowns ] ) , we get @xmath202 { \mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ { 22}(0)&={\mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ m(0)+{\mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ d(0 ) \cos2(\theta+\zeta ) = a+b \cos 2\theta -c \sin 2\theta , \\[1 mm ] { \mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ { 12}(0)&=-{\mbox{$ { t } \hspace{-.25 cm } \mbox{\raisebox{1.6ex}{$\circ$}}$ } } _ d(0 ) \sin2(\theta+\zeta ) = -b \sin 2\theta -c \cos 2\theta . \nonumber\end{aligned}\ ] ] then for a given @xmath155 we can express @xmath115 and thence also @xmath200 , through ( [ stress - rho ] ) , in terms of @xmath203 and the known material parameters , texture coefficients , and @xmath170 . thus for a given @xmath155 the right - hand side of ( [ 0orderterm ] ) is a nonlinear function of @xmath204 , and @xmath205 , which are the unknowns . let @xmath206 and @xmath207 be the zeroth - order term for the phase velocities of rayleigh waves with propagation direction @xmath155 along the free surface of the sample in state 0 and 1 , respectively ; here the superscript @xmath208 for @xmath209 denotes the state 0 or 1 . let @xmath210 on one hand , @xmath211 in ( [ deltav ] ) can be measured by experiments for various @xmath155 at such high frequencies that @xmath212 and @xmath213 appear to become constant . on the other hand , by using the formula ( [ 0orderterm ] ) to compute @xmath211 we can see that it can be expressed in terms of @xmath214 which are the unknown parameters ( [ unkowns ] ) for state 1 . thus , we can use maple to apply the levenberg - marquardt method to estimate the parameters @xmath214 with 1 as their initial guess to fit these experimental data . in other words , we can recover @xmath215 at the surface of the material for state 1 . from ( [ rotation ] ) , we can also evaluate the values of @xmath180 , @xmath181 , @xmath182 for state 1 . to find the depth profile of the new prestress , we assume that the components of the new prestress can be fitted by some cubic polynomials ( cf . ( [ stress - polynomial ] ) ) @xmath220 here @xmath8 denotes the depth ; @xmath180 , @xmath181 , and @xmath182 have already been determined by the method discussed in the preceding subsection ; @xmath221 @xmath222 are the parameters to be determined . to start with , let us choose a specific propagation direction @xmath155 . by corollary 3.2 , by applying the algorithm for solving the direct problem , we can get a parametric dispersion relation of the form @xmath223 where @xmath132 is linear in @xmath183 for @xmath224 . indeed the procedure and formulas for the computations of the functions @xmath147 , @xmath150 , and @xmath225 are explicitly given in @xcite . note that while the values of @xmath226 , @xmath227 , and @xmath228 in ( [ asymptotic ] ) can be determined from experimental data on @xmath229 over suitable frequency windows , it is not enough to determine all nine parameters @xmath216 , @xmath217 , @xmath218 for @xmath230 from the system @xmath231 if we use just one @xmath155 . hence we consider three different @xmath155 s to get a complete system . to determine the nine parameters @xmath221 , first we solve for @xmath232 from the system for @xmath147 , which is linear with respect to @xmath232 . then substituting the values of @xmath232 into the system for @xmath150 to get a linear system with respect to @xmath233 , we can solve for @xmath233 quickly . after that , continue to substitute the values of @xmath234 into the system for @xmath225 . doing so leads to a linear system associated with @xmath235 , from which the parameters @xmath235 can be easily determined . to get better estimates of the parameters @xmath221 , we can consider several different groups of @xmath155 with each group containing 3 different @xmath155 s . after determining the parameters @xmath221 for each group , we take the average of the parameters from these groups as our estimated values of the parameters @xmath221 for @xmath236 . the residual stress @xmath238 induced by lpb treatment on the aa 7075-t651 sample ( state 0 ) was measured by x - ray diffraction ( and supplemented by information gathered from hole - drilling ) up to a depth of 1.25 mm from the treated surface . the depth profiles of the principal stresses are depicted in figure [ state0 ] , where the top and bottom curves pertain to the principal stresses @xmath163 and @xmath164 , respectively . since @xmath238@xmath239 is also of the form ( [ prestress ] ) under the @xmath64 material coordinate system , the principal directions @xmath240 and @xmath241 of the stress are defined with respect to the material coordinate system as soon as the angle @xmath165 between @xmath241 and the @xmath158-direction is specified . in the stress measurements it was found that @xmath242 for @xmath243 mm . as shown in fig . [ state0 ] , @xmath244 for @xmath245 mm . hence we may take @xmath242 for @xmath245 mm , as @xmath165 is , to within experimental error , arbitrary there . in our computations below , however , only the information on @xmath238@xmath246 will be used to calculate @xmath247 . ( top curve ) and @xmath248 ( bottom curve ) in sample at state 0 . ( reprinted from @xcite , with permission from elsevier.),scaledwidth=50.0% ] suppose the residual stress has changed to some unknown @xmath131 ( state 1 ) after the sample is put into service for a period of time , but crystallographic texture and other material parameters of the sample remain the same as before . as an exercise to see if we could determine the unknown @xmath131 by boundary measurement of rayleigh - wave dispersion , let us consider a specific instance where the original residual stress is relaxed so that the depth profile of the principal stresses @xmath249 and @xmath248 become those given in figure [ state1 ] . moreover the angle @xmath250 varies with depth as shown in table 1 . ( top curve ) and @xmath248 ( bottom curve ) in sample at state 1.,scaledwidth=50.0% ] .change of @xmath250 with respect to depth . [ cols="^,^,^,^,^,^,^,^,^,^,^",options="header " , ] we use the average values of @xmath221 as our simulated results . the components of the corresponding @xmath0 are shown below : @xmath251 where the stresses are in mpa and @xmath252 denotes the depth in units of mm . comparisons between the components of our simulated @xmath131 ( green curves ) and those of the `` real '' residual stress ( black dots ) in the sample at state 1 are shown in figure [ stresscomp1 ] , where the red curves are the fitting curves for the `` real '' stresses . a comparison of the corresponding principal stresses is illustrated in figure [ principal ] . in some applications , information on the stress from the surface to a depth of about @xmath253 mm would be sufficient . moreover , in section [ testsample ] the lower bound of the frequency window is @xmath254 mhz , which could be too low in practice because diffraction errors are much larger for frequencies lower than @xmath254 or @xmath255 mhz . in this section we will truncate the asymptotic expansion ( [ nth - v - esp ] ) for the rayleigh - wave velocity @xmath117 at the order @xmath256 and use a frequency window of lower boundary @xmath257 mhz . the same aluminum sample as in section [ testsample ] is considered . the depth profiles of residual stresses @xmath238@xmath239 and @xmath131 for state 0 and 1 , respectively , are the same as section [ testsample ] . however , here we focus only on the parts corresponding to the range of @xmath258 in @xmath259 $ ] mm . the simulated experimental " data on @xmath117 at various frequencies for various propagation directions @xmath155 are the same as in section [ testsample ] , but we will only use the data within the frequency window from @xmath257 mhz to @xmath260 mhz . unlike the third - order approximation in section [ testsample ] , here we obtain the dispersion curves by applying the least square method to fit the simulated data points with the quadratic form @xmath261 as illustration , the fitting curve and the simulated experimental " data for @xmath262 are shown in figure [ simudisq ] . nd - order fitting curve for @xmath262 in the frequency window from @xmath257 mhz to @xmath260 mhz by steps of @xmath253 mhz.,scaledwidth=50.0% ] the horizontal asymptotes of the fitting curves are treated as the simulated data for @xmath263 for @xmath155 from @xmath264 to @xmath265 in steps of @xmath266 . just as what we did in section [ testsample ] , the fitted values of the parameters @xmath267 in ( [ deltav ] ) are determined in the same way . here , in the second - order approximation , we have @xmath268 mpa , @xmath269 mpa , @xmath270 mpa . therefore from ( [ unkowns ] ) , we obtain @xmath271 mpa , @xmath272 mpa and @xmath273 at the surface @xmath95 of the sample at state 1 . it follows that @xmath274 mpa and @xmath275 mpa at the free surface . the surface residual stress @xmath11 has the components @xmath276 mpa , @xmath277 mpa and @xmath278 mpa . in this section , we assume that the components of the relaxed prestress can be fitted by some quadratic forms @xmath279 for the second - order approximation , we just need to follow the algorithm given in @xcite to determine @xmath280 in terms of @xmath281 @xmath282 . from corollary 3.2 , @xmath147 is of first - order in @xmath283 and @xmath150 is of first - order in @xmath284 . just as what we did in section [ testsample ] , we use waves of three different propagation directions , say @xmath285 , and compare the quadratic - fitting dispersion curves in the form of ( [ quadraticapp ] ) with the parametric dispersion curves @xmath286 to determine @xmath283 first , and then @xmath284 . for instance , the components of the corresponding @xmath131 as obtained from simulated data that pertain to the propagation directions @xmath285 are : @xmath287 where @xmath167 are in units of mpa and @xmath252 denotes the depth in units of mm . a comparison between our simulated prestresses ( green curves ) and the `` real '' prestresses ( black dots ) in state 1 are shown in figure [ stresscompnorelaxq ] , where the red curves are the fitting curves for the `` real '' prestresses . + a comparison of the corresponding principal stresses is illustrated in figure [ principalstressrelaxq ] , where the green curves are the simulated principal prestresses . from figure [ stresscompnorelaxq ] and [ principalstressrelaxq ] , we see that the quadratic approximation gives good estimates of the stress profiles up to a depth of about @xmath253 mm . in this paper we study the inverse problem on inferring depth profile of near - surface residual stress in a weakly anisotropic medium by using the algorithm given in @xcite for finding each term of a high - frequency asymptotic formula for rayleigh - wave dispersion . we show that , after the zeroth order terms are determined , theorem 3.1 and corollary 3.2 reduce the inverse problem to the routine work of solving iteratively systems of linear equations . we apply the theory to the practical problem on monitoring retention of residual stress induced by the surface - conditioning treatment of low plasticity burnishing on an aluminum thick plate sample if all other material parameters ( including texture coefficients ) remain unchanged . our study suggests that , if measurement of rayleigh - wave velocity has an accuracy of @xmath288% , then the depth profile of the residual stress from the surface to a depth of 0.5 to 0.7 mm with the upper limit depending on the frequency window where velocity measurements have @xmath288% accuracy can be recovered . in practice , besides accuracy of the velocity measurements , there is one more crucial question to be answered for our proposed method to be applicable . it is whether we can identify a frequency window @xmath289 $ ] which satisfies the following three conditions : ( i ) @xmath290 should be high enough for some computable truncated version of the high - frequency formula derived in @xcite to be valid ; ( ii ) @xmath291 should be sufficiently low that the effects of surface roughness on rayleigh - wave dispersion can be ignored for waves with frequencies within the window ; ( iii ) dispersion of rayleigh waves with frequencies within the window should be sufficiently pronounced that information on near - surface stress can be extracted from the dispersion data . all these depend on the specific material medium and sample in question . for example , condition ( iii ) depends most notably on the size of the acoustoelastic effect for rayleigh waves propagating in that medium , and condition ( ii ) depends on the surface finish of the sample . hence whether the proposed method would work for a specific application can only be decided after careful study on a case - by - case basis . chen conducted his work in this research project while he was a postdoctoral scholar at department of mathematics , university of kentucky , lexington , usa . the research efforts of tanuma was partially supported by jsps kakenhi grant number jp26400157 . 99 bell j f 1973 the experimental foundations of solid mechanics _ mechanics of solids vol 1 ( encyclopedia of physics vol via / i ) _ ed c truesdell ( berlin : springer ) pp 1813 bidenharn l c and louck j d 1984 _ angular momentum in quantum physics _ ( cambridge : cambridge university press ) biot m a 1965 _ mechanics of incremental deformations _ ( new york : wiley ) bunge h j 1982 _ texture analysis in materials science _ ( london : butterworth s ) du w and man c - s material tensors and pseudotensors of weakly - textured polycrystals with orientation distribution function defined on the orthogonal group _ j. elasticity _ submitted hoger a ( 1986 ) on the determination of residual stress in an elastic body _ j. elasticity _ * 16 * 303324 lavrentyev a i and veronesi w a 2001 ultrasonic measurement of residual stress in shot peened aluminum alloy in : thompson d o and chimenti d e ( eds ) _ review of progress in quantitative nondestructive evaluation _ * 20 * ( melville , new york : american institute of physics ) 14721479 man c - s 1998 on the constitutive equations of some weakly - textured materials _ arch . rational mech . * 143 * 77103 man c - s 1998 hartig s law and linear elasticity with initial stress _ inverse problems _ * 14 * 313319 man c - s 1999 effects of crystallographic texture on the acoustoelastic coefficients of polycrystals _ nondestr . test . eval _ * 15 * 191214 man c - s and carlson d e 1994 on the traction problem of dead loading in linear elasticity with initial stress _ arch . rational mech . _ * 128 * 223247 man c s and lu w y 1987 towards an acoustoelatic theory for measurement of residual stress _ j. elasticity _ * 17 * 159182 man c - s , koo l and shepard m j 2002 dispersion of rayleigh waves in titanium alloy resulting from inhomogeneous residual stress induced by low plasticity burnishing in : thompson d o and chimenti d e ( eds ) , _ review of progress in quantitative nondestructive evaluation _ * 21 * ( melville , new york : american institute of physics ) 16511658 man c - s , li , j , lu w - y and fan x 2000 ultrasonic measurement of through - thickness stress gradients in textured sheet metals . in : thompson d o and chimenti d e ( eds ) _ review of progress in quantitative nondestructive evaluation _ * 19 * ( melville , new york : american institute of physics ) 16131620 man c - s , nakamura g , tanuma k and wang s 2015 dispersion of rayleigh waves in vertically - inhomogeneous prestressed elastic media _ i m a j. appl . math . _ * 80 * 4784 man c - s and paroni r 1996 on the separation of stress - induced and texture - induced birefringene in acoustoelasticity _ j. elast . _ * 45 * 91116 moreau a and man c - s 2006 laser - ultrasonic measurements of residual stresses in a 7075-t651 aluminum sample surface - treated with low plasticity burnishing . in : thompson d o and chimenti d e ( eds ) _ review of progress in quantitative nondestructive evaluation _ * 25 * ( melville , new york : american institute of physics ) 14341441 paroni r and man c - s 2000 two micromechanical models in acoustoelasticity : a comparative study _ j. elast . _ * 59 * 145173 radovic m , lara - curzio e and riester l 2004 comparison of different experimental techniques for determination of elastic properties of solids _ mater . a _ * 368 * 5670 roe r - j 1965 description of crystallite orientation in polycrystalline materials . general solutions to pole figure inversion _ j. appl . phys . _ * 36 * 20242031 ruiz a and nagy p b 2003 saw dispersion measurements for ultrasonic characterization of surface - treated metals _ instrumentation , mesure , mtrologie _ * 3 * 5985 sarma v p n and reddy p j 1972 third - order elastic constants of aluminium _ phys . ( a ) _ * 10 * 563 - 567 tanuma k and man c - s 2002 angular dependence of rayleigh - wave velocity in prestressed polycrystalline media with monoclinic texture _ j. elast . _ * 69 * 181214 tanuma k , man c - s and chen y 2015 dispersion of rayleigh waves in weakly anisotropic media with vertically - inhomogeneous initial stress _ int . * 92 * 6382 tanuma k , man c - s and du w 2013 perturbation of phase velocity of rayleigh waves in pre - stressed anisotropic media with orthorhombic principal part _ math . mech . solids _ * 18 * 301322 the aluminum association 2000 _ aluminum standards and data 2000 _ ( washington , dc : the aluminum association ) thomas j f 1968 third order elastic constants of aluminum _ phys . rev . _ * 175 * 955962 varshalovich , d a , moskalev , a n , & khersonskii , v k 1988 _ quantum theory of angular momentum _ ( singapore : world scientific ) in this appendix we provide the details that complete the constitutive equation of the 7075-t651 aluminum sample studied in this paper . in our computations we take @xmath292 gpa and @xmath293 gpa , which correspond to the mean values of @xmath51 and young s modulus @xmath294 gpa obtained by radovic et al . @xcite in their rus ( resonant ultrasound spectroscopy ) measurements on sixteen 7075-t651 samples . as for the other 10 parameters , we are not aware of any experimentally determined value reported in the literature . hence we adopt the values predicted by the man - paroni model ( @xcite ) from second - order and third - order elastic constants of single - crystal pure aluminum reported by @xcite and @xcite , respectively : @xmath295 gpa , @xmath296 , @xmath297 , @xmath298 , @xmath299 , @xmath300 , @xmath301 , @xmath302 , @xmath303 and @xmath304 . the texture coefficients of the sample that pertain to the treated surface and several depths ( up to 0.225 mm ) were determined by x - ray diffraction and serial sectioning . they were found to be largely constant for the depths examined . in this paper we simply take the texture coefficients to be constant for the entire sample . the values which refer to the material @xmath64 coordinate system are : * @xmath305 , @xmath306 , @xmath307 , @xmath308 , @xmath309 , @xmath310 , and @xmath311 . all components of tensors below refer to the coordinate system @xmath14 defined in section 2 . the material coordinate system @xmath64 has its @xmath65-axis agree with the @xmath63-axis . let @xmath155 be the angle of rotation about the @xmath63 axis which brings the @xmath313 axis to the @xmath67 axis ( see fig . 1 in section 4 ) . an @xmath314-th order tensor @xmath315 is said to be harmonic if it is totally symmetric and traceless , i.e. , its components @xmath316 satisfy @xmath317 for each permutation @xmath318 of @xmath319 , and @xmath320 for any pair of distinct indices @xmath321 and @xmath322 . for example , for @xmath323 we have @xmath324 , etc . from total symmetry , and @xmath325 , etc . from the traceless condition the fourth - order tensor @xmath74 and the sixth - order tensor @xmath79 are harmonic . all the non - trivial components of @xmath74 can be obtained from the following five through the total symmetry of and the traceless condition on the harmonic tensor @xmath74 : @xmath326 the non - trivial components of @xmath79 can be obtained from the following seven by using the total symmetry of and the traceless condition on @xmath79 : 2 _ 222211 & = -w_600 - w_6202 + w_6404 + w_6606 , & + _ 222233 & = 6w_600 + w_6202 + 2w_6404 , & + _ 333311 & = -8w_600 + w_6202 , _ 333322 = -8w_600 - w_6202 , & + _ 122222 & = w_6202 + 2w_6404 + w_6606 , & + _ 122233 & = -w_6202- 2w_6404 , _ 123333 = w_6202 . & the components of the sixth - order tensors @xmath327 are given in terms of those of the harmonic tensor @xmath74 by the following formulae : @xmath328 where @xmath329 is the kronecker delta . to prove theorem 3.1 , we start by observing how @xmath330 are affected by the first and the higher order @xmath8-derivatives of @xmath131 at @xmath95 . for @xmath134 , @xmath111 depends on @xmath11 and on the @xmath8-derivatives of @xmath131 at @xmath95 up to those of order @xmath135 ; in particular , @xmath111 is of first order in the @xmath135th - order @xmath8-derivative of @xmath131 at @xmath95 . _ proof of lemma b.1 . _ we solve lyapunov - type equations ( 20 ) and ( 21 ) in @xcite to obtain @xmath110 and solve equations ( 23 ) to ( 25 ) in @xcite to obtain @xmath331 . by the chain rule of differentiation for the composite function @xmath332 , @xmath333 is of first order in the first order @xmath8-derivative of @xmath131 at @xmath95 , whereas @xmath334 depends on the @xmath8-derivatives of @xmath131 at @xmath95 up to those of order @xmath135 and is of first order in the @xmath135th - order @xmath8-derivative of @xmath131 at @xmath95 . therefore , ( 19 ) of @xcite implies that the right hand sides of ( 20 ) and ( 21 ) in @xcite are of first order in the first order @xmath8-derivative of @xmath131 at @xmath95 , whereas the right hand sides of ( 23 ) to ( 25 ) in @xcite depends on the @xmath8-derivatives of @xmath131 at @xmath95 up to those of order @xmath135 , and are of first order in the @xmath135th - order @xmath8-derivative of @xmath131 at @xmath95 . hence the arguments ( 26 ) through ( 27 ) in @xcite proves the lemma . @xmath154 _ proof of theorem 3.1 . _ the expression of @xmath132 in terms of @xmath335 was given by section 6 of @xcite for @xmath336 ( see also ( 37 ) and ( 38 ) of @xcite ) . these expressions were obtained from ( [ nth - approximatesecular ] ) and the implicit function theorem , through which we also have for a general @xmath135 @xmath337 where @xmath338 n_m & = \frac{\partial^m r } { \partial \varepsilon^m}\bigr|_{\scriptstyle{v = v_0,}\atop{\!\!\!\!\scriptstyle{\varepsilon=0 } } } + \sum_{k+l = m , \atop{1\le l\le m , \atop { 0\le k\le m-1 } } } \!\!\frac{m!}{k!\ l!}\ \frac{\partial^{k+l } r } { \partial \varepsilon^k\ , \partial v^l}\cdot \bigl(\frac{d v_r } { d \varepsilon}\bigr)^l \bigr|_{\scriptstyle{v = v_0,}\atop{\!\!\!\!\scriptstyle{\varepsilon=0 } } } + \cdots \quad ( m=2,3,\cdots , n ) \label{nm}\end{aligned}\ ] ] and @xmath339 on the right hand side of the preceding equation denotes a linear combination of the terms included in @xmath340 it then follows from ( [ nth - approximatesecular ] ) that @xmath341 which does not depend on any component of the @xmath8-derivatives of @xmath131 at @xmath95 , and it also follows that @xmath342\bigr|_{\scriptstyle{v = v_0,}\atop{\!\!\!\!\scriptstyle{\varepsilon=0 } } } \ \ ( m=1,2,\cdots , n ) . \label{e - mth - derivative - r}\ ] ] using the component - wise expression @xmath343 we observe from the definition of the determinant of a @xmath106 matrix that @xmath344 where @xmath345 is the set of all permutations of @xmath346 . hence , @xmath347 where the second term on the right hand side is neglected when @xmath348 . the first term is linear in @xmath111 , whereas the second term is nonlinear function of @xmath349 . moreover , any order of the partial derivative of @xmath350 with respect to @xmath102 does not affect the linearity in @xmath111 . hence only the first term on the right hand side of ( [ nm ] ) depends on ( and is of first order in ) @xmath111 and the remaining terms are nonlinear functions of @xmath349 . this , combined with the preceding lemma and ( [ v1vm ] ) , proves the theorem .

we propose a supersymmetric model of quasi - degenerate neutrinos which predicts the effective neutrino mass relevant for neutrinoless double beta decay to be proportional to the neutrino mass scale , thereby allowing its determination approximately independently of unknown majorana phases . such a natural quasi - degeneracy is achieved by using family symmetry ( as an example of a non - abelian family symmetry with real triplet representations ) to enforce a contribution to the neutrino mass matrix proportional to the identity . tri - bimaximal neutrino mixing as well as quark cp violation with and a leptonic cp phase arise from the breaking of the family symmetry by the vacuum expectation values of four `` flavon '' fields pointing in specific postulated directions in flavour space . mpp-2010 - 52 + * measurable neutrino mass scale in * + s. antusch , stephen f. king , m. spinrath , + _ max - planck - institut fr physik ( werner - heisenberg - institut ) , _ _ fhringer ring 6 , d-80805 mnchen , germany _ _ school of physics and astronomy , university of southampton , _ _ so17 1bj southampton , united kingdom _

1005.0708

over the past dozen years or so our knowledge of the neutrino sector has increased dramatically with the discovery of neutrino mass and mixing in atmospheric and solar neutrino oscillations , followed by the observation of terrestrial neutrino oscillations in long baseline neutrino experiments which have confirmed and refined the earlier results @xcite . yet , despite this progress , which may even be termed a `` neutrino revolution '' , there are many questions about neutrinos which remain unanswered . perhaps the most pressing of these is the origin , nature and magnitude of neutrino mass , since neutrino oscillations only provide information about the squared mass differences between neutrino species which are independent of the absolute neutrino mass scale or the nature of the neutrino mass ( i.e. dirac or majorana ) . in the absence of any confirmed experimental signal from either beta decay end - point experiments or neutrinoless double beta decay experiments , the most stringent limits on the absolute neutrino mass scale comes indirectly from cosmology where one typically obtains the limit on the absolute neutrino mass scale expressed in terms of the lightest neutrino mass as @xmath8 ev @xcite . thus , there remains the interesting possibility that neutrinos are quasi - degenerate , which one may roughly define as @xmath9 ev , where the lower limit is approximately set equal to the square root of the atmospheric neutrino mass squared difference . the current generation of running or planned neutrinoless double beta decay experiments are capable of discovering quasi - degenerate neutrinos , as defined above , within the next years . such a discovery would herald a new neutrino revolution to rival the last one , and would lead to an explosion of interest in theoretical models capable of accounting for quasi - degenerate neutrinos . in general having quasi - degenerate neutrinos does not lead to a sharp prediction for the neutrinoless double beta decay observable @xmath1 as a function of @xmath10 due to the presence of unknown phases in the neutrino mass matrix @xcite . the general conclusion that unknown phases enter the prediction for @xmath1 also remains valid in models which combine the experimental observation of ( at least approximate ) tri - bimaximal ( tb ) @xcite lepton mixing with the possibility of a quasi - degenerate neutrino mass spectrum . the reason is that the relevant phases entering @xmath1 are majorana phases . allowing for arbitrary majorana phases and considering a quasi - degenerate neutrino mass spectrum and tb mixing , @xmath1 can still be in the approximate interval @xmath11 $ ] . what would we learn about the origin of neutrino mass from the discovery of quasi - degenerate neutrinos in neutrinoless double beta decay ? clearly this would imply that neutrinos are majorana , and possibly ( but not necessarily ) that would indicate that a seesaw mechanism is at work , but what kind of seesaw mechanism , i.e. is it type i or ii ? there are known examples of type i and type ii seesaw models which can lead to quasi - degenerate neutrinos as well as tb lepton mixing , so clearly quasi - degenerate neutrinos would not distinguish different types of seesaw mechanism . for example , the supersymmetric ( susy ) grand unified theory ( gut ) based on @xmath12 with family symmetry @xmath13 proposed in @xcite is based on the type ii seesaw mechanism and leads to tb mixing and allows quasi - degenerate neutrinos . on the other hand the susy @xmath2 model in @xcite based on the type i seesaw mechanism also leads to tb mixing and allows quasi - degenerate neutrinos . interestingly , the susy @xmath14 model with type i seesaw mechanism does not favour quasi - degenerate neutrinos @xcite , whereas a related model with a type ii seesaw mechanism does allow quasi - degenerate neutrinos @xcite . more generally , there is a huge literature on family symmetry models based on @xmath2 @xcite or other symmetries @xcite . however , to our knowledge , in all above examples , quasi - degenerate neutrinos are subject to the mentioned phase uncertainties in the prediction of @xmath1 as a function of @xmath15 . it is interesting to ask , in what class of theories would we learn the most about the neutrino mass scale @xmath16 from the discovery of a measurement of @xmath1 in neutrinoless double beta decay ? clearly the answer would be those theories which predict @xmath1 uniquely as a function of @xmath16 without ambiguities from unknown phases , but the next question is do such theories exist ? perhaps surprisingly the answer is in the affirmative , and , even more surprisingly , the class of theories which have this property turn out to suggest the way that the neutrino mass matrix is generated , namely by a usual type i seesaw contribution with two or three right - handed neutrinos , plus an additional contribution proportional to the unit matrix . in @xcite , two of us proposed a class of theories of exactly this kind which we referred to as a `` type ii upgrade of type i seesaw models '' . in this class of models the additional contribution to the neutrino mass matrix was realised by an additional type ii seesaw . the type i seesaw part of the neutrino mass matrix , which controls the mass squared differences and mixing angles , was governed by sequential right - handed neutrino dominance @xcite . the effect of such an additional unit matrix structure implies that for quasi - degenerate neutrino masses the majorana cp phases are small and thus @xmath17 . although the class of models was specified , no realistic `` type ii upgrade '' model has ever been proposed . in this paper we shall propose a model following the idea of an additional contribution the neutrino mass matrix proportional to the unit matrix based on @xmath2 family symmetry with @xmath18 grand unification . the model contains tri - bimaximal neutrino mixing after the @xmath2 family symmetry is broken as an indirect result of the assumed aligned `` neutrino flavons '' in the type i seesaw sector via constrained sequential dominance @xcite . these `` neutrino flavons '' break the @xmath2 symmetry , being assumed to be aligned along the columns of the tb mixing matrix , but quadratic combinations of the neutrino flavons respect accidentally the neutrino flavour symmetry as discussed in @xcite . further `` quark flavons '' are assumed to be misaligned compared to the `` neutrino flavons '' and are , together with the `` neutrino flavons '' , responsible for quark and charged lepton masses and quark mixings . as expected , due to a possibly large type ii seesaw contribution , or alternatively due to an additional type i seesaw contribution from an additional triplet representation of right - handed neutrinos , the model can naturally predict the neutrinoless double beta decay mass observable to be approximately equal to the neutrino mass scale . we also make a detailed fit to quark masses and mixing using the misaligned quark flavons and show that a simple ansatz for the phase of one of the misaligned quark flavons leads to successful quark cp violation . in order for radiative corrections not to modify too much the tb mixing for quasi - degenerate neutrinos @xcite , we shall restrict ourselves to low values of the ratio of higgs vacuum expectation values ( vevs ) @xmath19 . for such low @xmath20 , a viable gut scale ratio of @xmath21 is achieved within susy @xmath18 guts using a clebsch factor of @xmath22 , as proposed recently by two of us in @xcite . for the third generation we use @xmath23-@xmath24 yukawa coupling unification @xmath25 at the gut scale which is viable for low @xmath26 ( see , e.g. , @xcite ) . the layout of the remainder of the paper is as follows . in section 2 we present the model . in section 3 we perform a numerical fit to the quark and charged lepton masses and quark mixing angles and cp violating phase and discuss the neutrino masses and lepton mixing angles . section 4 summarises and concludes the paper . in appendix a we give a renormalisable superpotential and explicit expressions for the effective couplings and appendix b contains a possible vacuum alignment . in this section , we propose and describe a susy gut model based on the unified @xmath18 gauge group as well as on the family symmetry @xmath2 amended by some discrete @xmath27 symmetries and an @xmath28 symmetry as specified in tab . [ tab : symmetries ] . cccccccc & @xmath18 & @xmath2 & @xmath30 & @xmath31 & @xmath32 & @xmath33 & @xmath28 + + @xmath34 & @xmath35 & @xmath36 & + & 0 & + & 0 & 1 + @xmath37 , @xmath38 , @xmath39 & @xmath40 , @xmath40 , @xmath40 & @xmath41 , @xmath41 , @xmath41 & + , + , - & 0 , 1 , 0 & + , + , + & 1 , 0 , 0 & 1 , 1 , 1 + @xmath42 , @xmath43 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 0 , 1 & + , + & 1 , 0 & 1 , 1 + + @xmath44 , @xmath45 , @xmath46 & @xmath41 & @xmath36 & + , + , - & 0 , 3 , 0 & + , + , + & 3 , 0 , 0 & 0 , 0 , 0 + @xmath47 & @xmath48 & @xmath36 & + & 3 & - & 0 & 0 + @xmath49 , @xmath50 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 0 , 0 & + , + & 0 , 0 & 0 , 0 + @xmath52 , @xmath53 & @xmath54 , @xmath55 & @xmath41 , @xmath41 & + , + & 0 , 0 & + , + & 0 , 0 & 0 , 0 + @xmath56 , @xmath57 & @xmath58 , @xmath59 & @xmath41 , @xmath41 & + , + & 0 , 0 & - , - & 0 , 0 & 0 , 0 + + @xmath60 , @xmath61 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 1 , 3 & - , - & 0 , 0 & 1 , 1 + @xmath62 , @xmath63 & @xmath40 , @xmath64 & @xmath36 , @xmath36 & + , + & 0 , 0 & + , + & 0 , 0 & 1 , 1 + @xmath65 & @xmath41 & @xmath36 & + & 0 & + & 0 & 1 + + @xmath66 , @xmath67 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 2 , 2 & + , + & 0 , 0 & 0 , 2 + @xmath68 , @xmath69 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 0 , 0 & + , + & 2 , 2 & 2 , 0 + @xmath70 , @xmath71 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 2 , 2 & + , + & 0 , 0 & 2 , 0 + let us start introducing the model by specifying the field content and the symmetries . the standard model matter fields fit nicely into the two representations @xmath72 , which we call @xmath34 , and @xmath40 , which we call @xmath73 . explicitly they are given as @xmath74 where the lower indices @xmath75 , @xmath66 and @xmath76 denote the quark colours and @xmath77 is the family index . in our model , we consider that the three generations @xmath78 form a triplet representation @xmath79 of an @xmath2 family symmetry whereas the three generations @xmath80 form singlets @xmath41 under @xmath2 . , would in principle be suitable for the construction of models with additional contributions to the neutrino mass matrix proportional to the unit matrix . in this paper we focus on @xmath2 as a specific example . ] in the following , we suppress the @xmath2 indices . in addition , we consider two right - handed neutrinos , singlets under su(5 ) as well as under @xmath2 , labeled by @xmath42 and @xmath43 . we furthermore consider fifteen - dimensional higgs representations @xmath52 , @xmath53 which contain @xmath81-triplet higgs fields that obtain induced vevs after electroweak symmetry breaking . @xmath52 induces in this way a type ii seesaw contribution to the neutrino mass matrix which is , to leading order , proportional to the unit matrix and can increase the neutrino mass scale without modifying the values for the leptonic mixing angles . @xmath18 is spontaneously broken by the vev of the @xmath47 field , electroweak symmetry is broken by the vevs of the higgs fields @xmath49 , @xmath82 , @xmath56 , @xmath83 and @xmath2 is spontaneously broken by the vevs of the flavon fields , i.e. the family symmetry breaking higgs fields @xmath45 , @xmath44 , @xmath46 and @xmath47 . we comment below on the specific directions in which we assume @xmath2 to be broken by the flavons . on top of that , we consider additional `` messenger '' fields which are heavy and which , after effectively integrating them out of the theory , give rise to higher - dimensional operators generating the yukawa coupling matrices as well as the mass matrix of the gauge singlet ( right - handed ) neutrinos @xmath84 . the field content of our model as well as the symmetries are specified in tab . [ tab : symmetries ] . we note that it is always possible to replace any product of commuting discrete symmetries by a single abelian group @xmath85 with a suitable choice of charges for the fields , so it is possible to replace the @xmath27 symmetry by a single @xmath85 symmetry , with an appropriate choice of charges . indeed many models in the literature use an abelian @xmath85 symmetry rather than a product of @xmath86 symmetries to control the operators . although this looks simpler , it should be remarked that firstly an abelian symmetry has infinitely many more group elements than any discrete symmetry , and secondly one must then confront the question of goldstone bosons once the assumed global abelian symmetry is broken . if the abelian symmetry is gauged one must further complicate the model by ensuring that it is anomaly free . therefore an auxiliary discrete symmetry , even a large one , has definite advantages over an abelian symmetry . furthermore discrete symmetries are ubiquitous in string theory constructions . finally , the auxiliary discrete symmetry used here is rather a simple one consisting of a product of @xmath30 and @xmath33 parity factors . thus we regard the use of the discrete @xmath30 and @xmath33 symmetries as being a well motivated , simple and attractive alternative to the use of an abelian @xmath85 symmetry . we would like to remark that we do not explicitly consider the full flavour and gut higgs sector of the model and just assume that the @xmath18 and @xmath2 breaking vevs are aligned in the desired directions of field space . we assume that in these sectors issues like doublet - triplet splitting are resolved . without specifying these sectors , a reliable calculation of the proton decay rate must also be beyond the scope of the present paper . the focus of the present paper is thus to illustrate that quasi - degenerate light neutrino masses can be realised together with a type ii seesaw in a @xmath18 gut framework . we would furthermore like to remark that in addition to the type ii seesaw contribution there is a possible additional contribution to the neutrino mass matrix proportional to the unit matrix from the messenger field @xmath87 which is a singlet under @xmath18 and a triplet under @xmath2 . when it is integrated out , it also induces a contribution to the neutrino mass operator which is proportional to the unit matrix . the renormalisable superpotential resulting from tab . [ tab : symmetries ] is given in the appendix . integrating out the heavy messenger superfields denoted by @xmath89 , @xmath66 and @xmath90 , the feynman diagrams in figs . [ fig : messenger_dl ] , [ fig : messenger_u ] and [ fig : messenger_n ] then lead to the effective non - renormalisable superpotential terms in the @xmath18 and @xmath2 unbroken phase : @xmath91 after gut symmetry breaking the @xmath81 doublet components from @xmath49 and @xmath56 respectively @xmath50 and @xmath57 mix and only the light states acquire the @xmath81 breaking vevs which give the fermion masses , as discussed in appendix a where the effective couplings @xmath92 appearing in the effective superpotential are also explicitly given . ] ] ] in the following we assume that the vevs of the @xmath2 breaking flavon fields point in the following directions in field space such that @xmath93 the @xmath94 are defined in appendix a. these relations also define the quantities @xmath95 , @xmath96 and @xmath97 . the breaking of @xmath2 along the field directions of @xmath45 and @xmath44 allows us to realise tri - bimaximal neutrino mixing via constrained sequential dominance ( csd ) @xcite . it is also worth noting that the flavon vevs @xmath98 and @xmath99 are orthogonal , causing some of the terms in the superpotential to give a vanishing contribution to the mass matrices . in the following , we assume that cp is only broken spontaneously by the vev of the flavon @xmath47 . for the flavon @xmath47 one may suppose _ a priori _ a less constrained alignment , @xmath100 however _ empirically _ we find that the numerical fit to quark masses and mixings , in particular quark cp violation , seems strongly to prefer that the vacuum alignment of the flavon @xmath47 has its second component along the imaginary direction . to simplify the results of the numerical fit we shall restrict ourselves to the case : @xmath101 in some future more ambitious theory one may attempt to reproduce eq . as a result of some special vacuum alignment , but here we shall simply regard it as a special choice , or ansatz , which leads to a successful fit to quark cp violation . in appendix b , we will discuss another possibility , namely to realise the flavon @xmath47 effectively by splitting it up into two flavons @xmath102 and @xmath103 , where one gets a purely real and the other a purely imaginary vev . for the effective superpotential in eqs . ( [ eq : yl ] ) - ( [ eq : mr ] ) this would correspond to simply replacing @xmath104 and @xmath105 , with an additional higgs field @xmath106 . the field content of this extended version of the model that now includes also a vacuum alignment sector , is presented in appendix b in tab . [ tab : symmetries2 ] . the predictions of the two model variants are identical at the level of precision discussed here . we define our conventions for the yukawa matrices such that the operators of the form @xmath107 and @xmath108 give the following yukawa terms in the lagrangian : @xmath109 where the @xmath18 relation @xmath110 would be fulfilled , if all clebsch gordan factors were one . the convention we use here is the same as the one used by the particle data group @xcite . from eqs . , , , and the yukawa matrix coupling the up - type quarks to the light up - type higgs doublet with the @xmath23 coefficients as defined in appendix a is given as @xmath111 whereas the yukawa matrices coupling the down - type quarks and charged leptons to the light down - type higgs doublet are given as @xmath112 where @xmath113 , @xmath114 , @xmath115 and @xmath116 are the clebsch gordan factors arising from gut symmetry breaking , see , e.g. , @xcite . we have used the orthogonality of @xmath99 and @xmath98 and considered the above described notation for the flavon vevs . we note that in the definition for the yukawa matrices we have introduced a complex conjugation why here appears a phase factor of @xmath117 in the 2 - 2 elements of the down - type quark and charged lepton yukawa matrices . with the given representations of the flavons , we obtain the following clebsch gordan coefficients @xmath118 for small values of @xmath26 as we consider , the 1-loop susy threshold corrections are small and , taking the actual experimental values of the strange quark and muon masses into account , the gut scale value of @xmath119 prefers @xmath120 , as argued in @xcite ( see also @xcite ) . from the charged lepton yukawa matrix we can derive the following approximate relations for the eigenvalues @xmath121 furthermore , since there is no 1 - 2 mixing from the up - sector , the mixing angle @xmath122 is approximately given as @xmath123 from those four equations the four @xmath124 s can be calculated and the relation for the ckm phase gives at the gut scale @xmath125 the rg evolution of the measured value for @xmath126 gives a gut scale value of @xmath127 . so the value for the ckm phase ( based on our assumed vacuum alignment ) is already remarkably good if we only take the lepton masses and the value for @xmath122 into account which are measured to high accuracy . and @xmath128 and with the small @xmath129 , the model might also provide a solution to the strong cp problem , along the lines discussed in @xcite . ] .the model parameters for @xmath130 and @xmath131 gev from a fit to the experimental data . [ tab : parameters ] [ cols="^,^",options="header " , ] for the detailed fit of the model to the data we applied the following procedure : we have taken the gut scale yukawa matrices from eqs . , and and calculated their rg evolution down to the scale @xmath132 for @xmath130 are somewhat constrained due to bounds on the higgs mass . however , we emphasize that our model may well be formulated in the context of the nmssm or other non - minimal susy models where @xmath26 of order one can readily be realised without these constraints . ] and @xmath131 gev with the reap software package @xcite . at the low scale we performed a @xmath133 fit to the quark masses and mixing and charged lepton masses depending on the parameters of the gut scale yukawa matrices . the fit gave a total @xmath133 of about 3.5 where we have assumed a relative error of 1 % for the charged lepton masses and for the other observables we have taken the experimental errors . since we have 11 parameters and 13 observables this corresponds to a @xmath134 of about 1.6 . this is a good fit since we have neglected theoretical uncertainties like , e.g. , threshold corrections which could be treated as additional errors on the data lowering the total @xmath133 . . the straight blue lines give the 1 % ( 1@xmath135 ) bound while the dashed lines give the 2 % ( 2@xmath135 ) bound . the red crosses denote our fit results . [ fig : fitresultsplot ] ] the results for the gut scale parameters are listed in tab . [ tab : parameters ] . we would like to remark that these parameters depend on @xmath26 and @xmath136 and also are subject to several theoretical uncertainties . for example , we note that the higgs fields @xmath52 and @xmath53 containing the higgs triplets of the type ii seesaw mechanism have masses at intermediate energy scale between @xmath137 and @xmath138 . their effects are not included in the rg analysis . the effects are small and may be neglected if @xmath139 is small , but they could be sizable if @xmath139 is large . gives a contribution proportional to the unit matrix , it affects only the rg evolution of the mass eigenvalues , but not of the mixing angles . nevertheless , the possibility of additional rg effects from @xmath139 provides a theoretical uncertainty in our setup . ] due to the additional theoretical uncertainties we do not explicitly give the errors on the high energy parameters or low energy fit results . the important input parameters for us are the charged lepton masses and quark mixing angles which have a experimental error much smaller than these uncertainties . in tab . [ tab : fitresults ] the low energy results are shown and compared to experimental data . a graphical illustration is given in fig . [ fig : fitresultsplot ] . they illustrate that our minimal example model , with the assumed vacuum alignment of eq . and , can fit well the data . we turn now to the results for the neutrino sector . the neutrino yukawa matrix is obtained from eq . as @xmath140 additionally we have a diagonal mass matrix for the two right - handed neutrinos from eq . , @xmath141 and contribution proportional to the unit matrix coming from eqs . and , @xmath142 using the seesaw relation @xmath143 we obtain for the neutrino mass matrix @xmath144 with @xmath145 in our model we therefore identify the neutrino masses as @xmath146 , @xmath147 , @xmath148 , where without loss of generality we can take @xmath149 to be positive and real while @xmath150 are real but can take either sign . with @xmath151 , @xmath152 a quasi - degenerate mass spectrum of the light neutrinos can be explained in a natural way . in the following , we will mainly restrict ourselves to this case . from the structure of these matrices we see that we obtain tri - bimaximal mixing in the neutrino sector @xmath153 from the lepton sector we get the additional mixing contributions @xmath154 there is also a complex phase introduced by the charged lepton yukawa matrix which can be calculated in the same way as in the quark sector @xmath155 for the approximate calculation of the mns mixing parameters at the gut scale we can use @xcite : @xmath156 where we have already discarded all trivial phases and rg corrections which we will discuss later . for the total leptonic mixing angles we obtain @xmath157 for the phases we have @xmath158 , @xmath159 and @xmath160 from which the final mns phases can be calculated according to @xcite @xmath161 where @xmath162 and @xmath163 are the majorana phases as in the pdg parameterization where they are contained in a diagonal matrix @xmath164 . we note that with the mixing pattern of our model , i.e. tri - bimaximal mixing produced in the neutrino sector , and charged lepton mixing corrections only from @xmath165 , the leptonic mixing angles and the dirac cp phase @xmath166 satisfy the lepton mixing sum rule @xcite @xmath167 the approximately maximal cp violation , i.e. @xmath4 , affects that although the charged lepton corrections generate @xmath168 , the solar mixing angle remains very close to its tri - bimaximal value of @xmath169 . so far , we have discussed the neutrino mixing parameters at the gut scale . to calculate the low energy values we have to take rg running of the parameters into account . for a quasi - degenerate neutrino mass spectrum , rg corrections to the neutrino parameters can in principle change the high scale values dramatically . however , as has been discussed for type ii upgraded seesaw models in @xcite and more generally in @xcite , for small @xmath26 and small neutrino yukawa couplings ( in our example model they are much smaller than @xmath170 ) the corrections to the mixing angles and cp phases are under control . setting the small majorana phases to zero and with @xmath4 , we can estimate in leading order @xcite @xmath171 where @xmath172 is the renormalisation scale . in the case of quasi - degenerate neutrino masses we can further use the approximation @xmath173 . integrating these equations approximately with the parameters on the right side taken constant and equal to their gut scale values , and plugging in these numbers , we obtain the estimated low energy values of the mixing angles as @xmath174 in the last equation , the `` + '' applies for a normal neutrino mass ordering , whereas the `` - '' applies for an inverse mass odering , i.e. the case @xmath175 . it is important to note that both mass orderings can be realised in our model . the strong suppression for the rg running of @xmath176 is caused by the particular values of the cp violating phases in our model . for similar reasons the running of the cp phases themselves is also suppressed , as can be seen using the analytical results in @xcite . in summary , rg corrections are under control in our setup and only cause comparatively small corrections to the mixing parameters in the lepton sector . in summary , the leptonic mixing parameters in our model are compatible with the experimental 1@xmath135 ranges at low energy which are : @xmath177 , @xmath178 and @xmath179 , taken from @xcite , as long as @xmath149 is smaller than the cosmological bounds suggest , @xmath180 ev @xcite . the values for the leptonic mixing angles and dirac cp phase @xmath166 , resulting from our assumed vacuum alignment and stated in eqs . and can be tested accurately by ongoing and future precision neutrino oscillation experiments @xcite . the effective mass relevant for neutrinoless double beta decay is @xmath181 while the kinematic mass accessible in the single beta decay end - point experiment katrin is @xmath182 for quasi - degenerate neutrino mass spectrum ( @xmath183 ) we obtain that @xmath184 directly gives information about the absolute neutrino mass scale . on the other hand , due to the phases appearing in eq . there is typically a sizable ambiguity in the relation between @xmath1 and @xmath149 , as long as the majorana cp phases are not predicted . allowing , for instance , for arbitrary majorana phases and considering a quasi - degenerate neutrino mass spectrum ( @xmath183 ) and with small @xmath176 , @xmath1 can still be in the approximate interval @xmath11 $ ] . in our setup relevant for neutrinoless double beta decay as a function of the mass @xmath185 of the lightest neutrino , for an inverted neutrino mass ordering ( @xmath186 , upper line ) and for a normal mass ordering ( @xmath187 , lower line ) . the bands represent the experimental uncertainties of the mass squared differences . the mass bounds from cosmology @xcite and from the heidelberg - moscow experiment @xcite are displayed as grey shaded regions . the red lines show the expected sensitivities of the gerda experiment in phase i and ii @xcite . [ fig : meeplot ] ] this ambiguity is resolved in our model , since the large contribution to the neutrino mass matrix proportional to the unit matrix ( with @xmath151 , @xmath152 ) results in small majorana cp phases and thus we predict : @xmath188 the assumed dominance of @xmath149 in our model allows to realise a quasi - degenerate neutrino spectrum ( with normal or inverse mass ordering ) in a natural way , without any tuning of parameters . the possible values for @xmath1 as function of the lightest neutrino mass @xmath16 is shown in fig . [ fig : meeplot ] . we would like to remark that for smaller @xmath149 one can also naturally extend the model to hierarchical or inverted hierarchical neutrino masses without changing the leptonic mixing angles . with @xmath189 or @xmath190 or both we then also encounter cases where the majorana phases are close to @xmath191 . for a quasi - degenerate spectrum the model disfavours these unnatural cases since they would correspond to heavily fine - tuned parameters of the model . similarly , an inverse strongly hierarchical spectrum would require unnatural tuning between @xmath149 and @xmath192 to make @xmath193 very small . by contrast , for a typical parameter choice of the model , a normally ordered hierarchical spectrum simply corresponds to @xmath194 , @xmath195 and does not require any tuning at all . it is also interesting to note that with the phases in our model there is _ no _ possibility to have cancellations in eq . that could make @xmath1 vanish exactly . ev . ] neutrinoless double beta decay is thus , also for smaller @xmath149 , an unavoidable consequence in this class of models . we have proposed a model of quasi - degenerate neutrinos which predicts the neutrinoless double beta decay mass observable to be approximately equal to the neutrino mass scale @xmath16 , thereby allowing its determination approximately independently of unknown majorana phases . in general such quasi - degenerate neutrino masses may be naturally realised if there exists a non - abelian family symmetry with real triplet representations that enforces an additional contribution to the neutrino mass matrix proportional to the unit matrix , hence determining the neutrino mass scale . in our model , the additional contribution was generated by a type ii seesaw , or , alternatively , by another type i seesaw contribution from an additional triplet representation of right - handed neutrinos ( or neutrino messenger fields ) . in addition , the standard type i seesaw contribution determines the neutrino mixing angles . although such a mechanism ( called a `` type ii upgrade '' in @xcite ) has been known for some time , the model in this paper is the first of its kind to combine this mechanism with tri - bimaximal mixing arising from @xmath2 family symmetry together with a @xmath18 susy gut . the susy @xmath0 model considered here has several attractive features . the full renormalisable superpotential for the coupling of matter to flavons , messenger fields and higgs fields is specified , and only discrete auxiliary symmetries are introduced , rather than the more common continuous abelian symmetry that is typically invoked in such models . in the considered model , the neutrino mass scale can either originate from the induced vacuum expectation value of a @xmath196 triplet contained in a fifteen - dimensional representation of @xmath18 , or from an additional contribution to the neutrino mass operator induced by @xmath18 singlet messenger fields in the triplet representation of @xmath2 . since the type i seesaw contribution involves very hierarchical additional neutrino mass contributions , responsible for the tiny mass splittings between the quasi - degenerate neutrinos , we use the constrained sequential dominance mechanism which is well suited for achieving such strongly hierarchical contributions . in such a model the neutrino flavour symmetry associated with tri - bimaximal mixing is achieved indirectly from the family symmetry , given the assumed vacuum alignment . we have included renormalisation group corrections to the mixing angles and shown that they are under control for small values of @xmath26 in our framework and do not significantly modify the leptonic mixing angles . we have therefore considered small values of @xmath26 throughout our study . the model has several interesting phenomenological features which emerge from our numerical fit to the quark and lepton masses and quark mixing . in addition to @xmath23-@xmath24 unification that may be viable for small @xmath26 , the model realises the gut scale relation @xmath197 proposed in @xcite , which is favourable for small values of @xmath26 . in the quark sector , we observe that the correct @xmath126 ( corresponding to the right unitarity triangle with @xmath3 ) can be realised for the simple ansatz in eq . . we remark that this simple ansatz leads to an interesting alternative texture ( different from @xcite ) that gives rise to a right - angled unitarity triangle with @xmath198 . for the leptonic mixing angles and cp phases we find @xmath168 , @xmath199 , @xmath200 and @xmath4 . the leptonic mixing angles satisfy the `` lepton mixing sum rule '' proposed in @xcite with only small theoretical errors since the 1 - 3 mixing in the charged lepton mass matrix is very small and the 2 - 3 mixing vanishes . the majorana cp phases are small for quasi - degenerate neutrino masses via a large additional contribution proportional to the unit matrix , as expected , and the model thus predicts the neutrinoless double beta decay mass observable to be approximately equal to the neutrino mass scale , or lightest neutrino mass , i.e. @xmath201 . in conclusion , if neutrinoless double beta decay were observed in the near future then this would herald another neutrino revolution in which neutrino masses would be quasi - degenerate . amongst the many possible models of quasi - degenerate neutrinos , the model with an additional contribution to neutrino mass matrix proportional to the unit matrix , as considered here , are distinguished by their prediction of that the neutrinoless double beta decay mass observable is approximately equal to the neutrino mass scale @xmath16 . we have proposed the first realistic model of this kind involving @xmath2 and susy @xmath18 guts . the @xmath2 family symmetry has the dual effect of enforcing on the one hand quasi - degeneracy via the additional unit matrix contribution to the neutrino mass matrix , and on the other hand tri - bimaximal mixing via the type i seesaw mechanism and constrained sequential dominance . a numerical fit to quark masses and mixing angles reveals that a simple ansatz describing quark cp violation with @xmath3 also leads to the leptonic phase @xmath4 . in such models the absolute neutrino masses could be directly measurable by experiment quite soon . we would like to thank michal malinsk for valuable discussions , especially during the initial stages of the project , and christoph luhn for valuable discussions on possibilities for the vacuum alignment . we also thank vinzenz maurer for very valuable comments on the manuscript . we are indebted to nordita for the hospitality and support during the programme `` astroparticle physics a pathfinder to new physics '' held in stockholm in march 30 - april 30 , 2009 during which part of this study was performed . s. a. and m. s. acknowledge partial support by the dfg cluster of excellence `` origin and structure of the universe . '' s. f. k. acknowledges partial support from the stfc rolling grant no . st / g000557/1 and a royal society leverhulme trust senior research fellowship . with the field content and symmetries specified in tab . [ tab : symmetries ] the superpotential contains the following renormalisable terms : @xmath202 as discussed in the main text , after gut symmetry breaking the @xmath81 doublet components from @xmath49 and @xmath56 respectively @xmath50 and @xmath57 mix and only the light states acquire the @xmath81 breaking vevs which give the fermion masses . we parameterise the higgs mixing with the mixing angles @xmath203 respectively @xmath204 @xmath205 where we have used the common abbreviation @xmath206 and similar for the sine and the other angle @xmath204 . the light higgs doublets are denoted with an index @xmath207 while the heavy higgs doublets are denoted with an index @xmath208 . the effective couplings @xmath92 appearing in the effective superpotential in the main text can be expressed in term of the fundamental couplings from eq . and the messenger masses from @xmath209 @xmath210 @xmath211 @xmath212 the effective couplings @xmath23 appearing in the yukawa couplings can be expressed in terms of the couplings @xmath92 and the higgs mixing angles as @xmath213 @xmath214 @xmath215 @xmath216 ccccccccc & @xmath18 & @xmath2 & @xmath30 & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath28 + + @xmath34 & @xmath35 & @xmath36 & + & 0 & 0 & 0 & 0 & 1 + @xmath37 , @xmath38 , @xmath39 & @xmath40 , @xmath40 , @xmath40 & @xmath41 , @xmath41 , @xmath41 & + , + , - & 0 , 1 , 0 & 0 , 0 , 0 & 0 , 0 , 0 & 1 , 0 , 0 & 1 , 1 , 1 + @xmath42 , @xmath43 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 0 , 1 & 0 , 0 & 0 , 0 & 1 , 0 & 1 , 1 + + @xmath44 , @xmath45 , @xmath46 & @xmath41 , @xmath41 , @xmath41 & @xmath36 , @xmath36 , @xmath36 & + , + , - & 0 , 3 , 0 & 0 , 0 , 0 & 0 , 0 , 0 & 3 , 0 , 0 & 0 , 0 , 0 + @xmath221 , @xmath102 , @xmath103 & @xmath41 , @xmath48 , @xmath41 & @xmath36 , @xmath36 , @xmath36 & - , + , + & 3 , 3 , 3 & 3 , 3 , 0 & 3 , 0 , 3 & 0 , 0 , 0 & 0 , 0 , 0 + @xmath222 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + & 1 , 3 & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 + + @xmath49 , @xmath50 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 + @xmath223 , @xmath224 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 0 , 0 & 0 , 0 & 3 , 1 & 0 , 0 & 0 , 0 + @xmath52 , @xmath53 & @xmath54 , @xmath55 & @xmath41 , @xmath41 & + , + & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 + @xmath56 , @xmath57 & @xmath58 , @xmath59 & @xmath41 , @xmath41 & + , + & 0 , 0 & 3 , 1 & 0 , 0 & 0 , 0 & 0 , 0 + + @xmath60 , @xmath61 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 1 , 3 & 1 , 3 & 0 , 0 & 0 , 0 & 1 , 1 + @xmath225 , @xmath226 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 1 , 3 & 0 , 0 & 1 , 3 & 0 , 0 & 1 , 1 + @xmath62 , @xmath63 & @xmath40 , @xmath64 & @xmath36 , @xmath36 & + , + & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 & 1 , 1 + @xmath65 & @xmath41 & @xmath36 & + & 0 & 0 & 0 & 0 & 1 + + @xmath66 , @xmath67 & @xmath51 , @xmath35 & @xmath41 , @xmath41 & + , + & 2 , 2 & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 2 + @xmath227 , @xmath228 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 0 , 0 & 0 , 0 & 0 , 0 & 2 , 2 & 2 , 0 + @xmath229 , @xmath230 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 2 , 2 & 0 , 0 & 0 , 0 & 0 , 0 & 2 , 0 + @xmath68 , @xmath69 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 2 , 2 & 2 , 2 & 2 , 2 & 0 , 0 & 2 , 0 + @xmath70 , @xmath71 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 2 , 2 & 2 , 2 & 0 , 0 & 0 , 0 & 2 , 0 + @xmath231 , @xmath232 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 2 , 2 & 0 , 0 & 2 , 2 & 0 , 0 & 2 , 0 + + @xmath233 , @xmath234 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & + , + & 0 , 0 & 0 , 0 & 0 , 0 & 0 , 0 & 2 , 2 + @xmath235 , @xmath236 & @xmath41 , @xmath41 & @xmath36 , @xmath36 & + , + & 2 , 0 & 0 , 0 & 0 , 0 & 0 , 0 & 2 , 2 + @xmath237 , @xmath238 , @xmath239 & @xmath41 , @xmath41 , @xmath41 & @xmath36 , @xmath36 , @xmath36 & + , + , + & 2 , 2 , 2 & 2 , 2 , 0 & 2 , 0 , 2 & 0 , 0 , 0 & 2 , 2 , 2 + @xmath240 , @xmath241 , @xmath242 & @xmath48 , @xmath41 , @xmath48 & @xmath41 , @xmath41 , @xmath41 & - , - , + & 2 , 2 , 2 & 2 , 1 , 1 & 1 , 2 , 1 & 0 , 0 , 0 & 2 , 2 , 2 + @xmath243 , @xmath244 & @xmath41 , @xmath48 & @xmath41 , @xmath41 & - , - & 1 , 1 & 1 , 1 & 1 , 0 & 0 , 0 & 2 , 2 + @xmath245 , @xmath246 & @xmath41 , @xmath41 & @xmath41 , @xmath41 & - , + & 1 , 1 & 1 , 0 & 1 , 0 & 0 , 1 & 2 , 2 + in this section of the appendix , we will discuss a possibility to extend the model in the main part by a viable vacuum alignment . this possibility is based on realising the flavon @xmath47 effectively by splitting it up into two flavons @xmath102 and @xmath103 , which get purely real or purely imaginary vevs by a vacuum alignment as described below , i.e. : @xmath247 for the effective superpotential in eqs . ( [ eq : yl ] ) - ( [ eq : mr ] ) this would mean to simply replace @xmath104 and @xmath105 , with an additional higgs field @xmath106 . the field content and the symmetries of this extended version of the model is given in tab . [ tab : symmetries2 ] . we note that , at the level of precision discussed here , the model predictions are the same as in the main part of the paper . we now turn to the discussion of the vacuum alignment sector with the desired purely real or imaginary alignment ( see also @xcite ) . at the effective theory level the new operators involving , the `` driving fields '' @xmath234 , @xmath248 and @xmath249 , that are generated by the fields and symmetries specified in tab . [ tab : symmetries2 ] are given by ( dropping @xmath250 coupling constants ) : @xmath251 the minimisation of the resulting f - term potential leads to the desired vacuum alignment for all flavons . we note that the new flavons @xmath252 and @xmath253 are only `` auxiliary '' in the sense that they do not couple to the matter sector but are only relevant for the flavon alignment . the flavon @xmath222 leads to an additional term contributing to the charm mass , however its effect can be absorbed in the messenger masses such that the predictions of the model are unchanged . the four parts @xmath254 , @xmath255 , @xmath256 and @xmath257 have the following meaning : * the terms in @xmath254 enforce non - vanishing vevs of the three flavons @xmath221 , @xmath102 and @xmath103 that are orthogonal to each other and have only one nonzero element , due to the effects of the terms with the driving fields @xmath258 . here the @xmath259 denotes the symmetric triplet combination of @xmath2 ( see e.g. @xcite ) . with real @xmath260 due to the assumed spontaneous cp violation @xmath261 are forced to be real which means that the vevs @xmath262 are either purely real or imaginary . we will choose a vacuum where @xmath102 is imaginary and @xmath103 is real . the vev of @xmath253 can be either real or imaginary and since its phase does not affect the results . * the part @xmath255 leads to the desired alignment for the flavon vev @xmath263 along the @xmath264 direction ( see also @xcite ) in the minima of the potential with @xmath265 and @xmath266 . again , due to the @xmath267 in the term with @xmath268 the vev can be either real or imaginary , and we choose the real vacuum in the @xmath269 direction . * the terms in @xmath256 and @xmath257 finally provide the alignment for the flavons @xmath270 and @xmath44 . the terms with the driving fields @xmath249 force the vev of @xmath270 to be orthogonal to @xmath271 and @xmath272 , and the vev of @xmath44 to be orthogonal to @xmath271 and @xmath263 , leading to the alignments in the desired direction in flavour space . we can again choose the purely real vacuum as explained above . we note that the driving fields @xmath234 and @xmath233 have the same quantum numbers and therefore , a priori , a linear combination of such fields couples to each of the above terms @xmath273 and @xmath274 . without loss of generality we have written the driving fields here already in the field basis where only one driving field couples to one of these terms . this can always be achieved by a proper field redefinition . s. f. king , arxiv:0712.1750 [ physics.pop-ph ] ; s. f. king , rept . phys . * 67 * ( 2004 ) 107 [ arxiv : hep - ph/0310204 ] ; r. n. mohapatra _ et al . _ , rept . * 70 * ( 2007 ) 1757 [ arxiv : hep - ph/0510213 ] ; r. n. mohapatra and a. y. smirnov , ann . rev . nucl . * 56 * ( 2006 ) 569 [ arxiv : hep - ph/0603118 ] ; c. h. albright , arxiv:0905.0146 [ hep - ph ] ; g. altarelli and f. feruglio , arxiv:1002.0211 [ hep - ph ] . s. f. king and c. luhn , arxiv:0912.1344 [ hep - ph ] . g. altarelli and f. feruglio , nucl . b * 741 * ( 2006 ) 215 [ arxiv : hep - ph/0512103 ] . g. altarelli , f. feruglio and c. hagedorn , jhep * 0803 * ( 2008 ) 052 [ arxiv:0802.0090 [ hep - ph ] ] ; t. j. burrows and s. f. king , arxiv:0909.1433 [ hep - ph ] . p. ciafaloni , m. picariello , e. torrente - lujan and a. urbano , phys . d * 79 * ( 2009 ) 116010 [ arxiv:0901.2236 [ hep - ph ] ] ; p. ciafaloni , m. picariello , a. urbano and e. torrente - lujan , phys . rev . d * 81 * ( 2010 ) 016004 [ arxiv:0909.2553 [ hep - ph ] ] . e. ma and g. rajasekaran , phys . d * 64 * ( 2001 ) 113012 [ arxiv : hep - ph/0106291 ] ; e. ma , arxiv : hep - ph/0701016 ; e. ma , mod . phys . lett . a * 22 * ( 2007 ) 101 [ arxiv : hep - ph/0610342 ] ; e. ma , mod . a * 21 * ( 2006 ) 2931 [ arxiv : hep - ph/0607190 ] ; e. ma , h. sawanaka and m. tanimoto , phys . b * 641 * ( 2006 ) 301 [ arxiv : hep - ph/0606103 ] ; e. ma , phys . d * 73 * ( 2006 ) 057304 ; b. adhikary , b. brahmachari , a. ghosal , e. ma and m. k. parida , phys . b * 638 * ( 2006 ) 345 [ arxiv : hep - ph/0603059 ] ; e. ma , mod . phys . a * 20 * ( 2005 ) 2601 [ arxiv : hep - ph/0508099 ] ; e. ma , phys . rev . d * 72 * ( 2005 ) 037301 [ arxiv : hep - ph/0505209 ] ; s. l. chen , m. frigerio and e. ma , nucl . phys . b * 724 * ( 2005 ) 423 [ arxiv : hep - ph/0504181 ] ; w. rodejohann and z. z. xing , phys . b * 601 * ( 2004 ) 176 [ arxiv : hep - ph/0408195 ] . e. ma , phys . d * 70 * ( 2004 ) 031901 [ arxiv : hep - ph/0404199 ] ; f. feruglio , c. hagedorn , y. lin and l. merlo , nucl . b * 775 * ( 2007 ) 120 [ arxiv : hep - ph/0702194 ] ; x. g. he , nucl . * 168 * ( 2007 ) 350 [ arxiv : hep - ph/0612080 ] ; s. f. king and m. malinsky , phys . b * 645 * ( 2007 ) 351 [ arxiv : hep - ph/0610250 ] ; i. k. cooper , s. f. king and c. luhn , arxiv:1004.3243 [ hep - ph ] . r. barbieri , l. j. hall , g. l. kane and g. g. ross , arxiv : hep - ph/9901228 ; i. de medeiros varzielas and g. g. ross , nucl . b * 733 * ( 2006 ) 31 [ arxiv : hep - ph/0507176 ] ; s. f. king and m. malinsky , jhep * 0611 * ( 2006 ) 071 [ arxiv : hep - ph/0608021 ] ; i. de medeiros varzielas , s. f. king and g. g. ross , phys . lett . b * 644 * ( 2007 ) 153 [ arxiv : hep - ph/0512313 ] ; i. de medeiros varzielas , s. f. king and g. g. ross , phys . b * 648 * ( 2007 ) 201 [ arxiv : hep - ph/0607045 ] ; e. ma , mod . phys . lett . a * 21 * ( 2006 ) 1917 [ arxiv : hep - ph/0607056 ] ; e. ma , arxiv:0709.0507 [ hep - ph ] ; c. luhn , s. nasri and p. ramond , phys . lett . b * 652 * ( 2007 ) 27 [ arxiv:0706.2341 [ hep - ph ] ] ; f. plentinger and g. seidl , phys . rev . d * 78 * ( 2008 ) 045004 [ arxiv:0803.2889 [ hep - ph ] ] ; c. csaki , c. delaunay , c. grojean and y. grossman , arxiv:0806.0356 [ hep - ph ] ; m .- c . chen and k. t. mahanthappa , phys . b * 652 * ( 2007 ) 34 [ arxiv:0705.0714 [ hep - ph ] ] ; m .- c . chen and k. t. mahanthappa , arxiv:0710.2118 [ hep - ph ] ; m .- c . chen and k. t. mahanthappa , arxiv:0812.4981 [ hep - ph ] ; r. n. mohapatra , s. nasri and h. b. yu , phys . b * 639 * ( 2006 ) 318 [ arxiv : hep - ph/0605020 ] ; r. n. mohapatra and h. b. yu , phys . b * 644 * ( 2007 ) 346 [ arxiv : hep - ph/0610023 ] ; a. aranda , arxiv:0707.3661 [ hep - ph ] ; a. h. chan , h. fritzsch and z. z. xing , arxiv:0704.3153 [ hep - ph ] ; z. z. xing , phys . b * 618 * ( 2005 ) 141 [ arxiv : hep - ph/0503200 ] ; z. z. xing , h. zhang and s. zhou , phys . lett . b * 641 * ( 2006 ) 189 [ arxiv : hep - ph/0607091 ] ; s. k. kang , z. z. xing and s. zhou , phys . d * 73 * ( 2006 ) 013001 [ arxiv : hep - ph/0511157 ] ; s. luo and z. z. xing , phys . b * 632 * ( 2006 ) 341 [ arxiv : hep - ph/0509065 ] ; m. hirsch , e. ma , j. c. romao , j. w. f. valle and a. villanova del moral , phys . d * 75 * ( 2007 ) 053006 [ arxiv : hep - ph/0606082 ] ; n. n. singh , m. rajkhowa and a. borah , arxiv : hep - ph/0603189 ; x. g. he and a. zee , phys . b * 645 * ( 2007 ) 427 [ arxiv : hep - ph/0607163 ] ; n. haba , a. watanabe and k. yoshioka , phys . * 97 * ( 2006 ) 041601 [ arxiv : hep - ph/0603116 ] ; z. z. xing , phys . lett . b * 533 * ( 2002 ) 85 [ arxiv : hep - ph/0204049 ] ; y. lin , nucl . phys . b * 813 * ( 2009 ) 91 [ arxiv:0804.2867 [ hep - ph ] ] ; l. yin , arxiv:0903.0831 [ hep - ph ] ; s. f. king and c. luhn , arxiv:0905.1686 [ hep - ph ] ; c. hagedorn , s. f. king and c. luhn , arxiv:1003.4249 [ hep - ph ] . s. f. king , phys . b * 439 * ( 1998 ) 350 [ arxiv : hep - ph/9806440 ] ; s. f. king , nucl . b * 562 * ( 1999 ) 57 [ arxiv : hep - ph/9904210 ] ; s. f. king , nucl . b * 576 * ( 2000 ) 85 [ arxiv : hep - ph/9912492 ] ; s. f. king , jhep * 0209 * ( 2002 ) 011 [ arxiv : hep - ph/0204360 ] ; s. antusch and s. f. king , new j. phys . * 6 * ( 2004 ) 110 [ arxiv : hep - ph/0405272 ] ; s. antusch , s. boudjemaa and s. f. king , arxiv:1003.5498 [ hep - ph ] . s. antusch and m. spinrath , phys . d * 79 * ( 2009 ) 095004 [ arxiv:0902.4644 [ hep - ph ] ] . g. ross and m. serna , phys . b * 664 * ( 2008 ) 97 [ arxiv:0704.1248 [ hep - ph ] ] . r. howl and s. f. king , arxiv:0908.2067 [ hep - ph ] . b. bajc , p. fileviez perez and g. senjanovic , arxiv : hep - ph/0210374 . s. antusch and m. spinrath , phys . d * 78 * ( 2008 ) 075020 [ arxiv:0804.0717 [ hep - ph ] ] . a. e. nelson , phys . b * 136 * ( 1984 ) 387 ; s. m. barr , phys . d * 30 * ( 1984 ) 1805 . s. antusch , s. f. king , m. malinsky and m. spinrath , arxiv:0910.5127 [ hep - ph ] . z. z. xing , h. zhang and s. zhou , phys . d * 77 * ( 2008 ) 113016 [ arxiv:0712.1419 [ hep - ph ] ] . s. f. king , jhep * 0508 * ( 2005 ) 105 [ arxiv : hep - ph/0506297 ] ; i. masina , phys . b * 633 * ( 2006 ) 134 [ arxiv : hep - ph/0508031 ] ; s. antusch and s. f. king , phys . b * 631 * ( 2005 ) 42 [ arxiv : hep - ph/0508044 ] ; s. antusch , p. huber , s. f. king and t. schwetz , jhep * 0704 * ( 2007 ) 060 [ arxiv : hep - ph/0702286 ] . s. antusch , j. kersten , m. lindner and m. ratz , nucl . b * 674 * ( 2003 ) 401 [ arxiv : hep - ph/0305273 ] . s. antusch , j. kersten , m. lindner , m. ratz and m. a. schmidt , jhep * 0503 * ( 2005 ) 024 [ arxiv : hep - ph/0501272 ] . s. antusch , s. f. king and m. malinsky , nucl . b * 820 * ( 2009 ) 32 [ arxiv:0810.3863 [ hep - ph ] ] ; s. antusch , s. f. king and m. malinsky , jhep * 0805 * ( 2008 ) 066 [ arxiv:0712.3759 [ hep - ph ] ] ; s. antusch , s. f. king and m. malinsky , phys . lett . b * 671 * ( 2009 ) 263 [ arxiv:0711.4727 [ hep - ph ] ] ; s. boudjemaa and s. f. king , arxiv:0808.2782 [ hep - ph ] ; s. f. king , phys . lett . b * 659 * ( 2008 ) 244 [ arxiv:0710.0530 [ hep - ph ] ] . h. v. klapdor - kleingrothaus _ et al . _ , phys . j. a * 12 * ( 2001 ) 147 [ arxiv : hep - ph/0103062 ] . a. a. smolnikov [ gerda collaboration ] , arxiv:0812.4194 [ nucl - ex ] . a. osipowicz _ et al . _ [ katrin collaboration ] , arxiv : hep - ex/0109033 . m. c. gonzalez - garcia , m. maltoni and j. salvado , arxiv:1001.4524 [ hep - ph ] . c. amsler _ et al . _ [ particle data group ] , phys . b * 667 * ( 2008 ) 1 .

different observation data for cool star procyon ( obs_ids of 63 , 1461 and 1224 ) available from _ chandra data public archive _ were co - added and analyzed . emissivities of emission lines of highly charged silicon ions ( si vii si xii ) were calculated over temperatures by adopting the published data of liang et al . ( 2007 , _ atom . data and nucl . data tables _ , * 93 * , 375 ) . using the emission measure derived by raassen et al . ( 2002 , a&a , * 389 * , 228 ) , the theoretical line fluxes are predicted , and the theoretical spectra are constructed by assuming the gaussian profile with instrumental broadening ( 0.06 ) . by detailed comparison between observation and predictions , several emissions lines are identified firstly such as emissions at 43.663 ( si xi ) , 45.550 ( si xii ) , 46.179 ( si viii ) , 50.874 ( si x ) , 64.668 ( si ix ) , and 73.189 ( si vii ) etc . several emission lines are re - assigned in this work , such as the emission line at 52.594 to si x ( 52.612 ) , at 69.641 to the blending of si vii ( 69.632 ) and si viii ( 69.664 ) lines , as well as at 70.050 to si vii ( 70.027 ) . the prediction reveals the large discrepancies between the line ( 63.715 ) _ versus _ line ( 61.012 ) again for lower charge stage si viii . solar flare observation is also added for the assessment of present calculation . different assignments for some lines between procyon and solar flare , have been found to be due to the hotter emitter in sun than in procyon coronae . [ firstpage ] line : identification method : analytical stars : coronae x - rays : general

0711.2215

since the launch of _ chandra _ and _ xmm_-newton x - ray missions , a large quantity of high quality x - ray spectra with high - resolution has been obtained for nearly all classes of astrophysical x - ray sources @xcite , which allows detailed plasma diagnostics to be performed for a wide range of celestial objects . in the spectra of cool stars , rich emission lines of highly charged iron ions were detected due to its high abundance and high effective collecting area ( 200 @xmath3 at 8.5 for heg / meg+acis - s instrument ; 35 @xmath3 at 10.0 for letg+hrc - s instrument ) between 618 . in this wavelength region , a good agreement between predictions and observations is achieved so far @xcite , which profits from the iron - project @xcite and earlier running program opacity project @xcite . moreover , many laboratory experiments have been performed for iron by adopting grating and crystal spectrometers with high - resolution . the laboratory platform can be early vacuum spark , tokamak used for magnetic confinement fusion study , intense laser used for inertial confinement fusion research , and newly developed facility electron beam ion trap ( ebit , beiersdorfer 2003 ) . though the difference of @xmath0@xmath1 line intensity _ versus _ @xmath2@xmath1 line intensity of fe xvii is a debating problem , a good theoretical performance is obtained when compared with spectra of highly charged sulfur and argon , as revealed by experimental measurements at livermore ebit @xcite . besides rich emission lines of highly charged fe ions , some emission lines of l - shell calcium , argon , sulfur and silicon have been identified for cool stars . yet detailed analysis for spectra of these ions receive a scant attention due to the low effective area over the wavelength range of 35100 spanned by lines of l - shell si , ar and ca ions , and the absence of accurate atomic data of these ions . several literatures @xcite also reveal the great potentials of l - shell silicon ions . such as , the line intensity ratio @xmath4(50.524)/@xmath4(50.691 ) of si x , @xmath4(52.306)/@xmath4(46.391 ) of si xi are good diagnostic methods for the electron density ; the line ratio @xmath4(52.306)/@xmath4(43.743 ) of si xi shows a good performance for the electron temperature ( @xmath5 ) determination . similar characteristics were explored for s x , ar xiv and ca xvi etc . . furthermore , spatial information of coronae could be assessed indirectly using correlation ( @xmath6 ) of the electron density with emission measure . a laboratory measurement for silicon has been performed at the livermore ebit - ii . unfortunately , only the spectra between 8090 is shown @xcite , yet it is enough to demonstrate the large differences between the measurement and theory . recently , accurate atomic data including energy levels and spontaneous radiative decay rates were calculated for si ix si xi by consideration of large configuration interaction and relativistic effects @xcite . a self - consistent electron impact excitation rates over a large temperature grids are available , which are based on relativistic distort - wave ( rdw ) method . a laboratory measurement of soft x - ray spectra ( 40180 ) of highly charge silicon ions were also performed by irradiating the silicon target by intense laser beam at institute of physics , chinese academy of sciences , which satisfactorily benchmarks the theoretical modelling at high density @xcite . in this work , we calculate the line emissivities of highly charged silicon ions ( si vi si xii ) by adopting new available and accurate atomic data . the available observation data for cool star procyon ( obs_ids 63 , 1461 and 1224 ) are co - added and analyzed . by detailed comparison between the predictions and the co - added spectrum . emissions lines of highly charged silicon are investigated . the observation of solar flare is also summarized for the assessment of the calculation . procyon ( f5 iv v ) is a solar - like star at a distance of 3.5 pc with mass of 1.75@xmath7 and radius of 2.1@xmath8 , which has been observed by every x - ray space missions such as _ chandra _ and _ xmm_-newton , for calibration . from _ public data archive , four observations for procyon can be available , with one observed by hrc - i instrument and none grating . for other three observations , hrc - s instrument was used in combination with low energy transmission grating ( letg ) , which covers a wavelength range of 6176 . the description of the data sets is listed in table 1 . @xmath9{ccccc } \hline\hline { \rm seq . } & { \rm obs\_id } & { \rm instr./ } & { \rm exposure } & { \rm start } \\ { \rm num . } & & { \rm grating } & { \rm time~(ks ) } & { \rm time } \\ \hline 280174 & 1224 & { \rm hrc - s / letg } & 20.93 & 1999 - 11 - 08 \\ 280411 & 1461 & { \rm hrc - s / letg } & 70.25 & 1999 - 11 - 07 \\ 290032 & 63 & { \rm hrc - s / letg } & 70.15 & 1999 - 11 - 06 \\ \hline \end{array}\ ] ] in this work , the reduction of the data sets uses _ sherpa _ software package in ciao , version 3.3 , with the science threads for hcr - s / letg observations . the three spectra ( with obs_ids of 63 , 1461 and 1224 ) are co - added with the add_grating_spectra tool to improve the signal - to - noise ratio , which results in a spectrum with total exposure time of 159.5 ks after times of bad counts were excluded . similarly , the associate auxiliary response files ( arfs ) are averaged by this tool . here , the positive and negative spectra are analyzed separately by consideration of the different chip gaps for the two diffraction orders . 13 [ histogram curves ] show the co - added spectra for positive and negative diffraction orders in wavelength range of 4388 spanned by emission lines of highly charged silicon ions ( si vi si xii ) . line fluxes are determined by modelling the spectra locally with narrow gaussian profiles and constant value representing background and ( pseudo-)continuum emissions determined in line - free region . the observed line width is about 0.06 over the interested region , which is comparable with the broadening of instrument for point - like source . the fluxes have been obtained after correction for the effective area . in the fitting , 1@xmath10 uncertainty was adopted to determine the statistical errors for the line fluxes . here , only the line fluxes of highly charged silicon ions are listed in table 2 . for comparison , the line fluxes from raassen et al . are listed , which reveals a good agreement for most peaks , although different observations are used . additionally , solar observation data in this wavelength region are also listed for completeness . the wavelength and line intensity ( in unit of photons@xmath11@xmath12s@xmath13 acrsec@xmath13 ) are from work of acton et al . , whereas the line intensities being less than 10 photons @xmath12s@xmath13 acrsec@xmath13 have not been given . collisional - radiative ( cr ) models for highly charged si vii si xii ions are constructed based on the accurate atomic data from work of liang et al . and their unpublished data . these data are generated with the flexible atomic code ( fac ) provided by gu . 878 , 312 , 560 , 320 and 350 energy levels have been included in predictions of line emissivities of si vii si xi , respectively . these levels belongs to not only singly excited configurations , but also some doubly excited configurations for accounting for configuration interaction as fully as possible . some energy levels are replaced by available experimental values from the national institute of science and technology ( nist ) database . for si xii , 40 energy levels are available from chianti database @xcite , and used here . all possible decay channels among above listed levels by e1 , m1 , e2 and m2 type transitions , have been included in the present model for each charge states . a self - consistent calculation of electron impact excitation has been performed for the five ions ( si vii si xi ) , which based on the rdw method @xcite . these data has been assessed for si ix si xi in our previous work @xcite . for @xmath14 transitions among levels of ground and lower excited configurations , the excitation data is replaced by available @xmath15matrix data , which properly considers resonant effects in the threshold region . to our best knowledge , the @xmath15matrix calculation is not available for si vii . so present rdw calculation of electron impaction excitation is used for si vii by further consideration of self - consistency . bell et al . calculate the electron impact excitation by using _ ab initio r-_matrix method for excitations among levels of ground configuration @xmath16 of si viii . these data replaces our results in predictions of line emissivities . for si ix , resonant effects among levels of ground configuration , have been considered in predictions of level populations by using the data of aggarwal . zhang et al . performed @xmath15matrix calculation for 105 transitions among the 15 fine - structure energy levels belongings to @xmath17 , @xmath18 and @xmath19 configurations of si x. however , an error was noted by keenan , who re - calculated the electron impact excitation rates for the 105 transitions with the @xmath15matrix method . these data is adopted here for si x. @xmath15matrix results for 29 independent transitions among 10 fine - structure energy levels belongings to @xmath20 , @xmath21 and @xmath22 configurations of si xi were presented by berrington et al . . these accurate data replace the rdw calculations in the predictions of level populations of each charge states . by adopting the atomic data described in above section , line emissivities of si vii si xii are calculated at an electron density of 3.0@xmath23 @xmath24 and electron temperatures ranging 0.15 mk . the electron density is a typical value for cool star with lower activities as revealed by previous works @xcite . the silicon abundance adopts solar photospheric value , while the ionization equilibrium uses the result of mazzotta et al . . previous works have revealed that the emission measure ( @xmath25 ) of procyon has a continuous distribution and mainly dominates around 13 mk [ see fig . 3 in _ ref . _ raassen et al . ] . the theoretical line fluxes of highly charged silicon ions , are derived by combining the line emissivities and the emission measure . the predicted line fluxes are listed along with the observed values for comparison in table 2 . furthermore , the theoretical line fluxes couple the effective area extracted from arfs files , and are folded by gaussian profiles with full - width at half - maximum ( fwhm ) of 0.06 representing the observed line width . 13 show the theoretical spectra by color curves for different charge states . in these figures , the positive and negative diffraction are analyzed , separately . for most strong emission lines , the theoretical calculations agree well with the observations within the statistical uncertainty . in solar case , we calculate the line intensity at a higher electron density of 5.0@xmath26 @xmath24 and temperatures of peak fractions for each charge state in the ionization equilibrium @xcite . further the predictions are scaled by solar values at 44.20 ( si xii ) , 46.40 ( si xi ) , 50.52 ( si x ) , 55.41 ( si ix ) , 61.09 ( si viii ) and 70.05 ( si vii ) for each charge state , respectively . in the following , we discuss the calculation and comparison in sequence of charge states . @xmath27{lccccccclcccllll } \hline\hline & & & & & & & & & \multicolumn{3}{c}{\rm this~work } & \multicolumn{4}{c}{\rm transitions } \\ { \rm i d } & { \rm \lambda_{-}~(\aa ) } & { \rm flux_{- } } & { \rm \lambda_{+}~(\aa ) } & { \rm flux_{+ } } & { \rm rma02 } & { \rm \lambda_{\sun } } & f_{\rm \sun } & { \rm ions } & { \rm \lambda~(\aa ) } & f_{\tiny 3cie } & f^p_{\tiny \sun } & \multicolumn{2}{c}{\rm upper~level } & \multicolumn{2}{c}{\rm lower~level } \\ \hline 1 & 43.663 & 0.10(9 ) & 43.656 & 0.11(7 ) & & 43.65 & 21 & { \rm si~xi } & 43.683 & 0.08 & 7 & 2s3p & ^3p_1 & 2s^2 & ^1s_0 \\ 2 & 43.753 & 0.61(9 ) & 43.731 & 0.55(7 ) & 0.54(8 ) & 43.75 & 90 & { \rm si~xi } & 43.763 & 0.39 & 35 & 2s3p & ^1p_1 & 2s^2 & ^1s_0 \\ 3 & 44.029 & 0.49(8 ) & 44.002 & 0.33(6 ) & 0.43(8 ) & 44.02 & 138 & { \rm si~xii } & 44.019 & 0.43 & 125 & 3d & ^2d_{3/2 } & 2p & ^2p_{1/2 } \\ 4 & 44.183 & 0.73(9 ) & 44.131 & 0.60(7 ) & 0.67(10 ) & 44.16 & 224 & { \rm si~xii } & 44.165 & 0.76 & 224 & 3d & ^2d_{5/2 } & 2p & ^2p_{3/2 } \\ 5a & 44.275 & 0.28(8 ) & 44.193 & 0.64(7 ) & 0.52(10 ) & 44.20 & 292 & { \rm si~ix } & 44.213 & 0.08 & 3 & 2s^22p4d & ^3d_1 & 2s^22p^2 & ^3p_2 \\ 5b & & & & & & & & { \rm si~ix } & 44.215 & 0.10 & 4 & 2s^22p4d & ^3d_2 & 2s^22p^2 & ^3p_1 \\ 5c & & & & & & & & { \rm si~ix } & 44.249 & 0.15 & 6 & 2s^22p4d & ^3d_3 & 2s^22p^2 & ^3p_2 \\ 6 & { \bf 45.550 } & 0.14(8 ) & 45.490 & 0.12(6 ) & & 45.51 & 57 & { \rm si~xii } & 45.521 & 0.22 & 67 & 3s & ^2s_{1/2 } & 2p & ^2p_{1/2 } \\ 7 & 45.694 & 0.34(9 ) & 45.664 & 0.22(7 ) & 0.20(4 ) & 45.68 & 109 & { \rm si~xii } & 45.691 & 0.45 & 135 & 3s & ^2s_{1/2 } & 2p & ^2p_{3/2 } \\ 8a & { \bf 46.179 } & 0.16(5 ) & 46.110 & 0.25(6 ) & & & & { \rm si~viii } & 46.133 & 0.09 & 1 & 2s^22p^2(^3p)5d & ^4p_{3/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 8b & & & & & & & & { \rm si~viii } & 46.148 & 0.05 & 1 & 2s^22p^2(^3p)5d & ^4p_{5/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 9a & 46.301 & 0.31(5 ) & 46.272 & 0.31(6 ) & 0.25(7 ) & 46.30 & 55 & { \rm si~xi } & 46.263 & 0.08 & 8 & 2s3d & ^3d_1 & 2s2p & ^3p_0 \\ 9b & & & & & & & & { \rm si~xi } & 46.298 & 0.15 & 15 & 2s3d & ^3d_2 & 2s2p & ^3p_1 \\ 9c & & & & & & & & { \rm si~xi } & 46.313 & 0.06 & 6 & 2s3d & ^3d_1 & 2s2p & ^3p_1 \\ 10 & 46.407 & 0.49(5 ) & 46.366 & 0.42(6 ) & 0.40(8 ) & 46.40 & 50 & { \rm si~xi } & 46.399 & 0.41 & 50 & 2s3d & ^3d_3 & 2s2p & ^3p_2 \\ 11 & 49.217 & 1.40(8 ) & 49.190 & 1.27(6 ) & 1.44(14 ) & 49.22 & 118 & { \rm si~xi } & 49.222 & 1.06 & 94 & 2s3d & ^1d_2 & 2s2p & ^1p_1 \\ 12 & 49.700 & 0.22(7 ) & 49.673 & 0.32(6 ) & 0.29(7 ) & 49.71 & 43 & { \rm si~x } & 49.701 & 0.09 & 1 & 2s2p(^1p)3d & ^2f_{7/2 } & 2s2p^2 & ^2d_{5/2 } \\ 13 & 50.361 & 0.42(10 ) & 50.334 & 0.44(8 ) & 0.51(8 ) & 50.35 & 195 & { \rm si~x } & 50.333 & 0.20 & 1 & 2s2p(^3p)3d & ^4d_{7/2 } & 2s2p^2 & ^4p_{5/2 } \\ 14 & 50.525 & 1.61(10 ) & 50.512 & 1.38(9 ) & 1.68(15 ) & 50.52 & 12 & { \rm si~x } & 50.524 & 1.45 & 12 & 2s^23d & ^2d_{3/2 } & 2s^22p & ^2p_{1/2 } \\ 15a & 50.692 & 1.43(10 ) & 50.676 & 1.04(8 ) & 1.30(14 ) & 50.69 & 74 & { \rm si~x } & 50.691 & 1.08 & 19 & 2s^23d & ^2d_{5/2 } & 2s^22p & ^2p_{3/2 } \\ 15b & & & & & & & & { \rm si~x } & 50.703 & 0.29 & 2 & 2s^23d & ^2d_{3/2 } & 2s^22p & ^2p_{3/2 } \\ 16 & { \bf 50.874 } & 0.23(9 ) & 50.828 & 0.15(8 ) & & & & { \rm si~x } & 50.824 & 0.20 & 0 & 2s2p(^3p)3d & ^4f_{3/2 } & 2s2p^2 & ^4p_{1/2 } \\ 17 & 52.307 & 0.87(7 ) & 52.304 & 0.88(13 ) & 0.75(11 ) & 52.30 & 88 & { \rm si~xi } & 52.298 & 0.94 & 85 & 2s3s & ^1s_0 & 2s2p & ^1p_1 \\ 18 & { \bf 52.453 } & 0.29(7 ) & 52.473 & 0.40(11 ) & & 52.48 & 15 & { \rm si~x } & 52.484 & 0.13 & 3 & 2s2p(^3p)3d & ^2f_{7/2 } & 2s2p^2 & ^2d_{5/2 } \\ 19 & 52.594 & 0.30(7 ) & & & 0.35(8 ) & 52.61 & 40 & { \rm si~x } & 52.612 & 0.33 & 3 & 2s2p(^3p)3d & ^2f_{5/2 } & 2s2p^2 & ^2d_{3/2 } \\ 20a & & & 55.078 & 0.67(7 ) & 0.68(15 ) & 55.11 & 17 & { \rm si~ix } & 55.094 & 0.15 & 6 & 2s^22p3d & ^3p_0 & 2s^22p^2 & ^3p_1 \\ 20b & & & & & & & & { \rm si~ix } & 55.116 & 0.34 & 14 & 2s^22p3d & ^3p_1 & 2s^22p^2 & ^3p_1 \\ 21a & & & 55.246 & 1.19(8 ) & 0.88(25 ) & 55.23 & < 10 & { \rm si~ix } & 55.234 & 0.31 & 13 & 2s^22p3d & ^3p_1 & 2s^22p^2 & ^3p_2 \\ 21b & & & & & & 55.28 & 28 & { \rm si~ix } & 55.272 & 0.51 & 21 & 2s^22p3d & ^3p_2 & 2s^22p^2 & ^3p_2 \\ 22a & & & 55.347 & 2.26(9 ) & 2.14(27 ) & 55.37 & 22 & { \rm si~ix } & 55.356 & 0.97 & 39 & 2s^22p3d & ^3d_2 & 2s^22p^2 & ^3p_1 \\ 22b & & & & & & & & { \rm si~ix } & 55.383 & 0.20 & 8 & 2s^22p3d & ^3d_1 & 2s^22p^2 & ^3p_1 \\ 22c & & & & & & 55.41 & 43 & { \rm si~ix } & 55.401 & 1.06 & 43 & 2s^22p3d & ^3d_3 & 2s^22p^2 & ^3p_2 \\ 23 & & & 56.017 & 0.56(20 ) & 0.19(11 ) & 56.04 & 20 & { \rm si~ix } & 56.027 & 0.07 & 3 & 2s2p^2(^2s)3p & ^3p_0 & 2s^22p^2 & ^1d_2 \\ 24 & & & 57.196 & 0.25(7 ) & & 57.20 & 25 & { \rm si~x } & 57.208 & 0.29 & 6 & 2s2p(^3p)3s & ^2p_{3/2 } & 2s2p^2 & ^2d_{5/2 } \\ 25 & & & 57.309 & 0.35(8 ) & & 57.35 & 17 & { \rm si~x } & 57.366 & 0.62 & 4 & 2s2p(^3p)3s & ^2p_{3/2 } & 2s2p^2 & ^2d_{3/2 } \\ 26a & 61.012 & 1.66(14 ) & & & 1.41(25 ) & 61.03 & 17 & { \rm si~viii } & 60.989 & 0.58 & 8 & 2s^22p^2(^3p)3d & ^4p_{1/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 26b & & & & & & & & { \rm si~viii } & 61.022 & 1.14 & 17 & 2s^22p^2(^3p)3d & ^4p_{3/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 27 & 61.090 & 1.53(14 ) & & & 1.38(24 ) & 61.09 & 24 & { \rm si~viii } & 61.032 & 1.65 & 24 & 2s^22p^2(^3p)3d & ^4p_{5/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 28 & 61.611 & 0.79(13 ) & & & 0.52(17 ) & 61.61 & 18 & { \rm si~ix } & 61.600 & 0.07 & 3 & 2s^22p3s & ^3p_1 & 2s^22p^2 & ^3p_0 \\ 29 & 61.847 & 0.85(13 ) & & & 0.67(11 ) & 61.85 & 21 & { \rm si~ix } & 61.844 & 0.08 & 3 & 2s^22p3s & ^3p_1 & 2s^22p^2 & ^3p_1 \\ 30 & 61.937 & 0.89(13 ) & 61.971 & 1.11(14 ) & 0.55(17 ) & 61.92 & 32 & { \rm si~viii } & 61.792 & 0.18 & 3 & 2s^22p^2(^1d)3d & ^2d_{5/2 } & 2s^22p^3 & ^2d_{3/2 } \\ 31 & 63.715 & 0.91(15 ) & 63.777 & 0.57(11 ) & 0.58(11 ) & 63.72 & 352 & { \rm si~viii } & 63.716 & 0.17 & 2 & 2s^22p^2(^3p)3d & ^2f_{7/2 } & 2s^22p^3 & ^2d_{5/2 } \\ 32 & & & { \bf 64.668 } & 0.27(7 ) & & & & { \rm si~ix } & 64.815 & 0.19 & 8 & 2s2p^2(^4p)3s & ^3p_2 & 2s2p^3 & ^3d_3 \\ 33 & & & { \bf 64.767 } & 0.27(7 ) & & & & { \rm si~ix } & 64.964 & 0.40 & 16 & 2s2p^2(^4p)3s & ^3p_1 & 2s2p^3 & ^3d_2 \\ 34a & 67.152 & 0.73(16 ) & 67.143 & 0.56(25 ) & 0.48(11 ) & 67.15 & 17 & { \rm si~ix } & 67.066 & 0.04 & 2 & 2s2p^2(^4p)3s & ^3p_2 & 2s2p^3 & ^3p_1 \\ 34b & & & & & & & & { \rm si~ix } & 67.071 & 0.11 & 4 & 2s2p^2(^4p)3s & ^3p_2 & 2s2p^3 & ^3p_2 \\ 35a & 67.259 & 0.83(16 ) & 67.287 & 0.58(14 ) & 0.87(18 ) & 67.24 & 42 & { \rm si~ix } & 67.222 & 0.07 & 3 & 2s2p^2(^2d)3s & ^3d_1 & 2s2p^3 & ^3p_1 \\ 35b & & & & & & & & { \rm si~ix } & 67.224 & 0.10 & 4 & 2s2p^2(^2d)3s & ^3d_1 & 2s2p^3 & ^3p_0 \\ 35c & & & & & & & & { \rm si~ix } & 67.227 & 0.10 & 4 & 2s2p^2(^2d)3s & ^3d_1 & 2s2p^3 & ^3p_2 \\ 36a & 67.417 & 0.72(15 ) & 67.369 & 0.39(12 ) & 0.68(14 ) & 67.38 & 41 & { \rm si~viii } & 67.478 & 0.20 & 3 & 2s2p^3(^5s)3d & ^4d_{1/2 } & 2s2p^4 & ^4p_{1/2 } \\ 36b & & & & & & & & { \rm ne~viii } & 67.382 & & & 4p & ^2p_{3/2 } & 2s & ^2s_{1/2 } \\ 37 & 68.132 & 0.41(7 ) & & & & & & { \rm si~vii } & 68.148 & 0.38 & 39 & 2s^22p^3(^2p)3d & ^3d_3 & 2s^22p^4 & ^3p_2 \\ 38a & 69.641 & 2.19(11 ) & 69.694 & 2.31(14 ) & 2.03(21 ) & 69.65 & 256 & { \rm si~viii } & 69.632 & 0.70 & 10 & 2s^22p^2(^3p)3s & ^4p_{5/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 38b & & & & & & & & { \rm si~vii } & 69.664 & 0.46 & 47 & 2s^22p^3(^2d)3d & ^3p_2 & 2s^22p^4 & ^3p_2 \\ 39 & 69.797 & 1.22(11 ) & 69.864 & 1.23(12 ) & 1.05(14 ) & 69.84 & 14 & { \rm si~viii } & 69.790 & 0.38 & 6 & 2s^22p^2(^3p)3s & ^4p_{3/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 40a & { \bf 69.909 } & 0.67(9 ) & 69.993 & 0.73(12 ) & & 69.89 & 14 & { \rm si~ix } & 69.896 & 0.46 & 19 & 2s^22p3p & ^3d_3 & 2s2p^3 & ^3d_2 \\ 40b & & & & & & & & { \rm si~viii } & 69.905 & 0.20 & 3 & 2s^22p^2(^3p)3s & ^4p_{1/2 } & 2s^22p^3 & ^4s_{3/2 } \\ 41 & 70.050 & 0.80(9 ) & 70.102 & 0.48(12 ) & 0.70(11 ) & 70.05 & 62 & { \rm si~vii } & 70.027 & 0.61 & 62 & 2s^22p^3(^2d)3d & ^3d_3 & 2s^22p^4 & ^3p_2 \\ \hline \end{array}\ ] ] @xmath27{lccccccclcccllll } \hline\hline & & & & & & & & & \multicolumn{3}{c}{\rm this~work } & \multicolumn{4}{c}{\rm transitions } \\ { \rm i d } & { \rm \lambda_{-}~(\aa ) } & { \rm flux_{- } } & { \rm \lambda_{+}~(\aa ) } & { \rm flux_{+ } } & { \rm rma02 } & { \rm \lambda_{\sun } } & f_{\rm \sun } & { \rm ions } & { \rm \lambda~(\aa ) } & f_{\tiny 3cie } & f^p_{\tiny \sun } & \multicolumn{2}{c}{\rm upper~level } & \multicolumn{2}{c}{\rm lower~level } \\ \hline 42 & 72.443 & 0.36(15 ) & 72.469 & 0.17(17 ) & & 72.41 & 19 & { \rm si~viii } & 72.421 & 0.15 & 2 & 2s^22p^2(^3p)3s & ^2p_{1/2 } & 2s^22p^3 & ^2d_{3/2 } \\ 43 & { \bf 73.189 } & 0.40(12 ) & 73.124 & 0.50(22 ) & & & & { \rm si~vii } & 73.123 & 0.64 & 65 & 2s^22p^3(^4s)3d & ^3d_3 & 2s^22p^4 & ^3p_2 \\ 44 & { \bf 76.010 } & 0.90(18 ) & 76.038 & 0.56(18 ) & 0.77(13 ) & 76.03 & 55 & { \rm si~viii } & 75.988 & 0.71 & 10 & 2s2p^3(^5s)3s & ^4s_{3/2 } & 2s2p^4 & ^4p_{5/2 } \\ 45 & 76.157 & 0.25(15 ) & 76.186 & 0.71(18 ) & & 76.16 & 21 & { \rm si~viii } & 76.196 & 0.46 & 7 & 2s2p^3(^5s)3s & ^4s_{3/2 } & 2s2p^4 & ^4p_{3/2 } \\ \hline \end{array}\ ] ] * si xii * three emission lines at 44.029 , 44.183 and 45.694 were identified in work of raassen et al . . these lines are reproduced again in present prediction , and their intensities are slightly higher than the observations , yet are still within 1@xmath10 statistical error . the scaled theoretical line intensities are also in agreement with the solar observation as shown in table 2 . furthermore , we notice an emission at 45.550 , its intensity agrees with the procyon observation in the both diffractions , as shown in fig . 1 . so we assign the emission to the @xmath28@xmath29 transition with wavelength of 45.521 . for the stellar spectra , this assignment is the first time to our best knowledge . this assignment is confirmed in solar observation with higher resolution ( 0.02 ) . * si xi * the present prediction satisfactorily reproduces the emission lines at 43.753 , 46.301 , 46.407 , 49.217 and 52.307 as illustrated in table 2 and fig . 1 . around the peak at 43.753 , a weak line with wavelength of 43.663 is predicted , which can explain the left - side wing of the emission at 43.753 . in solar flare observation , this line is clearly resolved with a wavelength of 43.65 . yet it still has not been identified so far . for the emission line at 52.307 , the calculation agrees with the observed flux in both diffractions , as shown in table 2 and fig . 1 . however , acton et al . pointed that there is a contamination from al xi line at 52.244 . present prediction indicates that the contribution from al xi line can be negligible for procyon and solar . the line ratios among these emission lines show powerful diagnostic potentials for the electron density and temperature as revealed in work of liang & zhao . for example , the intensity of 43.753 line is sensitive to the electron temperature , and increases relative to the intensity of the 52.307 line with increasing the electron temperature [ see fig . 2 in ref . liang & zhao ] . the larger discrepancy ( @xmath3050% ) of this line between the prediction and solar value reveals the higher electron temperature in solar flare . * si x * the two strong emission lines at 50.525 and 50.692 show an excellent @xmath31diagnostic potential for inactive cool stars as reported in our previous work @xcite . they are satisfactorily reproduced at the electron density of 3.0@xmath23 @xmath24 and the em distribution determined by raassen et al . . for the emission at 50.361 , raassen et al . assigned it to si x according to the line - list of kelly database @xcite , whereas the database is based upon the high - density plasma such as laser plasma . fortunately , present work also predicts the emission with wavelength of 50.333 , but it contributes about @xmath3048% in both diffractions . this means that there is unknown contamination from other lines . in solar flare observation , acton et al . assign the observed emission at 50.35 to fe xvi line with wavelength of 50.350 . so the left contribution at 50.361 maybe from the fe xvi in procyon coronae . however , in solar flare , the prediction reveals that the contribution from si x at 50.35 of spectrum can be negligible . for peaks at 50.874 and 52.594 , the predicted line fluxes from the @xmath32@xmath33 ( 50.824 ) and @xmath34@xmath35 ( 52.612 ) transitions are comparable with the observed values within 1@xmath10 error , which indicates the contamination from ni xviii as revealed by raassen et al . is negligible . but , in solar flare , the contribution from ni xviii is dominant , because si x contributes only @xmath308% . the present calculation predicts another emission lines at 52.484 . moreover it intensity is 33% and 43% when compared to the observed flux in negative and positive diffractions , respectively . in solar flare , the contribution from si x is less than 7% at 52.48 that is the emission line is still an unknown line . at the emission of 49.700 , the contribution from si x ( 49.701 ) is predicted to be @xmath3036% and 2% , in procyon and sun , respectively . in wavelength range of 57.057.5 , two emission lines ( @xmath3657.196 and 57.309 ) are detected in the positive spectrum , whereas they havent been identified so far . in negative order spectrum , there are no counts between 55.758.5 , because of the chip gap . present work predicts line flux at wavelengths of 57.208 being consistent with the observed values at 57.196 , while the flux at 57.366 being higher than observed value at 57.309 by 75% . so we tentatively assign the two emission lines to the @xmath37 @xmath38 and @xmath39 @xmath35 transitions of si x. in solar flare , 24% contribution of si x is predicted for the two emissions . * si ix * present calculation satisfactorily reproduces the line emissions at 55.246 and 55.347 . the line intensity ratio between them is sensitive to the electron density ( @xmath40 ) , whereas insensitive to the electron temperature ( @xmath5 ) with variation less than 3% over log@xmath41 ( k)=5.96.3 [ see fig . 3 in ref . liang & zhao ] . the contribution of si ix to the observed flux at 55.078 occupies a considerable part . when 14% contribution from si x is considered , the contamination from mg ix pointed out by raassen et al . can be negligible . at the peak at 56.017 , we found the contribution from si ix is very small . the major component maybe from s ix ( 56.081 ) and ni xiii ( 56.000 ) as pointed out by raassen et al . . in solar flare , the contribution from si ix ( 56.027 ) is similarly very small ( @xmath3015% ) . at the position about 44.218 , a large difference is found between the both diffractions , which is due to the low statistical performance in the interested wavelength region . the contribution from si ix is estimated to be @xmath30100% and 50% in the negative and positive diffractions , respectively . additionally , contamination from second - order diffraction of o vii line ( [email protected] ) may be the origin of the difference , as pointed out by acton et al . in the observation of solar flare . the two @xmath0@xmath1 transition lines at 61.611 and 61.847 are underestimated greatly in procyon and solar flare . the large discrepancies compared with the @xmath2@xmath1 lines at 55.359 , maybe from other unknown contamination . indirect processes for populations on @xmath0 and @xmath2 levels and opacity effects also maybe the reasons of the deviations as in the debated case of fe xvii ( * ? ? ? * ; * ? ? ? * ect . ) . the theoretical calculation also predicts an considerable line fluxes at 64.815 and 64.964 being from @xmath43@xmath44 and @xmath45@xmath46 transitions . but , no emission lines are detected at the two positions . we notice that the predicted fluxes agree with the observed values at measured wavelength of 64.668 and 64.767 . so we tentatively assign the two emission lines from @xmath0@xmath1 transitions of si ix by consideration of the 1% uncertainty of level energies for double excited levels . though the lines at 67.152 and 67.259 were assigned to be mg ix ( 67.132 ) and ne viii ( 67.350 ) , respectively , by raassen et al . , we also notice the contamination ( @xmath302040% ) from si ix in procyon and solar flare . the contribution of si ix ( 69.896 ) is up to @xmath3070% at the peak of 69.909 as shown in fig . 2 . the left contribution is from si viii ( 69.905 ) . in solar flare , the contribution from si viii is predicted to be less than 20% . * si viii * the line ratio between @xmath2@xmath1 ( 61.022 ) and @xmath0@xmath1 ( 69.646 ) lines is predicted higher than observation again as revealed in si ix @xcite and fe xvii ( * ? ? ? * and references therein ) . the predicted line fluxes at 61.012 and 61.090 are slightly overestimated , but agree with observations within 1@xmath10 error for procyon and solar flare observations . this reveals that the contamination from mg ix ( 61.088 ) can be negligible . at peaks of 69.641 and 69.797 , the contribution from @xmath0@xmath1 transition of si viii is about 32% . table 2 indicates that the line intensity of si vii line ( 69.664 ) is @xmath3065% relative to the intensity of si viii line ( 69.632 ) . the total contribution from si viii ( 69.632 ) and si vii ( 69.664 ) are predicted to be @xmath3050% and @xmath3025% for procyon and solar flare , that is there is an unknown contamination besides the minor contribution of fe xiv as reported by acton et al . . we also notice that the prediction at 75.988 and 76.196 agree well with the observed fluxes at 76.010 and 76.157 . when a deviation of 1% in the double excited energy levels is considered , we tentatively assign the two emissions to si viii lines with wavelengths of 75.988 and 76.196 . in solar flare , the contribution from si viii is estimated to be @xmath302040% . acton et al . assign the two emission lines to fe xiv ( 76.023 ) and fe xiii ( 76.152 ) , respectively . moreover , 10 photons@xmath11@xmath12s@xmath47acrsec@xmath13 at 68.85 is detected in solar flare , which was assigned to the @xmath48@xmath49 transition of si viii with wavelength of 68.853 by acton et al . . yet , no dominant flux of si viii is predicted at this wavelength region . the lower contributions from si viii for solar flare , is due to the solar flare is hotter than procyon coronae . * si vii * as illustrated in fig . 3 and table 2 , the line flux of si vii ( 70.027 ) occupies about nearly 80% , that is the contribution from fe xii ( 70.010 ) and fe xv ( 70.054 ) is very small . moreover , in the positive spectrum , the observed wavelengths around this region systematically longer than calculations and the observed values in the negative spectrum by @xmath300.06 . in solar flare , the contribution from fe xv become strong as the assignment by acton et al . . at the peak around 68.132 of the negative spectrum of procyon , the predicted intensity agrees well with observed value . so we tentatively assign the emission line to @xmath50@xmath51 transition ( 68.148 ) of si vii . this assignment is the first identification for stellar coronal spectra . at peaks of 73.189 , 81.61 and 86.876 , the calculation satisfactorily reproduces the procyon observations . however , the former two lines havent been identified so far , and the last emission line is assigned to mg viii by according to the line - lists of kelly database . in the low - density astrophysical plasma , the contribution from mg viii can be negligible . at the peak of 81.61 in the positive spectrum , the observed wavelength is shorter than that in negative spectrum and the calculation by @xmath300.06 . this maybe due to the wavelength calibration in the positive diffraction . in summary , available observation data ( obs_ids of 63 , 1461 and 1224 ) for cool star procyon are co - added and analyzed . by adopting our published atomic data and some unpublished data , line emissivities of si vi si xii ions are calculated at an electron density of 3.0@xmath23 @xmath24 ( typical value for cool stars ) and temperatures over 0.15.0 mk . in this work , energy levels of 878 , 312 , 560 , 320 , 350 and 40 have been included for si vii si xii , respectively . moreover , decay rates of transitions ( including e1 , m1 , e2 and m2 ) up to tens of thousand are included . some electron impact excitation data are replaced by available @xmath15matrix data to take resonant effects into account as far as possible . based upon the em derived by raassen et al . and the calculated emissivities , we estimated the theoretical line fluxes of highly charged si vii si xii ions . by detailed comparison between the observations and predictions , several emissions lines [ with bold font in second or forth column of table 2 ] are identified firstly to our best knowledge . the identification is assessed by comparing with solar spectrum with higher resolution ( 0.02 ) . the prediction indicates the large discrepancies between the @xmath0@xmath1 line ( 69.641 ) _ versus _ @xmath2@xmath1 line ( 61.012 ) again for lower charge stage si viii . this reveals that more accurate calculation of cross section in electron - ion interactions is very necessary . for emission lines at 52.594 ( si x ) , 55.078 ( si ix ) and 70.027 ( si vii ) , this work indicates that the contamination from ni xviii ( 52.615 ) , mg ix ( 55.060 ) and fe xii ( 70.010 and 70.054 of fe xv ) , respectively , can be negligible for procyon . however , in solar flare , the contaminations from mg , fe and ni ions become dominant . this is due to the solar flare is hotter than the procyon coronae . for the peak around 69.641 , si vii line has the comparable contribution with the si viii lines identified by raassen et al . . however , their total contribution is less than 25% in solar flare . the discrepancies and the identification strongly suggest the benchmark from laboratory measurements at low - density plasmas . this work was supported by the national natural science foundation of china under grant nos . 10603007 and 10521001 , as well as national basic research program of china ( 973 program ) under grant 2007cb815103 . 99 acton l.w . , bruner m.e . , brown w.a . , et al . , 1985 , apj , 291 , 865 aggarwal k.m . , 1983 , j. phys . b , 16 , l59 audard m. , behar e. , g@xmath52del m. , et al . , 2001 , a@xmath53a , 365 , l329 behar e. , cottam j. , kahn s. m. , et al . , 2001 , apj , 548 , 966 beiersdorfer p. , 2003 astrophys . , 41 , 343 beiersdorfer p. , bitter m. , von goeler s. , hill k.w . , 2004 , apj , 610 , 616 bell k.l . , matthews a. , ramsbottom c.a . , 2001 , mnras , 322 , 779 berrington k.a . , burke p.g . , dufton p.l . , kingston a.e . , 1985 , atom . data and nucl . data tables , 33 , 195 brinkman a. c. , behar e. , g@xmath52del m. , et al . , 2001 , a@xmath53a , 365 , l324 brinkman a. c. , gunsing c. j. t. , kaastra j. s. , et al . , 2000 , apj , 530 , l111 canizares c. r. , huenemoerder d. p. , davis d. s. , et al . , 2000 , apj , 539 , l41 doron r. , behar e. , 2002 , apj , 574 , 518 doyle j.g . , keenan f.p . , ryans r.s.i . , et al . , 1999 , solar physics , 188 , 73 flanagan k.a . , canizares c.r . , dewey d. , et al . , 2004 , apj , 605 , 230 gu m. f. , 2003 , apj , 582 , 1241 hummer d.g . , berrington k.a . , eissner w. , et al . , 1993 , a&a , 279 , 298 keenan f.p . , 2000 , mnras , 315 , 450 keenan f.p . , aggarwal k.m . , katsiyannis a. c. , et al . , 2003 , solar physics , 217 , 225 keenan f. p. , aggarwal k.m . , williams d. r. , et al . , 2001 , mnras , 326 , 1387 keenan f.p . , conlon e.s . , foster v.j . , et al . , 1993 , solar physics , 145 , 291 keenan f.p . , oshea e. , thomas r.j . , et al . , 2000 , mnras , 315 , 450 keenan f.p . , pinfield d.j . , mathioudaskis m. , et al . , 2000 , solar physics , 197 , 253 keenan f.p . , pinfield d.j . , woods v.j . , et al . , 1998 , apj , 503 , 953 kelly r.l . 1987 , j.phys.chem.ref.data , 16 , suppl . landi e. , del zanna g. , young p.r . , dere k.p . , mason h.p . , landini m. , 2006 , apjs , 162 , 261 lepson j.k . beiersdorfer p. , behar e. , kahn s.m . , 2003 , apj , 590 , 604 lepson j.k . beiersdorfer p. , behar e. , kahn s.m . , 2005a , apj , 625 , 1045 lepson j.k . beiersdorfer p. , behar e. , kahn s.m . , 2005b , nucl . inst . and meth . research b , 235 , 131 liang g.y . , zhao g. , 2006a , aj , 132 , 1547 liang g.y . , zhao g. , shi j.r . , 2006b , aj , 132 , 371 liang g.y . , zhao g. , shi j.r . , 2006c , mnras , 368 , 196 liang g.y . , zhao g. , 2007 , new astronomy , 12 , 435 liang g.y . , zhao g. , zeng j.l . , 2007 , atom . data and nucl . data tables , 93 , 375 liang g.y . , zhao g. , zhong j.y . , et al . , 2007 , apj ( in re - review ) mazzotta p. , mazzitelli g. , colafrancesco s. , vittorio n. , 1998 , apjs , 133,403 ness j. -u . , mewe r. , schmitt j. h. m. m. , et al . , 2001 , a@xmath53a , 367 , 282 ness j. -u . , schmitt j. h. m. m. , audard , m. , g@xmath54del m. , mewe r. , 2003 , a&a , 407 , 347 ness j. -u . , schmitt j. h. m. m. , burwitz v. , et al . , 2002 , a&a , 394 , 911 raassen , a. j. j. , mewe , r. , audard , m. , et al . , 2002 , a&a , 389 , 228 rasmussen a. p. , behar e. , kahn s. m. , et al . , 2001 , a@xmath53a , 365 , l231 seaton m.j . , yan y. , mihalas d. , pradhan a.k . , 1994 , mnras , 266 , 805 zhang h.l . , graziani m. , pradhan a.k . , 1994 , a&a , 283 , 319

modeling species abundance patterns using local environmental features is an important , current problem in ecology . the cape floristic region ( cfr ) in south africa is a global hot spot of diversity and endemism , and provides a rich class of species abundance data for such modeling . here , we propose a multi - stage bayesian hierarchical model for explaining species abundance over this region . our model is specified at areal level , where the cfr is divided into roughly one minute grid cells ; species abundance is observed at some locations within some cells . the abundance values are ordinally categorized . environmental and soil - type factors , likely to influence the abundance pattern , are included in the model . we formulate the empirical abundance pattern as a degraded version of the potential pattern , with the degradation effect accomplished in two stages . first , we adjust for land use transformation and then we adjust for measurement error , hence misclassification error , to yield the observed abundance classifications . an important point in this analysis is that only of the grid cells have been sampled and that , for sampled grid cells , the number of sampled locations ranges from one to more than one hundred . still , we are able to develop potential and transformed abundance surfaces over the entire region . in the hierarchical framework , categorical abundance classifications are induced by continuous latent surfaces . the degradation model above is built on the latent scale . on this scale , an areal level spatial regression model was used for modeling the dependence of species abundance on the environmental factors . to capture anticipated similarity in abundance pattern among neighboring regions , spatial random effects with a conditionally autoregressive prior ( car ) were specified . model fitting is through familiar markov chain monte carlo methods . while models with car priors are usually efficiently fitted , even with large data sets , with our modeling and the large number of cells , run times became very long . so a novel parallelized computing strategy was developed to expedite fitting . the model was run for six different species . with categorical data , display of the resultant abundance patterns is a challenge and we offer several different views . the patterns are of importance on their own , comparatively across the region and across species , with implications for species competition and , more generally , for planning and conservation . , , , .

1011.3327

ecologists increasingly use species distribution models to address theoretical and practical issues , including predicting the response of species to climate change [ @xcite , @xcite , @xcite ] , designing and managing conservation areas [ @xcite ] , and finding additional populations of known species or closely related sibling species [ @xcite , @xcite ] . in all these applications the core problem is to use information about where a species occurs and about relevant environmental factors to predict how likely the species is to be present or absent in unsampled locations . the literature on species distribution modeling covers many applications ; there are useful review papers that organize and compare model approaches [ @xcite , @xcite , @xcite , @xcite , @xcite ] . most species distribution models ignore spatial pattern and thus are based implicitly on two assumptions : ( 1 ) environmental factors are the primary determinants of species distributions and ( 2 ) species have reached or nearly reached equilibrium with these factors [ @xcite , @xcite ] . these assumptions underlie the currently dominant species distribution modeling approaches generalized linear and additive models ( glm and gam ) , species envelope models such as bioclim [ @xcite ] , and the maximum entropy - based approach maxent [ @xcite ] . the statistics literature covers glm and gam models extensively . the latter tends to fit data better than the former since they employ additional parameters but lose simplicity in interpretation and risk overfitting and poor out - of - sample prediction . climate envelope models and the now increasingly - used maximum entropy methods are algorithmic and not of direct interest here . in addition to the fundamental ecological issues mentioned above , complication arises in various forms in modeling abundance from imperfect survey data such as observer error [ @xcite , @xcite ] , variable sampling intensity , gaps in sampling , and spatial misalignment of distributional and environmental data [ @xcite ] . first , since a region is almost never exhaustively sampled , individuals not exposed to sampling will be missed . second , it may be that potentially present individuals are undetected [ @xcite ] and , possibly vice versa , for example , a false positive misclassification error with regard to species detection [ @xcite ] . a third complication is that ecologists and field workers are biased against absences ; they tend to sample where species are , not where they are nt . such _ preferential sampling _ and its impact on inference is discussed in @xcite . further complications arise with animals due to their mobility . previous work has developed spatial hierarchical models that accommodate some of these difficulties , fitting these models to presence / absence data in a bayesian framework [ @xcite , @xcite ( @xcite ) , @xcite ] . the species distribution modeling discussed above is either in the presence / absence or presence - only data settings ; there is relatively little work on spatial abundance patterns , despite their theoretical and practical importance [ @xcite , @xcite ] . our primary contribution here is to develop a hierarchical modeling approach for ordinal categorical abundance data , explained by the suitability of the environment , the effect of land use / land transformation , and potential misclassification error . ordinal classifications are often the case in ecological abundance data , especially for plants [ @xcite , @xcite ] . from a stochastic modeling perspective , categorical data can be viewed as the outcome of a multinomial model , with the cell probabilities dependent on background features . within a bayesian framework such modeling is often implemented using data augmentation [ @xcite ] , introducing a latent hierarchical level . there , the ordered classification is viewed as a clipped version of a single _ latent _ continuous response , introducing cut points . see also @xcite and @xcite . at the latent level , suitability of the environment can be modeled through regression . availability in terms of land use degrades suitability . that is , an important feature of our modeling , from an ecological point of view , is that it deals with transformation of the study area by human intervention . in much of the region , the `` natural '' state of areas has been altered to an agricultural or urban state , or the vegetation has been densely colonized by alien invasive plant species . so , we can not treat the entire region as equally _ available _ to the plant species we are modeling . we must introduce a contrast between the current abundance of species ( their _ transformed _ or _ adjusted _ abundance ) and their potential distributions in the absence of land use change ( _ potential _ abundance ) . these notions are formally defined at the areal unit level in section [ sec3 ] . a further degradation enabling the possibility of misclassification and/or observer error in the data collection procedure can be accounted for as _ measurement error _ in the latent surface . there is a substantial literature on measurement error modeling for continuous observations , for example , @xcite , @xcite , and @xcite . in our modeling we impose a hard constraint : with no potential presence ( i.e. , an unsuitable environment ) , we can observe only zero abundance . we enforce this constraint on the latent scale . with cut points , modeled as random , we provide an explanatory model for the observed categorical abundance data . furthermore , we can invert from the latent abundance scale to the categorical abundances to predict abundance for unsampled cells and also to predict abundance in the absence of land use transformation . with spatial data collection , we anticipate spatial pattern in abundance and thus introduce spatial structure into our modeling . that is , causal ecological explanations such as localized dispersal , as well as omitted ( unobserved ) explanatory variables with spatial pattern such as local smoothness of geological or topographic features , suggest that , at sufficiently high resolution , abundance of a species at one location will be associated with its abundance at neighboring locations [ @xcite ] . moreover , through spatial modeling , we can provide spatial adjustment to cells that have not been sampled , accommodating the gaps in sampling and irregular sampling intensity mentioned above . in particular , we create a latent _ process _ model through a trivariate spatial process specification , with truncated support , to capture potential abundance , land transformation - adjusted abundance , and measurement error - adjusted abundance . since our environmental information is available at grid cell level , we use markov random field ( mrf ) models [ @xcite , @xcite ] to capture spatial dependence and to facilitate computation . however , we work with a large landscape of approximately @xmath0 grid cells which leads to very long run times in model fitting and so we introduce a novel parallelized computing strategy to expedite fitting . there have been other recent developments in modeling of species abundances , some using bayesian hierarchical models . first , there has been some work on developing models that deal almost exclusively with animal census data , including count data and mark - recapture data [ @xcite , @xcite , @xcite ] . @xcite provide an overview of abundance modeling , in fact , five regression models ( poisson , negative binomial , quasi - poisson , the hurdle model , and the zero - inflated poisson ) fitted for one particular plant example . these models focus on correcting observer error and bias as well as under - detection ( the species is present but not observed ) , whence the `` true '' abundance is virtually always higher than observed [ @xcite , @xcite ] . we note some very recent work on working with ordinal species abundance in plant data by @xcite . this approach takes ordinally scored abundances and uses an ordered logit hierarchical bayes model to infer potential abundances for species that are still spreading across the landscape . the advantages of working within the bayesian framework with markov chain monte carlo ( mcmc ) model fitting are familiar by now full inference about arbitrary unknowns , that is , functions of model parameters and predictions , can be achieved through their posterior distributions , and uncertainty can be quantified exactly rather than through asymptotics . in this application we work with the disaggregated data at the level of individual species and sites to present spatially resolved abundance `` surfaces '' and to capture uncertainty in model parameters . doing this turns out to be more difficult than might be expected , as we reveal in our model development section . the key modeling issues center on careful articulation of the definition of events and associated probabilities , the misalignment between the sampling for abundance ( at the relatively small sampling sites ) and the available environmental data layers ( at a scale of minute by minute grid cells , roughly 1.55 [email protected] km over the region ) , the sparseness of observations in terms of the entire landscape ( with uneven sampling intensity including many `` holes '' ) , the occurrence of considerable human intervention with regard to land use across the landscape ( `` transformation '' ) , and the need for spatially explicit modeling . the format of the paper is as follows . section [ sec2 ] describes the motivating data set . section [ sec3 ] develops the multi - level abundance model . section [ sec4 ] details the computational and inference issues . in section [ sec5 ] we present an analysis of the data from the cape floristic region ( cfr ) and conclude with some discussion and future extensions in section [ sec6 ] . the focal area for this abundance study is the cape floristic kingdom or region ( cfr ) , the smallest of the world s six floral kingdoms ( figure [ fig1 ] ) . as noted above , it encompasses a small region of southwestern south africa , about 90,000 km@xmath3 , including the cape of good hope , and is partitioned into 36,907 minute - by - minute grid cells of equal area . it has long been recognized for high levels of plant species diversity and endemism across all spatial scales . the region includes about 9000 plant species , @xmath4 of which are found nowhere else [ @xcite ] . globally , this is one of the highest concentrations of endemic plant species in the world . it is as diverse as many of the world s tropical rain forests and apparently has the highest density of globally endangered plant species [ @xcite ] . the plant diversity in the cfr is concentrated in relatively few groups , such as the icon flowering plant family of south africa , the proteaceae . we focus on this family because the data on species distribution and abundance patterns are sufficiently rich and detailed to allow complex modeling . the proteaceae have also shown a remarkable level of speciation , with about 400 species across africa , of which 330 species are @xmath5 restricted to the cfr . of those 330 species , at least 152 are listed as `` threatened '' with extinction by the international union for the conservation of nature . proteaceae have been unusually well sampled across the region by the protea atlas project of the south african national biodiversity institute [ @xcite ] . data were collected at _ record localities _ : relatively uniform , geo - referenced areas typically 50 to 100 m in diameter . in addition to the presence ( or absence ) at the locality of protea species , abundance of each species along with selected environmental and species - level information were also tallied [ @xcite ] . to date , some 60,000 localities have been recorded ( including null sites ) , with a total of about 250,000 species counts from among some 375 proteas [ @xcite ] . km@xmath6 region was divided into @xmath7 @xmath8-minute cells for modeling . ] abundance is given for a sampling locality . evidently , there is no notion of abundance at a point ; however , with roughly @xmath9 sites sampled over the entire cfr , the relative scale of the protea atlas observations is small enough when compared to our areal units to be considered as `` points . '' in the literature , abundance is sometimes measured as percent cover [ @xcite ] . in our data set , abundance is recorded as an ordinal categorical classification of the count for the species with four categories : category 0@xmath10 none observed , category 1@xmath10 110 observed , category 2@xmath10 11100 observed , category 3@xmath10 @xmath11100 observed . such categorization is fast and efficient for studying many species and many sampling locations but is certainly at risk for measurement error in the form of misclassification . additionally , a large number of cells were not sampled at all . in fact , only 10,158 , that is , @xmath1 , were sampled at one or more sites . even among cells sampled , some have just one or two sites while others have more than 100 , reflecting the irregular and opportunistic nature of the sampling rather than an experimentally designed sampling plan . turning to the covariates , in @xcite ( @xcite ) @xmath12 explanatory environmental variables were studied , capturing climate , soil , and topographic features ( further detail is provided there ) . here , we confine ourselves to the six most significant variables from that study , which are evapotranspiration ( apan.mean ) , july ( winter ) minimum temperature ( min07 ) , january ( summer ) maximum temperature ( max01 ) , mean annual precipitation ( mean.an.pr ) , summer soil moisture days ( sumsmd ) , and soil fertility ( fert1 ) . transformed areas ( by agriculture , reforestation , alien plant infestation , and urbanization ) were obtained as a gis data layer from r. cowling ( private communication ) . figure [ fig2 ] shows the pattern of transformation across the cfr . approximately @xmath13 of the cape has been transformed , mainly in the lowlands on more fertile soils where rainfall is adequate [ @xcite ] . most of the transformation outside of these areas , on the infertile mountains , is due to dense alien invader species , which are currently a major threat to fynbos vegetation and , in particular , to the proteaceae . in section [ subsec31 ] we briefly review the earlier work on hierarchical modeling for presence / absence data , presented in @xcite ( @xcite ) , in order to reveal how we have generalized it for the abundance problem . section [ subsec32 ] develops our proposed probability model for the categorical abundance data . in section [ subsec33 ] discrete probability distributions are replaced using latent continuous variables . in section [ subsec34 ] we discuss bias issues associated with modeling abundance data and , in particular , how they affect our setting . section [ subsec35 ] deals with explicit model details for the likelihood and posterior . in @xcite ( @xcite,@xcite ) the authors model at the scale of the grid cells in the cfr and provide a block averaged binary process presence / absence model at this scale . in particular , let @xmath14 denote the geographical region corresponding to the @xmath15th grid cell and @xmath16 the event that a randomly selected location within @xmath17 is suitable ( 1 ) or unsuitable ( 0 ) for species @xmath18 . set @xmath19 . then , @xmath20 is conceptualized by letting @xmath21 be a binary process over the region indicating the suitability ( 1 ) or not ( 0 ) of location @xmath22 for species @xmath18 and taking @xmath20 to be the block average of this process over unit @xmath15 . that is , @xmath23 where @xmath24 denotes the area of @xmath17 . from equation ( [ eq311 ] ) , the interpretation is that the more locations within @xmath17 with @xmath25 , the more suitable @xmath17 is for species @xmath18 , that is , the greater the chance of potential presence in @xmath17 . the collection of @xmath20 s over the @xmath17 is viewed as the potential distribution of species @xmath18 . let @xmath26 denote the event that a randomly selected location in @xmath17 is suitable for species @xmath18 in the presence of transformation of the landscape . let @xmath27 be an indicator process indicating whether location @xmath22 is transformed ( @xmath8 ) or not ( 0 ) . then , at @xmath22 , both @xmath28 ( availability ) and @xmath25 ( suitability ) are needed in order that location @xmath22 is suitable under transformation . therefore , @xmath29 if , for each pixel , availability is uncorrelated with suitability , then equation ( [ eq312 ] ) simplifies to @xmath30 , where @xmath31 denotes the proportion of area in @xmath17 which is untransformed , @xmath32 . next , assume that @xmath17 has been visited @xmath33 times in untransformed areas within the cell . further , let @xmath34 be the observed presence / absence status of the @xmath18th species at the @xmath35th sampling location within the @xmath15th unit . the @xmath36 are modeled as i.i.d . bernoulli trials with success probability @xmath37 , that is , for a randomly selected location in @xmath17 , @xmath37 is the probability of species @xmath18 being present given the location is both suitable and available . of course , given @xmath38 , @xmath39 with probability 1 . then , we have that @xmath40 . @xcite ( @xcite ) model the @xmath20 and @xmath37 using logistic regressions . in fact , they use environmental variables and spatial random effects to model the @xmath20 s , the probabilities of potential presence , and , to facilitate identifiability of parameters , use species level attributes to model the @xmath37 s . we first define what _ categorical abundance _ means at an areal scale using the four ordinal categories from section [ sec2 ] . suppressing the species index , let @xmath41 denote the classification for a randomly selected location in @xmath17 and define @xmath42 @xmath43 @xmath44 for @xmath45 . if @xmath46 is a four - colored process taking values @xmath47 , then @xmath48 . that is , @xmath42 is the proportion of area within @xmath17 with color @xmath49 , equivalently , the proportion in abundance class @xmath49 . the @xmath42 denote the potential abundance probabilities , that is , in the absence of transformation . we describe land transformation percentage @xmath50 as a block average of a binary availability process @xmath27 over @xmath17 . it can also be interpreted as the probability that a randomly selected site within @xmath17 is transformed . at a transformed location abundance _ must _ be @xmath51 . thus , as in equation ( [ eq312 ] ) , in the presence of transformation , we revise @xmath42 to @xmath52 . under independence of abundance and land transformation , we obtain @xmath53 . the @xmath54 denote the transformed abundance probabilities for @xmath55 . the probability of abundance class @xmath51 under transformation is evidently @xmath56 . let @xmath57 denote the abundance class probabilities in the presence of transformation . finally , suppose there is an observed categorical abundance at location @xmath35 within @xmath17 , say , @xmath58 . there is an associated conceptual @xmath59 and an observed @xmath60 . then , @xmath61 if there has been transformation degradation at location @xmath35 , unless @xmath62 . furthermore , if there has been a misclassification error at @xmath35 , @xmath63 unless @xmath62 . let @xmath64 denote the abundance class probabilities associated with the observed abundances . in section [ subsec33 ] we specify a latent trivariate continuous abundance model that produces the @xmath65 s , @xmath66 s , and @xmath67 s by integrating over appropriate intervals . this latent model can be viewed as the _ process model _ for our setting . the data set consists of observed abundances across several sampling sites within the cfr . let @xmath68 denote our cfr study domain so @xmath68 is divided into @xmath69 grid cells of equal area . for each cell @xmath70 , we are given information on @xmath65 covariates as @xmath71 . within @xmath17 , the abundance category of a species was recorded at each of @xmath72 sampling sites . for many cells @xmath73 . for site @xmath35 in @xmath17 we observe @xmath58 as a multinomial trial , that is , @xmath74 . we have a large number of unsampled cells , that is , @xmath75 . in fact , out of 36,907 cells , only @xmath76 ( @xmath1 ) were sampled at one or more sites . figure [ fig3 ] indicates locations of sampled cells . for the unsampled cells there are no @xmath58 s in the data set . hence , the inference problem involves estimation of probabilities over the observed cells as well as _ prediction _ over the unsampled region . prediction of a categorical response distribution for unsampled locations in a point level model was discussed in @xcite and @xcite . in our areal setup with only areal level @xmath77 s , we address this problem with a mrf model , as described in section [ subsec33 ] . again , we seek to infer about the @xmath65 s , @xmath66 s , and @xmath67 s given the observed @xmath78 s for a subset of cells and @xmath77 s known for all cells . it is now common to model the probability mass function of a scalar _ ordinal _ categorical variable through an underlying univariate continuous distribution . in a more general setup , @xcite and @xcite used latent random vectors to define the categorical probabilities in terms of these vectors taking values within a specific set . in a similar spirit , corresponding to an observed abundance category variable @xmath58 , we introduce a continuous latent variable @xmath79 such that @xmath80 where @xmath81 are an increasing sequence of cut points . for identifiability and without loss of generality , we can set @xmath82 and interpret @xmath83 as an absence , @xmath84 as a presence . we have @xmath85 . so @xmath64 will be determined by the probability model specified for the @xmath79 s . we will introduce spatial dependence between @xmath79 s below but , for now , to simplify notation , we drop the subscript . a simple model would put a gaussian distribution on these latent @xmath86 s whose means are linear functions of the associated @xmath77 s . this would provide a routine categorical regression model but ignores known land transformation and potential measurement / ecological error in the recorded abundance categories . instead , we introduce @xmath87 to provide the @xmath88 s and @xmath89 to provide the @xmath90 s . we need a joint distribution to relate the @xmath91 , @xmath92 , and @xmath86 . from a process perspective in terms of the proposed degradation , it seems natural to specify this distribution in the form @xmath93 . since ( @xmath94 ) capture the sequential degradation of an associated categorical abundance distribution , we need to use the same set of @xmath95 s to produce meaningful ( @xmath96 ) respectively . now , we propose ( and clarify ) the following dependence structure . define @xmath97 , where @xmath98 and @xmath99 are the standard normal p.d.f . and c.d.f . respectively . note that @xmath100 so @xmath101 for all @xmath102 . let @xmath103\\[-8pt ] f(z_{t}|z_{p},u ) & \sim & u\delta_{z_{p } } + ( 1-u)\delta_{c(z_{p})},\nonumber\\ f(z_{p}|v,\beta,\tau^2 ) & = & \phi(z_{p};v^t \beta + \theta,1).\nonumber\end{aligned}\ ] ] again , the conditional modeling above is motivated by the degradation perspective . to model the latent @xmath91 surface , we use the covariate information , that is , climate and soil features that are believed to influence the abundance distribution of different species in different ways . we also add a spatial random effect ( @xmath104 ) in the mean function to account for spatial association that may arise from factors , apart from included covariates , that may have a spatial pattern . the covariate effects @xmath105 as well as the spatial random effects @xmath104 are species - specific . variances are fixed at 1 for identifiability ( see section [ subsec35 ] ) . since we are working at areal scale , we assign each cell a single @xmath104 with the prior on @xmath106 specified using a gaussian markov random field ( mrf ) [ @xcite ] with first - order adjacency proximities . see @xcite for details as well as further references . next , the @xmath107 surface is degraded by land transformation . a random location inside @xmath17 is untransformed with probability @xmath108 . then , @xmath109 , that is , a degenerate distribution at @xmath107 given @xmath107 . if it is transformed , the degradation occurs so that the @xmath110 corresponds to the zero abundance category . for simplicity ( with further discussion below ) , we make this a degenerate distribution at @xmath111 , whence @xmath112 becomes a two point distribution as above . again , transformation is equivalent to absence and since @xmath82 is the upper threshold for that classification , we need @xmath113 for a transformed location . when a cell is completely transformed , from equation ( [ eq331 ] ) we have @xmath113 w.p . @xmath8 . for @xmath114 ( complete availability ) @xmath110 and @xmath107 are the same . for any @xmath115 , we get @xmath116 also , since @xmath117 , @xmath118 , which is essential in the sense that transformation can only degrade abundance [ and clarifies our choice for @xmath119 . posterior summaries of @xmath110 measure the prevailing abundance under transformation within the cfr . [ in appendix [ a1 ] we show that @xmath120 . ] the two - point mixture distribution also implies the probability of abundance class @xmath51 , @xmath121 , that is , no matter how large the potential abundance is within a cell , for any @xmath122 there is a positive probability that transformed abundance may fall below @xmath51 at a random location within the cell . other choices for the @xmath123 specification besides a point mass at @xmath124 include putting a point mass at some arbitrary point @xmath125 , or using a truncated normal @xmath123 on @xmath126 . in the first case , it is not ensured whether @xmath127 ( it depends on whether there are cells with @xmath128 ) , while the second choice adds complication for no benefit , is less interpretable , and does not ensure @xmath129 with probability @xmath8 . also , in section [ sec4 ] we show that , in terms of fitting the model , the specification used in equation ( [ eq331 ] ) is the same as using a truncated normal distribution for land transformation . next , we modify the \{@xmath92 } surface to produce \{@xmath86}. with regard to measurement error , the recorded category of abundance at a particular location can be different from the prevailing category due to inaccuracy in field assessment of species quantity . however , we assume that when the potential abundance was zero , one could not record a nonzero abundance category for it [ no false positives , see @xcite in this regard ] . this puts a directional constraint on the effect of noise . a specification for @xmath130 which is coherent with this restriction has , with @xmath131 ( i.e. , a presence ) , @xmath132 . this is a usual measurement error model ( _ mem _ ) specification . for a site with no presence @xmath133 , our assumption says there can not be any measurement error , thus , in equation ( [ eq331 ] ) , for simplicity , we set @xmath134 to be the same as @xmath110 . again , other choices of @xmath135 can be considered for the @xmath136 event , but they will not have any impact on estimation of the @xmath107 surface , as we clarify in section [ sec4 ] . this sequential dependence structure , @xmath137 , implies that if @xmath138 so is @xmath110 and @xmath134 . hence , if a site is not suitable for a species , at no intermediate stage of the model can the site have any positive probability of species occurrence . a change in category between _ actual _ and _ observed _ arises when the noise pushes @xmath92 to the other side of some cut point to produce @xmath86 . and , because of the truncation structure , that shift can not happen from the left of @xmath139 to the right . an alternative way to jointly model @xmath140 could use a bivariate normal distribution with support truncated to @xmath141 $ ] . however , this specification fails to produce an @xmath142 which match our intuition about how the degradation took place . also , from the distributional perspective , the truncated normal redistributes the mass contained inside the left - out region uniformly across the support , whereas the specification in equation ( [ eq331 ] ) shifts the mass only to @xmath143 , which is more in agreement with modeling a data set such as ours where we have an inflated number of reported zero abundances . the simple dependence structure for @xmath123 allows us to marginalize over @xmath110 and work with @xmath107 and @xmath144 as our joint latent distribution . we have @xmath145 rewriting equation ( [ eq332 ] ) in a simpler form , we get @xmath146 + ( 1-u)\delta_{c(z_{p})}.\end{aligned}\ ] ] moreover , equation ( [ eq333 ] ) has a nice interpretation in the sense that , first , it indicates whether the land is transformed or not with probability @xmath147 . if the land is transformed , it sets observed abundance to be @xmath148 . in the case of available land , if there is a potential presence , it allows for a mem around @xmath107 ; in the case of absence , it stays fixed at @xmath107 . since @xmath134 is related to the observed data and @xmath107 is our surface of interest , the marginalization removes one stage of hierarchy from our model fitting and thus reduces correlation , yielding better behaved mcmc in model fitting . furthermore , we can retrieve the @xmath92 surface after the fact since @xmath149 . in the we noted that measurement error and bias typically occur with ecological survey data . it can manifest itself in the form of detection error , spatial coverage bias [ @xcite ] , and under - reporting of absences . how do these biases arise in our modeling ? noteworthy points here are ( i ) the difference between obtaining abundance as actual counts as opposed to through ordinal classifications and ( ii ) what `` no abundance '' means across our collection of grid cells . nondetection bias ( i.e. , undetected individuals in a sampled location ) tends to be discussed more with regard to animal abundance [ @xcite , @xcite , @xcite ] . using counts , evidently observed abundance is at most true abundance ; error can occur in only one direction . with ordinal counts , the bias is still expected to reflect under - reporting but , depending upon the categorical definitions , will be much less frequent and need not be absolutely so . for example , in our data set , plant population size is visually estimated and an observation , especially of large populations , could potentially have error in either direction . in our modeling , `` true '' abundance is not `` potential '' abundance . for us , one could envision true abundance on the latent scale as a `` true '' transformed abundance , say , @xmath150 with measurement error yielding @xmath86 . then , one might insist that our measurement error model requires @xmath151 . under our measurement error formulation using @xmath92 , we even allow @xmath152 to account for potential overestimation of abundance . evidently , since @xmath153 may occasionally be less than the potential classification @xmath154 at that location , we may be slightly underestimating potential abundance . we do nt expect this to be consequential and , in any event , with no knowledge about the incidence of _ under - classification _ in our setting , we have no sensible way to correct for this bias . turning to spatial coverage bias ( i.e. , individuals not exposed to sampling will be missed ) , for us , with only @xmath1 of grid cells sampled , we certainly are subject to this . however , the spatial modeling we introduce helps in this regard . the mean of @xmath87 is @xmath155 regardless of whether we collected any data in @xmath17 . so , the regression is expected to find the appropriate level for the cell and the spatial smoothing associated with the @xmath156 is expected to provide suitable local adjustment . we could argue that , if sampling of grid cells is random , this bias can be ignored . perhaps the most difficult bias to address is the under - sampling of absences . this bias counters the previous ones ; under - sampling of absences will tend to produce over - estimates of potential abundance . in our setting , under - sampling of absences is reflected in the decision - making that leads to only @xmath1 of cells being sampled , that is , it is not a random @xmath1 that have been sampled . different from spatial coverage bias , in this context , the ecologist expresses confidence that the species is not present in some of the unsampled cells . if this is so and we were to set some additional abundances to @xmath51 , this would assert that these `` 0 ' 's are not nondetects and would diminish potential abundance , opposite to the case of nondetects . of course , in the absence of actual data collection , we would not see any of these @xmath51 s and would adopt model - based inference regarding potential abundance for these cells . in any event , with no explicit knowledge of how sampling sites were chosen , we are unable to attempt correction for this bias . possibly , approaches to address the effects of preferential sampling [ @xcite ] could be attempted here . the posterior distributions of interest , @xmath65 and @xmath66 , will be constructed in the post mcmc analysis ( discussed in detail in section [ subsec43 ] ) . from the conditional structure we first write @xmath157 . so the likelihood function for a single sample @xmath78 turns out to be @xmath158 . now @xmath159 can be written as in equation ( [ eq333 ] ) . again , we have @xmath160 cells with @xmath72 sampling sites within @xmath17 . for each @xmath58 we introduce a corresponding @xmath161 and hence a pair of @xmath162 , to represent the event happening at the @xmath35th sampling site within @xmath17 . we work directly with the @xmath144 structure . since we are interested in the areal level abundance distribution and have covariates at areal resolution , we assume for fixed @xmath15 , @xmath87 @xmath163 @xmath164 . it is also assumed that the @xmath79 s are conditionally independent given the @xmath87 s . without loss of generality , re - index cells so that the first @xmath165 of them are sampled and the last @xmath166 are not . the latter cells have no contribution to the @xmath78 column and , hence , no associated @xmath86 appears in the likelihood . using a nonspatial model , we would work with a posterior on the domain of sampled cells only . but assuming a car prior structure with adjacency proximity matrix @xmath167 for the @xmath104 over the whole domain enables us to learn about @xmath107 for unsampled cells . in summary , the posterior distribution takes the following form , up to proportionality , with @xmath168 : @xmath169\\[-8pt ] \pi(\theta ) & = & \pi(\bolds{\alpha})\pi(\bolds{\beta})\pi(\bolds{\theta}),\nonumber \\ \pi(\theta_{1,2,\ldots , i } ) & = & \operatorname{car}(\eta_0,w).\nonumber\end{aligned}\ ] ] we turn to the identifiability of the set of parameters under the hierarchical dependent latent structure . first , with a latent continuous process yielding an ordinal categorical variable , the mean and scale of the distribution can be identified only up to their ratio . in equation ( [ eq331 ] ) the dependence across @xmath170 is specified through conditional means . hence , all gaussian distributions there are specified with standard deviation @xmath8 . four categories of abundance allow three free probabilities and the corresponding four latent surfaces will also have @xmath171 degrees of freedom . as noted above , we set @xmath172 with @xmath173 as free parameters . we also need to ensure that all three @xmath174 surfaces can be distinguished from each other . since transformation percentage @xmath147 is given a priori , it is straightforward to separate @xmath107 and @xmath110 . we turn to the joint distribution for @xmath175 given as @xmath176 . with fixed variances and no constraint on measurement error , there would be no need to bring in @xmath107 ; it is redundant , there is no way to distinguish between @xmath107 and @xmath134 , and one can use the marginal @xmath177 . now the constraint comes into the picture ; it makes the @xmath134 surface non - gaussian though the @xmath107 surface is . the greater the measurement error , the more departure from gaussianity in the marginal distribution of @xmath134 . again , the measurement error can not be estimated on any absolute scale , since the latent @xmath174 scales are fixed for identifiability . it will be controlled by parameters like @xmath105 and @xmath104 . to compare the relative effect of measurement error across different species , under fixed scale parameters , @xmath178 is a candidate but other model features can be informative as well . finally , the full model specification , described in equation ( [ eq331 ] ) , can be represented through a graphical model , shown in figure [ fig4 ] . within cell @xmath17 . @xmath174 s denote latent abundance processes , observed ( o ) , transformed ( t ) , and potential ( p ) ; @xmath78 s denote interval - censored abundances , observed ( o ) , and potential ( p ) ; @xmath179 is proportion of land untransformed , @xmath77 s are covariates , @xmath105 s are regression coefficients , @xmath95 s are cut points for @xmath174 scale , and @xmath104 s are spatial random effects . ] here , we describe how to design a computationally efficient mcmc algorithm for the model . we then discuss how to summarize the posterior samples to estimate important model features . introduction of latent layers , although increasing the parameter dimension in the model , makes the posterior full conditionals standard and easy to sample from . our goal is to efficiently estimate components of @xmath180 which control potential abundance . we rewrite equation ( [ eq333 ] ) as follows : @xmath1811_{z_{p } \leq0 } \nonumber\\[-8pt]\\[-8pt ] & & { } + ( 1-u)\delta_{c(z_{p})}1_{z_p>0},\nonumber\end{aligned}\ ] ] and work with equation ( [ eq411 ] ) to implement the computation for the model fitting . we start with updating all @xmath182 using @xmath183 and then drawing components of @xmath180 from their respective posterior full conditionals based on @xmath184 @xmath185 . given the draw from @xmath91 , sampling the components of @xmath180 is standard as in almost any spatial regression analysis ( see appendix [ a3 ] ) . for the set of @xmath104 s , after sampling them sequentially , we need to `` center them on the fly '' [ @xcite , @xcite ] . the more challenging part is to update @xmath186 . in @xcite the latent variables were sampled in the mcmc from mutually independent truncated gaussian full conditionals , with the support determined by the corresponding classification . for our model , the posterior full conditional for any @xmath86 is @xmath187 we take two different strategies to update @xmath188 depending on the observed @xmath78 . for any site with nonzero @xmath78 we have ( with @xmath139 ) @xmath189 , which amounts to sampling first a univariate normal @xmath190 truncated within @xmath191 and then from @xmath192 @xmath193 which is also gaussian [ with location @xmath194 and scale @xmath195 where @xmath196 $ ] . for a site with observed @xmath197 the case is more complicated , with details provided in appendix [ a2 ] . all of the sampling distributions required in mcmc are listed in appendix [ a3 ] . the algorithm described above is computationally demanding as we have two latent variables to sample at each sampling site and one spatial parameter for each of the grid cells . however , since @xmath198 are independent across cells given @xmath180 , we can update them all at once . the problematic part is sampling the spatial effects , with approximately 37,000 grid cells . to handle this issue , we used a parallelization method where @xmath68 is divided into disjoint and exhaustive subregions @xmath199 along with a resultant set of _ boundary _ cells @xmath200 arising through the car proximity matrix . thus , once @xmath201 is updated conditional on the rest , then @xmath202 given @xmath201 can be updated in parallel . this algorithm is illustrated in figure [ fig5 ] , where we have a @xmath203 rectangular region with an adjacency structure @xmath167 which puts weight only on the cells sharing a common boundary . sequential updating would have required @xmath204 steps . we constructed a set of @xmath205 boundary cells @xmath200 ( the dark cells in figure [ fig5 ] ) . it divides the rectangle into @xmath206 segments of @xmath207 cells each , so that conditional on @xmath208 , those segments can be updated independently of each other so we need only @xmath209 updating steps . this is only an illustrative example , but for large regions , this may significantly improve the run time . however , the time required for communication and data assimilation is an issue for this parallelization method . with increasing @xmath210 , although the time required for the sequential updating within each @xmath211 goes down , the size of @xmath200 increases as does the amount of communication required within the parallel architecture . so a trade - off must be determined for choosing @xmath210 ; in our setting @xmath212 worked well . steps in each iteration . by dividing the rectangle into @xmath206 segments of @xmath207 cells each with @xmath205 boundary cells ( shown in dark grey ) , each segment can be updated independently ( conditional on the boundaries ) . in this example the parallelization results in only @xmath209 updating steps . ] there are several ways to summarize inference about the @xmath65 and @xmath66 distributions . according to our model , for @xmath17 , @xmath213 . posterior samples of @xmath214 enable us to compute samples of the @xmath215 . a posterior sample of @xmath216 can be constructed using the relation @xmath217 . additionally , we can calculate the mean as well as the uncertainty from these samples , enabling maps for transformed abundance ( @xmath66 ) and potential ( @xmath65 ) abundance . for each of @xmath215 and @xmath216 , we have @xmath218 submaps , one for each abundance category . this is useful in terms of assessing high and low abundance regions for the species . the @xmath105 s provide ( up to fixed scale ) the effect of a particular climate or soil - related factor on the abundance of a particular species . comparison of the @xmath65 and @xmath66 maps informs about the effect of land transformation . one may also be interested in capturing @xmath65 or @xmath66 through a single summary feature rather than all @xmath218 categorical probabilities . grouped mean abundance ( expectation with respect to the @xmath65 or @xmath66 distribution ) can be used with suitable categorical midpoints . we note that the posterior inference can also be summarized on the latent scale using posterior samples of the @xmath174 s . however , working on the @xmath174 scale can only provide relative comparison . @ld2.5d2.5d2.5d2.5d2.5d2.5@ & & & & & & + prpunc & 1.2275 & -0.9436 & -0.8248 & 0.2439 & 0.1834 & 0.0306 + & ( 0.3809 ) & ( 0.2768 ) & ( 0.1143 ) & ( 0.1158 ) & ( 0.2006 ) & ( 0.1089 ) + prrepe & 0.6825 & -0.4512 & -0.0864 & 0.1753 & -0.2958 & 0.0566 + & ( 0.1710 ) & ( 0.1179 ) & ( 0.0612 ) & ( 0.0673 ) & ( 0.0996 ) & ( 0.0455 ) + we have implemented the described model on abundance data for several different plant species over the whole _ cfr_. we centered and scaled all the @xmath77 s before using them in the model . as priors we used @xmath219 , @xmath220 with large @xmath221 . for @xmath104 , we used @xmath222 and @xmath167 to be a binary matrix with @xmath223 _ iff _ @xmath224 . the threshold @xmath225 was used to provide an @xmath226 nearest neighbor structure for most of the cells . however , for boundary cells , the number of neighbors varies from @xmath171 to @xmath207 . the parallelization algorithm was implemented inside r ( http://www.r-project.org ) using @xmath227 . the run time for an individual species was about @xmath228 iterations@xmath229day . the outputs presented below are created by first running @xmath230 iterations of mcmc , discarding the initial @xmath231 samples , and thinning the rest at every fifth sample . the @xmath105 s were quick to converge , but the @xmath95 sequences were highly autocorrelated and moved more slowly in the space . here we consider two species , _ protea punctata _ ( prpunc ) and _ protea repens _ ( prrepe ) . a summary of the model output is presented through the following table and diagrams . table [ table1 ] provides the mean covariate effects for both species along with the @xmath232% equal tail credible interval width ( in parentheses ) . considering @xmath232% equal tail credible interval , all the covariate effects are significant _ except _ fert1 for _ p. punctata_. ) for _ protea punctata _ ( prpunc ) and _ protea repens _ ( prrepe ) . these effects offer local adjustment to potential abundance . cells with values greater than zero represent regions with larger than expected populations , conditional on the other covariates . ] means the probability the species is absent , while classes @xmath8@xmath171 indicate estimated abundance from @xmath8@xmath233 , @xmath234@xmath235 , @xmath236 individuals , respectively . ] means the probability the species is absent , while classes @xmath8@xmath171 indicate estimated abundance from @xmath8@xmath233 , @xmath234@xmath235 , @xmath237 individuals , respectively . ] -scale , `` mean @xmath238 $ ] '' refers to the potential abundance , while `` mean @xmath239 $ ] '' refers to the potential abundance corrected for habitat transformation . the _ grouped mean abundance _ rescales the mean @xmath238 $ ] surface to the expected potential size of a population in a grid cell ( using the observed abundance classes : absent , @xmath8@xmath233 , @xmath233@xmath235 , @xmath236 ) . the _ group mean transformed abundance _ shows the expected size of a population after correcting for habitat transformation . ] -scale , `` mean @xmath238 $ ] '' refers to the potential abundance , while `` mean @xmath239 $ ] '' refers to the potential abundance corrected for habitat transformation . the _ grouped mean abundance _ rescales the mean @xmath238 $ ] surface to the expected potential size of a population in a grid cell ( using the observed abundance classes : absent , @xmath8@xmath233 , @xmath233@xmath235 , @xmath236 ) . the _ group mean transformed abundance _ shows the expected size of a population after correcting for habitat transformation . ] [ fig10 ] the mean posterior spatial effects are shown in figure [ fig6 ] . note that the spatial effects for the two species have quite different patterns , with _ protea repens _ having a region of low values in the northeast and larger values elsewhere , while protea punctata is more even across the landscape , but with lower values toward the edges of the _ cfr_. these surfaces capture the spatial variability in abundance that is not explained by the other covariates within the model . this suggests that the covariates predict higher abundance of _ p. repens _ in the northwest than what was observed , perhaps indicating some unobserved limiting factor ( such as unsuitable soils , more extreme seasonality in rainfall , or dispersal limitations ) . similarly for _ p. punctata _ , the covariates may over - predict abundances at the edges of the cfr where many environmental factors change as one transitions to other biome types . figures [ fig7 ] and [ fig8 ] show the mean posterior abundance category probabilities ( potential and transformed ) for _ p. punctata _ ( figure [ fig7 ] ) and _ p. repens _ ( figure [ fig8 ] ) . comparing these plots among rows contrasts the probabilities associated with each abundance class for the species , while comparing between columns shows the effects of landscape transformations on abundance class probabilities . both species show higher predicted abundances coinciding with mountainous areas of the _ cfr_. this is where the fynbos biome dominates the landscape and where proteas are characteristically the dominant , indicator species [ @xcite ] . note that _ p. punctata _ , a less common species , is only slightly affected by landscape transformation , while there are dramatic differences for _ p. repens _ ( figures [ fig9 ] and [ fig10 ] ) . this is because _ p. punctata _ is mostly limited to dry , rocky , or shale slopes [ @xcite ] which are less suitable for agriculture or development and thus mostly untransformed . _ p. repens _ , on the other hand , is much more ubiquitous across the region and can frequently occur in lowland areas that have been largely transformed by human activities [ @xcite , @xcite ] . it is also useful to summarize these data through mean potential abundance and mean transformed abundance ( see section [ subsec43 ] ) as in figures [ fig9 ] and [ fig10 ] . these figures allow inspection of the underlying latent surfaces that are of interest to ecologists as a continuous relative representation of species abundances . however , the latent `` @xmath174 '' scales may be difficult to interpret ecologically and , thus , estimated potential and transformed abundance ( using the grouped mean ) are also shown . these represent the expected abundance ( with respect to the @xmath65 s or @xmath66 s ) of a species at a randomly selected sample location in that grid . the associated display makes it easy to visualize the effects of habitat transformation on protea populations . _ p. punctata _ shows almost no effects of landscape transformation , while large differences are apparent for _ p. repens . _ note the large transformed regions in the south and west where the expected abundance of plants has dropped from more than 50 to near zero . it is also apparent that , across the landscape , _ p. punctata _ tends to have a higher expected mean abundance at any given sample point than does _ p. repens _ ( figures [ fig9 ] and [ fig10 ] ) . building on previous efforts that have addressed the presence / absence of species , we have presented a modeling framework for learning about potential patterns for species abundance not degraded by land transformation and potential measurement error . the model was built using a hierarchical latent abundance specification , incorporating spatial structure to capture anticipated association between adjacent locations . along with potential pattern , we also have an estimate of transformed abundance pattern . comparison of these two patterns is helpful for understanding the effect of land transformation on species presence and abundance and , in particular , for disentangling these effects from those of other environmental factors . this may facilitate designing strategies for species conservation as well as understanding the overall effects of climate change . this work has applications in biogeography and in conservation biology [ @xcite , @xcite ] . we can now develop predictive maps of `` high quality '' habitat sites within a species range , based on high predicted abundances . this will help identify prime locations for effective conservation efforts . we can also estimate the impact of habitat transformation on the size of the population using the information from figures [ fig8 ] and [ fig10 ] , and thus identify threats to conservation . predictive abundance maps will also be useful to explore patterns in biodiversity and species abundances . do species abundances tend to peak in the middle of the species range [ @xcite ] ? do areas of high biodiversity tend to have lower species abundances ? are there areas that are rich in both abundance and biodiversity ( perhaps identifying ideal regions for conservation efforts ) ? there are several natural extensions . one is to study the temporal change in abundance . with abundance data collected over time as well as associated environmental factors such as rainfall and temperature , dynamic modeling of species abundance with changing environmental factors may give a clearer picture of how a species is responding to climate change . indeed , when connected to future climate scenarios , we may attempt to forecast prospective species abundance . similarly , if the transformation data is also time varying , we could illuminate the effect of land transformation in greater detail . the current model uses transformation percentage @xmath240 in a deterministic way ( transformation having a binary effect on potential abundance ) . in other cases ( e.g. , to study abundance pattern of animals ) it may be reasonable to treat transformation as another covariate influencing species habitat . also , it may be imagined that the relationship between potential abundance and environmental variables is not linear as specified in equation ( [ eq331 ] ) , for example , environmental variables may affect larger abundance classes differently from smaller abundance classes ; piecewise linear specification , introducing different regression coefficients over the different abundance classes , could be explored . another possible extension lies in joint modeling of two or more species . one may wish to learn whether two plant varieties are sympatric or allopatric and whether or not there is evidence for competitive interactions or facilitation . such modeling can be done by extending our model to have multiple @xmath241 surfaces , where @xmath18 is the species indicator . dependence can be introduced across @xmath242 surfaces by modeling @xmath243 using an mcar [ @xcite , @xcite ] . fitting such models will be very challenging if there are many grid cells . instead of taking an areal level approach , if covariate information is available at point level ( where sampling sites are viewed as `` points '' within the large region , @xmath68 ) , one may consider a point - level model . this amounts to replacing the _ car _ model with a gaussian process prior for the spatial effects . with many sampling sites , we will need to use appropriate approximation techniques [ @xcite ] . @xmath245 . assuming @xmath246 , it is enough to show @xmath247 . consider the quantity @xmath248 , if @xmath249 , it goes to @xmath51 . when @xmath250 , we have @xmath251 so @xmath252 , thus , we can get @xmath253 , such that @xmath254 for all @xmath255 . hence , the result follows . we subdivide by considering the ways that we can generate a @xmath51 realization of @xmath78 based on equation ( [ eq333 ] ) [ one may also use equation ( [ eq331 ] ) to do this ] : a. the area is untransformed , the species was potentially there , but missed during data collection or it was absent at that time instance ; the event is @xmath256 with prior probability @xmath257 . b. potentially the species was absent there ; the event is @xmath258 with prior probability @xmath259 . c. the species was potentially there @xmath260 , but the area was transformed ; the event has prior probability @xmath261 these three events are exhaustive and mutually exclusive for the event @xmath262 . thus , @xmath263 is a @xmath171-component mixture . to draw a ( @xmath198 ) pair from this distribution amounts to first choosing a component and then drawing a pair ( @xmath198 ) from that component distribution . by bayes rule , conditional on observed @xmath262 , these three cases can happen with posterior probability @xmath264 . so we use a multinomial to select which of these events took place . before going into case by case details , it is worth mentioning that in all these cases the sampling from the joint density of chosen mixture component was done via the marginal @xmath265 followed by @xmath266 . the advantage of this scheme is that we do nt need to draw from the latter because @xmath134 s corresponding to @xmath197 are not involved in posterior full conditionals of any other parameters in the model ( as @xmath267 , fixed ) . if the second case is selected , then @xmath2681_{z_{o } \leq0}1_{z_{p } \leq0}\phi ( z_p)$ ] and thus marginalizing over @xmath86 , we get @xmath269 which is a truncated gaussian on @xmath270 . similarly under case ( iii ) , we need to simulate @xmath107 from @xmath271 , a gaussian truncated on @xmath272 . in case ( i ) , @xmath273 , so marginalizing over @xmath86 we get @xmath274 . an efficient way to draw from this density is to propose a @xmath107 from a truncated normal on @xmath272 and do a metropolis hastings update with an independent proposal , using the quantity @xmath275 . however , all sampling distributions are summarized in appendix [ a3 ] below . * if @xmath276 , draw @xmath277 . draw @xmath278 . * if @xmath279 , compute @xmath280\times [ - \infty,0 ] ; \mu_{ij},\sigma_0),1-\phi(v_i^t\beta+ \theta),(1-u)\phi(v_i^t\beta+ \theta_i))$ ] , where @xmath281 and @xmath282 are the location and dispersion parameters for bivariate normal joint prior distribution of @xmath283 . draw @xmath284 . if @xmath285 , propose @xmath286 and do a metropolis hastings sampler using @xmath275 . else if @xmath287 , draw @xmath288 , else draw @xmath289 . * draw @xmath290 , @xmath291 . * draw @xmath292 @xmath293 . * draw @xmath294 @xmath295 for @xmath296 . draw @xmath294 @xmath297 for @xmath298 . the authors thank guy midgley and anthony rebelo for useful discussions . busby , j. r. ( 1991 ) . bioclim : a bioclimatic analysis and predictive system . in _ nature conservation : cost effective biological surveys and data analysis _ ( c. r. margules and m. p. austin , eds . ) 6468 . csiro , canberra , australia . conroy , m. j. , runge , j. p. , barker , r. j. , schofield , m. r. and fonnesbeck , c. j. ( 2008 ) . efficient estimation of abundance for patchily distributed populations via two - phase , adaptive sampling . _ ecology _ * 89 * 33623370 . cressie , n. , calder , c. a. , clark , j. s. , ver hoef , j. m. and wikle , c. k. ( 2009 ) . accounting for uncertainty in ecological analysis : the strengths and limitations of hierarchical statistical modeling . _ ecological applications _ * 19 * 553570 . elith , j. , graham , c. h. , anderson , r. p. , dudk , m. , ferrier , s. , guisan , a. , hijmans , r. j. , huettmann , f. , leathwick , j. r. , lehmann , a. , li , j. , lohmann , l. g. , loiselle , b. a. , manion , g. , moritz , c. , nakamura , m. , nakazawa , y. , overton , j. mcc . , peterson , a. t. , phillips , s. j. , richardson , k. s. , scachetti - pereira , r. , schapire , r. e. , sobern , j. , williams , s. , wisz , m. s. and zimmermann , n. e. ( 2006 ) . novel methods improve prediction of species distributions from occurrence data . _ ecography _ * 29 * 129151 . fitzpatrick , m. c. , gove , a. d. , nathan , j. , sanders , n. j. and dunn , r. r. ( 2008 ) . climate change , plant migration , and range collapse in a global biodiversity hotspot : the banksia ( proteaceae ) of western australia . _ global change biology _ * 14 * 13371352 . gelfand , a. e. , silander , j. a. , jr . , wu , s. , latimer , a. m. , lewis , p. , rebelo , a. g. and holder , m. ( 2005a ) . explaining species distribution patterns through hierarchical modeling . _ bayesian anal . _ * 1 * 4292 . gelfand , a. e. , schmidt , a. m. , wu , s. , silander , j. a. , jr . , latimer , a. m. and rebelo , a. g. ( 2005b ) . modelling species diversity through species level hierarchical modeling . c _ * 54 * 120 . guisan , a. , lehman , a. , ferrier , s. , austin , m. p. , overton , j. m. c. , aspinall , r. and hastie , t. ( 2006 ) . making better biogeographical predictions of species distributions . _ journal of applied ecology _ * 43 * 386392 . ibez , i. , silander , j. a. , jr . , allen , j. m. , treanor , s. and wilson , a. ( 2009 ) . identifying hotspots for plant invasions and forecasting focal points of further spread . _ journal of applied ecology _ * 46 * 12191228 . le loch , g. and galli , a. ( 1997 ) . truncated plurigaussian method : theoretical and practical points of view . in _ geostatistics wollongong96 _ ( e. y. baafi and n. a. schofield , eds . ) * 1 * 211222 . kluwer , dordrecht , the netherlands . raxworthy , c. j. , martinez - meyer , e. , horning , n. , nussbaum , r. a. , schneider , g. e. , ortega - huerta , m. a. and peterson , a. t. ( 2003 ) . predicting distributions of known and unknown reptile species in madagascar . _ nature _ * 426 * 837841 . rebelo , a. g. ( 2002 ) . the state of plants in the cape flora . in _ proceedings of a conference held at the rosebank hotel in johannesburg _ ( g. h. verdoorn and j. le roux , eds . ) * 18*. the state of south africa s species , endangered wildlife trust . rebelo , a. g. , boucher , c. , helme , n. , mucina , l. and rutherford , m. c. ( 2006 ) . fynbos biome . in _ the vegetation of south africa , lesotho and swaziland . streltzia _ ( l. micina and m. c. rutherford , eds . ) * 19*. south african national biodiversity institute , pretoria , south africa . rouget , m. , richardson , d. m. , cowling , r. m. , lloyd , j. w. and lombard , a. t. ( 2003 ) . current patterns of habitat transformation and future threats to biodiversity in terrestrial ecosystems of the cape floristic region , south africa . _ biological conservation _ * 112 * 6383 . ver hoef , j. m. , cressie , n. , fisher , r. n. and case , t. j. ( 2001 ) . uncertainty and spatial linear models for ecological data . in _ spatial uncertainty in ecology _ ( c. t. hunsaker , m. f. goodchild , m. a. friedl and t. j. case , eds . ) 214237 . springer , new york . wisz , m. s. , hijmans , r. j. , li , j. , peterson , a. t. , graham , c. h. and guisan , a. ( 2008 ) . effects of sample size on the performance of species distribution models . _ diversity and distributions _ * 14 * 763773 .

for any finite graph and any field of characteristic unequal to we construct an algebraic variety over whose-points parameterise-lie algebras generated by extremal elements , corresponding to the vertices of the graph , with prescribed commutation relations , corresponding to the non - edges . after that , we study the case where is a connected , simply laced dynkin diagram of finite or affine type . we prove that is then an affine space , and that all points in an open dense subset of parameterise lie algebras isomorphic to a single fixed lie algebra . if is of affine type , then this fixed lie algebra is the split finite - dimensional simple lie algebra corresponding to the associated finite - type dynkin diagram . this gives a new construction of these lie algebras , in which they come together with interesting degenerations , corresponding to points outside the open dense subset . our results may prove useful for recognising these lie algebras .

0707.2927

an _ extremal element _ of a lie algebra @xmath4 over a field @xmath1 of characteristic unequal to @xmath2 is an element @xmath5 for which @xmath6\subseteq kx$ ] . a _ sandwich element _ is an @xmath7 satisfying the stronger condition @xmath6=0 $ ] . the definition of extremal elements in characteristic @xmath2 is more involved , which is one reason for restricting ourselves to characteristics unequal to @xmath2 here . extremal elements and sandwich elements play important roles in both classical and modern lie algebra theory . in complex simple lie algebras , or their split analogues over other fields , extremal elements are precisely the elements that are long - root vectors relative to some maximal torus . sandwich elements are used in the classification of simple lie algebras in small characteristics @xcite ; they occur in the modular lie algebras of cartan type , such as the witt algebras . sandwich elements were originally introduced in relation with the restricted burnside problem @xcite . an important insight for the resolution of this problem is the fact that a lie algebra generated by finitely many sandwich elements is necessarily finite - dimensional . while this fact was first only proved under extra assumptions , in @xcite it is proved in full generality . we will use this result in what follows . the prominence of extremal elements in the work of kostrikin and zelmanov and in modular lie algebra theory led to the natural problem of describing all lie algebras generated by a fixed number of extremal elements @xcite . example suppose that we want to describe all lie algebras @xmath4 generated by two extremal elements @xmath8 and @xmath9 . since @xmath10 $ ] is a scalar multiple @xmath11 of @xmath8 and @xmath12=-[y,[y , x]]$ ] is a scalar multiple @xmath13 of @xmath9 , @xmath14 is spanned by @xmath15 $ ] . there may be linear dependencies between these elements , but let us assume that they are linearly independent . then @xmath16=[y,[x,[x , y]]]=[[y , x],[x , y]]+[x,[y,[x , y]]]=0-b[x , y]=b[y , x],\ ] ] and since we have assumed that @xmath17 \neq 0 $ ] , we find that @xmath18 . hence three - dimensional lie algebras with a distinguished pair of extremal generators are parameterised by the single number @xmath19 . moreover , all algebras with @xmath20 are mutually isomorphic and isomorphic to the split simple lie algebra of type @xmath21 , while the algebra with @xmath22 is nilpotent and isomorphic to the three - dimensional heisenberg algebra . this is a prototypical example of our results . the next smallest case of three generators is treated in @xcite and also by our results below . there the generic lie algebra is split of type @xmath23 and more interesting degenerations exist . we now generalise and formalise this example to the case of more generators , where we also allow for the flexibility of prescribing that certain generators commute . thus let @xmath0 be a finite simple graph without loops or multiple edges . let @xmath24 be the vertex set of @xmath0 and denote the neighbour relation by @xmath25 . fixing a field @xmath1 of characteristic unequal to @xmath2 , we denote by @xmath26 the quotient of the free lie algebra over @xmath1 generated by @xmath24 modulo the relations @xmath27=0 \text { for all } x , y \in \pi \text { with } x\not \sim y.\ ] ] so @xmath26 depends both on @xmath0 and on @xmath1 , but we will not make this dependence explicit in the notation . we write @xmath28 for the space of all @xmath1-linear functions @xmath29 . for every @xmath30 , also written @xmath31 , we denote by @xmath32 the quotient of @xmath26 by the ideal @xmath33 generated by the ( infinitely many ) elements @xmath34-f_x(y)x,\ x \in \pi , y \in { \mathcal{f}}.\ ] ] by construction @xmath32 is a lie algebra generated by extremal elements , corresponding to the vertices of @xmath0 , which commute when they are not connected in @xmath0 . the element @xmath35 is a parameter needed to express the extremality of @xmath36 . if @xmath0 is not connected , then both @xmath26 and @xmath32 naturally split into direct sums over all connected components of @xmath0 , so it is no restriction to assume that @xmath0 is connected ; we will do so throughout this paper . in the lie algebra @xmath37 the elements of @xmath24 map to sandwich elements , hence by @xcite this lie algebra is finite - dimensional . for general @xmath30 it turns out that @xmath38 ; see @xcite or the proof of theorem [ thm : variety ] below . it is therefore natural to focus on the lie algebras @xmath32 of the maximal possible dimension @xmath39 . this leads us to define the set @xmath40 the parameter space for all maximal - dimensional lie algebras of the form @xmath32 . example in the two - generator case above @xmath0 is the graph with two vertices joined by an edge . the sandwich algebra @xmath37 is the three - dimensional heisenberg algebra , and the condition that @xmath41 corresponds to our assumption above that @xmath15 $ ] be linearly independent . this linear independence forced the parameters @xmath19 and @xmath42 to be equal . here @xmath3 is the affine line with coordinate @xmath19 . all lie algebras corresponding to points @xmath20 are mutually isomorphic . our first main result is that @xmath3 carries a natural structure of an affine algebraic variety . to specify this structure we note that @xmath43 is a homogeneous ideal relative to the natural @xmath44-grading that @xmath26 inherits from the free lie algebra generated by @xmath24 . [ thm : variety ] the set @xmath3 is naturally the set of @xmath1-rational points of an affine variety of finite type defined over @xmath1 . this variety can be described as follows . fix any finite - dimensional homogeneous subspace @xmath45 of @xmath26 such that @xmath46 . then the restriction map @xmath47 maps @xmath3 injectively onto the set of @xmath1-rational points of a closed subvariety of @xmath48 . this yields a @xmath1-variety structure on @xmath3 which is independent of the choice of @xmath45 . we prove this theorem in section [ sec : moduli ] . in section [ sec : sandwich ] we first derive some relations between the sandwich algebra @xmath37 and the positive part of the _ complex _ kac - moody algebra of type @xmath0 . then we determine @xmath37 explicitly in the case where @xmath0 is a simply laced dynkin diagram of finite or affine type ; by this we mean any of the diagrams in figure [ fig : diagrams ] without or with vertex @xmath49 , respectively . see theorems [ thm : finite ] and [ thm : affine ] . in section [ sec : parameter ] we study the variety @xmath3 . after some observations for general @xmath0 , we again specialise to the diagrams of figure [ fig : diagrams ] . for these we prove that @xmath3 is an affine space , and that for @xmath50 in an open dense subset of @xmath3 the lie algebra @xmath32 is isomorphic to a fixed lie algebra ; see theorems [ thm : genericfinite ] and [ thm : genericaffine ] . the latter of these theorems is our second main result , and we paraphrase it here . let @xmath0 be any of the simply laced dynkin diagrams of affine type in figure [ fig : diagrams ] , let @xmath51 be the finite - type diagram obtained by removing vertex @xmath49 from @xmath0 , and let @xmath52 be the edge set of @xmath0 . then @xmath3 is isomorphic to the affine space of dimension @xmath53 over @xmath1 , and for @xmath50 in an open dense subset of @xmath3 the lie algebra @xmath32 is isomorphic to the chevalley algebra of type @xmath51 . by the _ chevalley algebra of type @xmath51 _ we mean the lie algebra obtained by tensoring a certain @xmath54-form of the complex simple lie algebra of type @xmath51 with the field @xmath1 ; see subsection [ ssec : finitesandwich ] for details . this lie algebra is often simple , but not always ; see ( * ? ? ? * chapter 4 ) and @xcite . we conclude the paper with remarks on applications and related work in section [ sec : notes ] . recall the notation from section [ sec : introduction ] : @xmath0 is a connected finite graph without loops or multiple edges , @xmath1 is a field of characteristic unequal to @xmath2 , and @xmath3 is the set of all @xmath55 such that @xmath32 has the maximal possible dimension , namely that of @xmath37 . to avoid formulas with many lie brackets , we write @xmath56 for the expression @xmath57\ldots]]$ ] . such an element is called a _ monomial _ in the @xmath58 of _ degree _ @xmath59 . in the proof of theorem [ thm : variety ] we use the @xmath44-grading @xmath60 of @xmath26 where @xmath61 is the span of all monomials of degree @xmath59 in the elements of @xmath24 . we also use the following terminology : a subspace @xmath45 of @xmath26 is called _ homogeneous _ if it equals @xmath62 . a single element of @xmath26 is _ homogeneous _ if it lies in some @xmath61 . to any subspace @xmath45 of @xmath26 we associate the homogeneous subspace @xmath63 of @xmath26 spanned by all @xmath64 as @xmath65 runs through @xmath45 . if @xmath45 is an ideal then so is @xmath63 . let @xmath45 be a finite - dimensional homogeneous subspace of @xmath26 such that @xmath66 ; such a subspace exists as @xmath37 is finite - dimensional @xcite and @xmath43 is homogeneous . note that the theorem only requires that @xmath67 ; we will argue later why this suffices . observe that @xmath45 contains the image of @xmath24 in @xmath14 : the abelian lie algebra spanned by @xmath24 is clearly a quotient of @xmath37 , so the component of @xmath43 in degree @xmath68 is trivial . from the shape of the generators it is clear that the homogeneous ideal @xmath69 associated to @xmath33 contains @xmath43 , so that @xmath70 for _ all _ @xmath50 , and @xmath71 if and only if @xmath72 . we will argue that the map @xmath73 is injective , and that its image is a closed subvariety of @xmath48 . for each @xmath74 let @xmath75 be the projection onto @xmath45 along @xmath76 . we prove two slightly technical statements : first , for all @xmath77 there exists a polynomial map @xmath78 such that @xmath79 and second , for @xmath36 and @xmath77 there exists a polynomial @xmath80 such that @xmath81 we proceed by induction on the degree of @xmath82 : assume that both statements are true in all degrees less than @xmath59 , and write @xmath83 where @xmath84 has degree less than @xmath59 , @xmath85 , and @xmath86 . then @xmath87 can be written as a sum of terms of the form @xmath88 with @xmath89 and @xmath90 of degree @xmath91 . modulo @xmath33 for @xmath74 this term is equal to @xmath92 where we used the induction hypothesis for @xmath90 and @xmath93 . hence a @xmath94 of the form @xmath95 has the required property . similarly , for @xmath36 and @xmath74 we have @xmath96 and since @xmath97 we conclude that @xmath98 hence we may define @xmath99 by @xmath100 this shows the existence of @xmath94 and @xmath99 . the injectivity of @xmath101 is now immediate : any @xmath74 is determined by its restriction to @xmath45 by @xmath102 . we now show that @xmath103 is closed . for any tuple @xmath104 one may try to define a lie algebra structure on @xmath45 by setting @xmath105_h:=p_{[u , v]}(h ) , \quad u , v \in v.\ ] ] by construction , if @xmath106 for some @xmath74 , then this turns @xmath45 into a lie algebra isomorphic to @xmath32 . moreover , in this case the lie bracket has the following two properties : 1 . if @xmath107 is expressed as a linear combination @xmath108 of monomials in the elements of @xmath24 , where the lie bracket is taken in @xmath26 , then the expression @xmath109_h\ldots]_h]_h$ ] also equals @xmath110 ; and [ it : gen ] 2 . @xmath111_h]_h = q_{x , u}(h ) x$ ] for all @xmath112 . [ it : ext ] conversely , suppose that @xmath113_h$ ] indeed defines a lie algebra on @xmath45 satisfying and . then @xmath114_h)$ ] is a lie algebra of dimension @xmath39 that by is generated by the image of @xmath24 , and by this image consists of extremal elements . hence there exists an @xmath74 corresponding to this lie algebra , and its restriction to @xmath45 is @xmath115indeed , @xmath116 is the coefficient of @xmath8 in @xmath117_h]_h$ ] , which is @xmath118 for @xmath119 . finally , all stated conditions on @xmath115the fact that @xmath113_h$ ] satisfies the jacobi identity and anti - commutativity , together with and are closed ; here we use the polynomiality of @xmath94 and @xmath99 . this proves that @xmath103 is closed . now if @xmath120 is any homogeneous subspace containing @xmath45 , then the restriction map @xmath121 is clearly also injective . moreover , an @xmath122 lies in the image of this map if and only if @xmath123 lies in @xmath124 and @xmath125 for all @xmath126 . thus @xmath127 is closed and the maps @xmath128 and @xmath129 with @xmath130 are inverse morphisms between @xmath124 and @xmath127 . similarly , if @xmath131 is any other homogeneous vector space complement of @xmath43 contained in @xmath120 , then the restriction map @xmath132 induces an isomorphism between the images of @xmath3 in these spaces . this shows that the variety structure of @xmath3 does not depend on the choice of @xmath45 . finally , all morphisms indicated here are defined over @xmath1 . we conclude that we have a @xmath1-variety structure on @xmath3 which is independent of the choice of @xmath45 . the type of reasoning in this proof will return in section [ sec : parameter ] : in the case where @xmath0 is a dynkin diagram we will show that for @xmath74 the restriction @xmath133 actually depends polynomially on even fewer values of the @xmath35 , thus embedding @xmath3 into smaller affine spaces . that these embeddings are closed can be proved exactly as we did above . for now , @xmath0 is an arbitrary finite graph ( not necessarily a dynkin diagram ) . the lie algebra @xmath134 is the so - called _ sandwich algebra _ corresponding to @xmath0 . it is a finite - dimensional nilpotent lie algebra , and carries an @xmath135-grading defined as follows . the _ weight _ of a word @xmath136 over @xmath24 is the element @xmath137 whose @xmath8-coordinate equals @xmath138 for all @xmath36 . for such a word the corresponding monomial @xmath139 lives in the free lie algebra on @xmath24 , but we use the same notation for its images in @xmath140 and @xmath141 when this does not lead to any confusion . we will sometimes say that a monomial @xmath142 has weight @xmath143 ; then we mean that the word @xmath136 has weight @xmath143the monomial @xmath56 itself might be @xmath49 . now the free lie algebra is graded by weight , and this grading refines the grading by degree . like the grading by degree , the grading by weight is inherited by @xmath134 as all relations defining @xmath134 are monomials . we write @xmath144 for the space of weight @xmath145 and call @xmath146 the _ multiplicity _ of @xmath143 . for @xmath36 let @xmath147 be the element with a @xmath68 on position @xmath8 and zeroes elsewhere ; that is , @xmath147 is the weight of the word @xmath148 . we define a symmetric @xmath54-bilinear form @xmath149 on @xmath150 by its values on the standard basis : for @xmath151 we set @xmath152 the matrix @xmath153 is called the _ cartan matrix _ of @xmath0 . the _ height _ of an element of @xmath150 is by definition the sum of the coefficients of the @xmath154 , in it . in what follows we often need to show that certain monomials @xmath155 are zero in @xmath134 . lemmas [ lm : atleasttwo][lm : veryreal ] show how the bilinear form comes into play . but first we recall an elementary property of sandwich elements , to which they owe their name . [ lm : sandwich ] let @xmath8 be a sandwich element in a lie algebra @xmath4 and let @xmath156 be arbitrary . then @xmath157 . we have @xmath158xz+yxxz=[x , y]xz=-[x , xy]z+x[x , y]z = zxxy+x[x , y]z\\ & = x[x , y]z = xxyz - xyxz =- xyxz.\qedhere\end{aligned}\ ] ] note that we have used here that the characteristic is not @xmath2 . in the case of characteristic @xmath2 , this lemma should be taken as part of the definition of a sandwich element @xcite . [ lm : atleasttwo ] let @xmath159 be a word over @xmath24 and let @xmath160 . let @xmath58 , @xmath161 be consecutive occurrences of @xmath8 in @xmath162 ( i.e. , @xmath163 , @xmath164 , and @xmath165 for all @xmath166 strictly between @xmath167 and @xmath168 ) . suppose that the letters in @xmath162 strictly between @xmath58 and @xmath161 contain at most @xmath68 occurrence of a @xmath0-neighbour of @xmath8 , i.e. , the set @xmath169 has cardinality at most @xmath68 . then @xmath170 is @xmath49 in @xmath134 . set @xmath171 . first , using the fact that on @xmath140 the linear map @xmath172 commutes with @xmath173 for any @xmath174 with @xmath175 , we can move @xmath58 in @xmath176 to the right until it is directly to the left of either @xmath161 or the unique @xmath177 between @xmath58 and @xmath161 , so we may assume that this was already the case to begin with . if @xmath178 then either @xmath179 and @xmath176 is zero by anti - commutativity , or @xmath180 and the monomial @xmath181 is zero by the sandwich property of @xmath8 . suppose , on the other hand , that @xmath182 . then if @xmath179 and @xmath183 the monomial @xmath176 is zero since @xmath184 is indeed , @xmath185 . on the other hand , if @xmath186 or @xmath180 then we can move @xmath161 in @xmath176 to the left until it is directly to the right of @xmath187 . so again , we may assume that it was there right from the beginning . but now @xmath188 if @xmath180 , then this monomial equals zero by lemma [ lm : sandwich ] ; and if @xmath179 , then it is zero by the sandwich property of @xmath8 . [ lm : reduce ] let @xmath189 be a word with @xmath190 over @xmath24 and suppose that the weight @xmath143 of @xmath191 satisfies @xmath192 . then @xmath193 in @xmath134 . set @xmath194 and @xmath195 . the condition on the bilinear form can be written as follows : @xmath196 first we note that if the right - hand side is @xmath49 , then @xmath176 is trivially zero : then all @xmath58 with @xmath197 commute with @xmath198 , and there at least @xmath199 such factors . so we may assume that the right - hand side is positive , and hence so is the left - hand side . let the set in the left - hand side of this inequality consist of the indices @xmath200 ; by the above @xmath201 is positive . in the word @xmath162 there are @xmath201 pairs @xmath202 satisfying the conditions of lemma [ lm : atleasttwo ] with @xmath203 , namely , @xmath204 @xmath205 @xmath206 @xmath207 . now if for some such @xmath202 there are less than two @xmath0-neighbours of @xmath198 in the interval between @xmath58 and @xmath161 , then @xmath208 by lemma [ lm : atleasttwo ] . so we may assume that each of these @xmath201 intervals contains at least two @xmath0-neighbours of @xmath198 . but then , by the above inequality , these exhaust all @xmath0-neighbours of @xmath198 in @xmath162 , so in particular there are exactly @xmath2 @xmath0-neighbours of @xmath198 between @xmath209 and @xmath210 , and none to the right of @xmath210 . now if @xmath211 , then @xmath176 is zero because @xmath210 commutes with everything to the right of it . hence assume that @xmath212 , and note that @xmath213 . if @xmath214 , then again @xmath176 is trivially @xmath49 , so assume that @xmath215 is a @xmath0-neighbour of @xmath216 . then we have @xmath217 but in the monomial on the right there is only one @xmath0-neighbour of @xmath198 between @xmath209 and @xmath218hence it is zero by lemma [ lm : atleasttwo ] . [ lm : multone ] let @xmath36 and @xmath219 satisfy @xmath220 . then @xmath221 $ ] . let @xmath222 be a word over @xmath24 of weight @xmath223 . we show that in @xmath134 the monomial @xmath224 is a scalar multiple of some monomial of the form @xmath225 , where @xmath226 is a monomial of weight @xmath227 . obviously , @xmath8 occurs in @xmath162 ; let @xmath166 be maximal with @xmath228 . if @xmath229 , then we may interchange @xmath230 and @xmath231 in @xmath176 at the cost of a minus sign ( note that @xmath190 ) , so we may assume that @xmath232 . suppose first that there occur @xmath0-neighbours of @xmath233 to the left of @xmath234 in @xmath162 . we claim that then @xmath208 . indeed , let @xmath235 be the weights of @xmath236 and @xmath237 , respectively . then we have @xmath238 where in the last inequality we use that there are occurrences of neighbours of @xmath234 , but none of @xmath234 itself , in the word @xmath237 . now we find @xmath239 by lemma [ lm : reduce ] ( note that @xmath240 ) , hence @xmath208 as claimed . so we can assume that there are no @xmath0-neighbours of @xmath234 to the left of @xmath234 in @xmath162 . then we may move @xmath234 in @xmath176 all the way to the left , hence @xmath176 is indeed equal to @xmath241 for some monomial @xmath226 of weight @xmath227 . recall the definition of the kac - moody algebra @xmath242 over @xmath243 corresponding to @xmath0 : it is the lie algebra generated by latexmath:[$3 \cdot @xmath36 , modulo the relations @xmath245 endow @xmath242 with the @xmath150-grading in which @xmath246 have weights @xmath247 , respectively . let @xmath248 be the _ root system _ of @xmath242 ; it is equal to the disjoint union of its subsets @xmath249 and contains the _ simple roots _ @xmath250 ; we refer to @xcite for the theory of kac - moody algebras . in what follows we will compare the multiplicities of weights in the @xmath1-algebra @xmath37 and the @xmath243-algebra @xmath242 . [ lm : roots ] for @xmath251 we have @xmath252 . we proceed by induction on the height of @xmath227 . the proposition is trivially true for @xmath227 of height @xmath68 . suppose now that it is true for height @xmath199 , and consider a word @xmath159 of weight @xmath253 . set @xmath254 . if @xmath255 , then @xmath256 by the induction hypothesis , so we may assume that @xmath257 . this together with @xmath258 implies ( by elementary @xmath259-theory in @xmath242 ) that @xmath192 . now lemma [ lm : reduce ] shows that @xmath260 . for another relation between weight multiplicities in @xmath37 and @xmath242 , recall that a root in @xmath261 is called _ real _ if it is in the orbit of some simple root under the weyl group @xmath262 of @xmath242 . in that case it has multiplicity @xmath68 in @xmath242 . we now call a root @xmath263 _ very real_this is non - standard terminology if it can be written as @xmath264 , for some @xmath265 , such that for all @xmath266 we have @xmath267 ( this implies that @xmath268 , where the @xmath269 are the fundamental reflections corresponding to the @xmath36 . ) [ lm : veryreal ] any very real @xmath263 has multiplicity at most @xmath68 in @xmath37 . this follows by induction on the height of @xmath270 , using lemma [ lm : multone ] for the induction step . in this section we assume that @xmath0 is a dynkin diagram of finite type , i.e. , one of the diagrams in figure [ fig : diagrams ] with vertex @xmath49 removed . then @xmath242 is a finite - dimensional simple lie algebra over @xmath243 . now @xmath242 has a _ chevalley basis _ * section 4.2 ) . this basis consists of the images of the @xmath271 and one vector @xmath272 for every root @xmath273 , where @xmath274 may be taken as @xmath275 , respectively . an important property that we will need is that @xmath276 = \pm e_{\alpha+\beta}$ ] for all roots @xmath277 such that @xmath278 is a root ; here we use that the @xmath279 in ( * ? ? ? * theorem 4.2.1 ) is @xmath49 in the simply laced case . the chevalley basis spans a @xmath54-subalgebra of @xmath242 . let @xmath280 be the @xmath1-algebra obtained by tensoring this @xmath54-form with @xmath1 , and let @xmath281 be the images in @xmath280 of @xmath246 . the lie algebra @xmath280 has a triangular decomposition @xmath282 where @xmath283 . we will refer to @xmath280 as the _ chevalley algebra of type @xmath0_. one can also define @xmath280 as the lie algebra of the split simply connected algebraic group over @xmath1 of type @xmath0 . [ thm : finite ] let @xmath0 be a simply laced dynkin diagram of finite type , obtained from a diagram in figure [ fig : diagrams ] by removing vertex @xmath49 . let @xmath280 be the corresponding chevalley algebra over the field @xmath1 of characteristic unequal to @xmath2 , and let @xmath284 be the subalgebra generated by the @xmath285 . then the map sending @xmath160 to @xmath285 induces a ( necessarily unique ) isomorphism @xmath286 . in the proof of this theorem we use the following well - known facts about simply laced kac - moody algebras of finite type : first , @xmath287 only takes the values @xmath288 on @xmath289 , and second , all roots in @xmath290 are very real . to prove the existence of a _ homomorphism _ @xmath291 sending @xmath8 to @xmath285 , we verify that the relations defining @xmath37 hold in @xmath284 . that is , we have to prove that @xmath292=0\textrm { for all } x , y\in\pi \text { with } x \not\sim y,\\ & { \ensuremath{\mathrm{ad}}}(e_x^0)^2 z=0\textrm { for all } x\in\pi\textrm { and all } z\in{\ensuremath{\mathfrak{n}}}_+.\end{aligned}\ ] ] the first statement is immediate from the relations defining @xmath242 . for the second relation , if @xmath293 is a root vector with root @xmath263 , then @xmath294 by the above , so that @xmath295 and therefore @xmath296 , so that @xmath297 . as root vectors span @xmath284 , we have proved the existence of @xmath291 ; uniqueness is obvious . now we have to show that @xmath291 is an isomorphism . it is surjective as @xmath284 is generated by the @xmath285 ; this follows from the properties of the chevalley basis in subsection [ ssec : finitesandwich ] . hence it suffices to prove that @xmath298 . but by lemma [ lm : roots ] , lemma [ lm : veryreal ] and the fact that all roots are very real we have @xmath299 for all @xmath255 and @xmath300 for all @xmath263 . this concludes the proof . suppose now that @xmath0 is a simply laced dynkin diagram of affine type from figure [ fig : diagrams ] . recall that the cartan matrix @xmath301 has a one - dimensional kernel , spanned by a unique primitive vector @xmath302 . here primitive means that the greatest common divisor of the coefficients of @xmath303 on the standard basis is @xmath68 ; indeed , there always exists a vertex @xmath304 ( labelled @xmath49 in figure [ fig : diagrams ] ) with coefficient @xmath68 in @xmath303 , and all such vertices form an @xmath305-orbit . for later use , we let @xmath115 be the _ coxeter number _ , which is the height of @xmath303 . write @xmath306 , @xmath51 for the induced subgraph on @xmath307 ( which is a dynkin diagram of finite type ) , and @xmath308 for the root system of the chevalley algebra @xmath280 of type @xmath51 defined in subsection [ ssec : finitesandwich ] . this root system lives in the space @xmath309 , which we identify with the elements of @xmath310 that are zero on @xmath311 . retain the notation @xmath312 from subsection [ ssec : finitesandwich ] . consider the semi - direct product @xmath313 , where the second summand is endowed with the trivial lie bracket and the natural @xmath284-module structure . this @xmath314 is clearly a nilpotent lie algebra ; we will prove that it is isomorphic to @xmath37 . in our proof we use the following @xmath150-grading of @xmath314 : the root spaces in @xmath284 have their usual weight in @xmath315 , while the image of @xmath316 in @xmath317 has weight @xmath318 . thus the set of all weights ocurring in @xmath314 is @xmath319 [ thm : affine ] let @xmath0 be a simply laced dynkin diagram of affine type from figure [ fig : diagrams ] , let @xmath51 be the subdiagram of finite type obtained by removing vertex @xmath49 , and let @xmath280 be the chevalley algebra of type @xmath51 over a field of characteristic unequal to @xmath2 . for @xmath320 let @xmath321 be the element of the chevalley basis of @xmath280 with simple root @xmath147 and for the lowest root @xmath322 let @xmath323 be the image of the element in the chevalley basis of weight @xmath324 . then the map sending @xmath320 to @xmath285 and @xmath311 to @xmath325 induces a @xmath150-graded isomorphism @xmath326 of lie algebras . over @xmath243 one can argue directly in the kac - moody algebra @xmath242 . then @xmath37 is also the quotient of the positive nilpotent subalgebra of @xmath242 by the root spaces with roots of height larger than the coxeter number @xmath115 . in the proof one uses the root multiplicities of ( * proposition 6.3 ) . one might also pursue this approach in positive characteristic using the results of @xcite , but we have chosen to avoid defining the kac - moody algebra in arbitrary characteristic and use the chevalley basis instead . the proof is close to that of theorem [ thm : finite ] . we start by verifying that the relations defining @xmath37 hold in @xmath327 . first , @xmath285 and @xmath328 with @xmath329 commute when they are not connected in @xmath0 ; this follows from the defining equations of @xmath242 . second , @xmath285 and @xmath330 commute if @xmath320 is not connected to @xmath311 , as @xmath331 is then not in @xmath308 . third , each @xmath285 is a sandwich element in @xmath314 : for its action on @xmath284 this follows as in the proof of theorem [ thm : finite ] and for its action on @xmath332 it follows from the fact that @xmath333 . fourth , @xmath330 is a sandwich element as @xmath334 maps @xmath314 into @xmath332 , which has trivial multiplication . this shows the existence of a homomorphism @xmath335 . moreover @xmath291 is graded ; in particular , the weight of @xmath330 is @xmath336 . the @xmath337 generate @xmath284 and @xmath330 generates the @xmath284-module @xmath332 . these statements follow from properties of the chevalley basis in subsection [ ssec : finitesandwich ] , and imply that @xmath291 is surjective . so we need only show that @xmath338 ; we prove this for each weight in @xmath339 . first , the roots in @xmath308 are very real , so their multiplicities in @xmath37 are at most @xmath68 by lemma [ lm : veryreal ] . second , we claim that all roots of the form @xmath340 with @xmath341 are also very real . this follows by induction on the height of @xmath270 : for @xmath342 it is clear since @xmath336 . for @xmath343 it is well - known that there exists an @xmath320 such that @xmath344 . then we have @xmath345 where @xmath346 and @xmath347 ; here we use that @xmath303 is in the radical of the form @xmath149 . by induction , @xmath348 is very real , hence so is @xmath318 . this shows that also the roots of the form @xmath318 with @xmath349 have multiplicity at most @xmath68 in @xmath37 , again by lemma [ lm : veryreal ] . next we show that @xmath303 has multiplicity at most @xmath350 in @xmath37 . indeed , we claim that @xmath351 is contained in @xmath352.\ ] ] then , by the above , each of the summands has dimension at most @xmath68 , and we are done . the claim is true almost by definition : any monomial of weight @xmath303 must start with some @xmath36 , so we need only show that monomials starting with @xmath311 are already contained in the sum above . consider any monomial @xmath224 of weight @xmath303 , where @xmath353 . as the coefficient of @xmath311 in @xmath303 is @xmath68 , none of the @xmath58 with @xmath197 is equal to @xmath311 . but then an elementary application of the jacobi identity and induction shows that @xmath176 is a linear combination of monomials that do _ not _ start with @xmath311 . finally we have to show that @xmath354 is @xmath49 for @xmath355 ( we already have @xmath299 for @xmath356 by lemma [ lm : roots ] ) . but lemma [ lm : reduce ] and the fact that @xmath357 for all @xmath36 together imply that @xmath358=0 $ ] for all @xmath36 . so it suffices to show that if @xmath257 is not in @xmath339 , then `` @xmath143 can only be reached through @xmath303 '' . more precisely : if @xmath136 is any word over @xmath24 such that @xmath359 and @xmath360 for all @xmath361 , then there exists an @xmath167 such that @xmath362but this follows immediately from the fact that @xmath303 is the only root of height @xmath115 ( * ? ? ? * proposition 6.3 ) . we find that every monomial corresponding to such a word is zero , and this concludes the proof of the theorem . so far we have only considered the lie algebras @xmath37 . now we will be concerned with the variety @xmath3 of all parameters @xmath363 for which @xmath364 . we collect some tools for determining @xmath3 in the case of simply laced dynkin diagrams . first let @xmath0 be arbitrary again , not necessarily a dynkin diagram . scaling of the generators @xmath58 has an effect on @xmath3 : given @xmath365 in the torus @xmath366 there is a unique automorphism of @xmath26 that sends @xmath36 to @xmath367 . this gives an action of @xmath368 on @xmath26 , and we endow @xmath369 with the contragredient action . finally , we obtain an action of @xmath368 on @xmath3 by @xmath370 indeed , note that with this definition the automorphism of @xmath26 induced by @xmath371 sends @xmath372 to @xmath373 and hence the ideal @xmath76 defining @xmath32 to @xmath374 . therefore , this automorphism of @xmath26 induces an isomorphism @xmath375 . this scaling action of @xmath368 on @xmath3 will make things very easy in the case of simply laced dynkin diagrams , where @xmath3 will turn out to be isomorphic to an affine space with linear action of @xmath368 , in which the maximal - dimensional orbits have codimension @xmath49 , @xmath68 or @xmath2 . observe that the one - parameter subgroup @xmath376 satisfies @xmath377 for all @xmath74 . this shows that all irreducible components of @xmath3 contain @xmath49 ; in particular , @xmath3 is connected . in @xcite it is proved that on any lie algebra over a field of characteristic unequal to @xmath2 that is generated by finitely many extremal elements , there is a unique bilinear form @xmath378 such that @xmath379 for all extremal @xmath8 . moreover , it is shown there that @xmath378 is symmetric and associative : @xmath380 and @xmath381 for all @xmath382 . we call @xmath378 the _ extremal form_. for the chevalley algebra @xmath280 of subsection [ ssec : finitesandwich ] the extremal form is non - zero on @xmath383 if and only if @xmath384 . the form may have a radical contained in the cartan subalgebra @xmath385 . on the other hand , for any @xmath363 ( not necessarily in @xmath3 ) the lie algebra @xmath32 is generated by the images of the elements of @xmath24 , which are extremal elements . in particular , for @xmath36 we have in @xmath32 @xmath386 where @xmath378 is the extremal form on @xmath32 . so if the image of @xmath8 in @xmath32 is non - zero , then @xmath387 . our arguments showing that certain monomials @xmath388 are zero in the sandwich algebra @xmath37 always depended on the sandwich properties : @xmath389 and @xmath157 whenever @xmath8 is a sandwich element and @xmath390 are arbitrary elements of the lie algebra . the premet relations of the following lemma translate such a statement into the following statement : in @xmath32 the monomial @xmath201 can be expressed in terms of monomials of degree less than @xmath391 and values of @xmath392 on monomials of degree less than @xmath391 . let @xmath8 be a non - zero extremal element of a lie algebra @xmath14 , and let @xmath393 be the linear function with @xmath394 . then we have @xmath395 in characteristic @xmath2 the definition of an extremal element @xmath8 involves the existence of a function @xmath396 such that @xmath397 , i.e. , @xmath396 plays the role of @xmath398 . see ( * ? ? ? * definition 14 ) . recall from section [ sec : moduli ] that the restriction map @xmath399 is injective and has a closed image ; a key step in the proof was showing that for @xmath74 the values @xmath400 depend polynomially on @xmath133 . in what follows , this will be phrased informally as _ @xmath50 can be expressed in @xmath133 _ or _ @xmath133 determines @xmath50_. in this phrase we implicitly make the assumption that @xmath74 , i.e. , that @xmath32 has the maximal possible dimension . in the case of dynkin diagrams , we will exhibit a small number of values of @xmath50 in which @xmath50 can be expressed . for this purpose , the following lemma , which also holds for other graphs , is useful . [ lm : expressible ] let @xmath401 be a monomial of degree @xmath190 and weight @xmath270 , and let @xmath402 be such that @xmath403 . then @xmath404 can be expressed in the parameters @xmath405 with monomials @xmath201 of degree less than @xmath391 and @xmath36 . first , if @xmath198 is not a @xmath0-neighbour of @xmath176 in @xmath0 , then @xmath406 and we are done . so assume that @xmath198 is a @xmath0-neighbour of @xmath176 . now @xmath407 in both cases we have used that the images of @xmath176 and @xmath198 are non - zero in @xmath32 for @xmath74 ; see subsection [ ssec : extremalform ] . now @xmath408 , so lemma [ lm : reduce ] says that @xmath409 can be expressed in terms of smaller monomials and values @xmath405 for @xmath36 and monomials @xmath201 of degree less than @xmath391 . then by linearity of @xmath392 the last expression above can also be expressed in terms of values @xmath405 with @xmath36 and @xmath201 of degree less than @xmath391 . suppose that @xmath0 is a simply laced dynkin diagram of finite type . let @xmath280 be the chevalley algebra with dynkin diagram @xmath0 of subsection [ ssec : finitesandwich ] . we identify @xmath150 with the character group of @xmath410 in the natural way : we write @xmath143 for the character that sends @xmath371 to @xmath411 . furthermore , let @xmath52 be the set of edges of @xmath0 , and write @xmath412 for @xmath413 . [ thm : genericfinite ] let @xmath414 be a simply laced dynkin diagram of finite type , obtained from a diagram in figure [ fig : diagrams ] by removing vertex @xmath49 . let @xmath280 be the chevalley algebra of type @xmath0 over the field @xmath1 of characteristic unequal to @xmath2 , and set @xmath415 . then the variety @xmath3 is , as a @xmath368-variety , isomorphic to the vector space @xmath416 on which @xmath368 acts diagonally with character @xmath417 on the component corresponding to @xmath418 . for @xmath50 corresponding to any element in the dense @xmath368-orbit @xmath419 the lie algebra @xmath32 is isomorphic to a fixed lie algebra . we first need a lemma that will turn out to describe the generic @xmath32 . we retain the notation @xmath420 and @xmath284 from subsection [ ssec : finitesandwich ] . we denote by @xmath421 the variety of tuples @xmath422 with @xmath423 extremal ; @xmath421 is an irreducible variety . [ lm : genericfinite ] for generic @xmath424 the lie subalgebra @xmath425 of @xmath280 generated by the @xmath426 has dimension @xmath427 , and moreover @xmath428 is a non - zero multiple of @xmath426 for all @xmath429 . by definition @xmath425 is generated by extremal elements , hence it has dimension at most that of @xmath37 , which is isomorphic to @xmath284 by theorem [ thm : finite ] . the condition that the @xmath426 generate a lie algebra of dimension less than @xmath427 is closed , and the tuple @xmath430 does not fulfill it . hence using the irreducibility of @xmath421 we find that for @xmath431 in an open dense subset of @xmath421 the lie algebra @xmath425 has @xmath432 . this proves the first statement . the second statement follows directly from the same statement for @xmath0 of type @xmath23 , i.e. , for @xmath433 , where it boils down to the statement that the two copies of @xmath259 in @xmath434 corresponding to the simple roots are not mutually perpendicular relative to the extremal form in @xmath434 . by lemma [ lm : expressible ] and theorem [ thm : finite ] , any @xmath74 is determined by its values @xmath405 with @xmath36 and monomials @xmath201 of weights @xmath263 such that either @xmath270 has height @xmath68 or @xmath435 . but since @xmath270 is a positive root , the latter inequality can not hold . hence @xmath270 has height @xmath68 , so that @xmath436 , and the only @xmath36 for which @xmath437 are the neighbours of @xmath201 . moreover , from the symmetry of the extremal form we conclude that @xmath438 for @xmath439 neighbours in @xmath0 . we have thus found a closed embedding @xmath440 sending @xmath50 to @xmath441 . now if we let @xmath368 act on @xmath442 through the homomorphism @xmath443 then @xmath101 is @xmath368-equivariant by the results of subsection [ ssec : scaling ] . note that @xmath368 acts by the character @xmath417 on the component corresponding to @xmath418 . the fact that @xmath0 is a tree readily implies that the characters @xmath444 are linearly independent over @xmath54 in the character group of @xmath368 , so that the homomorphism @xmath445 is surjective . but then @xmath368 has finitely many orbits on @xmath442 , namely of the form @xmath446 with @xmath447 . now as @xmath448 is a closed @xmath368-stable subset of @xmath442 we are done if we can show that @xmath449 is non - empty . but this is precisely what lemma [ lm : genericfinite ] tells us : there exist lie algebras @xmath425 generated by extremal elements that have the largest possible dimension , and where all coordinates @xmath450 with @xmath429 are non - zero . this concludes the proof . the proof above also implies that all lie algebras described in lemma [ lm : genericfinite ] are isomorphic . more generally : for any two lie algebras @xmath425 and @xmath451 with tuples of distinguished , extremal generators @xmath452 and @xmath453 such that 1 . @xmath454 and @xmath455 for @xmath456 , and 2 . @xmath457 if and only if @xmath458 for @xmath429 , and 3 . @xmath459 , there exists an isomorphism @xmath460 mapping each @xmath461 to a scalar multiple of @xmath462 . suppose now that @xmath0 is a simply laced dynkin diagram of affine type . we retain the notation from subsection [ ssec : affinesandwich ] . in particular , let @xmath280 be the chevalley algebra of type @xmath51 , the graph induced by @xmath0 on @xmath463 . to state the analogue of theorem [ thm : genericfinite ] we again identify @xmath150 with the character group of @xmath410 and retain the notation @xmath464 for @xmath418 , the edge set of @xmath0 . [ thm : genericaffine ] let @xmath414 be a simply laced dynkin diagram of affine type from figure [ fig : diagrams ] and let @xmath51 be the finite - type diagram obtained by deleting vertex @xmath49 from @xmath0 . let @xmath280 be the chevalley algebra of type @xmath51 over the field @xmath1 of characteristic unequal to @xmath2 , and set @xmath415 . then the variety @xmath3 is , as a @xmath368-variety , isomorphic to the vector space @xmath465 on which @xmath368 acts diagonally with character @xmath417 on the component corresponding to @xmath418 , and with character @xmath466 on the last component . for all @xmath74 corresponding to points in some open dense subset of @xmath467 the lie algebra @xmath32 is isomorphic to @xmath280 . [ re : rank ] unlike for diagrams of finite type , it is _ not _ necessarily true that @xmath368 has only finitely many orbits on @xmath45 . indeed , the following three situations occur : 1 . the characters @xmath468 are linearly independent . this is the case for @xmath469 , @xmath470 , and @xmath471 . then @xmath368 has finitely many orbits on @xmath45 . 2 . the characters @xmath472 are linearly independent , but @xmath303 is in their @xmath473-linear span . this is the case for @xmath474 , @xmath475 and @xmath476 . now the orbits of @xmath368 in @xmath477 have codimension @xmath68 . for the character @xmath303 has full support when expressed in the @xmath464 ; and this readily implies that @xmath368 has finitely many orbits on the complement of @xmath478 . for @xmath479 with @xmath480 odd , however , @xmath481 edge characters get coefficient @xmath49 when @xmath303 is expressed in them , and therefore @xmath368 still has infinitely many orbits on said complement . 3 . the characters @xmath472 are linearly dependent . this is the case only for @xmath482 , and in fact @xmath303 is then also in the span of the @xmath464 . now the @xmath368-orbits in @xmath477 have codimension @xmath2 , and in the complement there are still infinitely many orbits . this gives some feeling for the parameter space @xmath3 . it would be interesting to determine exactly all isomorphism types of lie algebras @xmath32 with @xmath74but here we confine ourselves to those with @xmath50 in some open dense subset of @xmath467 . the proof is very similar to that of theorem [ thm : genericfinite ] . again , we first prove a lemma that turns out to describe the generic @xmath32 . retain the notation @xmath483 for @xmath320 . moreover , denote the lowest weight by @xmath484 , let @xmath485 be the elements of the chevalley basis of weights @xmath324 and @xmath486 , respectively , and set @xmath487 $ ] . write @xmath421 for the irreducible variety of tuples @xmath422 with @xmath488 extremal . [ lm : genericaffine ] for generic @xmath489 the @xmath426 generate @xmath280 , and moreover @xmath428 is a non - zero multiple of @xmath426 for all @xmath429 . the first statement is true for @xmath490 ; this follows from the properties of the chevalley basis in subsection [ ssec : finitesandwich ] . hence by the irreducibility of @xmath421 it is true generically . the second statement follows , as in lemma [ lm : genericfinite ] , from the same statement in @xmath434 . in the following proof we will show that the choice of @xmath491 as in lemma [ lm : genericaffine ] already gives generic points in @xmath3 , except for the case where @xmath0 is of type @xmath492 , for which we give another construction . by lemma [ lm : expressible ] and theorem [ thm : affine ] , any @xmath74 is determined by its values @xmath405 with @xmath36 and monomials @xmath201 of weights @xmath493 such that either @xmath270 has height @xmath68 or @xmath435 . in contrast with the case of finite - type diagrams , there do exist pairs @xmath494 with this latter property , namely precisely those of the form @xmath495 . for all @xmath36 , let @xmath496 be a monomial that spans the weight space in @xmath37 of weight @xmath497 ; this space is @xmath68-dimensional by theorem [ thm : affine ] . we claim that the @xmath498 can all be expressed in terms of @xmath499 and values @xmath500 with @xmath402 and @xmath501 of degree less than @xmath502 . indeed , if @xmath503 , then @xmath311 occurs exactly once in @xmath496 ; and writing @xmath504 with @xmath505 we find @xmath506)\\ & = ( -1)^d f_{x_0}([x_1 \cdots x_d x , y_e \cdots y_1]),\end{aligned}\ ] ] and the expression @xmath507 $ ] can be rewritten in terms of @xmath508 and shorter monomials , using values @xmath500 with @xmath501 of degree less than @xmath509 . we have now found a closed embedding @xmath510 which sends @xmath50 to @xmath511 ; for ease of exposition we will view @xmath3 as a closed subset of @xmath467 . the theorem follows once we can realise generic parameter values in @xmath467 with extremal elements that generate @xmath280 . to this end , choose a generic tuple @xmath491 in @xmath421 . by lemma [ lm : genericaffine ] these generate @xmath280 and they clearly satisfy @xmath512 for @xmath513 . hence they yield a point in @xmath3 with @xmath514 for @xmath429 . furthermore , the parameter @xmath499 equals the extremal form evaluated on @xmath515 and the monomial @xmath508 evaluated in the @xmath426 . express that monomial in the @xmath426 as @xmath516 plus a term perpendicular to @xmath517 , and write @xmath515 as @xmath518 . for the degenerate case where @xmath519 for all @xmath8 we have @xmath520 and @xmath521 ( that monomial is a non - zero scalar multiple of the highest root vector , @xmath522 ) , so that @xmath523 . therefore , this parameter is non - zero generically . hence we have found a point @xmath524 . along the lines of remark [ re : rank ] we now distinguish three cases : 1 . if the @xmath525 and @xmath303 are linearly independent , then @xmath368 acts transitively on @xmath477 , and we are done . if the @xmath525 are linearly independent , but @xmath303 lies in their span , then we now show that we can alter the point @xmath50 above in a direction transversal to its @xmath368-orbit . let @xmath526 be the torus in the adjoint group of @xmath280 whose lie algebra is @xmath385 , and consider the effect on @xmath50 of conjugation of @xmath515 with an element @xmath527 , while keeping the other @xmath426 fixed . this transforms @xmath528 in @xmath529 , and therefore it transforms @xmath499 into @xmath530 while it keeps the parameters @xmath531 with @xmath532 unchanged : these only depend on @xmath533 . this shows that we can indeed move @xmath50 inside @xmath3 in a direction transversal to its @xmath368-orbit , and we are done . + [ it : conjugate ] 3 . finally , in the case of @xmath534 with @xmath480 even we first show that tuples in @xmath421 only give points in a proper closed subset of @xmath467 . here @xmath535 and @xmath0 is an @xmath480-cycle ; label its points @xmath536 . relative to the usual choices of @xmath537 the element @xmath538 is a matrix with @xmath539-block @xmath540 on the diagonal in rows ( and columns ) @xmath167 and @xmath541 and zeroes elsewhere . we count the rows and columns modulo @xmath480 so that row @xmath49 is actually row @xmath480 . but then we have @xmath542 , and this implies @xmath543 so the tuple of parameter values of the tuple @xmath544 lies in a proper closed subset @xmath545 of @xmath467 . + we therefore allow the tuple @xmath546 to vary in a slightly larger variety @xmath547 as follows : the conditions on @xmath548 remain the same , but @xmath549 is now allowed to take the shape @xmath550 ( which is extremal since it has rank @xmath68 and trace @xmath49 ) , subject to the equations @xmath551 which ensure that @xmath549 commutes with @xmath552 . still , any tuple in an open neighbourhood @xmath553 of our original tuple @xmath546 ( with generic @xmath554 and @xmath555 but all @xmath556 equal to @xmath49 ) generates @xmath557 . we now argue that the differential @xmath59 at @xmath546 of the map @xmath558 sending a tuple to the parameters that it realises has rank @xmath53 , as required . indeed , the @xmath368-action already gives a subspace of dimension @xmath559 , tangent to @xmath545 . making @xmath560 ( and hence all @xmath556 ) non - zero adds @xmath561 to @xmath562 and @xmath563 to @xmath564 , and it fixes all other @xmath565 . we show that this infinitesimal direction is not tangent to @xmath545 : it adds @xmath566 to the left - hand side of , and @xmath567 to the right - hand side . dividing these expressions by common factors , the first becomes @xmath568 and the second @xmath569 . these expressions are not equal generically , even modulo the equations relating the @xmath556 to the @xmath554 and @xmath555 ; indeed , these equations do not involve @xmath570 . + note that varying @xmath560 may also effect the parameter @xmath499 , but in any case the above shows that the composition of the differential @xmath59 with projection onto @xmath442 is surjective . on the other hand , conjugation with the torus @xmath526 as in case yields a vector in the image of @xmath59 which is supported only on the factor @xmath1 corresponding to @xmath303 . this concludes the proof that @xmath59 has full rank . going through the proof that @xmath3 is an affine variety , one observes that the map @xmath571 is not only injective on @xmath3 , but even on @xmath572 the same is true for the map @xmath573 in the case where @xmath0 is a dynkin diagram of finite type , and for the map @xmath574 in the case where @xmath0 is a dynkin diagram of affine type . this shows that , for these dynkin diagrams , @xmath575 is actually _ equal _ to @xmath3 , whence the following theorem . suppose that @xmath0 is a dynkin diagram of finite or affine type . let @xmath14 be any lie algebra , over a field of characteristic unequal to @xmath2 , which is generated by non - zero extremal elements @xmath576 , in which the commutation relations @xmath512 for @xmath456 hold . define @xmath577 by the condition that @xmath578 holds in @xmath14 . then @xmath74 and @xmath14 is a quotient of @xmath32 . this theorem could well prove useful for recognising the chevalley algebras @xmath280 : if @xmath50 corresponds to a point in the open dense subset of @xmath467 referred to in theorem [ thm : genericaffine ] , then one concludes that @xmath14 is a quotient of @xmath280 . hence if @xmath280 is a simple lie algebra , then @xmath14 is isomorphic to @xmath280 . it is not clear to us whether , for general @xmath0 , the image of @xmath575 in @xmath48 is closed ; this is why we chose to work with @xmath3 instead . our methods work very well for dynkin diagrams , but for more general graphs new ideas are needed to determine @xmath579 and @xmath32 for @xmath74 . the relation with the kac - moody algebra of @xmath0 may be much tighter than we proved in subsection [ ssec : kacmoody ] . general questions of interest are : is @xmath3 always an affine space ? is there always a generic lie algebra ? we expect the answers to both questions to be negative , but do not have any counterexamples . the references @xcite contain other series of graphs which exhibit the same properties as we have proved here : the variety @xmath3 is an affine space , and generic points in it correspond to simple lie algebras of types @xmath580 . in fact , the graph that they find for @xmath581 is just the finite - type dynkin diagram of type @xmath582 . this also follows easily from our results : take @xmath583 generic extremal elements @xmath584 in @xmath585 as in lemma [ lm : genericfinite ] . these generate a subalgebra of @xmath585 of dimension @xmath586 by that same lemma , and if we consider them as matrices , their images span a subspace @xmath262 of dimension @xmath583 in @xmath587 . it is not hard to write down an explicit , non - degenerate skew symmetric form on @xmath262 with respect to which the @xmath426 are skew hence the lie algebra generated by them is @xmath588 . dan a. roozemond , _ lie algebras generated by extremal elements _ , master s thesis , technische universiteit eindhoven , 2005 , available from ` http://alexandria.tue.nl/extra1/afstversl/wsk-i/roozemond2005.pdf ` . efim i. zelmanov and alexei i. kostrikin , _ a theorem on sandwich algebras _ , trudy mat . inst . steklov . * 183 * ( 1990 ) , 106111 , 225 , translated in proc . steklov inst . 1991 * , no . 4 , 121126 , galois theory , rings , algebraic groups and their applications ( russian )

we consider the _ lower - bounded facility location _ ( ) problem ( also sometimes called _ load - balanced facility location _ ) , which is a generalization of _ uncapacitated facility location _ ( ) , where each open facility is required to serve a certain _ minimum _ amount of demand . more formally , an instance of is specified by a set of facilities with facility - opening costs , a set of clients , and connection costs specifying the cost of assigning a client to a facility , where thes form a metric . a feasible solution specifies a subset of facilities to open , and assigns each client to an open facility so that each open facility serves _ at least clients _ , where is an input parameter . the cost of such a solution is , and the goal is to find a feasible solution of minimum cost . the current best approximation ratio for is . we substantially advance the state - of - the - art for by devising an approximation algorithm for that achieves a significantly - improved approximation guarantee of . our improvement comes from a variety of ideas in algorithm design and analysis , which also yield new insights into . our chief algorithmic novelty is to present an improved method for solving a more - structured instance obtained from via a bicriteria approximation algorithm for , wherein all clients are aggregated at a subset of facilities , each having at least co - located clients ( for some $ ] ) . one of our key insights is that one can reduce the resulting instance , denoted , to a problem we introduce , called _ capacity - discounted _ ( ) . is a special case of capacitated facility location ( ) where facilities are either uncapacitated , or have finite capacity and zero opening costs . circumventing the difficulty that inherits the intractability of with respect to lp - based approximation guarantees , we give a simple local - search algorithm for based on add , delete , and swap moves that achieves the same approximation ratio ( of ) as the corresponding local - search algorithm for . in contrast , the algorithm in proceeds by reducing to , whose current - best approximation ratio is worse than that of our local - search algorithm for , and this is one of the reasons behind our algorithm s improved approximation ratio . another new ingredient of our -algorithm and analysis is a subtly different method for constructing a bicriteria solution for ( and hence , ) , combined with the more significant change that we now choose a _ random _ from a suitable distribution . this leads to a surprising degree of improvement in the approximation factor , which is reminiscent of the mileage provided by random-points in scheduling problems .

1104.3128

facility location problems have been widely studied in the operations research community ( see , e.g. , @xcite ) . in its simplest version , _ uncapacitated facility location _ ( ) , we are given a set of facilities with opening costs , and a set of clients , and we want to open some facilities and assign each client to an open facility so as to minimize the sum of the facility - opening and client - assignment costs . this problem has a wide range of applications . for example , a company might want to open its warehouses at some locations so that its total cost of opening warehouses and servicing customers is minimized . we consider the _ lower - bounded facility location _ ( ) problem , which is a generalization of where each open facility is required to serve a certain _ minimum _ amount of demand . more formally , an instance @xmath0 is specified by a set @xmath1 of facilities , and a set @xmath3 of clients . opening facility @xmath6 incurs a _ facility - opening cost _ @xmath20 , and assigning a client @xmath5 to a facility @xmath6 incurs a _ connection cost _ @xmath7 . a feasible solution specifies a subset @xmath21 of facilities , and assigns each client @xmath5 to an open facility @xmath9 so that _ each open facility serves at least @xmath10 clients _ , where @xmath10 is an input parameter . the cost of such a solution is the sum of the facility - opening and connection costs , that is , @xmath11 , and the goal is to find a feasible solution of minimum cost . as is standard in the study of facility location problems , we assume throughout that @xmath7s form a metric . we use the terms connection cost and assignment cost interchangeably in the sequel . can be motivated from various perspectives . this problem was introduced independently by karger and minkoff @xcite , and guha , meyerson , and munagala ( who called the problem _ load - balanced facility location _ ) @xcite ( see also @xcite ) , both of whom arrived at as a means of solving their respective buy - at - bulk style network design problems . arises as a natural subroutine in such settings because obtaining a near - optimal solution to the buy - at - bulk problem often entails aggregating a certain minimum demand at certain hub locations , and then connecting the hubs via links of lower per - unit - demand cost ( and higher fixed cost ) . also finds direct applications in supply - chain logistics problems , where the lower - bound constraint can be used to model the fact that it is not profitable or feasible to use services unless they satisfy a certain minimum demand . for example ( as noted in @xcite ) , lim , wang , and xu @xcite , use to abstract a transportation problem faced by a company that has to determine the allocation of cargo from customers to carriers , who then ship their cargo overseas . here the lower bound arises because each carrier , if used , is required ( by regulation ) to deliver a minimum amount of cargo . also , is an interesting special case of _ universal facility location _ ( ) @xcite a generalization of where the facility cost depends on the number of clients served by it with non - increasing facility - cost functions . with arbitrary non - increasing functions is not a well - understood problem , clearly , with @xmath22 is simply , and hence , is _ np_-hard ; consequently , we are interested in designing approximation algorithms for . the first constant - factor approximation algorithm for was devised by svitkina @xcite , whose approximation ratio is @xmath12 . prior to this , the only known approximation guarantees were _ bicriteria guarantees_. @xcite and @xcite independently devised @xmath23-approximation algorithms via a reduction to : these algorithms return a solution of cost at most @xmath24 times the optimum where each open facility serves at least @xmath15 clients [ [ our - results - and - techniques . ] ] our results and techniques . + + + + + + + + + + + + + + + + + + + + + + + + + + + we devise an approximation algorithm for that achieves a substantially - improved approximation guarantee of @xmath13 ( theorem [ mainthm ] ) , thus significantly advancing the state - of - the - art for . our improvement comes from a combination of ideas in algorithm design and analysis , and yields new insights about the approximability of . in order to describe the ideas underlying our improvement , we first briefly sketch svitkina s algorithm . svitkina s algorithm begins by using the reduction in @xcite to obtain a bicriteria solution for @xmath0 , which is then used to convert @xmath0 into an instance @xmath25 with facility - set @xmath26 having the following structure : ( i ) all clients are aggregated at @xmath14 with each facility @xmath27 having @xmath28 co - located clients ; ( ii ) all facilities in @xmath14 have zero opening costs ; and ( iii ) near - optimal solutions to @xmath25 translate to near - optimal solutions to @xmath0 ( and vice versa ) . the goal now is to identify a subset of @xmath14 to close , such that transferring the clients aggregated at these closed facilities to the remaining ( open ) facilities in @xmath14 ensures that each remaining facility serves at least @xmath10 demand ( and the cost incurred is `` small '' ) . @xcite shows that one can achieve this by solving a suitable instance . essentially the idea is that a facility @xmath6 that remains open corresponds to a _ demand point _ in the instance that requires @xmath29 units of demand , and a facility @xmath6 that is closed maps to a _ supply point _ in the instance having @xmath30 units that can be supplied to demand points ( i.e. , open facilities ) . of course , one does not know beforehand which facilities will be closed and which will remain open ; so to encode this correspondence in the instance , we create at every location @xmath27 , a supply point with ( suitable opening cost and ) capacity @xmath10 , and a demand point with demand @xmath29 if @xmath31 ( so the supply point at @xmath6 has @xmath30 residual capacity after satisfying this demand ) . ( assume @xmath31 for simplicity ; facilities with @xmath32 are treated differently . ) finally , @xcite argues that a -solution ( where a supply point may end up sending _ less _ then @xmath30 supply to other demand points ) can be mapped to a solution to @xmath25 without increasing the cost incurred by much ; since admits an @xmath33-approximation algorithm , one obtains an @xmath33-approximate solution to @xmath25 , and hence to the original instance @xmath0 . our algorithm also proceeds by ( a ) obtaining an instance @xmath25 satisfying properties ( i)(iii ) mentioned above , ( b ) solving @xmath25 , and ( c ) mapping the @xmath25-solution to a solution to @xmath0 , but our implementation of steps ( a ) and ( b ) differs from that in svitkina s algorithm . these modified implementations , which are independent of each other and yield significant improvements in the overall approximation ratio even when considered in isolation , result in our much - improved approximation ratio . we detail how we perform step ( a ) later , and focus first on describing how we solve @xmath25 , which is our chief algorithmic contribution . our key insight is that one can solve the instance @xmath25 by reducing it to a new problem we introduce that we call _ capacity - discounted _ ( ) , which closely resembles and admits an algorithm ( that we devise ) with a much better approximation ratio than . a -instance has the property that every facility is either uncapacitated ( i.e. , has infinite capacity ) , or has finite capacity and _ zero _ facility cost . the instance we construct consists of the same supply and demand points as in the reduction of @xmath25 to in @xcite , except that all supply points with non - zero opening cost are now uncapacitated . ( an interesting consequence is that if all facilities in @xmath25 have @xmath31 , the instance is in fact a -instance ! ) we prove two crucial algorithmic results . it is not hard to see that the `` standard '' integrality - gap example for the natural lp - relaxation of can be cast as a instance , thus showing that the natural lp - relaxation for has a large integrality gap ( see appendix [ cduflgap ] ) ; in fact , we are not aware of any lp - relaxation for with constant integrality gap . circumventing this difficulty , we devise a local - search algorithm for based on add , swap , and delete moves that achieves the _ same performance guarantees _ as the corresponding local - search algorithm for @xcite ( see section [ cdufl - ls ] ) . the local - search algorithm yields significant dividends in the overall approximation ratio because not only is its approximation ratio for better than the state - of - the - art for , but also because it yields separate ( asymmetric ) guarantees on the facility - opening and assignment costs , which allows one to perform a tighter analysis . second , we show that any near - optimal -solution can be mapped to a near - optimal solution to @xmath25 ( see section [ i2-soln ] ) . as before , it could be that in the -solution , a supply point @xmath6 ( which corresponds to facility @xmath6 being closed down ) sends less than @xmath30 supply to other demand points , so that closing down @xmath6 entails transferring its residual clients to open facilities . but since some supply points are now uncapacitated , it could also be that @xmath6 sends more than @xmath30 supply to other demand points . we argue that this artifact can also be handled without increasing the solution cost by much , by opening the facilities in a carefully - chosen subset of @xmath34 and closing down the remaining facilities . for _ every value of @xmath19 _ ( recall that the instance @xmath25 is specified in terms of a parameter @xmath19 ) , the resulting approximation factor for @xmath25 ( theorem [ i2thm ] ) is better than the guarantee obtained for @xmath25 in svitkina s algorithm ; this in turn translates ( by choosing @xmath19 suitably ) to an improved solution to the original instance . we now discuss how we implement step ( a ) , that is , how we obtain instance @xmath25 . as in @xcite , we arrive at @xmath25 by computing a bicriteria solution to , but we obtain this bicriteria solution in a different fashion ( see section [ bicriteria ] ) . the reduction from to in @xcite proceeds by setting the opening cost of facility @xmath6 to @xmath35 , where @xmath36 is the set of @xmath10 clients closest to @xmath6 , solving the resulting instance , and postprocessing using ( single - facility ) delete moves if such a move improves the solution cost . we modify this reduction subtly by creating a instance , where facility @xmath6 s opening cost is instead set to @xmath37 , where @xmath38 is the distance between @xmath6 and the @xmath15-closest client to it . as in the case of the earlier reduction , we argue that each open facility @xmath6 in the resulting solution ( obtained by solving and postprocessing ) serves at least @xmath15 clients . the overall bound we obtain on the total cost now includes various @xmath38 terms . instead of plugging in the ( weak ) bound @xmath39 ( which would yield the same guarantee as that obtained via the earlier reduction ) , we are able to perform a tighter analysis by choosing @xmath19 from a suitable distribution and leveraging the fact that @xmath40 . ( this can easily be derandomized , since there are only @xmath10 combinatorially distinct choices for @xmath19 . ) these simple modifications ( in algorithm - design _ and _ analysis ) yield a surprising amount of improvement in the approximation factor , which is reminiscent of the mileage provided by ( random ) @xmath19-points for various scheduling problems ( see , e.g. , @xcite ) and @xcite . also , we observe that one can obtain further improvements by using the local - search algorithm of @xcite to solve the above instance : this is because the resulting solution is then already postprocessed , which allows us to exploit the asymmetric bounds on the facility - opening and assignment costs provided by the local - search algorithm via scaling , and improve the approximation ratio . finally , we remark that the study of may provide useful and interesting insights about . is a special case of that despite its special structure inherits the intractability of with respect to lp - based approximation guarantees . if one seeks to develop lp - based techniques and algorithms for ( which has been a long - standing and intriguing open question ) , then one needs to understand how one can leverage lp - based techniques for , and it is plausible that lp - based insights developed for may yield similar insights for ( and potentially lp - based approximation guarantees for ) . [ [ related - work . ] ] related work . + + + + + + + + + + + + + as mentioned earlier , was independently introduced by @xcite and @xcite , who used it as a subroutine to solve the ( _ rent - or - buy _ and hence , the ) _ maybecast _ problem , and the _ access network design _ problem respectively . their ideas , which lead to bicriteria guarantees for , play a preprocessing role both in svitkina s algorithm for @xcite and ( slightly indirectly ) in our algorithm . there is a large body of literature that deals with approximation algorithms for ( metric ) , and its variants ; see @xcite for a survey on . the first constant approximation guarantee for was obtained by shmoys , tardos , and aardal @xcite via an lp - rounding algorithm , and the current state - of - the - art is a 1.488-approximation algorithm due to li @xcite . local - search techniques have also been utilized to obtain @xmath33-approximation guarantees for @xcite . we apply some of the ideas of @xcite in our algorithm . starting with the work of korupolu , plaxton , and rajaraman @xcite , various local - search algorithms with constant approximation ratios have been devised for , with the current - best approximation ratio being @xmath41 @xcite . local - search approaches are however not known to work for ; in appendix [ app : lbfllocgap ] , we show that local search based on @xmath42 , @xmath43 , and @xmath44 moves yields poor approximation guarantees . universal facility location ( ) , where the facility cost is a non - decreasing function of the number of clients served by it , was introduced by @xcite , and @xcite gave a constant approximation algorithm for this . we are not aware of any work on with arbitrary non - increasing functions ( which generalizes ) . @xcite give a constant approximation for the case where the cost - functions do not decrease too steeply ( the constant depends on the steepness ) ; notice that does not fall recall that we have a set @xmath1 of facilities with facility - opening costs @xmath2 , a set @xmath3 of clients , metric connection ( or assignment ) costs @xmath4 specifying the cost of assigning client @xmath5 to facility @xmath6 , and a ( integer ) parameter @xmath10 . our objective is to open a subset @xmath8 of facilities and assign each client @xmath5 to an open facility @xmath9 , so that at least @xmath10 clients are assigned to each open facility , and the total cost incurred , @xmath11 , is minimized . we use @xmath0 to denote this instance . let @xmath45 and @xmath46 denote respectively the facility - opening and assignment cost of an optimal solution to @xmath0 ; we will often refer to this solution as `` the optimal solution '' in the sequel . we sometimes abuse notation and also use @xmath45 to denote the set of open facilities in this optimal solution . let @xmath47 denote the total optimal cost . for a facility @xmath48 , let @xmath36 be the set of @xmath10 clients closest to @xmath6 , and @xmath38 denote the distance between @xmath6 and the @xmath49-closest client to @xmath6 ; that is , if @xmath50 , where @xmath51 , then @xmath52 ( for @xmath53 ) . let @xmath54 . observe that each @xmath38 is an increasing function of @xmath19 , @xmath40 , and @xmath55 . hence , @xmath56 is an increasing function of @xmath19 , @xmath57 , and @xmath58 . we now give a high - level description of our algorithm using certain building blocks that are supplied in the subsequent sections . let @xmath0 denote the instance . ( ) = 0.5ex * obtaining a bicriteria solution . * construct a instance with the same set of facilities and clients , and the same assignment costs as @xmath0 , where the opening cost of facility @xmath6 is set to @xmath59 . use the local - search algorithm for in @xcite or @xcite with scaling parameter @xmath60 to solve this instance . ( we set @xmath61 suitably to get the desired approximation ; see theorem [ mainthm ] . ) let @xmath26 be the set of facilities opened in the -solution . claim [ bicrit - lb ] and lemma [ bicrit - perf ] * transforming to a structured instance . * we use the bicriteria solution obtained above to transform @xmath0 into another structured instance @xmath25 as in @xcite . in the instance @xmath25 , we set the opening cost of each @xmath27 to zero , and we `` move '' to @xmath6 all the @xmath62 clients assigned to it , that is , all these clients are now co - located at @xmath6 . so @xmath25 consists of only the points in @xmath14 ( which forms both the facility - set and client - set ) . we will sometimes use the notation @xmath17 to indicate explicitly that @xmath25 s specification depends on the parameter @xmath19 . solve @xmath25 using the method described in section [ i2solve ] . obtain a solution to @xmath0 by opening the same facilities and making the same client assignments as in the solution to @xmath25 . [ [ analysis . ] ] analysis . + + + + + + + + + our main theorem is as follows . [ mainthm ] for any @xmath63 $ ] and @xmath60 , the above algorithm returns a solution to @xmath0 of cost at most @xmath64\ ] ] where @xmath65 . thus , we can compute efficiently a solution to @xmath0 of cost at most : ( ) = 0ex = 0ex @xmath66 , by setting @xmath67 ; @xmath68 , by letting @xmath69 be a suitable ( efficiently - computable ) function of @xmath19 , and choosing @xmath19 randomly from the interval @xmath70 $ ] according to the density function @xmath71 . the roadmap for proving theorem [ mainthm ] is as follows . we first bound the cost of the bicriteria solution obtained in step ( 1 ) in terms of @xmath72 ( lemma [ bicrit - perf ] ) . this will allow us to bound the cost of an optimal solution to @xmath25 , and argue that mapping an @xmath25-solution to a solution to @xmath0 does not increase the cost by much ( lemma [ i2mapping ] ) . the only missing ingredient is a guarantee on the cost of the solution to @xmath25 found in step ( 3 ) , which we supply in theorem [ i2thm ] , whose proof appears in section [ i2solve ] . the following claim follows from essentially the same arguments as in @xcite . [ bicrit - lb ] let @xmath73 be a _ delete - optimal _ solution to the above instance ; that is , the total -cost does not decrease by deleting any open facility of @xmath73 . then , each facility of @xmath73 serves at least @xmath15 clients . the local - search algorithms for in @xcite have the same performance guarantees and both include a delete - move as a local - search operation , so upon termination , we obtain a delete - optimal solution . observe that opening the same facilities and making the same client assignments as in the optimal solution to @xmath0 yields a solution @xmath74 to the instance constructed in step ( 1 ) of the algorithm with facility cost @xmath75 and assignment cost @xmath76 . combined with the analysis in @xcite , this yields the following . ( for simplicity , we assume that all local - search algorithms return a local optimum ; standard arguments show that dropping this assumption increases the approximation by at most a @xmath77 factor . ) [ bicrit - perf ] for a given parameter @xmath60 , executing the local - search algorithm in @xcite on the above instance returns a solution with facility cost @xmath78 and assignment cost @xmath79 satisfying @xmath80 , where each open facility serves at least @xmath15 clients . [ i2mapping ] ( i ) the ( assignment ) cost @xmath81 of an optimal solution to @xmath25 is at most @xmath82 . \(ii ) any solution to @xmath25 of cost @xmath83 yields a solution to @xmath0 of cost at most @xmath84 . [ i2thm ] for any @xmath85 , there is a @xmath86-approximation algorithm for @xmath17 , where @xmath87 . [ i2remk ] our @xmath86-approximation ratio for @xmath17 improves upon the approximation obtained in @xcite by a factor of roughly 2 _ for all @xmath19_. thus , plugging in our algorithm for solving @xmath25 in the -algorithm in @xcite ( and choosing a suitable @xmath19 ) , already yields an improved approximation factor of @xmath88 for . theorem [ mainthm ] recall that @xmath89 . note that @xmath90 for all @xmath16 $ ] ; we use this upper bound throughout below . combining theorem [ i2thm ] and the bounds in lemmas [ bicrit - perf ] and [ i2mapping ] , we obtain a solution to @xmath0 of cost at most @xmath91 @xmath92 part ( i ) follows by plugging in the values of @xmath19 and @xmath69 , and using the bound @xmath58 . let @xmath93 . for part ( ii ) , we set @xmath94 , where @xmath95}}/ \bigl(\frac{\int_{\beta}^1{\ensuremath{h}}(x)dx}{1-\beta}\bigr)\right)^{\frac{1}{4}}$ ] . plugging in this @xmath69 , we see that the cost incurred is at most @xmath96 we now bound the expected cost incurred when one chooses @xmath19 randomly according to the stated density function . this will also yield an explicit expression for @xmath97 ( as a function of @xmath98 ) , thus showing that @xmath99 ( and hence , @xmath69 ) can be computed efficiently . we note that @xmath100}}\leq\sqrt{{\ensuremath{{\textstyle \operatorname*{e}_{{}}}\left[x\right]}}}$ ] and utilize chebyshev s integral inequality ( see @xcite ) : if @xmath101 and @xmath102 are non - increasing and non - decreasing functions respectively from @xmath103 $ ] to @xmath104 , then @xmath105 . observe that @xmath106 decreases with @xmath19 . recall that @xmath93 . we have the following . @xmath107 } } & = & { \ensuremath{c}}_2(\beta):= \bigl[\frac{4}{\beta}-4 + 8\sqrt{6}\bigl(\pi/4-\tan^{-1}(\sqrt{2\beta-1})\bigr)+ 2\ln\bigl(\frac{1}{2\beta-1}\bigr)+\ln(1/\beta)\bigr]/\ln(1/\beta ) \\[0.2ex ] { \ensuremath{{\textstyle \operatorname*{e}_{{\ensuremath{\alpha}}}}\left[{\ensuremath{\alpha}}mr^*({\ensuremath{\alpha}})\right ] } } & = & m\bigl(\int_{\beta}^1 r^*(x)dx\bigr)/\ln(1/\beta ) \leq c^*/\ln(1/\beta).\end{aligned}\ ] ] finally , using chebyshev s inequality , we obtain that @xmath108 } } \leq \bigl[m\bigl(\int_{\beta}^1 r^*(x)dx\bigr ) \tfrac{\int_{\beta}^1 dx\sqrt{{\ensuremath{h}}(x)}}{1-\beta}\bigr]/\ln(1/\beta ) \leq \bigl[c^*\sqrt{{\ensuremath{c}}_3(\beta)}\bigr]/\ln(1/\beta),\ ] ] where @xmath109/(1-\beta).\ ] ] the second inequality follows since @xmath110}}}]{\sqrt{{\ensuremath{h}}({\ensuremath{\alpha}})}}$ ] . plugging in these bounds , we get that @xmath111 and the total cost is at most @xmath112\ ] ] we now describe our algorithm for solving instance @xmath17 and analyze its performance guarantee , thereby proving theorem [ i2thm ] . as mentioned earlier , one of the key differences between our algorithm and the one in @xcite is that instead of reducing @xmath25 to capacitated facility location ( ) , we solve @xmath25 by reducing it to a new problem that we call _ capacity - discounted _ ( ) . is a special case of where all facilities with non - zero opening cost are uncapacitated ( i.e. , have infinite capacity ) . perhaps surprisingly , despite this special structure , inherits the intractability of with respect to lp - based approximation guarantees : there is no known lp - relaxation for that has constant integrality gap ; appendix [ cduflgap ] shows that the natural lp - relaxation for has bad integrality gap . however , as we show in section [ cdufl - ls ] , we can obtain a simple local - search algorithm for whose approximation ratio is better than the current - best approximation for . recall that @xmath25 has only the points in @xmath26 , and there are @xmath62 co - located clients at each @xmath27 . let @xmath113 . to avoid confusion , we refer to the facilities and clients in the instance as supply points and demand points respectively . the instance created to solve @xmath25 resembles the instance created in @xcite ; the difference is that all supply points with non - zero opening costs are now uncapacitated . more precisely , at each @xmath27 , we create an uncapacitated supply point with opening cost @xmath114 , where @xmath115 is a parameter we fix later . if @xmath32 we create a second supply point at @xmath6 with capacity @xmath116 and zero opening cost . if @xmath117 , we create a demand point at @xmath6 with demand @xmath29 . let @xmath118 denote this instance ( see fig . [ i2toip ] ) . let @xmath119 denote respectively the set of uncapacitated and capacitated supply points of @xmath118 . roughly speaking , satisfying a demand point @xmath6 by non - co - located supply points translates to leaving facility @xmath6 open in the @xmath25 solution ; hence , its demand is set to @xmath29 , which is the number of additional clients it needs . conversely , opening the uncapacitated supply point at @xmath6 and supplying demand points from @xmath6 translates to closing @xmath6 in the @xmath25 solution and transferring its co - located clients to other open facilities . [ cdufl - cost ] there exists a solution to @xmath118 with facility cost @xmath120 and assignment cost @xmath121 . [ cdufl - thm ] ( i ) given any instance , one can efficiently compute a solution with facility - opening cost @xmath122 and assignment cost @xmath123 , where @xmath124 and @xmath125 are the facility and assignment costs of an arbitrary solution to the instance . \(ii ) thus , lemma [ cdufl - cost ] implies that one can compute a solution to @xmath118 with facility cost @xmath126 and assignment cost @xmath127 satisfying @xmath128 . we defer the description of the local - search algorithm for , and the proof of theorem [ cdufl - thm ] to section [ cdufl - ls ] . we first describe how to convert an @xmath118-solution to a solution to @xmath25 with a small increase in cost , and show how this combined with theorem [ cdufl - thm ] leads to the approximation bound for @xmath25 stated in theorem [ i2thm ] . an @xmath118-solution need not directly translate to an @xmath25 solution because an open supply point @xmath6 may not supply ( and hence , transfer ) exactly @xmath30 units of demand ( see , e.g. , @xmath6 and @xmath129 in fig . [ i2toip](b ) ) . since we have uncapacitated supply points , we have to consider both the cases where @xmath6 supplies more than @xmath30 demand ( a situation not encountered in @xcite ) , and less than @xmath30 demand . suppose that we are given a solution @xmath74 to @xmath118 with facility cost @xmath130 and assignment cost @xmath131 ( see fig . [ i2toip](b ) ) . again , we abuse notation and use @xmath130 to also denote the set of supply points that are opened in @xmath74 . let @xmath132 initialized to @xmath30 keep track of the number of clients at location @xmath27 . our goal is to reassign clients ( using @xmath74 as a template ) so that at the end we have @xmath133 or @xmath134 for each @xmath27 . observe that once we have determined which facilities in @xmath14 will have @xmath134 ( i.e. , the facilities to open in the @xmath25-solution ) , one can find the best way of ( re)assigning clients by solving a min - cost flow problem . however , for purposes of analysis , it will often be convenient to explicitly specify a ( possibly suboptimal ) reassignment . we may assume that : ( i ) @xmath135 ; ( ii ) if @xmath74 opens an uncapacitated supply point located at some @xmath27 with @xmath32 , then the demand assigned to the capacitated supply point at @xmath6 equals its capacity @xmath116 ; ( iii ) for each @xmath27 with @xmath31 , if the supply point at @xmath6 is open then it serves the entire demand of the co - located demand point ; and ( iv ) at most one _ uncapacitated _ supply point serves , maybe partially , the demand of any demand point ; we say that this uncapacitated supply point satisfies the demand point . we reassign clients in three phases . a * ( removing capacitated supply points ) * consider any location @xmath27 with @xmath32 . let @xmath136 and @xmath137 denote respectively the capacitated and uncapacitated supply points located at @xmath6 . if @xmath136 supplies @xmath138 units to the demand point at location @xmath129 , we transfer @xmath138 clients from location @xmath6 to @xmath129 . now if @xmath136 has @xmath139 leftover units of capacity in @xmath74 , then we `` move '' @xmath140 clients to @xmath137 ( which is not open in @xmath74 ) . we update the @xmath132s accordingly . note that this reassignment effectively gets rid of all capacitated supply points . thus , there is now exactly one uncapacitated supply point and at most one demand point at each location @xmath27 ; we refer to these simply as supply point @xmath6 and demand point @xmath6 below . * ( taking care of @xmath141 and demand points satisfied by @xmath141 ) * for each @xmath142 , if @xmath6 supplies @xmath138 units to demand point @xmath129 , we move @xmath138 clients from @xmath6 to @xmath129 , and update @xmath143 . we now have @xmath144 residual clients at each @xmath142 , which we must reduce to 0 , or increase to at least @xmath10 . we follow the same procedure as in @xcite , which we sketch below . for each @xmath142 , we include an edge @xmath145 where @xmath146 is the facility nearest to @xmath6 ( recall that @xmath147 ) . we use an arbitrary but fixed tie - breaking rule here , so each component of the resulting digraph is a directed tree rooted at either ( i ) a node @xmath148 , or ( ii ) a 2-cycle @xmath149 , where @xmath150 . we break up each component @xmath151 into a collection of smaller components as follows . essentially , we move the residual clients of supply points in the component bottom - up from the leaves up to the root , cut off the component at the first node @xmath152 that accumulates at least @xmath10 clients , and recurse on the portion of the component not containing @xmath152 . more precisely , let @xmath153 denote the subtree of @xmath151 rooted at node @xmath154 ( if @xmath152 belongs to a 2-cycle then we do not include the other node of this 2-cycle in @xmath153 ) . =0ex = 0ex if @xmath155 , or if @xmath151 is of type ( i ) and all children @xmath152 of the root satisfy @xmath156 , we leave @xmath151 unchanged . otherwise , let @xmath152 be a deepest ( i.e. , furthest from root ) node in @xmath151 such that @xmath157 . we delete the arc leaving @xmath152 . if this disconnects @xmath152 from @xmath158 , then we recurse on @xmath158 . otherwise @xmath152 must belong to the root 2-cycle of @xmath151 . let @xmath159 be the other node of this 2-cycle . if @xmath160 , we delete @xmath159 s outgoing arc ( thus splitting @xmath151 into @xmath153 and @xmath161 ) . after applying the above procedure ( to all components ) , if we are left with a component of type ( ii ) with @xmath162 , we convert it to type ( i ) by arbitrarily deleting one of the arcs of the 2-cycle . thus , at the end of this process , we have two types of components . ( ) = 0ex a tree @xmath163 rooted at a node @xmath164 : we move the @xmath132 residual clients of each non - root node @xmath165 to @xmath164 . a type-(ii ) tree @xmath163 with root @xmath166 : we must have @xmath167 . let @xmath168 be the location nearest to @xmath166 ; we move the @xmath132 residual clients of each @xmath165 to @xmath129 . update the @xmath132s to reflect the above reassignment . observe that we now have @xmath133 or @xmath134 for each @xmath142 , and each @xmath169 has @xmath170 , or is a demand point satisfied by a supply point in @xmath171 . figure [ transition](a ) shows a snapshot after steps a1 and a2 have been executed on the solution shown in fig . [ i2toip](b ) . here @xmath172 has one client left after moving clients to the bottom two facilities , which is then transferred to @xmath173 . * ( taking care of @xmath171 and demand points satisfied by @xmath171 ) * for @xmath174 , let @xmath175 be the set of demand points @xmath176 satisfied by @xmath6 , and let @xmath177 . note that @xmath178 . phase a2 may only increase @xmath179 for all @xmath5 in @xmath180 , so @xmath181 for all @xmath182 . fix @xmath174 . we reassign clients so that @xmath183 or @xmath184 for all @xmath185 , without decreasing @xmath179 for @xmath186 . applying this procedure to all supply points in @xmath171 will complete our task . define @xmath187 ( which is at most @xmath188 ) for @xmath189 . we consider two cases . =0ex = 0ex @xmath190 . for each @xmath189 , if @xmath6 supplies @xmath138 units to @xmath5 , we transfer @xmath138 clients from @xmath6 to @xmath5 . if @xmath6 is now left with less than @xmath10 residual clients , we move these residual clients to the location in @xmath175 nearest to @xmath6 . @xmath191 ( see fig . [ transition ] ) . let @xmath192 , and @xmath193 , where @xmath194 . let @xmath195 , so @xmath196 ( and @xmath197 since @xmath198 ) . note that @xmath199 is the unique index such that @xmath200 . this enables us to transfer @xmath201 clients to each @xmath202 from the locations @xmath203we do this by transferring all clients of @xmath204 ( where @xmath205 ) before considering @xmath206and be left with at most @xmath10 residual clients in @xmath207 . we argue that these residual clients are all concentrated at @xmath208 and @xmath209 , with @xmath209 having at most @xmath210 residual clients . we transfer these residual clients to @xmath211 . [ i2map ] the above algorithm returns an @xmath25-solution of cost at most @xmath212 . thus , taking @xmath74 to be the solution mentioned in part ( ii ) of theorem [ cdufl - thm ] , and @xmath213 , we obtain a solution to @xmath17 satisfying the approximation bound stated in theorem [ i2thm ] . let @xmath214 denote the solution computed for @xmath25 . for a supply point @xmath6 opened in @xmath74 , we use @xmath215 to denote the cost incurred in supplying demand from @xmath6 to the demand points satisfied by @xmath6 ; so @xmath216 . at various steps , we transfer clients between locations according to the assignment in the solution @xmath74 , and the cost incurred in this reassignment can be charged against the @xmath215s of the appropriate supply points . so the cost of phase a1 is @xmath217 , and the cost of the first step of phase a2 is @xmath218 . as in @xcite , we can bound the remaining cost of phase a2 , incurred in transferring clients according to the tree edges by @xmath219 . when we move clients up to the root of a component , we move strictly less than @xmath10 clients along any edge @xmath145 in that component , and since @xmath142 , we pay at least @xmath220 opening cost for @xmath6 . the only unaccounted cost now is the cost incurred in step ( b ) of phase a2 , where we have a tree @xmath163 rooted at @xmath166 . let @xmath168 be the location nearest to @xmath166 , and ( say ) @xmath221 . note that we have already bounded the cost in transferring clients to @xmath164 , so we only need to bound the cost incurred in transferring at most @xmath10 clients from @xmath164 to @xmath129 . this is at most @xmath222 , because @xmath166 send @xmath223 units to demand points in @xmath224 , all of which are at distance at least @xmath225 from @xmath166 . finally , consider phase a3 and some @xmath174 . if @xmath190 , then the cost incurred is at most @xmath226 ( as @xmath227 ) . now consider the case @xmath191 . for any @xmath228 and any @xmath229 , we have @xmath230 , so the cost of transferring @xmath231 clients to each @xmath202 is at most @xmath232 . observe that @xmath233 , i.e. , @xmath234 , so after this reassignment , there are less than @xmath10 residual clients in @xmath235 . by our order of transferring clients , all these residual clients are at @xmath236 ( otherwise we would have at least @xmath198 residual clients ) with at most @xmath237 of them located at @xmath209 . the cost of reassigning these residual clients is at most @xmath238 , since @xmath215 is the total cost of supplying at least @xmath239 demand to each @xmath240 . the latter expression is at most @xmath241 , since @xmath242 , @xmath243 . ) thus , the cost of @xmath214 is at most @xmath244 so if @xmath74 is the solution given by part ( ii ) of theorem [ cdufl - thm ] , the cost of @xmath214 is at most @xmath245 , and plugging in the value of @xmath115 yields the @xmath87 approximation bound stated in theorem [ i2thm ] . we now describe our local - search algorithm for , which leads to the proof of theorem [ cdufl - thm ] . let @xmath246 be the facility - set of the instance , where @xmath247 . here , @xmath248 are the uncapacitated facilities with opening costs @xmath249 , and facilities in @xmath250 have ( finite ) capacities @xmath251 and zero opening costs . let @xmath252 be the set of clients and @xmath253 be the cost of assigning client @xmath5 to facility @xmath6 . the goal is to open facilities and assign clients to open facilities ( respecting the capacities ) so as to minimize the sum of the facility - opening and client - assignment costs . we can find the best assignment of clients to open facilities by solving a network flow problem , so we focus on determining the set of facilities to open . the local - search algorithm consists of three moves : @xmath254 , @xmath255 , @xmath256 , which respectively , add a facility @xmath129 not currently open , delete a facility @xmath6 that is currently open , and swap facility @xmath6 that is open with facility @xmath129 that is not open . we note that _ all _ previous ( local - search ) algorithms for that work with non - uniform capacities use moves that are more complicated than the moves above ( and involve adding and/or deleting multiple facilities at a time ) . the algorithm repeatedly executes the best cost - improving move ( if one exists ) until no such move exists . ( as mentioned earlier , to ensure polynomial time , we only consider moves that yield significant improvement and hence terminate at an approximate local optimum ; but this has only a marginal effect on the approximation bound . ) we assume for simplicity that each client has unit demand . this is without loss of generality because , even with non - unit client - demands , one can compute the best local - search move ( and hence run the algorithm ) , and for the purposes of analysis , one can always treat a client with integer demand @xmath257 as @xmath257 co - located unit - demand clients . [ [ analysis.-1 ] ] analysis . + + + + + + + + + let @xmath258 denote a local - optimum returned by the algorithm , with facility - opening cost ( and set of open facilities ) @xmath259 and assignment cost @xmath260 . let @xmath261 be an arbitrary solution , with facility - cost ( and set of open facilities ) @xmath124 and assignment cost @xmath125 . note that we may assume that @xmath262 . for a facility @xmath6 , we use @xmath263 and @xmath264 to denote respectively the ( possibly empty ) set of clients served by @xmath6 in @xmath258 and @xmath261 . for a client @xmath5 , let @xmath265 and @xmath266 be the assignment cost of @xmath5 in @xmath258 and @xmath261 respectively . we borrow ideas from the analysis of the corresponding local - search algorithm for in @xcite , but the presence of capacitated facilities means that we need to reassign clients more carefully to analyze the change in assignment cost due to a local - search move . in particular , unlike the analysis in @xcite , where upon deletion of a facility @xmath267 we reassign only the clients currently assigned to @xmath268 , in our case ( as in the analysis of local - search algorithms for ) , we need to perform a more `` global '' reassignment ( i.e. , even clients not assigned to @xmath268 may get reassigned ) along certain ( possibly long ) paths in a suitable graph . this also means that we need to construct a suitable mapping between paths instead of the client - mapping considered in @xcite . we construct a directed graph @xmath269 with node - set @xmath270 , and arcs from @xmath6 to all clients in @xmath263 and arcs from all clients in @xmath264 to @xmath6 , for every facility @xmath6 . via standard flow - decomposition , we can decompose @xmath269 into a collection of ( simple ) paths @xmath271 , and cycles @xmath272 , so that ( i ) each facility @xmath6 appears as the starting point of @xmath273 paths , and the ending point of @xmath274 paths , and ( ii ) each client @xmath5 appears on a unique path @xmath275 or on a cycle . let @xmath276 and @xmath277 denote respectively the collection of paths starting at @xmath268 and ending at @xmath278 , and @xmath279 . for a path @xmath280 , define @xmath281 , @xmath282 , and @xmath283 . a _ shift _ along @xmath284 means that we reassign client @xmath285 to @xmath286 for each @xmath287 ( opening @xmath278 if necessary ) . note that this is feasible , since if @xmath288 , we know that @xmath289 . let @xmath290 be the increase in assignment cost due to this reassignment , which is an upper bound on the actual increase in assignment cost if @xmath278 is added to @xmath259 . also , let @xmath291 . we define a shift along a cycle @xmath292 similarly , letting @xmath293 . by considering a shift operation for every path and cycle in @xmath294 ( i.e. , suitable @xmath42 moves ) , we get the following result . [ asgncost ] for every @xmath295 and any @xmath296 , we have @xmath297 for every cycle @xmath292 , we have @xmath298 . thus , we have @xmath123 . [ [ bounding - the - opening - cost - of - facilities - in - ensuremathwidehat - fensuremathsetminusfensuremathmathrmsol . ] ] bounding the opening cost of facilities in @xmath299 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for this , we only need paths that start at facility in @xmath299 . note that all facilities in @xmath300 are _ uncapacitated_. to avoid excessive notation , for a facility @xmath301 , we now use @xmath302 to refer to the collection of paths ending in @xmath278 that start in @xmath299 . ( as before , @xmath303 is the set of paths that start at @xmath268 and end at @xmath278 . ) for any @xmath301 , we can obtain a 1 - 1 mapping @xmath304 such that if @xmath305 and @xmath306 , then ( i ) if @xmath307 , we have @xmath308 ; ( ii ) if @xmath309 , then @xmath310 ; and ( iii ) @xmath311 . say that @xmath301 is _ captured _ by @xmath268 if @xmath312 . note that @xmath278 is captured by at most one facility in @xmath259 . call a facility in @xmath299 _ good _ if it does not capture any facility , and _ bad _ otherwise . [ goodf ] for any good facility @xmath268 , we have @xmath313 consider the move @xmath314 . we upper bound the increase in reassignment cost as follows . consider @xmath315 , and let @xmath316 . ( recall that @xmath275 is the unique path containing @xmath5 . ) if @xmath317 , then we perform a shift along @xmath275 . otherwise , let @xmath318 , where @xmath319 . we reassign all clients on @xmath275 except @xmath320 as in the shift operation , and reassign @xmath320 to @xmath321 . let @xmath322 . since @xmath323 , the increase in cost by reassigning clients on @xmath275 this way is at most @xmath324 . thus , the actual increase in cost due to this move , which should be nonnegative , is at most @xmath325 . \\[-20pt]\ ] ] now consider a bad facility @xmath268 . let @xmath326 be the facilities captured by @xmath268 , and let @xmath327 be the facility nearest to @xmath268 . [ badf ] for any bad facility @xmath268 , we have @xmath328 consider the move @xmath329 . we reassign client @xmath315 as follows . let @xmath316 . @xmath330 = 0ex if @xmath317 , or @xmath331 and @xmath332 , we perform a shift along @xmath275 . the increase in assignment cost is at most @xmath333 . otherwise , let @xmath318 . if @xmath334 ( so @xmath319 ) , we reassign @xmath335 as in the shift operation , and assign @xmath320 to @xmath321 . as in the proof of lemma [ goodf ] , the increase in assignment cost is at most @xmath336 . if @xmath332 ( so @xmath337 ) , we assign @xmath5 to @xmath338 . note that @xmath339 , so the increase in assignment cost is at most @xmath340 . this gives the inequality @xmath341 \\ & + \sum_{o\notin{\ensuremath{\widehat f } } : o\neq o_s}\sum_{p\in{\ensuremath{\mathcal p}}(s , o ) : \pi(p)=p}{\ensuremath{\mathit{cost}}}(p ) . \label{swap - ineq } \end{split}\ ] ] now consider the operation @xmath342 for all @xmath343 , and apply lemma [ asgncost ] taking @xmath344 . this yields the inequality @xmath345 for each @xmath346 . adding these inequalities to , and rearranging proves the lemma . theorem [ cdufl - thm ] we prove part ( i ) ; part ( ii ) follows directly from part ( i ) and lemma [ cdufl - cost ] . lemma [ asgncost ] bounds @xmath260 . consider adding for all good facilities and for all bad facilities , and the vacuous equality @xmath347 for all @xmath348 . the lhs of the resulting inequality is precisely @xmath259 . the @xmath349s on the rhs add up to give at most @xmath124 . we claim that each path @xmath350 contributes at most @xmath351 to the rhs . thus the rhs is at most @xmath352 , and we obtain that @xmath122 . each path @xmath284 in @xmath353 appears exactly once , either in or in , and contributes @xmath354 . now consider a path @xmath355 , and let @xmath306 . note that @xmath311 . if @xmath356 , then @xmath284 appears twice in our inequality - system : once in the inequality for @xmath268 contributing @xmath354 ( due to @xmath284 ) , and once in the inequality for @xmath321 contributing @xmath357 ( due to @xmath358 ) . if @xmath359 , then @xmath309 and @xmath268 is a bad facility ; now @xmath284 appears only in ( for @xmath268 ) and contributes either @xmath354 if @xmath331 , or @xmath360 otherwise . corollary of theorem [ cdufl - thm ] : there is a @xmath361-approximation algorithm for . we take @xmath261 in part ( i ) of theorem [ cdufl - thm ] to be an optimum solution ( with cost @xmath362 ) to the instance , and scale the facility costs by @xmath363 before running the local - search algorithm . the solution returned has cost @xmath364 . setting @xmath365 yields the result . 10 v. arya , n. garg , r. khandekar , a. meyerson , k. munagala , and v. pandit . local search heuristics for @xmath366-median and facility location problems . , 33(3):544562 , 2004 . m. charikar and s. guha . improved combinatorial algorithms for facility location problems . , 34(4):803824 , 2005 . s. guha , a. meyerson , and k. munagala . a constant factor approximation for the single sink edge installation problem . , 38(6):22462442 , 2009 . s. guha , a. meyerson , and k. munagala . facility location with demand dependent costs and generalized clustering . , 2000 . s. guha , a. meyerson , and k. munagala . hierarchical placement and network design problems . in _ proceedings of the 41st annual ieee symposium on foundations of computer science _ , pages 603612 , 2000 . m. hajiaghayi , m. mahdian , and v. mirrokni . the facility location problem with general cost functions . , 42:4247 , 2003 . g. h. hardy , j. e. littlewood , and g. plya . . cambridge university press , 1952 . d. r. karger and m. minkoff . building steiner trees with incomplete global knowledge . in _ proceedings of the 41st annual ieee symposium on foundations of computer science _ , pages 613623 , 2000 . m. r. korupolu , c. g. plaxton , and r. rajaraman . analysis of a local search heuristic for facility location problems . , 37(1):146188 , 2000 . a 1.488 approximation algorithm for the uncapacitated facility location problem . in _ proceedings of the 38th international colloquium on automata languages and programming _ , pages 7788 , 2011 . a. lim , f. wang , and z. xu . a transportation problem with minimum quantity commitment . , 40(1):117129 , 2006 . m. mahdian and m. pl . universal facility location . in _ proceedings of 11th esa _ , pages 409421 , 2003 . p. mirchandani and r. francis , editors . . john wiley and sons , inc . , new york , 1990 . d. b. shmoys . the design and analysis of approximation algorithms : facility location as a case study . in s. hosten , j. lee , and r. thomas , editors . , pages 8597 , 2004 . d. b. shmoys , . tardos , and k. i. aardal . approximation algorithms for facility location problems . in _ proceedings of the 29th annual acm symposium on theory of computing _ , pages 265274 , 1997 . m. skutella . list scheduling in order of @xmath19-points on a single machine . in e. bampis , k. jansen , and c. kenyon , editors . , pages 250291 , springer - verlag , berlin , 2006 . m. sviridenko . an improved approximation algorithm for the metric uncapacitated facility location problem . in _ proceedings of 9th ipco _ , pages 240257 , 2002 . z. svitkina . lower - bounded facility location . , 6(4 ) , j. zhang , b. chen , and y. ye . a multi - exchange local search algorithm for the capacitated facility location problem . , 30:389403 , 2005 . let @xmath367 be a instance with facility - set @xmath368 ( where @xmath369 for all @xmath370 , and @xmath371 for all @xmath372 ) , and client - set @xmath252 . we consider the following lp - relaxation . we use @xmath6 to index facilities , and @xmath5 to index clients . note that we may assume that all facilities in @xmath250 are open . @xmath373 & & x_{ij } & \leq y_i & & { \ensuremath{\text { for all } } } i\in{\ensuremath{{\textstyle \widehat{\mathcal f}^u } } } , j \\ & & \sum_j x_{ij } & \leq u_i & & { \ensuremath{\text { for all } } } i\in{\ensuremath{{\textstyle \widehat{\mathcal f}^c}}}\\[-1ex ] & & x_{ij},y_i & \geq 0 & & { \ensuremath{\text { for all } } } i , j.\end{aligned}\ ] ] here @xmath374 denotes if facility @xmath6 is open , and @xmath375 denotes if client @xmath5 is assigned to facility @xmath6 . ( we assume that each client has unit demand . ) now consider the following simple instance . we have two facilities @xmath6 and @xmath129 , and @xmath376 clients , all present at the same location . facility @xmath6 is uncapacitated and has opening cost @xmath101 , and facility @xmath129 has capacity @xmath152 ( and zero opening cost ) . any solution to must open facility @xmath6 and therefore incur cost at least @xmath101 . however , there is a feasible solution to of cost @xmath377 : we set @xmath378 , and @xmath379 . thus , the integrality gap of is at least @xmath376 . we show that the local - search algorithm based on @xmath42 , @xmath43 , and @xmath44 moves that is , adding / dropping one facility ( with @xmath42 permitted only if it preserves feasibility ) , or deleting one facility and adding another has a bad _ locality gap _ , which is the maximum ratio between the cost of a locally - optimal solution and that of an ( globally ) optimal solution . consider the instance shown below with facility - set @xmath380 , and client - set @xmath381 , where the @xmath382s are disjoint sets of size @xmath10 . the facility - opening costs are as follows : @xmath383 and @xmath384 for each @xmath385 . for each @xmath386 and each client @xmath387 , we have @xmath388 . all other distances are defined by taking the metric completion with respect to these @xmath7s . one can verify that the solution @xmath74 which opens the facilities @xmath389 is a local optimum . the cost of this solution is @xmath390 . however , the optimal solution opens facility @xmath391 , and incurs a total cost of @xmath392 . thus , the locality gap is at least @xmath393 . we can modify this example to show that the locality gap remains bad , even if aim for a bicriteria solution and consider an @xmath42 move to be permissible if every open facility can be assigned at least @xmath15 clients . the only change is that each set @xmath382 now has @xmath15 clients : @xmath74 is still a local optimum , and the locality gap is therefore at least @xmath394 . even in the setting where all facilities have zero opening cost ( as in the @xmath25 instance ) , we can construct bad examples for local - search based on @xmath42 , @xmath43 , and @xmath44 moves . for simplicity , first suppose that @xmath395 . consider a cycle with @xmath396 nodes , which are labeled @xmath397 . we have @xmath398 facilities @xmath399 , and @xmath398 clients @xmath400 ( see fig . [ locgapf0 ] ) . we define the following distances . the solution @xmath74 which opens facilities @xmath405 is a local optimum : no @xmath42 move is feasible , and it is easy to see that no @xmath43 move improves the cost . consider a swap move , which we may assume is of the form @xmath406 by symmetry . the new client - assignment will not necessarily assign the clients @xmath407 and @xmath408 ( which were previously assigned to @xmath409 ) to @xmath410 . however , the intuition is that the long cycle will lead to a large increase in assignment cost . the optimal way of reassigning clients is to assign @xmath411 to @xmath412 , assign @xmath413 to @xmath414 for @xmath415 ( which is empty if @xmath416 ) , and assign @xmath417 to @xmath414 for @xmath418 ( which is empty if @xmath419 ) . the cost increase due to this reassignment is @xmath420 . thus , @xmath74 is a local optimum . the cost of @xmath74 is @xmath421 . however , the optimal solution opens facilities @xmath422 , and has a total cost of @xmath398 . so this instance shows a locality gap of @xmath366 , and since @xmath366 can be made arbitrarily large , this shows an unbounded locality gap . the above example can be extended to all values of @xmath10 . for each @xmath10 , let @xmath423 be an @xmath10-regular bipartite graph with vertex set @xmath424 with a large girth @xmath163 . we use @xmath423 to construct the following instance . the set of facilities is @xmath425 . for each edge @xmath426 in @xmath423 , we create a client @xmath427 with @xmath428 and @xmath429 . as before , one can argue that the solution @xmath74 that opens facilities @xmath430 is a local optimum . the cost of this solution is @xmath431 , whereas the solution that opens facilities @xmath432 has total cost of @xmath433 . so the locality gap is @xmath163 , which can be made arbitrarily large .

we study scalar , electromagnetic , axial and polar gravitational perturbations of the four - dimensional reissner - nordstrm - like black holes with a _ tidal charge _ in the randall - sundrum braneworld in the first approximation when the tidal perturbations are not taken into account . the quasinormal modes of these perturbations have been studied in both normal and eikonal regimes . calculations have shown that the black holes on the randall - sundrum brane are stable against all kinds of perturbations . moreover , we determine the greybody factor , giving transmission and reflection of the scattered waves through the effective potentials . it has been shown that the scalar perturbative fields are the most favorite to reflect the wave as compared to the other fields . with increasing value of the tidal charge , the ability of the all perturbative potentials to reflect the waves decreases . our calculations in low- and high - frequency regimes have shown that black holes on the braneworld always have a bigger absorption cross section of massless scalar waves than the schwarzschild and standard reissner - nordstrm black holes .

1605.02058

an early attempt to unify theories of gravitation and electromagnetism was realized by kaluza @xcite in five - dimensional theory based on classical general relativity . that was later extended by klein @xcite with quantum interpretations . the idea of extra dimensions played a key role in development of an existing new fundamental theory of physics , superstring theory ( m - theory ) which requires the spacetime to have ten ( eleven ) dimensions . braneworld models were proposed in order to tackle the hierarchy problem . the question why there is such a large gap between the electroweak scale at @xmath0 tev and the planck scale at @xmath1 tev has been addressed in @xcite . the first string realization of low - scale gravity and braneworld models were given , pointing out the motivation of tev strings from the stabilization of mass hierarchy and the graviton emission in the bulk @xcite . randall and sundrum proposed their braneworld models @xcite where the hierarchy problem can be addressed , as the large size of the extra dimension plays crucial role to fill the gap between the electroweak and planck effective scales . according to the randall - sundrum model , our universe is a three - brane ( domain wall ) embedded in five - dimensional bulk spacetime ; one extra dimension is large , and the bulk is a slice of the anti - de sitter ( ads ) spacetime , i.e. , spacetime with a negative cosmological constant . later , this model was further extended by shiromizu et al . reissner - nordstrm - like static , a spherically symmetric black hole solution with a tidal charge parameter ( instead of electric charge ) , localized on a three - brane in five - dimensional gravity in the randall - sundrum model was obtained , without finding the bulk metric , by dadhich et al . so far , several black hole solutions @xcite , wormholes @xcite , and nonuniform stars and gravastars @xcite in the randall - sundrum model have been obtained ( see reviews @xcite and references therein . ) . the discovery of the braneworld model opens up a new window to test modified general relativity . thus , up to now , many physical effects related to test the particle motion and geodesic structure of braneworld black holes @xcite and neutron and compact stars @xcite have been studied . one of the most important properties of black holes is their characteristic oscillations , which are called _ quasinormal modes _ that carry information about them . the quasinormal modes determine , e.g. , ringdown of gravitational waves created while a black hole is born . they are characterized by black hole parameters being dependent on initial perturbations . they have complex frequencies real and imaginary parts of the quasinormal frequencies represent frequencies of the real oscillations and their dissipation rate , respectively . recently , the interferometric ligo detectors have measured the first ever gravitational wave signals from the merging of two black holes @xcite . later , it has been shown that current precision of the experiment leaves some possibilities for alternative theories of gravity @xcite . cardoso et al . @xcite stated that ringdown waveforms indicate the existence of the stable light rings regardless of existence of the horizons . according to @xcite , the ringdown part of the gw150914 signal has excluded formation of gravastar by the merger of two rotating compact objects . so far , characteristic ringdown signals ( quasinormal modes ) of the various black holes have been studied in great detail by a number of authors within perturbation theory@xcite . since there is still room open for alternative theories of gravity , we aim to study in this paper perturbations of the black holes localized in the randall - sundrum braneworld . we concentrate on their stabilities , scattering effects and quasinormal modes . after this work was almost completed , electromagnetic perturbations of current braneworld solution by molina et al . @xcite appeared in arxiv . despite the fact that some our results of the quasinormal frequencies of the electromagnetic perturbations are repeating the results of @xcite , we keep them in order to compare them with profiles of scalar and gravitational perturbations . moreover , we study scattering and absorption problem in the electromagnetic case as well . the paper is organized as follows : in section [ sec - spacetime ] we briefly describe the spacetime geometry and its main properties . in section [ sec - pert - eqs ] the equations for scalar , electromagnetic , and axial and polar gravitational perturbations are introduced . we give some numerical results , such as quasinormal frequencies obtained by the sixth order wkb method in low and large multipole number limits , and stability analysis in section [ results ] . in section [ scattering ] classical scattering problem is solved by using the standard @xmath2-matrix . in section [ absorption ] we study absorption cross section of massless scalar waves by the braneworld black hole in comparison with the schwarzschild and reissner - nordstrm black holes . finally , we present some concluding remarks in section [ summary ] . throughout the paper we use the geometric system of units @xmath3 and a spacelike signature @xmath4 . we focus on the static , spherically symmetric black hole geometry localized on a braneworld , described by the line element @xcite [ metric ] ds^2=-f(r)dt^2++r^2d^2+r^2 ^ 2d^2 , where [ lapse ] f(r)=1-+ , and @xmath5 is a constant parameter . one can see from ( [ lapse ] ) that if @xmath6 the spacetime metric ( [ metric ] ) reduces to the schwarzschild one . moreover , for @xmath7 the spacetime metric ( [ metric ] ) is identical to the reissner - nordstrm black hole one with two horizons which both are smaller than the one of the schwarzschild black hole ( @xmath8 ) . however , in the braneworld @xmath5 can have negative value ( @xmath9 ) too . in this case , black hole has only one horizon which is always bigger than the one of the schwarzschild black hole [ horizon ] r_+=m+ > 2 m . in this paper we consider the latter case , @xmath9 . therefore , in order to guarantee its negativity we introduce new notation @xmath10 , where @xmath11 is always positive , @xmath12 , and is called _ tidal charge ( brane tension ) parameter _ @xcite . black hole entropy is determined by horizon area as [ entropy ] s==r_+^2=(m+)^2 . where @xmath13 is horizon area . the hawking temperature is given by [ temperature ] t= . as one see from ( [ temperature ] ) , unlike in the case of the reissner - nordstrm black hole , that the nonvanishing tidal charge parameter leads to an increase of the black hole entropy and decrease of the hawking temperature . by considering perturbation terms dependent on time as @xmath14 and separating angular and radial perturbations by introducing the tortoise coordinate @xmath15 , we obtain a schrdinger - like wave equation as follows : [ ax-5 ] ( + ^2)z_s = v_s z_s , where @xmath16 represent scalar and electromagnetic perturbations , respectively . for the massive scalar perturbations with mass @xmath17 of the black hole on the brane we have the potential [ scalar - potential ] v_0&=&f&=&f , where prime ( `` @xmath18 '' ) denotes the derivative with respect to @xmath19 . here , @xmath20 is the multipole number which represents the spherical harmonic index and takes only nonnegative integers for scalar perturbations . for the electromagnetic perturbations the potential reads [ em - potential ] v_1=f where the multipole number @xmath20 for electromagnetic perturbations takes only natural numbers . the potentials for the scalar ( [ scalar - potential ] ) and electromagnetic ( [ em - potential ] ) perturbations can be written in a compact form as [ scalar - em ] v_s = f . where @xmath21 and @xmath22 correspond to the scalar and electromagnetic perturbations , respectively . of course , in the case of the electromagnetic perturbations in the standard situations we assume massless photons , i.e. , @xmath23 . it is well known that the simplest way of studying the gravitational perturbations around black holes is to introduce the first order perturbations . if the considered black hole is not a vacuum solution of the einstein equations , the perturbation equations are governed by the equation @xmath24 . in the case of a black hole localized on a three - brane in five - dimensional gravity , @xmath25 and @xmath26 are the ricci tensor and the projection of the five - dimensional weyl tensor on the brane , respectively . the gravitational perturbations of the higher dimensional black holes can be easily obtained by using the master equations presented in @xcite . however , the black hole solution in the brane @xcite was found without finding the 5-dimensional bulk metric . therefore , we adopt the simplifying assumption @xmath27 that can be justified at least in a region where the perturbation energy does not exceed the threshold of the kaluza - klein massive modes @xcite . moreover , the fact that the gravitational perturbative field can not travel deep into the bulk @xcite supports the above assumption . in this subsection we present a general formalism for gravitational perturbations in a static , spherically symmetric background , following chandrasekhar s method @xcite . with the notation @xmath28 , the spherically symmetric , time - independent metric with small perturbations can be written in the form [ pert - metric ] ds^2= -e^2(+)dt^2&+&e^2(+)(d - q_0 dt - q_2 dr - q_3 d)^2&+&e^2(_2+_2 ) dr^2 + e^2(_3+_3 ) d^2 , where & & e^2=f(r ) , e^2=r^2 ^ 2 , & & e^-2_2=f(r ) , e^2_3=r^2 , and @xmath29 , @xmath30 , @xmath31 , @xmath32 , @xmath33 , @xmath34 and @xmath35 are nonvanishing small perturbation terms . the first three small quantities ( @xmath29 , @xmath30 , @xmath31 ) characterize axial tensor perturbations with odd parity , while the small quantities ( @xmath32 , @xmath33 , @xmath34 , @xmath35 ) correspond to the polar tensor perturbations with even parity . below we briefly present these perturbations for the braneworld black hole @xcite . the axial perturbations with perturbation terms @xmath29 , @xmath30 and @xmath31 are governed by the relation [ ax-11 ] r_=0 . we write the linearized axial gravitational perturbation equations from ( [ ax-11 ] ) in the form [ ax-01 ] ( r^2f q_23 ^ 3)_,3=-r^4 ^ 3q_02,0 , ( r_12=0 ) , [ ax-02 ] ( r^2f q_23 ^ 3)_,2=q_03,0 , ( r_13=0 ) , where @xmath36 . by considering the perturbation terms depending on time as @xmath14 and introducing the new notation @xmath37 by the relation [ ax-2 ] q = r^2f q_23 ^ 3=r^2f ( q_2,3-q_3,2)^3 , we can write eqs . ( [ ax-01 ] ) and ( [ ax-02 ] ) as follows : [ ax-3.1 ] = -i q_0 , 2+^2q_2 , [ ax-3.2 ] = iq_0,3+^2q_3 . in order to eliminate @xmath38 from these equations , we differentiate eqs . ( [ ax-3.1 ] ) and ( [ ax-3.2 ] ) with respect to the coordinates @xmath39 ( @xmath40 ) and @xmath19 ( @xmath41 ) , respectively . then , we obtain [ ax-4 ] r^4()&+&^3 ( ) & & + ^2q=0 . ( [ ax-4 ] ) can be separated to the radial and angular variable differential equations by choosing the function @xmath37 as @xmath42 , where @xmath43 are the gegenbauer polynomials and related to the legandre function @xmath44 as [ gegenbauer ] c_l+2 ^ -3/2()=(p_l ,,- p_l,)^2 , by substituting @xmath45 and introducing the tortoise coordinate @xmath46 , we obtain a schrdinger - like wave equation [ ax-1 ] ( + ^2)z_2^()=v_2^()z_2^ ( ) , where @xmath47 and @xmath48 denote axial and polar gravitational perturbations , respectively . potential for the axial gravitational perturbations reads [ ax - potential1 ] v_2^(-)=f , where @xmath49 and @xmath50 are energy density and radial pressure of the fluid , respectively , with @xmath51 . here , @xmath52 . then , eq . ( [ ax - potential1 ] ) for the axial gravitational perturbations of the black hole on the brane takes the form : [ ax - potential2 ] v_2^(-)=f . in the limit @xmath53 this potential coincides with the regge - wheeler potential for the schwarzschild black hole @xcite . in the case of polar perturbations with even parity , @xmath32 , @xmath33 , @xmath34 and @xmath35 in the metric functions are considered nonvanishing . polar perturbations are examined by vanishing the following ricci and einstein tensors : @xmath54 , @xmath55 , @xmath56 , @xmath57 and @xmath58 ( see ref . @xcite for details ) . moreover , in order to separate @xmath19 and @xmath39 variables one may introduce the following notations : [ polar-1 ] & & = n(r)p_l ( ) , + & & _ 2=l(r)p_l ( ) , + & & _ 3=[t(r)p_l+v(r)p_l , , ] , + & & = [ t(r)p_l+v(r)p_l , ] . then , we obtain the following perturbation equations in terms of the new radial functions @xmath59 , @xmath60 , @xmath61 and @xmath62 in the form [ polar-2 ] & & n=a n + b l + c x , + & & l=(a-+)n+(b--)l+c x , + & & x=-(a-+)n-(b+-2)l&&-(c+-)x , where prime ( `` @xmath18 '' ) denotes the derivative with respect to @xmath19 and @xmath63 and @xmath64 . furthermore , [ polar - coefs ] & & a= , + & & b=--+++ , + & & c=-+++ after making some simplifications , we obtain the schrdinger - like equation for @xmath65 family in ( [ ax-1 ] ) , where the wave function @xmath66 reads [ polar-4 ] z_2^(+)=rv_2^(+)-k(l+nv_2^(+ ) ) , with [ polar-5 ] k= . the potential for the polar perturbations reads [ polar - potential ] v_2^(+)= & & , where [ polar-1 ] k= , and @xmath67 . for the gravitational perturbations , the multipole number @xmath20 takes all natural numbers starting from @xmath68 . for the braneworld black hole the potential of the polar gravitational perturbations reads [ polar - potential2 ] v_2^(+)= , where [ coefr ] a=9m^3 + 9m^2r+3m^2r^2+^2(+1)r^3 , [ coefq ] b&=&8q^*2 + 10q^*(3m+r)r&+&r^2[36m^2 + 2m(11 - 3)r+(-6)r^2 ] . in the limiting case @xmath53 one recovers the zerilli potential for the polar gravitational perturbations of the schwarzschild black hole @xcite . dependence of effective potentials for the scalar ( red ) , electromagnetic ( cyan ) , axial ( blue ) and polar ( black ) gravitational perturbations on the tortoise coordinate @xmath69 ( @xmath70 ) for the fixed values of the tidal charge parameter @xmath71 and multipole number @xmath72.,scaledwidth=43.0% ] one can see from fig . [ fig_veff ] that the effective potential for the scalar perturbations is dominant in comparison with the other ones , while axial and polar gravitational perturbations have effective potentials with the smallest height among them . it is well known that an increase in the value of the multipole number @xmath20 increases the height of the potential @xcite . the schrdinger - like wave equations ( [ ax-5 ] ) and ( [ ax-1 ] ) are solved as usual , imposing appropriate boundary conditions . considering the wave is purely incoming at the event horizon and outgoing at the spatial infinity : [ boundary_condition ] & & z(r)~e^-ix , at x0 ( rr_+ ) , & & z(r)~e^ix , at x ( r ) , [ cols="<,<,^,^,^,^,^",options="header " , ] solving the schrdinger - like wave equations with the effective potentials ( [ scalar - potential ] ) , ( [ em - potential ] ) , ( [ ax - potential2 ] ) and ( [ polar - potential2 ] ) analytically is impossible . therefore , we use the wkb method that was applied for the first time for calculation of quasinormal modes of black holes by schutz and will @xcite . afterwards , to increase its accuracy of the method was extended up to the third order by iyer and will @xcite and up to the sixth order by konoplya @xcite . the sixth order wkb method for solving the schrdinger - like wave equation governing the quasinormal modes of black holes implies the relation [ wkb ] + _ j=2 ^ 6_j = n+ , where @xmath73 is the value of the radial coordinate @xmath19 corresponding to the maximum of the potential @xmath62 , @xmath74 is the order of the wkb corrections and @xmath75 is the correction term corresponding to the @xmath74th order . one can find the expressions of @xmath75 in @xcite . the prime @xmath76 stands for the derivative with respect to the tortoise coordinate @xmath69 , and @xmath77 is the overtone number . in tables i and ii quasinormal frequencies of the scalar , electromagnetic , axial and polar gravitational perturbations of the braneworld black hole are given . one can see from the numerical results presented in tabs i , ii and fig . [ fig_qnm ] that an increase in the value of the tidal charge parameter decreases the frequency of the real oscillations and the damping rate . moreover , with increasing multipole number @xmath20 the frequency of the oscillations increases while the damping rate decreases . this means that quasinormal frequencies with higher multipole numbers are longer lived . however , from the general characteristics of the overtone number @xmath77 , we know that an increase of the value of the overtone number situation completely changes to the contrary , i.e. it implies the wave with lower oscillation frequency and higher damping rate . _ * evolution of the wave . * _ the temporal evolution of the gravitational perturbations is governed by the equation [ bwe1 ] -+=v . by turning into the null coordinates @xmath78 and @xmath79 , one can rewrite the wave equation ( [ bwe1 ] ) in the form [ bwe2 ] -4=v(x ) , discretization of eq . ( [ bwe2 ] ) allows us to calculate the values of wave function @xmath80 @xcite . we follow the method for numerical calculations presented in @xcite . semilog graphs of the absolute value of the wave function for the axial gravitational perturbations of the braneworld black holes evaluated at @xmath81 and @xmath82 . top panel : different values of the tidal charge parameter @xmath83 ( blue curve ) , and @xmath84 ( red curve ) in comparison with the schwarzschild one @xmath53 ( black curve ) in the fundamental mode @xmath72 . bottom panel : different values of the multipole number @xmath72 ( black curve ) , @xmath85 ( blue curve ) , and @xmath86 ( red curve).,title="fig:",scaledwidth=43.0% ] semilog graphs of the absolute value of the wave function for the axial gravitational perturbations of the braneworld black holes evaluated at @xmath81 and @xmath82 . top panel : different values of the tidal charge parameter @xmath83 ( blue curve ) , and @xmath84 ( red curve ) in comparison with the schwarzschild one @xmath53 ( black curve ) in the fundamental mode @xmath72 . bottom panel : different values of the multipole number @xmath72 ( black curve ) , @xmath85 ( blue curve ) , and @xmath86 ( red curve).,title="fig:",scaledwidth=43.0% ] in fig . [ time_domain ] the time - domain profiles for the evolution of the axial gravitational perturbations in the fundamental mode are presented for the braneworld black hole with mass @xmath82 at the radius @xmath81 in comparison with the schwarzschild black hole . one can see in these figures a monotonic decay of the signal ; the quasinormal mode signal dominates after @xmath87 . from the top panel of fig . [ time_domain ] we see that with increasing the tidal charge parameter damping rate of the signal decreases and the duration of dominance of the quasinormal mode signal decreases relative to the schwarzschild black hole . moreover , the bottom panel shows that an increasing multipole number @xmath20 increases the damping and the real oscillations time scales . * _ massive scalar field . _ * in the case of a massive scalar field around the black hole the region of the values of the parameters allowing for occurrence of the quasinormal modes is restricted @xcite . for small enough values of the scalar field mass parameter @xmath17 , the effective potential @xmath62 is in the form of a barrier . however , with increasing the value of @xmath17 , the asymptotical value of @xmath62 ( namely @xmath88 ) increases more rapidly than the peak of the potential [ because of @xmath89 . consequently , in the case of @xmath90 , the effective potential is not in the barrier form anymore . therefore , in that case quasinormal modes do not occur . below in figs . [ mass_limit ] these region of parameters are shown . possible values of the scalar field mass @xmath17 that give the limit of existence of quasinormal modes ( qnms ) in the field of black hole on the brane . shaded region represents part of parameter space where the qnm can occur . in the white region there are no qnm . top panel illustrates for the mass of the scalar field @xmath17 versus multipole number @xmath20 for the fixed value of the tidal charge parameter @xmath71 . in the bottom panel also limiting values of the scalar field mass @xmath17 enabling existence of qnms in the field of the black hole in the braneworld are given in dependence on the @xmath91 in the fundamental mode @xmath92.,title="fig:",scaledwidth=43.0% ] possible values of the scalar field mass @xmath17 that give the limit of existence of quasinormal modes ( qnms ) in the field of black hole on the brane . shaded region represents part of parameter space where the qnm can occur . in the white region there are no qnm . top panel illustrates for the mass of the scalar field @xmath17 versus multipole number @xmath20 for the fixed value of the tidal charge parameter @xmath71 . in the bottom panel also limiting values of the scalar field mass @xmath17 enabling existence of qnms in the field of the black hole in the braneworld are given in dependence on the @xmath91 in the fundamental mode @xmath92.,title="fig:",scaledwidth=43.0% ] contrary to the case of the standard reissner - nordstrm black hole , in the case of the braneworld black hole with increasing value of the tidal charge parameter , maximum possible value of the mass of the scalar field decreases . this is connected with the increase of the horizon radius . from fig . [ mass_limit ] one can see that with increasing multipole number @xmath20 the range of the possible values of the mass parameter increases . it is known that the wkb method has very good accuracy for large values of the multipole number @xmath20 . in the large multipole number limit one can solve the wave equation analytically by using the first order wkb approximation . to do this , we expand the expression ( [ wkb ] ) in powers of @xmath93 , and quasinormal frequencies tend to finite values . interestingly , for the both scalar and electromagnetic perturbations , as well as axial and polar gravitational perturbations , eikonal limits are the same and read [ eikonal1 ] & & _ r(l+ ) , + & & _ i-^1/2(n+ ) , where [ max_r ] r_0 . therefore , in the large multipole number @xmath20 case it is almost impossible to distinguish the types of perturbations . notice correspondence of @xmath73 and the radius of photon circular orbit @xcite dependence of the real ( black curve ) and imaginary ( blue curve ) parts of the quasinormal frequencies on the tidal charge parameter @xmath91 in the large multipole number @xmath20 limit.,scaledwidth=43.0% ] one can see from fig . [ large_l ] that in the limit of large multipole number @xmath20 real and imaginary parts of the quasinormal frequencies tend to finite values . since there is no upper limit on the value of the tidal charge @xmath91 , in the eikonal limit for large values of the tidal charge parameter quasinormal frequencies tend to zero . calculations have shown that imaginary part of the quasinormal frequencies is always negative , @xmath94 . moreover , effective potentials ( [ scalar - potential ] ) , ( [ em - potential ] ) , ( [ ax - potential2 ] ) and ( [ polar - potential2 ] ) are always positive - definite . furthermore , from fig . [ time_domain ] of time - domain profiles for the evolution of axial gravitational perturbations one can see that there is no indication of instability . one of the most important results of this paper is that in calculations made by the wkb method and the form of the wave in time domain profile , we have not observed any unstable mode . from these we can conclude that black holes on the brane ( [ metric ] ) are stable against scalar , electromagnetic and gravitational perturbations . in this section we study the greybody factor for the braneworld black holes . the greybody factor is understood as the probability for an outgoing wave in @xmath95-mode to reach infinity or , equivalently , the absorption probability for an incoming wave in @xmath95-mode to be absorbed by the black hole @xcite . in other words , the greybody factor is the tunneling probability of the wave through the barrier determined by the effective potential in the given black hole spacetime . in order to calculate the greybody factor , we write the boundary conditions for the ingoing and outgoing waves that are the solution of the schrdinger - like wave equations in the asymptotic form @xmath96 where @xmath97 and @xmath98 are the reflection and transmission coefficients , respectively . we write the above given boundary conditions ( [ svi1 ] ) for @xmath99 : @xmath100 from these two boundary conditions we can write the expression for the flux @xcite @xmath101.\end{aligned}\ ] ] using the boundary conditions ( [ svi1 ] ) and ( [ svi2 ] ) , we find the relation between the reflection and transmission coefficients from the flux conservation in the form @xmath102 if we consider @xmath95 real ( @xmath103 ) , we can write that @xmath104 , @xmath105 and @xmath106 . then the relation ( [ svi4 ] ) can be written in the form latexmath:[\[\begin{aligned } \label{svi5 } @xmath95 larger than the height of the potential barrier @xmath108 will not be ( classically ) reflected by the barrier . therefore , in this case reflection coefficient is close to zero . the incoming wave with smaller frequency than the height of the potential barrier , @xmath109 , is reflected partly , while the rest part is transmitted through the barrier by tunneling effect , depending on the values of @xmath95 and @xmath108 . this is why this case is more interesting to study . first , we consider the small frequency case when @xmath110 is much less than the height of the potential barrier @xmath108 ( @xmath111 ) . for such small values of @xmath110 , the wkb approximation has not high accuracy because of the large distance between two turning points . in this case transmission coefficient is given by the well - known formula @xmath112 and the reflection coefficient @xmath113 the radii @xmath114 and @xmath115 are the classical turning points that are the solutions of the equation @xmath116 . at the small values of @xmath95 , transmission coefficient is close to zero while the reflection coefficient is close to one @xcite . the second limit case is that the value of @xmath110 is of the same order as the maximum height of the potential barrier @xmath108 , i.e. @xmath117 . here , we can calculate the greybody factor by using the sixth order wkb approximation because of the small distance between the turning points @xcite . then we arrive at @xmath118 where @xmath119 is given by the formula of the sixth order wkb method ( [ wkb ] ) . from ( [ svi5 ] ) with ( [ svi8 ] ) one can write the expression for the transmission coefficient as @xmath120 , we give the reflection coefficients for the scalar , electromagnetic and axial gravitational fields in the mode @xmath72 for the braneworld black holes . reflection and transmission coefficients for the axial and polar gravitational perturbations are almost the same , i.e. it is impossible to distinguish difference in figures . one can see from fig . [ gbf ] that for the scalar perturbative fields , probability of the wave reflection by the potential barrier is larger than for the electromagnetic and gravitational ones . moreover , unlike the case of the standard reissner - nordstrm black hole , the presence of the non - vanishing tidal charge parameter decreases the reflection ability of the potential barrier , while it increases the transmission probabilities in comparison with the reissner - nordstrm and schwarzschild ones . absorption and scattering of test particles and fields in the black hole backgrounds are very important since they are relevant for observations related to accretion processes , deflection of light , and etc . so far , absorption and scattering properties of waves by various static spherically symmetric @xcite and axially symmetric black holes @xcite were studied . here we extend these studies for the analysis of the absorption cross sections of massless scalar field by braneworld black holes . the massless scalar field is defined by the equation ( [ ax-5 ] ) with potential ( [ scalar - potential ] ) . in order to demonstrate the effect of the tidal charge we compare the results of the braneworld black holes with the reissner - nordstrm black holes calculated in @xcite . * _ partial wave approach . _ * in the partial wave approach we should solve the schrdinger - like wave equation ( [ ax-5 ] ) with the scalar field potential ( [ scalar - potential ] ) by using the boundary conditions ( [ svi1 ] ) . for such scalar fields the total absorption cross section is defined by the sum of the partial absorption cross sections of the planar massless scalar waves [ abs1 ] _ abs()=_l=0^+_l ( ) , where @xmath121 represents the partial absorption cross section corresponding to the wave with angular momentum @xmath20 and energy @xmath95 , given by [ abs2 ] _ l()=(2l+1)|t_l()|^2 . @xmath122 is the transmission ( absorption ) coefficient for the wave with energy @xmath95 and angular momentum @xmath20 . it is known from the classical mechanics that @xmath123 is the absorption probability . by using the higher order wkb method one can write ( [ abs2 ] ) as [ abs3 ] _ l()=(2l+1 ) , where @xmath119 is given by the expression of the higher order wkb formulas ( [ wkb ] ) . * _ high - energy limit . _ * in the high - energy scale , the wavelength is almost negligible relative to the horizon scale of the black hole . therefore , in this regime massless scalar waves propagate along the null geodesics @xcite . therefore , one may use the classical capture cross section of the light by the black holes as the geometric cross section of the light rays . to do so , we consider the motion of the massless particle ( photon ) around the black hole confined to the equatorial plane ( @xmath124 ) , [ abs4 ] ^2=e^2-v_eff , v_eff = f , where @xmath125 and @xmath126 are energy and angular momentum of the photon , respectively , which are conserved quantities due to symmetries of the spacetime . however , for the photon motion only the impact parameter @xmath127 is relevant @xcite . the geometric cross section of the light rays is given by the expression @xmath128 , where @xmath129 is the critical impact parameter of the light defined by the ratio of angular momentum and energy of the photon moving along the circular photon orbit as @xmath130 . therefore , [ abs5 ] _ geo = b_ps^2=2 , where `` @xmath18 '' means the derivative with respect to @xmath19 , and @xmath131 is the radius of the photon sphere which is found from the equation [ abs6 ] 2f - rf=0 . the radius of the photon sphere for the braneworld black hole is given by the expression @xcite [ abs7 ] r_ps= the loci of the event horizons ( dashed ) and photon spheres ( solid ) of the black hole in braneworld ( black ) and reissner - nordstrm black holes ( blue).,scaledwidth=43.0% ] in fig . [ ps ] the loci of the event horizons and photon spheres of the braneworld and reissner - nordstrm black holes are shown for comparison . one can see from fig . [ ps ] that event horizon and photon spheres of the braneworld black holes never vanish for any values of the tidal charge parameter . by inserting ( [ abs7 ] ) into ( [ abs5 ] ) we obtain expression for the geometric absorption cross section of the massless scalar wave by the braneworld black hole in the form [ abs8 ] _ geo= . the geometric absorption cross section by black hole in braneworld ( black ) and the reissner - nordstrm black hole ( blue).,scaledwidth=43.0% ] in fig . [ geo_cs ] the geometric absorption cross section of the massless scalar wave by the braneworld black hole and the reissner - nordstrm black holes are shown . one can see that unlike the case of the reissner - nordstrm black hole , with increasing value of the tidal charge parameter the geometric absorption cross section of the massless scalar wave increases . in the paper from @xcite it has been shown that there are fluctuations ( regular oscillations ) of the high - energy ( frequency ) absorption cross section around the limiting value of the geometric cross section ; this is a universal property of the absorption cross section of scalar field in the high - energy regime in the field of spherically symmetric black holes . this oscillatory part of the absorption cross section of the massless scalar waves can be written as [ abs9 ] _ osc()=-8b_c_geoe^-b_c(2b_c ) , where @xmath132 and @xmath133 is the lyapunov exponent used for analysis of the instability of the null geodesics @xcite . the total absorption cross section of the massless scalar waves is the sum of the geometric ( [ abs5 ] ) and oscillatory ( [ abs9 ] ) cross sections [ abs10 ] _ abs_geo+_osc . absorption cross section of massless scalar waves by the braneworld for the values of the tidal charge parameter @xmath71 ( cyan ) , @xmath134 ( red ) and @xmath135 ( magenta ) . the schwarzschild ( black ) and reissner - nordstrm with electric charge @xmath136 ( blue ) black holes are shown for comparison , where horizontal lines represent the geometric absorption cross sections.,scaledwidth=43.0% ] in fig . [ abs_cs ] we show the total absorption cross sections of the massless scalar waves by the braneworld black holes for several values of the tidal charge parameter . for comparison , total absorption cross section obtained for the schwarzschild and reissner - nordstrm black holes are also shown . one can see from fig . [ abs_cs ] that with increasing tidal charge parameter the absorption cross section increases . by comparing the absorption cross sections to those related to the the schwarzschild and reissner - nordstrm black holes , we can conclude @xmath137 . * _ low - energy limit . _ * in the papers from @xcite it has been shown that for small values of the massless scalar wave frequency the absorption cross section by a black hole tends to the horizon area of the black hole ( @xmath138 ) . we know from ( [ horizon ] ) that the horizon of the braneworld black hole always increases with increasing tidal charge . then , in the low - frequency regime the absorption cross section of the massless scalar waves also increases appropriately to the increasing of the horizon . * _ scalar particle emission by hawking radiation . _ * as already stated by hawking @xcite , the particle absorption cross section is very relevant to the particle emission by black holes @xcite . the particle emission rate by hawking radiation is defined by the number of emitted particles by the black hole per unit time and per unit frequency . one should take into consideration also the spin of the emitted particles . here , we are considering a massless scalar field . it is well known that scalar particles have spin - zero ( bosons ) and their emission probabilities are defined by the bose - einstein distribution . then , the particle emission rate ( or emission spectrum ) by hawking scalar radiation reads [ hawking_emission ] = _ l=0^+= where @xmath139 is the hawking temperature given by ( [ temperature ] ) . in fig . [ emission ] we represent the massless scalar particle emission spectrum by the braneworld black holes in comparison with the reissner - nordstrm and schwarzschild black holes . massless scalar particle emission spectrum by the braneworld for the values of the tidal charge parameter @xmath71 ( cyan ) , @xmath134 ( red ) , and @xmath135 ( magenta ) in comparison with the schwarzschild ( black ) and reissner - nordstrm with electric charge @xmath136 ( blue ) black holes.,scaledwidth=43.0% ] one can see from fig . [ emission ] that unlike the case of the reissner - nordstrm black holes , an increase of the tidal charge parameter implies increase of the particle emission rate . furthermore , with increasing frequency the particle emission spectrum rises up to peak and falls to zero rapidly . at the frequencies @xmath140 , emission spectrum is almost zero regardless the value of the tidal charge parameter . from fig . [ emission ] one can deduce that at the high - frequency regime distinction of types of black holes is impossible . from this viewpoint , the low - frequency regime is more favorite to distinguish various black holes . we have studied the scalar , electromagnetic , axial and polar gravitational perturbations of the reissner - nordstrm black hole ( with tidal charge instead of electric charge ) in the randall - sundrum braneworld . by using the sixth order wkb method , we have calculated the quasinormal frequencies of these perturbations . results have shown that with increasing tidal charge parameter the frequency of the real oscillations decreases while damping rate increases , unlike the case of the standard reissner - nordstrm black hole . the reflection probability of the wave by the scalar potential barrier is larger than for the electromagnetic and gravitational ones ; i.e. , scalar fields are the most favorite in terms of reflection of waves , while the gravitational fields are the most favorite in terms of transmission of waves . with increasing tidal charge parameter reflection abilities of the perturbative fields decrease , as the radii of the horizon and the photon sphere increase with the tidal charge . however , we should note that because of no upper limit on the value of the tidal charge parameter , black holes on the brane always have event horizons and photon spheres which are located further away from the central object in comparison to the schwarzschild ones . we have studied also the absorption cross section of the massless scalar waves by the braneworld black hole in the low- and high - frequency regimes . calculations have shown that the braneworld black holes always have bigger absorption cross sections than the corresponding ones related to the schwarzschild and reissner - nordstrm black holes . one of the main results of this paper is that in the high - frequency regime distinction of the black holes from the particle emission spectrum is almost impossible . from this point of view the low - frequency regime is more significant . authors would like to thank the anonymous referee for his careful reading of our manuscript and many insightful comments and suggestions that have improved the paper . the authors would like to thank naresh dadhich for his useful discussions on the properties of the black hole solution on the brane and luciano rezzolla for noting a problem with our calculations in a previous version of the manuscript . b.t . , z.s . , and j.s . would like to express their acknowledgments for the institutional support of the faculty of philosophy and science of the silesian university in opava , the internal student grant of the silesian university ( grant no . sgs/23/2013 ) and the albert einstein centre for gravitation and astrophysics under the czech science foundation ( grant no . 14 - 37086 g ) . b.a . acknowledges the faculty of philosophy and science , silesian university in opava , czech republic , and the goethe university , frankfurt am main , germany , for their warm hospitality . the research of b.a . is supported in part by projects no . f2-fa - f113 , no . ef2-fa-0 - 12477 , and no . f2-fa - f029 of the uzas , and by the ictp through grants no . oea - prj-29 and no . oea - net-76 and by the volkswagen stiftung , grant no . 86 866 . r. emparan , g. t. horowitz , and r. c. myers , j. high energy phys . * 01 * ( 2000 ) 007 , [ http://arxiv.org/abs/hep-th/9911043 [ hep - th/9911043 ] ] ; a. chamblin , h. s. reall , h .- a . shinkai , and t. shiromizu , phys . d * 63 * , 064015 ( 2001 ) , [ http://arxiv.org/abs/hep-th/0008177[hep-th/0008177 ] ] ; n. dadhich and s. g. ghosh , phys . b * 518 * , 1 ( 2001 ) , [ http://arxiv.org/abs/hep-th/0101019[hep-th/0101019 ] ] . k. bronnikov , v. melnikov , and h. dehnen , phys . d * 68 * , 024025 ( 2003 ) , [ http://arxiv.org/abs/gr-qc/0304068[gr-qc/0304068 ] ] ; k. a. bronnikov and a. v. michtchenko , int . j. mod . a * 20 * , 2256 ( 2005 ) ; d .- c . dai and d. stojkovic , phys . b * 704 * , 354 ( 2011 ) , [ http://arxiv.org/abs/1004.3291[arxiv:1004.3291 ] ] . s. shankaranarayanan and n. dadhich , int . j. mod d * 13 * , 1095 ( 2004 ) , [ http://arxiv.org/abs/gr-qc/0306111[gr-qc/0306111 ] ] ; c. molina and j. c. s. neves , phys . d * 82 * , 044029 ( 2010 ) , [ http://arxiv.org/abs/1005.1319[arxiv:1005.1319 ] ] ; j. c. s. neves and c. molina , phys . d * 86 * , 124047 ( 2012 ) , [ http://arxiv.org/abs/1211.2848 [ arxiv:1211.2848 ] ] . k. a. bronnikov and s .- w . kim , phys . d * 67 * , 064027 ( 2003 ) , [ http://arxiv.org/abs/gr-qc/0212112[gr-qc/0212112 ] ] ; f. s. n. lobo , phys . d * 75 * , 064027 ( 2007 ) , [ http://arxiv.org/abs/gr-qc/0701133[gr-qc/0701133 ] ] ; c. molina and j. c. s. neves , phys d * 86 * , 024015 ( 2012 ) , [ http://arxiv.org/abs/1204.1291[arxiv:1204.1291 ] ] . j. ovalle , int . j. mod d * 18 * , 837 ( 2009 ) , [ http://arxiv.org/abs/0809.3547[arxiv:0809.3547 ] ] ; j. ovalle and f. linares , phys . rev . d * 88 * , 104026 ( 2013 ) , [ http://arxiv.org/abs/1311.1844[arxiv:1311.1844 ] ] ; j. ovalle , l. . gergely , and r. casadio , classical quantum gravity * 32 * , 045015 ( 2015 ) , [ http://arxiv.org/abs/1405.0252[arxiv:1405.0252 ] ] . j. schee and z. stuchlk , gen . . gravit . * 41 * , 1795 ( 2009 ) , [ http://arxiv.org/abs/0812.3017[arxiv:0812.3017 ] ] ; z. stuchlk and a. kotrlov , gen . * 41 * , 1305 ( 2009 ) , [ http://arxiv.org/abs/0812.5066[arxiv:0812.5066 ] ] ; a. n. aliev and a. e. gmrkolu , phys . d * 71 * , 104027 ( 2005 ) , [ http://arxiv.org/abs/hep-th/0502223[hep-th/0502223 ] ] . a. abdujabbarov and b. ahmedov , phys . d * 81 * , 044022 ( 2010 ) , [ http://arxiv.org/abs/0905.2730[arxiv:0905.2730 ] ] ; l. amarilla and e. f. eiroa , phys . d * 85 * , 064019 ( 2012 ) , [ http://arxiv.org/abs/1112.6349[arxiv:1112.6349 ] ] ; s. r. shaymatov , b. j. ahmedov , and a. a. abdujabbarov , phys . rev . d * 88 * , 024016 ( 2013 ) . a. kotrlov , z. stuchlk , and g. trk , classical quantum gravity * 25 * , 225016 ( 2008 ) , [ http://arxiv.org/abs/0812.0720[arxiv:0812.0720 ] ] ; b. j. ahmedov and f. j. fattoyev , phys . rev . d * 78 * , 047501 ( 2008 ) , [ http://arxiv.org/abs/gr-qc/0608039[gr-qc/0608039 ] ] ; f. x. linares , m. a. garca - aspeitia , and l. a. urea - lpez , phys . d * 92 * , 024037 ( 2015 ) , [ http://arxiv.org/abs/1501.04869[arxiv:1501.04869 ] ] . z. stuchlk , m. blaschke , and p. slan , classical quantum gravity * 28 * , 175002 ( 2011 ) , [ http://arxiv.org/abs/1108.0191[arxiv:1108.0191 ] ] ; j. hladk and z. stuchlk , j. cosmol . astropart * 7 * , 012 ( 2011 ) , [ http://arxiv.org/abs/1108.5760[arxiv:1108.5760 ] ] ; z. stuchlk , j. hladk , and m. urbanec , gen . . gravit . * 43 * , 3163 ( 2011 ) , [ http://arxiv.org/abs/1108.5767[arxiv:1108.5767 ] ] . a. o. starinets , phys d * 66 * , 124013 ( 2002 ) , [ http://arxiv.org/abs/hep-th/0207133[hep-th/0207133 ] ] ; e. berti , k. d. kokkotas , and e. papantonopoulos , phys . d * 68 * , 064020 ( 2003 ) , [ http://arxiv.org/abs/gr-qc/0306106[gr-qc/0306106 ] ] ; y. kurita and m. sakagami , prog . 148 * , 298 ( 2002 ) . j. f. vzquez - poritz , j. high energy phys . * 03 * ( 2002 ) 008 , [ http://arxiv.org/abs/hep-th/0110085[hep-th/0110085 ] ] ; y. kurita and m .- a . sakagami , phys . d * 67 * , 024003 ( 2003 ) , [ http://arxiv.org/abs/hep-th/0208063[hep-th/0208063 ] ] ; a. g. casali , a. elcio , and b. wang , phys . d * 70 * , 043542 ( 2004 ) , [ http://arxiv.org/abs/hep-th/0403155[hep-th/0403155 ] ] . k. maeda , m. natsuume , and t. okamura , phys . d * 72 * , 086012 ( 2005 ) , [ http://arxiv.org/abs/hep-th/0509079[hep-th/0509079 ] ] ; s. s. seahra , phys . d * 72 * , 066002 ( 2005 ) , [ http://arxiv.org/abs/hep-th/0501175[hep-th/0501175 ] ] ; s. fernando , gen . . gravit . * 37 * , 585 ( 2005 ) , [ http://arxiv.org/abs/hep-th/0407062[hep-th/0407062 ] ] . e. berti , v. cardoso , and m. casals , phys . d * 73 * , 024013 ( 2006 ) , [ http://arxiv.org/abs/gr-qc/0511111[gr-qc/0511111 ] ] ; phys . d * 73 * , 109902(e ) ( 2006 ) ; p. kanti , r. a. konoplya , and a. zhidenko , phys . rev . d * 74 * , 064008 ( 2006 ) , [ http://arxiv.org/abs/gr-qc/0607048[gr-qc/0607048 ] ] ; d. k. park , phys . b * 633 * , 613 ( 2006 ) , [ http://arxiv.org/abs/hep-th/0511159[hep-th/0511159 ] ] . s. chen , b. wang , and r. su , phys . b * 647 * , 282 ( 2007 ) , [ http://arxiv.org/abs/hep-th/0701209[hep-th/0701209 ] ] ; a. starinets , phys . b * 670 * , 442 ( 2009 ) , [ http://arxiv.org/abs/0806.3797[arxiv:0806.3797 ] ] ; m. nozawa and t. kobayashi , phys . d * 78 * , 064006 ( 2008 ) , [ http://arxiv.org/abs/0803.3317[arxiv:0803.3317 ] ] . a. zhidenko , phys . d * 78 * , 024007 ( 2008 ) , [ http://arxiv.org/abs/0802.2262[arxiv:0802.2262 ] ] ; h. t. cho , a. s. cornell , j. doukas , and w. naylor , phys . d * 77 * , 041502 ( 2008 ) , [ http://arxiv.org/abs/0710.5267[arxiv:0710.5267 ] ] ; u. a. al - binni and g. siopsis , phys . d * 76 * , 104031 ( 2007 ) , [ http://arxiv.org/abs/0708.3363[arxiv:0708.3363 ] ] . j. morgan , v. cardoso , a. s. miranda , c. molina , and v. t. zanchin , j. high energy phys . * 9 * ( 2009 ) 117 , [ http://arxiv.org/abs/0907.5011[arxiv:0907.5011 ] ] ; e. abdalla , owen pavel fernandez piedra , j. de oliveira , and c. molina , phys . d * 81 * , 064001 ( 2010 ) , [ http://arxiv.org/abs/0810.5489[arxiv:0810.5489 ] ] ; t. delsate , v. cardoso , and p. pani , j. high energy phys . * 6 * ( 2011 ) 55 , [ http://arxiv.org/abs/1103.5756[arxiv:1103.5756 ] ] . s. hod , classical quantum gravity * 28 * , 105016 ( 2011 ) , [ http://arxiv.org/abs/1107.0797[arxiv:1107.0797 ] ] ; h. chung , l. randall , m. j. rodriguez , and o. varela , http://arxiv.org/abs/1508.02611[arxiv:1508.02611 ] ; s. janiszewski and m. kaminski , phys . d * 93 * , 025006 ( 2016 ) , [ http://arxiv.org/abs/1508.06993[arxiv:1508.06993 ] ] . k. kokkotas and b. schmidt , living rev . relativity * 2 * , 2 ( 1999 ) , [ http://arxiv.org/abs/gr-qc/9909058[gr-qc/9909058 ] ] ; a. nagar and l. rezzolla , classical quantum gravity * 22 * , r167 ( 2005 ) , [ http://arxiv.org/abs/gr-qc/0502064[gr-qc/0502064 ] ] ; e. berti , v. cardoso , and a. o. starinets , classical quantum gravity * 26 * , 163001 ( 2009 ) , [ http://arxiv.org/abs/0905.2975[arxiv:0905.2975 ] ] ; r. a. konoplya and a. zhidenko , rev . mod 83 * , 793 ( 2011 ) , [ http://arxiv.org/abs/1102.4014[arxiv:1102.4014 ] ] . v. cardoso , m. cavagli , and l. gualtieri , phys . 96 * , 071301 ( 2006 ) , [ http://arxiv.org/abs/hep-th/0512002 [ hep - th/0512002 ] ] ; phys . lett . * 96 * , 219902(e ) ( 2006 ) ; r. a. konoplya and a. zhidenko , phys . d * 81 * , 124036 ( 2010 ) , [ http://arxiv.org/abs/1004.1284[arxiv:1004.1284 ] ] . t. r. choudhury and t. padmanabhan , phys . d * 69 * , 064033 ( 2004 ) , [ http://arxiv.org/abs/gr-qc/0311064[gr-qc/0311064 ] ] ; t. harmark , j. natario , and r. schiappa , adv . * 14 * , 727 ( 2010 ) , [ http://arxiv.org/abs/0708.0017[arxiv:0708.0017 ] ] . s. r. das , g. gibbons , and s. d. mathur , phys . lett . * 78 * , 417 ( 1997 ) , [ http://arxiv.org/abs/hep-th/9609052[hep-th/9609052 ] ] ; c. f. b. macedo , l. c. s. leite , and l. c. b. crispino , phys . d * 93 * , 024027 ( 2016 ) , [ http://arxiv.org/abs/1511.08781 [ arxiv:1511.08781 ] ] . l. c. b. crispino , a. higuchi , and g. e. a. matsas , phys . d * 82 * , 124038 ( 2010 ) , [ http://arxiv.org/abs/1004.4018 [ arxiv:1004.4018 ] ] ; l. c. b. crispino , a. higuchi , and e. s. oliveira , phys . d * 80 * , 104026 ( 2009 ) ; s. r. dolan , e. s. oliveira , and l. c. b. crispino , phys . d * 79 * , 064014 ( 2009 ) , [ http://arxiv.org/abs/0904.0010[arxiv:0904.0010 ] ] ; e. jung and d. k. park , nucl . b717 * , 272 ( 2005 ) , [ http://arxiv.org/abs/hep-th/0502002[hep-th/0502002 ] ] . c. l. benone , e. s. de oliveira , s. r. dolan , and l. c. b. crispino , phys . d * 89 * , 104053 ( 2014 ) , [ http://arxiv.org/abs/1404.0687[arxiv:1404.0687 ] ] ; c. f. b. macedo , e. s. de oliveira , and l. c. b. crispino , phys . d * 92 * , 024012 ( 2015 ) , [ http://arxiv.org/abs/1505.07014[arxiv:1505.07014 ] ] . k. glampedakis and n. andersson , classical quantum gravity * 18 * , 1939 ( 2001 ) , [ http://arxiv.org/abs/gr-qc/0102100[gr-qc/0102100 ] ] ; e. s. oliveira , s. r. dolan , and l. c. b. crispino , phys . d * 81 * , 124013 ( 2010 ) ; c. f. b. macedo , l. c. s. leite , e. s. oliveira , s. r. dolan , and l. c. b. crispino , phys . d * 88 * , 064033 ( 2013 ) , [ http://arxiv.org/abs/1308.0018[arxiv:1308.0018 ] ] . m. casals , s. dolan , p. kanti , and e. winstanley , j. high energy phys . * 06 * ( 2008 ) 071 , [ http://arxiv.org/abs/0801.4910[arxiv:0801.4910 ] ] ; p. kanti and n. pappas , phys . d * 82 * , 024039 ( 2010 ) , [ http://arxiv.org/abs/1003.5125[arxiv:1003.5125 ] ] .

we performed a detailed analysis of the process where we included all tree level feynman diagrams that contribute to this final state . we studied the sensitivity of this process to anomalous trilinear gauge boson couplings of the and vertices using two popular parametrizations . we used a maximum likelihood analysis of a five dimensional differential cross - section based on the and decay product angular distributions . we concentrated on lep-200 energies , taking gev , and energies appropriate to the proposed next linear collider ( nlc ) , a high energy collider with center of mass energies and 1 tev . at 175 gev , can be measured to about , to and to , to and to . at 95% c.l . assuming 500 pb integrated luminosity . although these will be improvements of existing measurements they are not sufficiently precise to test the standard model at the loop level and are unlikely to see deviations from sm expectations . at 500 gev with 50 fb integrated luminosity , can be measured to about , and to and and to at 95% c.l . while at 1 tev with 200 fb integrated luminosity , and can be measured to about and respectively . the 500 gev measurements will be at the level of loop contributions to the couplings and may show hints of new physics while the 1 tev should be sensitive to new physics at the loop level . w - pair production in the process + and measurement of the and couplings ottawa - carleton institute for physics + department of physics , carleton university , ottawa canada , k1s 5b6 dpartement de physique , universit du qubec montral + c.p . 8888 , succ . centre - ville , montral , qubec , canada , h3c 3p8

hep-ph9511204

@xmath5 colliders have made important contributions to our understanding of the electroweak interactions@xcite and it is expected that this tradition will continue with the advent of higher luminosity and higher energy machines . in the near future the cern lep-200 @xmath5 collider@xcite will begin operation and beyond that there is a growing effort directed towards the design and construction of future high energy @xmath5 linear colliders with @xmath22 gev which we will refer to generically as the next linear collider ( nlc)@xcite . one of the primary physics goals of lep-200 and an important goal of the nlc is to make precision measurements of @xmath3 boson properties including @xmath23 , @xmath24 , and @xmath3-boson interactions with fermions and the photon and @xmath25 . the latter measurements , that of the trilinear gauge boson vertices ( tgv s ) provides a stringent test of the gauge structure of the standard model@xcite . the current measurement of these couplings are rather weak . using a popular parametrization of the cp conserving gauge boson couplings , indirect measurements of tgv s via radiative corrections to precision electroweak measurements@xcite give the following limits@xcite : @xmath26 , @xmath27 , @xmath28 , @xmath29 , and @xmath30 . however , there are ambiguities in these calculations associated with running the couplings down from the scale of new physics to low energy so that these limits are not particularly rigorous and it is necessary to use direct measurements for more reliable bounds . the cdf and d0 collaborations at the tevatron @xmath31 collider at fermilab , using the processes @xmath32 , have obtained the direct 95% c.l . limits of @xmath33 , @xmath34 , @xmath35 , and @xmath36 @xcite . these measurements are quite weak but it is expected that they will improve as the luminosity of the tevatron increases . in the longer term measurements at the large hadron collider at cern will improve these limits considerably@xcite . it is expected that measurements at high energy @xmath5 colliders will surpass those at the hadron colliders . as a result , many processes have been studied to determine their usefulness for measuring the tgv s ; @xmath37 @xcite , @xmath38 @xcite , @xmath39@xcite , @xmath40 @xcite , @xmath41@xcite , @xmath42 @xcite and more detailed studies of various four fermion final states in the process @xmath43 @xcite . probably the most useful of the @xmath42 channels for these studies is @xmath44 . with only one unobserved neutrino this channel has several advantages : it can be fully reconstructed using the constraint of the initial beam energies , the @xmath45 and @xmath46 can be discriminated using lepton charge identification , it does not have the qcd backgrounds that plague the fully hadronic decay modes , and it offers much higher statistics than the fully leptonic modes . as a result of the importance of this channel there have been numerous studies of this process . in particular , there is a growing list of analysis of @xmath47 , @xmath5 to four fermion final state processes @xcite , single @xmath3 production@xcite , electroweak radiative corrections to these reactions including the important contribution from initial state radiation @xcite , and the sensitivity of these processes to anomalous @xmath1 and @xmath48 gauge boson couplings ( tgv s ) . in this paper we examine in detail the four fermion final state @xmath49 where @xmath50 is either @xmath51 or @xmath52 and @xmath53 can be either @xmath54 or @xmath55 . we study this process for the centre of mass energies @xmath4 gev appropriate to lep200 , and @xmath6 and 1000 gev appropriate to the nlc . to obtain results we included all tree level diagrams to the four fermion final states using helicity amplitude techniques . for the @xmath56 final state 10 diagrams contribute and for the @xmath57 final state 20 diagrams contribute . our primary interest is to study the sensitivity of these processes to anomalous gauge boson couplings . to do so we examined numerous distributions . for the purpose of comparing theory to experiment we also examined the question of whether the approximation of only including the resonant diagrams is adequate or whether the full four - fermion final state calculation is needed . using helicity amplitudes we are able to study the usefulness of initial state polarization in extracting the tgv s . in the @xmath58 final state single @xmath3 production can also be studied@xcite where the @xmath1 vertex can be isolated from the @xmath2 vertex by imposing an appropriate cut on the outgoing electron . in this case , when only hadronic jets are observed and not the outgoing lepton , there are ambiguities in identifying the charge of the @xmath3 besides problems with hadronic backgrounds which we do not deal with here . a detailed analysis of single @xmath3 production will be presented elsewhere @xcite . in the next section we discuss effective lagrangians and the various parametrizations used to describe gauge boson self interactions which have appeared in the literature . in section iii we describe our calculation . section iv comprises the bulk of the paper which is used to present and discuss our results . we summarize our conclusions in section v. the formalism of effective lagrangians provides a well - defined framework for investigating the physics of anomalous couplings and electroweak symmetry breaking @xcite . in this approach an infinite set of non - renormalizable operators , consistent with the unbroken symmetries and whose coefficients parametrize the low - energy effects of electroweak symmetry breaking or new physics are organized in an energy expansion . at low energy only a finite number of terms contribute to a given process . at higher energies more and more terms become important until the whole process breaks down at the scale of new physics . one focuses on the leading operators in the expansion . there are three main parametrizations of gauge boson couplings that appear in the literature . the characteristic distinguishing the approaches is the degree to which constraints are imposed in terms of the symmetry and particle content of the low energy theory . we summarize the most commonly used parametrizations below@xcite . the first approach is to describe the @xmath59 verticies using the most general parametrization possible that respects lorentz invariance , electromagnetic gauge invariance and @xmath60 invariance @xcite . this approach has become the standard parametrization used in phenomenology making the comparison of the sensitivity of different measurements to the tgv s straightforward . we do not consider cp violating operators in this paper as they are tightly constrained by measurement of the neutron electric dipole moment which constrains the two cp violating parameters to @xmath61 @xcite . with these constraints the @xmath1 and @xmath2 vertices have five free independent parameters , @xmath7 , @xmath10 , @xmath9 , @xmath13 and @xmath12 and is given by @xcite : @xmath62 where the subscript @xmath63 denotes either a photon or a @xmath25 , @xmath64 and @xmath65 represents the photon or @xmath25 and @xmath46 fields respectively , @xmath66 and @xmath67 and @xmath23 is the @xmath3 boson mass . ( @xmath68 is constrained by electromagnetic gauge invariance to be equal to 1 . ) the first two terms correspond to dimension 4 operators and the third term corresponds to a dimension 6 operator . the mass in the denominator of the dimension 6 term would correspond to the scale of new physics , typically of order 1 tev . however , it has become the convention to use @xmath23 so that the @xmath3 magnetic dipole and electric quadrupole can be written in a form similar to that of the muon . nevertheless , one expects the dimension 6 operator to be suppressed with respect to the dimension 4 operators by a factor of @xmath69 . higher dimension operators correspond to momentum dependence@xcite in the form factors which are not so important in the process we are considering so are not included . at tree level the standard model ( sm ) requires @xmath70 and @xmath71 . typically , radiative corrections from heavy particles will change @xmath18 by about @xmath72 and @xmath19 by about @xmath73 @xcite . in particular , the contributions from a 200 gev top quark and a 150 gev higgs boson to @xmath18 and @xmath19 are of order @xmath74 . the nearness of the @xmath75 parameter to 1 implies an @xmath76 invariance of the weak interaction . imposing this from the outset implies a relationship between the parameters reducing the number of parameters from 5 to 3 @xcite . @xmath76 invariance is the basis of the other two parametrizations we mention . the difference between the two is that in the first , the higgs boson is heavy so that goldstone bosons are nonlinearly realized while in the second , the higgs bosons are light , leading to linearly realized goldstone bosons . the second commonly used parametrization is the chiral lagrangian approach@xcite . a custodial @xmath76 is assumed which is supported to high accuracy by the nearness of the @xmath75 parameter to 1 . this approach assumes that the theory has no light higgs particles and the electroweak gauge bosons interact strongly with each other above approximately 1 tev . this can be described by a non - linear realization of the @xmath77 symmetry in a chiral lagrangian formalism leading to the effective lagrangian : @xmath78 -i g ' { { l_{9r}}\over{16\pi^2 } } tr [ b^{\mu\nu } d_\mu \sigma^\dagger d_\nu \sigma ] + g g'{{l_{10}}\over{16\pi^2 } } tr [ \sigma b^{\mu\nu } \sigma^\dagger w_{\mu\nu}]\ ] ] where @xmath79 and @xmath80 are the @xmath76 and @xmath81 field strength tensors given in terms of @xmath82 by @xmath83 ) \nonumber\\ b_{\mu\nu } & = & { 1\over 2 } ( \partial _ \mu b_\nu -\partial_\nu b_\mu ) \tau_3 , \end{aligned}\ ] ] @xmath84 , @xmath85 gev , @xmath86 are the would - be goldstone bosons that give the @xmath3 and @xmath87 their masses via the higgs mechanism , and the @xmath88 covariant derivative is given by @xmath89 . the feynman rules are found by going to the unitary gauge where @xmath90 . note that often in the literature the coefficient @xmath91 is replaced with @xmath92 . @xmath93 contributes to the gauge boson self energies where it is tightly constrained to @xmath94 @xcite so we will not consider it further . new physics contributions are expected to result in values of @xmath95 of order 1@xcite . in the linear realization scenario@xcite the higgs doublet field @xmath96 is included in the low energy particle content . this approach assumes that any deviations from the standard model due to new physics manifests themselves in @xmath97 gauge invariant singlet operators . there are 7 relevant operators of which four are stringently constrained by the high precision low energy and @xmath87 boson data @xcite . the remaining three can give rise to non - standard couplings @xmath98\end{aligned}\ ] ] it seems most likely that anomalous couplings in the light higgs linear scenario would best be studied by measuring the properties and couplings of the higgs boson directly . in any case the parameters from this approach can be rewritten in terms of the parameters of the first two lagrangians discussed . the parameters from the three lagrangians can be mapped onto each other : @xmath99 dropping the @xmath93 term , the linear and non - linear realizations are obtained from each other by identifying @xmath100 and @xmath101 . in the non - linear realization , the counterpart of @xmath102 is higher dimension . to study the process @xmath103 we included all tree level diagrams to the four fermion final states . there are 10 diagrams contributing to the @xmath104 final state which are shown in fig . the gauge boson coupling we are studying is present in diagram ( 1a ) . this , along with diagram ( 1b ) are the diagrams responsible for real @xmath3 production . for the @xmath57 final state the 10 diagrams shown in fig . 2 must also be included with those of fig . 1 for a total of 20 diagrams . diagram ( 2a ) includes a tgv . the diagrams with t - channel photon exchange make large contributions to single @xmath3 production due to the pole in the photon propagator which can be used to isolate the @xmath1 vertex from the @xmath2 vertex @xcite . we include final width effects by using vector boson propagators of the form @xmath105 which yields a gauge invariant result . strictly speaking we should have included a momentum dependent vector boson width but this leads to problems with gauge invariance@xcite . although a number of solutions to this problem have been discussed @xcite the difference between our treatment and more rigorous ones have a totally negligible effect on the tgv sensitivities we obtain from our analysis . a more rigorous treatment must of course be included in monte carlo simulations that will be used to analyze real experimental data . we will find that the non - resonant diagrams make non - negligible contributions to cross sections and are dependent on the kinematic cuts used in the analysis . these contributions are at least as important as electroweak radiative corrections . to evaluate the cross - sections and different distributions , we used the calkul helicity amplitude technique @xcite to obtain expressions for the matrix elements and performed the phase space integration using monte carlo techniques @xcite . the expressions for the helicity amplitudes are lengthy and unilluminating so we do not include them here . to obtain numerical results we used the values @xmath106 , @xmath107 , @xmath108 gev , @xmath109 gev , @xmath110 gev , and @xmath111 gev . in our results we included two generations of quarks and took them to be massless . in order to take into account finite detector acceptance we require that the lepton and quarks are at least 10 degrees away from the beam and have at least 10 gev energy unless otherwise noted . in principle we should include qed radiative corrections from soft photon emission and the backgrounds due to a photon that is lost down the beam pipe @xcite . these backgrounds are well understood and detector dependent . we assume the approach taken at lep , that these effects can best be taken into account by the experimental collaborations . in any case , although initial state radiation must be taken into account their inclusion does not substantially effect the bounds we obtain and therefore our conclusions . in figure 3 and table i we show the cross sections for the processes @xmath112 for different applications of cuts on the invariant mass of the lepton - neutrino pair and the @xmath53 pair ; @xmath113 gev where @xmath114 is the invariant mass of the @xmath115 and @xmath53 pair respectively . imposing the cut on one fermion pair gives the single @xmath3 cross section and imposing the cut on both fermion pairs gives the @xmath3-pair production cross section . in both cases the single @xmath3 and @xmath3 pair thresholds are clearly seen . although the single @xmath3 production cross section is nonzero below the @xmath3-pair production threshold , it is still too small to obtain adequate statistics to perform studies of @xmath3 boson properties . in general the electron mode has a larger cross - section than the muon mode . the difference is small at 175 gev but becomes increasingly larger at higher energy as the t - channel photon exchange becomes increasingly important , reaching a factor of 5 at 1 tev . for the muon mode the invariant mass cuts reduce the cross - section by 10% to 20% depending on @xmath116 irrespective of whether the cut is on @xmath117 or @xmath118 . the relatively small effect of these cuts verifies the dominance of the resonant diagrams on the total cross section . for the electron mode the results are similar when the @xmath119 gev cut is imposed which constrains the @xmath120 pair to be on the @xmath3 mass - shell . however when @xmath121 gev ( ie . @xmath122 ) the cross section is significantly larger than the previous case due to the enhancement arising from the non - resonant t - channel photon exchange diagrams of fig . 2 . with appropriate kinematic cuts this can be used to study single-@xmath3 production @xcite . despite the relative smallness of the off - resonance contributions to the muon mode they still contribute up to 30% of the cross section at 1 tev . clearly , they must be properly included when making high precision tests of standard model processes . for the electron mode they are even more important and are interesting in the context of single @xmath3 production . the above points can be amplified by examining kinematic distributions . in addition , since our goal is to extract measurements of the tgv s , we must explore which distributions are most sensitive to anomalous couplings . for descriptive purposes we will show various distributions for @xmath6 gev . we begin by showing in fig . 4 the invariant mass distributions for the @xmath120 ( @xmath123 ) and @xmath124 ( @xmath125 ) pairs for left and right handed initial electron polarization . as the unpolarized cross sections are dominated by the left handed electrons they are quite similar to them , so we do not include them separately . in addition , the @xmath126 invariant mass distributions for left handed initial electrons are similar to the @xmath125 distributions . since there are differences for the right handed initial electron distributions , these distributions are also included . the differences in these cross sections reflects the differences and relative importance in the feynman diagrams that contribute to a process . although the cross sections and the sensitivities to the tgv s are dominated by the production of real @xmath3 s one can see that off - resonance production of the @xmath127 final state can be quite sensitive to anomalous couplings . we will explore this in a later section . the effects are especially pronounced for the @xmath128 final state where there is the possibility of single @xmath3 production which is discussed elsewhere @xcite . we examined numerous distributions with the purpose of finding the distributions and isolating the regions of phase space most sensitive to anomalous couplings ; @xmath129 there is , of course , overlap among the regions of interest in these distributions . to gauge the sensitivity of these distributions to the tgv s we typically divided them into 4-bins and performed a @xmath130 analysis . for @xmath6 gev and integrated luminosity of 50 fb@xmath14 we found , for example , that the @xmath131 s could be measured to a couple of percent at 95% confidence level . it turns out that this is not competitive with a more sophisticated analysis of angular distributions we will describe below . generally , this is because the phase space regions with the highest statistics are least sensitive to anomalous couplings and tend to overwhelm deviations while the regions most sensitive to tgv s have poor statistics . for the purpose of understanding @xmath3-boson properties the most interesting distributions are the various angular distributions . to understand this better it is useful to first consider the @xmath3 pair production cross section without decays to fermions@xcite . to leading order the amplitude for @xmath3 pair - production is given by three diagrams ; via an s - channel photon , an s - channel @xmath25 and a t - channel neutrino exchange . the cross sections at different @xmath132 , 500 , and 1000 gev , for @xmath133 , @xmath134 , and @xmath135 , different initial state polarizations , and as a function of the @xmath3 scattering angle are shown in fig . 5 @xcite . for the initial state @xmath136 only the first two diagrams contribute which at high energy is dominated by longitudinal @xmath3 ( @xmath137 ) production . due to the delicate cancellations between the diagrams it is @xmath137 production which is most sensitive to anomalous couplings . in contrast , the cross section for the @xmath138 initial state produces both transverse and longitudinal @xmath3 bosons with comparable rates . the @xmath139 cross section is dominated by a peak in the forward direction with respect to the incoming @xmath140 associated with the t - channel neutrino exchange which is made up entirely of transverse @xmath3 production . this contribution is relatively insensitive to new physics . the cross sections in the backward direction includes sizable longitudinal @xmath3 production accounting for about 25% of the total cross section in the backward hemisphere . however , in the backward direction where the s - channel diagrams contribute substantially , the cross section for @xmath141 is always quite small . for @xmath141 there is a large change in the magnitude of the cross - section but only a small change in its shape . any disruption of the delicate gauge theory cancellations leads to large changes to the standard model results . for @xmath137 production amplitudes the enhancements can be a factor of @xmath142 . this is shown in fig . 6 where the angular distribution of the outgoing @xmath3 is plotted for several values of anomalous couplings at @xmath6 gev . because it is the longitudinal @xmath3 production which is most sensitive to anomalous couplings , and because the cross section is dominated by transverse @xmath3 production it is crucial to disentangle the @xmath137 from the @xmath143 _ background_. the most convenient means of doing so makes use of the angular distribution of the @xmath3 decay products . defining @xmath144 and @xmath145 as the angle between the @xmath50 or @xmath146 and the @xmath3 momentum measured in the @xmath3 rest frame , the angular distribution in @xmath147 peaks about @xmath148 for longitudinally polarized @xmath3 bosons and at forward or backward angles for transversely polarized bosons . in addition the parity violation of the @xmath3 couplings distinguishes the two polarization states adding to the effectiveness of the decay as a polarimeter . thus the angular distributions can be used to extract information about the @xmath3 boson polarizations . in fig . 7 we show the angular distributions for the outgoing quark with respect to the @xmath3 direction ( @xmath145 ) for the three bins in the @xmath3 scattering angle , @xmath149 , @xmath150 and @xmath151 , ( where we take @xmath152 to be the @xmath46 angle with respect to the incoming @xmath140 ) for the process @xmath153 at @xmath6 gev with the initial electron unpolarized . several values of @xmath18 and @xmath19 are included to demonstrate the sensitivity of the distributions to anomalous couplings . the figure shows the dominance of the transverse @xmath3 polarization at forward angles and the increasing importance of the @xmath3 longitudinal polarization at @xmath154 . note also the relative lack of sensitivity to anomalous couplings for the forward , dominantly transverse @xmath3 s and how the sensitivity increases as the scattering angle increases and longitudinal @xmath3 s contribute a larger fraction of the cross section . we have shown the @xmath155 distribution as it displays the most dramatic change in the shape of the distributions . however , interference between the transverse and longitudinal @xmath3 s also depends on the azimuthal angle so that the azimuthal angular distribution also shows changes in its shape , albeit smaller . one finds similar effects in the angular distributions for the decay @xmath156 . the approach which makes the most complete use of information in an event is the maximum likelihood method . based on the observations of the previous section we perform a maximum likelihood fit based on the 5 angles@xcite ; @xmath157 , the @xmath46 scattering angle with respect to the initial @xmath158 direction , @xmath159 , the polar decay angle of the @xmath146 in the @xmath46 rest frame using the @xmath46 direction as the quantization axis , @xmath160 , the azimuthal decay angle of the @xmath146 in the @xmath46 rest frame , and @xmath161 and @xmath162 are the analogous angles for the lepton in the @xmath45 rest frame . the azimuthal angles are defined as the angle between the normal to the reaction plane , @xmath163 and the plane defined by the @xmath3 decay products , @xmath164 . the angles are shown in fig . 8 . for the @xmath126 case there is an ambiguity since we can not tell which hadronic jet corresponds to the quark and which to the antiquark . we therefore include both possibilities in our analysis . to implement the maximum likelihood analysis we divided each of @xmath157 , @xmath159 , @xmath160 , @xmath161 , and @xmath162 into four bins so that the entire phase space was divided into @xmath165 bins . with this many bins some will not be very populated with events so that it is more appropriate to use poisson statistics rather than gaussian statistics . this leads naturally to the maximum likelihood method . the change in the log of the likelihood function is given by @xmath166\ ] ] where the sum extends over all the bins and @xmath167 and @xmath168 are the predicted number of non - standard model and standard model events in bin @xmath169 respectively , given by @xmath170 where @xmath171 is the expected integrated luminosity . the 68% and 95% confidence level bounds are given by the values of anomalous couplings which give a change in @xmath172 of 0.5 and 2.0 respectively . ideally , one would perform the analysis on an event by event basis but to simplify our calculations we used a five dimensional angular distribution . to check the sensitivity to binning we varied the number of dimensions and bins used in our fits . for this binning approach we found that the results converged to the tightest bounds using the five dimensional distribution and 4 bins per dimension . in a few test runs for special cases of kinematic cuts we calculated the likelihood function on an event by event basis and found that the sensitivities improved a small amount over the 5 dimensional distribution case described above . the results we obtained are based solely on the statistical errors based on the integrated lumininosity we assume for the various cases . to include the effects of systematic errors using the maximum likelihood approach requires an unweighted monte carlo simulation through a realistic detector . since we did not have the facilities to do this we attempted a simplified estimate of systematic errors using @xmath130 analysis to make our estimates . we assumed a systematic error of 5% of a measurement which we combined in quadrature with the statistical error . in general the systematic errors are negligible compared to the statistical errors . the only times they made a measurable difference was for the high luminosity cases of the 500 gev and 1 tev nlc , and even there the effect was quite small . it is straightforward to see why this is so ; with so many bins the number of events per bin is quite small resulting in a large statistical error . thus , it appears that the total errors will be dominated by the statistical errors but clearly , a full detector monte carlo must be performed to properly understand the situation . a thorough analysis of gauge boson couplings would allow all five parameters in the lagrangian to vary simultaneously to take into account cancellations ( and correlations ) among the various contributions . this approach is impractical , however , due to the large amount of computer time that would be required to search the parameter space . instead we show 2-dimensional contours for a selection of parameter pairs to give a sense of the correlations . for the case of the chiral lagrangian where the global su(2 ) symmetry imposes relations between the parameters and where we restrict ourselves to dimension four operators the parameter space reduces to 2 dimensions . for the lep200 collider we study the sensitivity to the gauge boson couplings using the expected machine parameters of @xmath4 gev and an integrated luminosity of 500 pb@xmath14 . these results do not have a cut on @xmath117 or @xmath118 since for these energies the cuts have virtually no effect on the sensitivities except for the electron mode involving the @xmath1 vertex where the effect is still quite small . the 95% confidence limits for the @xmath173 , @xmath174 , @xmath175 , and @xmath176 planes are shown in fig . 9 and for the @xmath177 plane is shown in fig . the sensitivities of the couplings , varying one parameter at a time , are summarized in table ii . in each of these figures , contours are shown for the muon mode alone and then for the combined results of the @xmath178 and @xmath179 modes with both charge possibilities . we also show contours for a reduced integrated luminosity of 300 fb@xmath14 . for the @xmath180 vs @xmath181 plot we show contours for both the electron and muon modes since there is a visible difference for the two modes . by combining the four lepton modes the couplings can be measured to @xmath182 , @xmath183 , @xmath184 , @xmath185 , @xmath186 , @xmath187 , and @xmath188 . if the results of the four lep experiments could be combined these results could be reduced further . nevertheless , these limits are at the very least an order of magnitude less sensitive than would be required to see the effects of new physics through radiative corrections and are comparable to the sensitivities that could be achieved at a high luminosity tevatron upgrade . it is therefore unlikely , that new physics will reveal itself at lep200 through precision measurements of the tgv s . for the @xmath6 gev nlc option we assume an integrated luminosity of 50 fb@xmath14 . at 500 gev the results are most sensitive when we impose that the @xmath3 s are on mass - shell ; ie . gev and @xmath190 gev . this is slightly more pronounced for the electron mode . after these cuts are imposed the electron and muon modes are essentially identical . the 95% confidence limits for the @xmath173 , @xmath174 , @xmath175 , and @xmath176 planes are shown in fig . 11 and for the @xmath177 planes in fig . 12 . in each of these plots we show results for combining the four final states and the @xmath191 mode alone . we also show contours for 10 fb@xmath14 of integrated luminosity to show the effects of reducing the collider luminosity . the sensitivities , varying one at a time , are included in table ii . using the maximum likelihood analysis we find that @xmath10 , @xmath9 , @xmath13 and @xmath12 can be measured to better than @xmath16 and @xmath192 to roughly @xmath193 at 95 % c.l . including the invariant mass cut and assuming 50 fb@xmath14 integrated luminosity . if the integrated luminosity were reduced to 10 fb@xmath14 the bounds become weaker by roughly a factor of two while combining all four modes improves the single mode bounds by roughly a factor of two . these measurements of @xmath192 , @xmath10 , and @xmath9 should be precise enough to probe loop radiative corrections to the couplings . on the other hand the measurements of @xmath13 and @xmath12 are still an order of magnitude too large to see expected deviations from tree level values due to radiative corrections . the @xmath177 contours are shown in fig . @xmath180 can be measured to @xmath194 and @xmath181 to @xmath195 using a single mode and to @xmath196 and @xmath197 respectively combining all four @xmath198 final states . for the 1 tev nlc collider we assume an integrated luminosity of 200 fb@xmath14 . the 95% c.l . sensitivity contours for the @xmath173 and @xmath174 are shown in fig . 13 . these results were obtained by imposing that the @xmath3 s be on mass shell ; @xmath189 gev and @xmath190 gev . we do not bother showing the @xmath200 contours since they are uncorrelated ( one parameter is least sensitive when the other is taken to be equal to zero ) so it is sufficient to give the sensitivity when all other parameters are set to zero . the @xmath177 contours are shown in fig 14 . with these collider parameters @xmath201 can be measured to about @xmath16 while @xmath9 and @xmath10 can be measured to about @xmath74 . these measurements will be sensitive enough to test the standard model at the level of radiative corrections . in the earlier discussion of angular distributions we pointed out that reactions with different initial electron polarizations have different dependences on anomalous couplings @xcite . in this section we explore the consequences of this behavior . we restrict our results to the dimension 4 operators where deviations are most likely to show up . for the 500 gev @xmath5 collider we took 25 fb@xmath14 of integrated luminosity per polarization . we only include results for the four combined lepton modes . the 95% c.l . contours for the @xmath202 , @xmath174 , and @xmath177 planes are shown in fig . 15 . shown are contours for @xmath203 , @xmath204 , and unpolarized initial states . for the @xmath205 plane there is not much difference in the shape of the contour for the different polarizations although the bounds improve slightly . on the other hand the different polarizations give much different dependences for the @xmath174 and @xmath177 contours . in the @xmath174 plane the right - handed electron polarizations give constraints orthogonal to the left - handed polarizations and unpolarized results . the unpolarized contours are aligned along the @xmath206 contours which is not too surprising considering that @xmath207 dominates the right - handed contribution in the unpolarized cross - section . for the @xmath177 plane the left and right handed polarizations also give different dependences which would further constrain @xmath180 . for the 1 tev @xmath5 collider we took 100 fb@xmath14 of integrated luminosity per polarization . the sensitivities are shown in fig . 16 . they are similar to , but more constraining , than the 500 gev case so we do not comment further . more important than the improvement in sensitivity is the usefulness of polarization for disentangling the nature of anomalous tgv s if deviations are observed . in this section we explore the information potential available from @xmath208 final states off the @xmath3 resonance . referring to the invariant mass distributions shown in fig . 4 one sees that there is considerable sensitivity to anomalous gauge boson couplings when the fermion pairs do not originate from real @xmath3 production . we do not perform a rigorous analysis here but demonstrate that there is considerable information in the non - resonant production . in particular we do not consider possible backgrounds to non - resonant events and do not make any effort to optimize our cuts to enhance deviations from sm results . we consider @xmath132 gev , 500 gev , and 1 tev for both the @xmath209 and @xmath210 final states and include initial state polarization when appropriate . we based our results on the total cross - section upon imposing the cuts @xmath211 gev and @xmath212 gev where @xmath213 is the invariant mass of the final state fermion pairs and @xmath214 stands for either @xmath215 or @xmath126 . these give rise to a large number of possibilities so we only present the `` best '' case when the four possible final states are combined for each energy . for @xmath132 gev we only considered unpolarized initial state electrons and positrons . the results here should be taken with a grain of salt due to the low number of events expected in these kinematic regions . for example , the standard model predicts , for an integrated luminosity of 500 fb@xmath14 and combining all 4 final states , only 40 events when either @xmath216 gev or @xmath217 gev . with this warning , the optimum results occur for @xmath216 gev and are given in table iii . the results are slightly weaker for the case @xmath218 gev . although for a specific case , sensitivities may differ between the @xmath179 and @xmath178 final states , they are generally quite similar . the case @xmath212 gev is not nearly as sensitive to anomalous couplings except for @xmath181 and @xmath10 . these results , along with the previous ones which concentrated on the real @xmath3 production , indicate that the results are dominated by real @xmath3 production . the off - resonance results are roughly a factor of two to three weaker than those given previously for real @xmath3 production and are not likely to contribute much to bounds on tgv s at lep200 . for @xmath6 gev and combining the four final states , the sensitivity is greatest when @xmath217 gev except for a few cases . the results for @xmath118 cuts are slightly less sensitive . considering either the @xmath51 or @xmath52 final states separately we find that for the @xmath51 final states the @xmath217 gev case is more sensitive than the @xmath219 gev case while for the @xmath52 final states they are comparable . with @xmath217 gev the @xmath51 final states are more sensitive to couplings involving photons than the @xmath52 final states . in both cases when we take @xmath217 gev the cross - section is dominated by the @xmath126 pair originating from an on - shell @xmath3 . this gives us the case of single-@xmath3 production which receives large contributions from t - channel photon exchange and hence it is more sensitive to the @xmath1 coupling . the nlc offers the possibility of initial electron polarization . we have included some representative results . an important difference between the two polarizations , which can be seen in fig . 4 , is that the cross - section with left handed electrons is about an order of magnitude larger than for right handed electrons . at the same time the right - handed cross - section is significantly more sensitive to anomalous couplings than the left - handed cross - section . there is therefore a tradeoff between sensitivity and statistics so that in some cases the bounds obtainable for the two polarizations are comparable . the unpolarized results offer no improvement over the polarized results since the right - handed cross - section is overwhelmed by the left - handed contribution . one exception to these comments is when we consider @xmath180 where the left - handed electrons are more constraining for @xmath180 . the results at 1 tev are qualitatively similar to those at 500 gev so we do not repeat the discussion of the previous section but only point out the few points that differ . again , the highest sensitivity is for the constraint @xmath217 gev . the achievable bounds for this case are included in table iii . they are typically 4 to 5 times more constraining than those obtainable at 500 gev ; less than 1% for @xmath10 and @xmath9 which is at the level of loop contributions from new physics . one interesting difference is that the muon mode for right - handed initial electrons provides the most stringent constraints for many of the tgv couplings . as before , the electron final state offers the best measurements of @xmath10 . from the above results it is clear that , although the constraints that could be obtained from off - resonance production are not as tight as those obtained from on - shell @xmath3 production , there is nevertheless considerable information contained in these events it appears to us that the method that makes optimal use of each event is to calculate the probability of each event , irrespective of where it appears in phase space , and compute a likelihood function for the combined probabilities . the only experimental cuts that should be included are those that represent detector acceptance and that are introduced to eliminate backgrounds . we performed a detailed analysis of the measurement of tri - linear gauge boson couplings in the process @xmath220 . we included all tree level contributions to this final state and included finite gauge boson width effects . the off - shell @xmath3 contributions contribute from 20% for the @xmath179 mode at lep200 to 30% and 100% for the electron mode at a 500 gev and 1 tev nlc respectively ( with the kinematic cuts we used ) . clearly , the non - resonant contributions must be included to properly account for the experimental situation . to gauge the sensitivity of this process to anomalous gauge boson couplings we used the @xmath3 decay distributions as a polarimeter to distinguish the longitudinal @xmath3 modes , which are more sensitive to anomalous coupings , from the transverse modes . we implemented this through the use of a quintic differential cross section , with each angular variable divided into 4 bins , and then calculating the likelihood function of non - standard model couplings as compared to the standard model . using this approach we found that at lep200 operating at 175 gev and assuming integrated luminosity of 500 pb@xmath14 , @xmath7 @xmath9 , @xmath10 , @xmath12 , and @xmath13 could be measured to roughly @xmath8 and @xmath180 and @xmath181 to @xmath221 and @xmath222 respectively . it is extremely unlikely that measurements of this precision would reveal anomalous couplings . at a 500 gev nlc with an integrated luminosity of 50 fb@xmath14 , @xmath7 , @xmath18 and @xmath19 , could be measured to roughly @xmath15 , @xmath16 and @xmath223 respectively and @xmath180 and @xmath181 to @xmath224 and @xmath197 respectively . at the 1 tev nlc with 200 fb@xmath14 the corresponding numbers are @xmath225 , @xmath226 , @xmath227 and @xmath228 . the 500 gev nlc measurements are sensitive enough that they should be sensitive to loop contributions to the tgv s while the 1 tev will be able to measure such effects . we studied the sensitivity of the off - mass shell cross sections to anomalous couplings by imposing kinematic cuts on the invariant mass distributions of the outgoing fermion pairs . a cursory analysis found that the off - resonance cross section is relatively sensitive to anomalous couplings and that useful information could be extracted from this region of phase space . although the inclusion of @xmath3 decays to fermions and the non - resonant diagrams does not alter the precision to which the tgv s can be measured they do change the cross sections and kinematic distributions at the same level as radiative corrections and must be taken into account for an accurate comparison between experiment and theory . the optimal strategy to maximize the information contained in each event is to construct a likelihood function based on the four vector of each of the outgoing fermions on an event by event basis , putting them through a realistic detector simulation . this would make the best use of the information whether it be on the @xmath3 resonance or not . kinematic cuts should only be introduced to reduce backgrounds . since the precision of these measurements is beyond the level of loop induced radiative corrections it is crucial that radiative corrections are well understood and included in event generators used in the study of these processes . progress is being made along these lines as exemplified by the monte carlo event generators ; excalibur @xcite , whopper , @xcite eeww @xcite , wwf @xcite , and wwgenp @xcite . the authors benefited greatly from many helpful conversations , communications , and suggestions during the course of this work with tim barklow , genevieve blanger , pat kalyniak , dean karlen , and paul madsen . this research was supported in part by the natural sciences and engineering research council of canada and les fonds fcar du quebec . d. schaile , _ proceedings of the xxvii int . conf . on high energy physics _ , bussey , i.g . knowles , glasgow , uk , 20 - 27 july 1994 ( inst . of physics publishing , 1995 ) p. 27 . k. hagiwara _ et al _ , nucl b282 * , 253 ( 1987 ) . d. zeppenfeld , phys . let . * 183b * , 380 ( 1987 ) ; d. treille _ et al _ , proceedings of the ecfa workshop on lep 200 , ed . a. bhm and w. hoogland , aachen ( 1986 ) , cern 87 - 08 , vol.2 , p.414 . dicus , k. kallianpur , phys . rev . * d32 * , 35 ( 1985 ) ; m.j . duncan , g.l . kane , phys . lett . * 55 * , 773 ( 1985 ) ; e.n.argyres and c.g.papadopoulos , phys . b263 * , 298(1991 ) . g.kane , j. vidal , c.p . yuan , phys . rev . * d39 * , 2617 ( 1990 ) , and references therein . some recent reviews on physics at high energy @xmath5 colliders are : m. peskin , proceedings of the 1987 slac summer institute on particle physics , slac - pub-4601 ( 1988 ) ; d. treille , proceedings of the workshop on physics and experiments with linear @xmath5 colliders , waikoloa hawaii , april 1993 ( world scientific ; in press ) ; a. djouadi and p.m. zerwas , proceedings of _ beyond the standard model iii _ , ed . s. godfrey and p. kalyniak , ottawa canada , june 1992 , ( world scientific , singapore , 1993 ) p. 204 . a. miyamoto , proceedings of workshop on physics and experiments with linear @xmath5 colliders , waikoloa , hawaii , ed . f.a . harris , s.l . olsen , s. pakvasa , and x. tata , ( world scientific , singapore , 1994 ) p 141 . b. wiik , proceedings of the workshop on physics and experiments with linear @xmath5 colliders , waikoloa hawaii , ed . f.a . harris , s.l . olsen , s. pakvasa , and x. tata , ( world scientific , singapore , 1994 ) . proceedings of the first workshop on japan linear collider ( jlc ) , kek , oct 24 - 25 , 1989 , kek report 90 - 2 ( 1990 ) ; proceedings of the second workshop on japan linear collider , ed . s. kawabata , nov . 6 - 8 , 1990 kek proceedings 91 - 10 nov . 1991 . c. ahn _ _ , _ opportunities and requirments for experimentation at a very high energy @xmath5 collider _ , slac report slac-0329 ( 1988 ) ; proceedings of the workshop on physics at future accelerators , cern yellow report 87 - 07 ( 1987 ) . for a recent comprehensive review of trilinear gauge boson couplings see : h. aihara _ et al . _ , to appear in _ electroweak symmetry breaking and beyond the standard model _ , eds . t. barklow , s. dawson , h. haber and j. siegrist ( world scientific ) , hep - ph/9503425 . for further recent reviews on tgv s see : i. hinchliffe , to appear in the proceedings of the international symposium on vector boson self interactions , ucla , feb 1 - 3 , 1995 , hep - ph/9504206 ; f. boudjema , to appear in the proceedings of _ beyond the standard model iv _ , lake tahoe , california , dec . 12 - 16 , 1994 , hep - ph/9504409 ; c.p . burgess , s. godfrey , h. knig , d. london , and i. maksymyk , phys . rev . * d49 * , 6115 ( 1994 ) . s. dawson and g. valencia , nucl . * b439 * , 3 ( 1995 ) . k. hagiwara , s. ishihara , r. szalapski , and d. zeppnefeld , phys . b283 * , 353 ( 1992 ) ; phys . rev . * d48 * , 2182 ( 1993 ) . ( cdf collaboration ) , phys . lett . * 74 * , 1936 ( 1995 ) ; j. ellison _ ( d0 collaboration ) , proceedings of the _ dpf94 conference _ , albuquerque , nm , august 1994 ; h. aihara , to appear in the proceedings of the _ international symposium on vector boson self - interactions _ , ucla , february 1995 ; t.a . fuess ( cdf collaboration ) , proceedings of the _ dpf94 conference _ , albuquerque , nm , august 1994 ; et al . _ , ( cdf collaboration ) , proceedings of the _ 27th international conference on high energy physics _ , glasgow , scotland , july 20 - 27 , 1994 ; f. abe _ et al . _ , ( cdf collaboration ) , fermilab - pub-95/036-e , submitted to phys ; s. abachi _ et al . _ , ( d0 collaboration ) , fermilab - pub-95/101-e , hep - ex/9505007 . u. baur and d. zeppenfeld , nucl . b308 * , 127 ( 1988 ) ; see also h. aihara _ et al . e. yehudai , phys . rev . * d 41 * , 33 ( 1990 ) ; * d44 * , 3434 ( 1991 ) ; s.y . choi and f. schrempp , phys . 272b * , 149 ( 1991 ) ; i.f . ginzburg , g.l . kotkin , s.l . panfil and v.g . serbo , nucl . phys . * b228 * , 285 ( 1983 ) s. godfrey and k.a . peterson , carleton university report ocip / c-92 - 7 , hep - ph/9302319 . g. couture , s. godfrey , p. kalyniak phys . rev . * d39 * , 3239 ( 1989 ) ; phys . rev . * d42*,1841 ( 1990 ) ; phys . lett . * b218 * , 361 ( 1989 ) . m. raidal , university of helsinki report hu - seft r 1994 - 16 , hep - ph/9411243 ; k. cheung , s. dawson , t. han , and g. valencia , phys . * d51 * , 5 ( 1995 ) ( hep - ph/9403358 ) ; a. queijeiro , phys . lett . * b193 * , 354 ( 1987 ) ; k.j . abraham and c.s . kim , phys . b301 * , 430 ( 1993 ) ; s.j . brodsky , t.g . rizzo , and i. schmidt , slac - pub-95 - 6904 , hep - ph/9505441 . d. choudhury and f. cuypers , hep - ph/9312308 . g. couture and g. blanger , phys . * d49 * , 5720 ( 1994 ) . g. couture , s. godfrey , and r. lewis , phys . * d45 * , 777 ( 1992 ) ; g. couture and s. godfrey , phys . rev . * d49 * , 5709(1994 ) . s. ambrosanio and b. mele , nucl . phys . * b374 * , 3 ( 1992 ) . k. hagiwara _ et al . _ , b365 * , 544 ( 1991 ) . g. couture and s. godfrey , phys . * d50 * , 5607(1994 ) ; k.j . abraham , j. kalinowski , and p. ciepko , proceedings of the xvii warsaw symposium on elementary particle physics , kazimierz , may 23 - 27 , 1994 , hep - ph/9407223 . borisov , v.n . larin , and f.f . tikhonin , zeit . c41 * , 287 ( 1988 ) . t. barklow , _ proceedings of the 1st workshop on physics with linear colliders _ , saariselka , finland , sept . 9 - 14 ( 1992 ) ; t. barklow , _ proceedings of the dpf94 conference : 1994 meeting of the division of particls and fields of the aps _ , albuquerque , nm , aug2 - 6 , 1994 . c.g . papadopoulos , cern report cern - th/95 - 46 , hep - ph/9503276 . sekulin , phys . lett . b338 * , 369 ( 1994 ) . m. bilenky , j.l . kneur , f.m . renard and d. schildknecht , nucl . phys . * b409 * , 22 ( 1993 ) ; g. gounaris , j.l . kneur , j. layssac , g. moultaka , f.m . renard , and d. schildknecht , _ @xmath5 collisions at 500 gev , the physics . p. zerwas , desy ( hamburg , 1991 ) ; k.j.f . gaemers and m.r . van velzen , z. phys . * c43 * , 103 ( 1989 ) ; m. diehl and o. nachtmann , z. phys . * c62 * , 397 ( 1994 ) . a.a . pankov and n. paver , phys . lett . * b324 * , 224 ( 1994 ) ; * b346 * , 115 ( 1995 ) ; a.a . likhoded , a.a . pankov , n. paver , m.v . shevlyagin and o.p . yushchenko , uts - dft-93 - 22 ( 1993 ) . for a detailed review see w. beenakker and a. denner , international journal of modern physics * a9 * , 4837(1994 ) . p. kalyniak , p. madsen , n. sinha , and r. sinha , phys . rev . * d48 * , 5081 ( 1993 ) . m. gintner and s. godfrey , phys . lett . * b * , ( in press ) , hep - ph/9510324 . y. kurihara , d. perret - gallix , and y. shimizu , phys . b349 * , 367 ( 1995 ) ( hep - ph/9412215 ) ; m. nowakowski and a. pilaftsis , z. phys . * c * ; a. aeppli , f. cuypers and g. j van oldenborgh , phys . lett . * b314 * , 413 ( 1993 ) . berends , r. pittau , and r. kleiss , nucl . b424 * , 308(1994 ) ; g. montagna , o. nicrosini , g. passarino , and f. piccinini , cern report cern - th.7497/94 ( 1994 , unpublished ) ; d. bardin , a. leike , and t. riemann , cern report cern - th.7478/94 ( 1994 , unpublished ) , hep - ph/9410361 ; g. montagna , o. nicrosini , g. passarino , and f. piccinini , cern report cern - th.7497/94 ( 1994 ; unpublished ) ; j. fujimoto _ et al . _ , kep report kek 94 - 46 ; t. ishikawa _ et al . _ , proceedings of the _ viith workshop on high energy physics and quantum field theory _ , sotchi , russia , oct . 7 - 14 1992 . h. iwasaki , int . journal of mod . * a7 * , 3291(1992 ) ; h. neufeld , z. phys . * c17 * , 145 ( 1983 ) ; e.n . argyres and c.g . papadopoulos , phys . * 263b * , 298 ( 1991 ) ; o. philipsen , z. phys . * c54 * , 643 ( 1992 ) ; g. couture and j.n . ng , z. phys . * c32 * , 579 ( 1986 ) ; j.c . romao and p. nogueira , z. phys . * c42 * , 263 ( 1989 ) ; o. cheyette , phys . lett . * 137b * , 431 ( 1984 ) ; c.g . papadopoulos , phys . b333 * , 202 ( 1994 ) ; e. gabrielli , mod . phys . lett . * a1 * , 465 ( 1986 ) . o. nicrosini and l. trendadue , nucl . phys . * b318 * , 1 ( 1989 ) ; l. trentadue _ et al . _ , in : z physics at lep1 , ed . g. altarelli , cern yellow report cern 89 - 08 ( cern , geneva , 1989 ) p129 ; d. bardin , m. bilenky , a. olchevski , and t. riemann , phys . lett . * b308 * , 403 ( 1993 ) ; f.a . berends , r. pittau , and r. kleiss , nucl . * b426 * , 344(1994 ) ; hep - ph/9409326 ; r. pittau , hep - ph/9406233 . w. beenakker , k. kolodziej , and t. sack , phys . lett . * b258 * , 469 ( 1991 ) ; w. beenakker and a. denner , desy 94 - 051 ; w. beenakker , f.a . berends , and t. sack , nucl . b367 * , 287 ( 1991 ) . j. fleischer , f. jegerlehner and m. zralek , z. phys . * c42 * , 409 ( 1989 ) ; j. fleischer , f. jegerlehner , k. koodziej , and g.j . van oldenborgh , preprint psi - pr-94 - 16 ( 1994 ; unpublished ) , hep - ph/9405380 . g. jan van oldenborgh , p.j . franzini , and a. borrelli , hep - ph/9402298 . d. bardin _ et al . _ , cern report cern - th.7295/94 , hep - ph/9406340 ; j. fujimoto _ et al . _ , hep - ph/9407308 ; g. montagna _ et al . _ , cern report cern - th.7497/94 , hep - ph/9411332 ; yu.l . dokshitzer , v.a . khoze , l.h . orr , and w.j . stirling , phys . * b313 * , 171 ( 1993 ) . m. gintner , s. godfrey and g. couture , in preparation . for recent reviews of effective lagrangians see : m.b . einhorn proceedings of the _ workshop on physics and experiments with linear @xmath5 colliders_,eds . f.a . harris _ et al . _ , waikoloa hawaii , april 26 - 30 , 1993 ( world scientific , 1994 ) p. 122 ; f. boudjema , _ ibid _ , p.712 . j. bagger , s. dawson , and g. valencia , nucl b399 * , 364 ( 1993 ) . k. gaemers and g. gounaris , z. phys . * c1 * , 259 ( 1979 ) . k .- i hikasa phys . rev . * d33 * , 3203 ( 1986 ) ; k. hagiwara _ et al _ , nucl . phys . * b282 * , 253 ( 1987 ) . marciano , a. queijeiro , phys . rev . * d 33 * , 3449 ( 1986 ) ; f. boudjema , k. hagiwara , c. hamzaoui , and k. numata , phys . rev . * d43 * , 2223 ( 1991 ) . u. baur and d. zeppenfeld , nucl . * b308 * , 127 ( 1988 ) . bardeen , r. gastmans , and b. lautrup , nucl . phys . * b46 * , 319 ( 1972 ) ; e.n . et al . , _ nucl . phys . * b391 * , 23 ( 1993 ) ; j. papavassiliou and k. philoppides , phys . rev . * d48 * , 4255 ( 1993 ) ; g. couture and j.n . ng , z. phys . * c35 * , 65 ( 1987 ) ; g. couture _ et al . _ , phys . rev . * d38 * , 860 ( 1988 ) . m. kuroda , f.m . renard , and d. schildnecht , phys . lett . * b183 * , 366 ( 1987 ) . a. longhitano , nucl . phys . * b188 * , 118 ( 1981 ) ; t. appelquist and c. bernard , phys . rev . * d22 * , 200 ( 1980 ) ; b. holdom , phys . lett . * b258 * , 156 ( 1991 ) ; a. falk , m. luke , and e. simmons , nucl . phys . * b365 * , 523 ( 1991 ) ; t. appelquist and g.h . wu , phys . rev . * d48 * , 3235 ( 1993 ) ; see also j.f . donoghue , e. golowich , and b.r . holstein , _ dynamics of the standard model _ , ( cambridge university press , 1994 ) . w. buchmller and d. wyler , nucl . phys . * b268 * , 621 ( 1986 ) ; a. de rjula , m.b . gavela , p. hernandez and e. mass , nucl . phys . * b384 * , 3 ( 1992 ) d. zeppenfeld , j.a.m . vermaseren and u. baur , nucl . phys . * b375 * , 3 ( 1992 ) . u. baur and d. zeppenfeld , mad / ph/878 . r. kleiss and w. j. stirling , nucl . * b262 * , 235 ( 1985 ) ; z.xu d .- h.zhang l. chang nucl . phys . * b291 * , 392 ( 1987 ) . see for example v. barger and r. phillips , _ collider physics _ , ( addison - wesley publishing company , 1987 ) . m. peskin , proceedings of the 1987 slac summer institute on particle physics , slac - pub-4601 ( 1988 ) . burke , proceedings of the 1990 slac summer institute on particle physics , slac - pub-5418 ( 1991 ) . h. anlauf _ proceedings of workshop on physics and experiments with linear @xmath5 colliders , waikoloa , hawaii , ed . f.a . harris , s.l . olsen , s. pakvasa , and x. tata , ( world scientific , singapore , 1994 ) p 708 . g. montagne , o. nicrosini , f. piccinini , comp . . commun . * 90 * , 141 ( 1995 ) . lccccccc mode & @xmath180 & @xmath181 & @xmath232 & @xmath233 & @xmath234 & @xmath12 & @xmath13 + + @xmath179 & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 + @xmath178 & @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & @xmath248 + combined & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 + + @xmath179 & @xmath256 & @xmath257 & @xmath258 & @xmath259 & @xmath16 & @xmath16 & @xmath260 + @xmath178 & @xmath256 & @xmath257 & @xmath261 & @xmath259 & @xmath16 & @xmath16 & @xmath260 + combined & @xmath262 & @xmath256 & @xmath263 & @xmath264 & @xmath223 & @xmath223 & @xmath223 + + @xmath179 & @xmath265 & @xmath266 & @xmath15 & @xmath267 & @xmath268 & @xmath267 & @xmath267 + @xmath178 & @xmath265 & @xmath266 & @xmath15 & @xmath267 & @xmath268 & @xmath267 & @xmath267 + combined & @xmath269 & @xmath270 & @xmath271 & @xmath268 & @xmath272 & @xmath273 & @xmath273 + ccccccccc initial state & cut & @xmath180 & @xmath181 & @xmath274 & @xmath233 & @xmath234 & @xmath12 & @xmath13 + + @xmath140 & @xmath275 gev & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & @xmath282 + + @xmath206 & @xmath283 gev & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 & @xmath289 & @xmath290 + @xmath291 & @xmath283 gev & @xmath292 & @xmath292 & @xmath293 & @xmath294 & @xmath295 & @xmath296 & @xmath297 + @xmath140 & @xmath283 gev & @xmath298 & @xmath299 & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath304 + + @xmath206 & @xmath283 gev & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 & @xmath310 & @xmath311 + @xmath291 & @xmath283 gev & @xmath292 & @xmath312 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & @xmath316 + @xmath140 & @xmath283 gev & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath316 & @xmath311 & @xmath311 +

we calculate transition form factors , , and , and study semileptonic decay of based on qcd sum rule method . we compare our results of the ratios of , , , and the total decay branching ratio of with experimental data , and find that they are consistent . ( 0,0 ) ( 340,230)bihep - th-2003 - 30 ( 340,210 ) * *

hep-ph0308259

semileptonic decay of charm meson is important for studying strong and weak interactions . it can be used to test techniques developed for solving perturbative and nonperturbative problems in quantum chromodynamics ( qcd ) , and to extract elements of cabibbo - kobayashi - maskawa ( ckm ) matrix . semileptonic decay is simpler than hadronic decay of charm meson because leptons do not involve strong interaction . the amplitude of semileptonic decay can be decomposed into several transition form factors due to lorentz property of the hadronic matrix element . the form factors include all the nonperturbative effects . several methods can be used to treat these problems , such as quark model , qcd sum rule and lattice qcd , among which , qcd sum rule and lattice qcd are based on the first principle of qcd . the method of qcd sum rules @xcite has been widely used in hadronic physics since its establishment in the late 1970s . for semileptonic decays of charm meson , @xmath9 was firstly studied in qcd sum rule method with three - point correlation function @xcite . several years later , qcd sum rule method was extended to semileptonic decays of @xmath10 meson , @xmath11 @xcite and @xmath12 @xcite . in these works , form factors @xmath13 and @xmath14 are calculated at the point @xmath15 , where @xmath16 is the momentum transfer squared . for the whole physical region of @xmath17 , the form factors are either assumed to be pole dominance @xmath18 , or a linear approximation was used . in refs @xcite , @xmath19 , @xmath20 and @xmath21 , @xmath22 were studied , where qcd sum rule method was extended to a very large value of @xmath16 with a careful treatment of non - landau - type singularities . @xmath23 decays to @xmath24 and @xmath25 final states were studied in @xcite . in this work , we study @xmath5 in qcd sum rule method . this decay mode has been measured in experiment long time before @xcite . now it is necessary to analyze it theoretically . we calculate up to contributions of operators of dimension 6 in the operator product expansion ( ope ) and keep the mass of s - quark . in our result , the large contributions come from unit operator @xmath26 ( result of perturbative diagram ) and condensate of operators of dimension 3 . operator of dimension 5 gives smaller contribution . the contributions of operators of dimension 6 are negligible . when calculating contribution of perturbative diagram and gluon condensate ( operator of dimension 4 ) , cutcosky s rule has been used . therefore subtraction of continuum contribution is conveniently performed not only for perturbative diagram but also for contribution of gluon condensate . after some long steps of calculation , we find the contributions of diagrams for the gluon condensate cancel each other , so there is no gluon - condensate contribution in @xmath0 transition . this is our new finding . there are two independent borel parameters @xmath27 and @xmath28 in manipulating three - point correlation functions . in general , to simplify the numerical analysis , a fixed ratio of @xmath29 was taken in recent references . in this work , we make the numerical analysis in the whole region of independent @xmath27 and @xmath28 to select the stability window " , so it is different from taking a fixed ratio of these two borel parameters . we calculate @xmath30 transition form factors @xmath1 , @xmath2 , @xmath3 , @xmath4 and the branching ratio of @xmath31 . our result of the ratios of @xmath6 , @xmath7 , @xmath8 ( @xmath32 and @xmath33 denote decay width of @xmath34 to @xmath35 meson in longitudinal and transverse polarization , respectively ) , and the total branching fraction are in agreement with experimental data . recently , just before this work is finished , we find that @xmath36 was also calculated in ref.@xcite . however , their analysis is very different from ours . first , by carefully choosing the requirement that the double borel parameters @xmath27 and @xmath28 should not be too large for keeping the continuum contribution small , and at the same time , @xmath27 and @xmath28 should not be too small for keeping the truncated ope series effective , i.e. , keeping the contributions of higher dimension operators small , we get very different stability window " for the borel parameters . second , our results of the transition form factors are different from theirs . especially for @xmath4 , they got negative value , however , we get positive . using their values of the form factors , although one can get the total branching ratio of @xmath5 to be compatible with experimental result , the ratio of @xmath8 will be too large . but in our case , @xmath37 which is consistent with world average @xmath38 . the paper is organized as follows . in section 2 , we briefly introduce the qcd sum rule method used in this work . section 3 is the calculation . section 4 is the numerical analysis and discussion . section 5 is devoted to the summary . to calculate the transition form factors of semileptonic @xmath23 meson decays , the standard procedure in qcd sum rule method is to consider the three - point correlation function defined as @xmath39 with the currents having the same quantum numbers as the relevant mesonic states under consideration , which are defined by : 1 ) the current of @xmath23 channel , @xmath40 ; 2 ) the current of weak transition : @xmath41 ; 3 ) the current of @xmath35 channel : @xmath42 . on one hand , inserting a complete set of intermediate hadronic states into the correlation function , and using the double dispersion relation , one can express the correlation function in terms of a set of hadronic states , @xmath43 with @xmath44 where @xmath45 and @xmath46 denote the complete set of hadronic states of @xmath35 and @xmath23 channels , respectively . @xmath47 and @xmath48 are the four - momentum of @xmath45 and @xmath46 states , @xmath49 , @xmath50 , and @xmath51 . integrate over @xmath52 and @xmath53 in eq . ( [ c2 ] ) , we can obtain @xmath54 separate the ground states of @xmath23 and @xmath35 channels apparently , the above equation becomes @xmath55 the weak transition matrix element @xmath30 can be decomposed as @xmath56 \varepsilon^*\cdot q \frac{a_2(q^2)}{m_{d_s}+m_{\phi } } \nonumber\\ & & -i\frac{2m_\phi \varepsilon^*\cdot q}{q^2}q_\mu a_0(q^2),\nonumber\end{aligned}\ ] ] where @xmath51 . the vacuum - to - meson transition amplitudes can be parameterized through defining the corresponding decay constants , @xmath57 finally the correlation function can be expressed in terms of meson decay constants and @xmath58 transition matrix element , @xmath59 on the other hand , the correlation function of eq . ( [ correlator ] ) can be evaluated at negative values of @xmath60 and @xmath61 by the operator - product expansion in qcd , in which the time - ordered current operators in eq . ( [ correlator ] ) is expanded in terms of a series of local operators with increasing dimensions , @xmath62 where @xmath63 s are wilson coefficients , @xmath26 is the unit operator , @xmath64 is the local fermion field operator of light quarks , @xmath65 is gluon strength tensor , @xmath66 and @xmath67 are the matrices appearing in the procedure of calculating the wilson coefficients . sandwich the left and right hand sides of eq . ( [ opef ] ) between two vacuum states , we get the correlation function in terms of wilson coefficients and condensates of local operators , @xmath68 for later convenience , we shall reexpress the above equation . in general , it can be expressed in terms of six independent lorentz structures @xmath69 each @xmath70 includes perturbative and condensate contributions @xmath71 where @xmath72 , @xmath73 , @xmath74 are contributions of condensates of dimension 3 , 4 , 5 , 6 , @xmath73 in eq . ( [ conden ] ) . in next section we can see that perturbative contribution and gluon condensate contribution can be finally written in the from of dispersion integration , @xmath75 we approximate the contribution of higher resonances and continuum states as integrations over some thresholds @xmath76 and @xmath77 in the above equations . then equate the two representations of the correlation function in eq . ( [ c3 ] ) and ( [ c4 ] ) , we can get an equation for the form factors . to improve such equation , we make borel transformation over @xmath60 and @xmath61 in both sides , which can further suppress higher resonance contribution . the definition of borel transformation to any function @xmath78 is @xmath79 some examples of borel transformation is given in the following , @xmath80 equating the two representations of the correlation function , subtracting the higher resonances and continuum contribution , and performing borel transformation in both variables @xmath60 and @xmath61 , we finally obtain the sum rules for the form factors , @xmath81 , \nonumber \end{aligned}\ ] ] where @xmath82 denotes borel transforming @xmath70 in both variables @xmath60 and @xmath61 , @xmath83 and @xmath84 are borel parameters . because we have subtracted the higher resonance and continuum contribution , now the dispersion integration for perturbative and gluon condensate contribution should be performed under the threshold , @xmath85 in the next section , we will explain the technique of calculating the wilson coefficients and given the resulted form of the sum rules for the form factors . in this work , we first calculate the wilson coefficients in the operator - product expansion @xcite , then extract the relevant terms @xmath70 s for the sum rules of the form factors in eq . ( [ formsr ] ) . we will not present the result of each wilson coefficient here because their forms are very tedious . we only give the results of the form factors according to the contribution of each condensate . the diagram for the perturbative contribution is depicted in fig.1 . the leading order in @xmath86 expansion is considered here . this contribution amounts to wilson coefficient @xmath87 in ope representation of the correlation function in eq . ( [ conden ] ) . we can write down this amplitude ( see fig.1 ) , ( 150,150)(-15,-25.98 ) ( -15,-25.98)(0,0)3 ( 0,0)(40,69.28 ) ( 0,0)(40,69.28)(40,69.28)(80,0)(80,0)(0,0 ) ( 80,0)(95,-25.98)3(40,69.28)(40,94.28)14 ( 41,70)@xmath88 ( 80.5,0.5)@xmath89 ( -5.5,0.5)@xmath90 ( -24,-16)@xmath91 ( 95,-16)@xmath92 ( 10,34)@xmath93 ( 38,-10)@xmath94 ( 64,34)@xmath94 @xmath95 . \label{pert1}\end{aligned}\ ] ] the diagrams for the contributions of @xmath107 and @xmath108 are shown in fig . 2 . the contribution of fig . 2(b ) is zero after double borel transformation in both variables @xmath60 and @xmath61 because only one variable appears in the denominator @xmath109 . so we will not consider fig . 2(b ) in the following . the contribution of fig . 2(a ) to the correlation function is ( 300,150)(-15,-25.98 ) ( -15,-25.98)(0,0)3 ( 0,0)(40,69.28 ) ( 0,0)(40,69.28)(40,69.28)(80,0)(80,0)(50,0)(30,0)(0,0 ) ( 80,0)(95,-25.98)3(40,69.28)(40,94.28)14(47,-2.7)@xmath110 ( 26.8,-2.7)@xmath110(41,70)@xmath88(80.5,0.5)@xmath89(-5.5,0.5)@xmath90 ( 185,-25.98)(200,0)3 ( 200,0)(240,69.28 ) ( 200,0)(240,69.28)(240,69.28)(255,43.3 ) ( 265,25.98)(280,0 ) ( 280,0)(200,0 ) ( 280,0)(295,-25.98)3(240,69.28)(240,94.28)14 ( 250.2,41.6)@xmath110(261,23.8)@xmath110 ( 241,70)@xmath88(280.5,0.5)@xmath89(194.5,0.5)@xmath90 ( 33,-40)@xmath111(233,-40)@xmath112 @xmath113 where @xmath114 and @xmath115 are the propagators of @xmath94 and @xmath93 quarks , respectively . move the quark field operators @xmath116 and @xmath117 together , we get @xmath118_{\alpha\beta } , \label{pi2a}\ ] ] where @xmath119 and @xmath120 are dirac spinor indices . the matrix element @xmath121 and @xmath122 using the technique explained in @xcite , @xmath123 , \label{qq } \end{aligned}\ ] ] @xmath124 and @xmath125 in the above are the color indices , @xmath126 is the quark mass , and the ellipsis stands for terms of higher orders in @xmath89 and @xmath90 expansion . from eq . ( [ qq ] ) we know that fig.2(a ) contributes to the coefficients of quark condensate @xmath127 , mixed quark - gluon condensate @xmath128 and the four - quark condensate @xmath129 . substitute eq . ( [ qq ] ) into ( [ pi2a ] ) and integrate over the coordinates @xmath89 and @xmath90 , we can obtain explicitly the coefficients of these condensates contributed by fig.2(a ) . the diagrams for the contribution of bi - gluon operator are depicted in fig.3 . they are calculated in the fixed - point gauge , in which the gauge fixing condition is taken to be @xmath131 @xcite . then the external gauge field can be expressed directly in terms of the color field strength tensor @xcite , @xmath132 which is expanded to the first order to be , @xmath133 in the following calculation , it is convenient to transform @xmath134 to the momentum space , @xmath135 ( 300,200)(-15,-25.98 ) ( -11.25,100.515)(0,120)3 ( 0,120)(30,171.96 ) ( 0,120)(30,171.96)(30,171.96)(60,120)(60,120)(0,120 ) ( 60,120)(71.25,100.515)3(30,171.96)(30,190.71)14 ( 0,120)(30,40,60)22(0,120)(30,0,20)22 ( 19.88,137,25)@xmath110(23.5,129)@xmath110 ( 88.75,100.515)(100,120)3 ( 100,120)(130,171.96 ) ( 100,120)(130,171.96)(130,171.96)(160,120)(160,120)(100,120 ) ( 160,120)(171.25,100.515)3(130,171.96)(130,190.71)14 ( 130,171.96)(30,240,260)22(130,171.96)(30,280,300)22 ( 120.4,138.8)@xmath110(131.3,138.8)@xmath110 ( 188.75,100.515)(200,120)3 ( 200,120)(230,171.96 ) ( 200,120)(230,171.96)(230,171.96)(260,120)(260,120)(200,120 ) ( 260,120)(271.25,100.515)3(230,171.96)(230,190.71)14 ( 260,120)(30,160,180)22(260,120)(30,120,140)22 ( 232.3,136)@xmath110(228.125,128)@xmath110 ( -11.25,-19.485)(0,0)3 ( 0,0)(30,51.96 ) ( 0,0)(30,51.96)(30,51.96)(60,0)(60,0)(0,0 ) ( 60,0)(71.25,-19.485)3(30,51.96)(30,70.71)14 ( 11.25,19.485)(-1.74,26.985)22(18.75,32.475)(5.76,39.975)22 ( -5.625,23.5)@xmath110(1.65,36.5)@xmath110 ( 88.75,-19.485)(100,0)3 ( 100,0)(130,51.96 ) ( 100,0)(130,51.96)(130,51.96)(160,0)(160,0)(100,0 ) ( 160,0)(171.25,-19.485)3(130,51.96)(130,70.71)14 ( 122.5,0)(122.5,-15)22(137.5,0)(137.5,-15)22 ( 119.875,-16.5)@xmath110(135.25,-16.5)@xmath110 ( 188.75,-19.485)(200,0)3 ( 200,0)(230,51.96 ) ( 200,0)(230,51.96)(230,51.96)(260,0)(260,0)(200,0 ) ( 260,0)(271.25,-19.485)3(230,51.96)(230,70.71)14 ( 248.75,19.485)(261.74,26.985)22(241.25,32.475)(254.24,39.975)22 ( 256.5,24.75)@xmath110(250,37.5)@xmath110 ( 23,90)(a ) ( 125,90)(b ) ( 225,90)(c ) ( 23,-35)(d ) ( 125,-35)(e ) ( 225,-35)(f ) then the amplitude can be written down in the momentum space by following the standard feynman rule . again , as what we did in the previous subsections , we move the gluon strength tensor operator together : @xmath136 . then using the following decomposition to obtain the bi - gluon condensate , @xmath137 in which @xmath138 is the abbreviation of @xmath139 . in the evaluation of the diagrams of fig.3 some types of loop integrals encountered are treated at first by derivatives with respect to the quark masses , then transform them to dispersion integrals by using cutkosky s rule and with help of @xmath26 , @xmath140 and @xmath141 functions given previously . for instance , @xmath142 ^ 2[(k+p_1)^2-m_1 ^ 2]^2}\nonumber\\ & = & \frac{\partial}{\partial m_2 ^ 2}\frac{\partial}{\partial m_1 ^ 2 } \int d^4 k \frac{k_\mu k_\nu}{(k^2-m^2)[(k+p_2)^2-m_2 ^ 2][(k+p_1)^2-m_1 ^ 2]}\nonumber\\ & = & -2\pi i\frac{\partial}{\partial m_2 ^ 2}\frac{\partial}{\partial m_1 ^ 2 } \int ds_1 ds_2 \frac{i_{\mu\nu}}{(s_1-p_1 ^ 2)(s_2-p_2 ^ 2)},\end{aligned}\ ] ] and @xmath143 ^ 2[(k+p_1)^2-m_1 ^ 2]^2}\nonumber\\ & = & \frac{\partial}{\partial m_2 ^ 2}\frac{\partial}{\partial m_1 ^ 2 } \int d^4 k \frac{k_\mu k_\nu k\cdot p_2}{(k^2-m^2)^2[(k+p_2)^2-m_2 ^ 2][(k+p_1)^2-m_1 ^ 2]}\nonumber\\ & = & -2\pi i\frac{\partial}{\partial m^2}\frac{\partial } { \partial m_2 ^ 2}\frac{\partial}{\partial m_1 ^ 2 } \int ds_1 ds_2 \frac{-\frac{1}{2}(s_2-m_2 ^ 2+m^2)i_{\mu\nu}}{(s_1-p_1 ^ 2)(s_2-p_2 ^ 2)},\end{aligned}\ ] ] where the term @xmath144 comes from the @xmath98 functions @xmath145 $ ] with the substitution @xmath106 when using the cutkosky s rule . after some long steps of calculation , we finally find that the contributions of the diagrams ( a ) to ( f ) in fig.3 cancel each other . therefore there are no contributions of gluon condensate in @xmath146 transition . the diagrams for quark - gluon mixing and four - quark contributions are depicted in fig.4 and fig.5 , respectively . the techniques are similar to that explained in previous subsections . we only give some different points here . ( 400,100)(-15,-25.98 ) ( -11.25,-19.485)(0,0)3 ( 0,0)(30,51.96 ) ( 0,0)(30,51.96)(30,51.96)(60,0)(60,0)(37.5,0)(22.5,0)(0,0 ) ( 60,0)(71.25,-19.485)3(30,51.96)(30,70.71)14(35.25,-2.025)@xmath110 ( 19,-2.025)@xmath110 ( 0,0)(30,20,60)23(25.5,10)@xmath110 ( 88.75,-19.485)(100,0)3 ( 100,0)(130,51.96 ) ( 100,0)(130,51.96)(130,51.96)(160,0)(160,0)(137.5,0)(122.5,0)(100,0 ) ( 160,0)(171.25,-19.485)3(130,51.96)(130,70.71)14(135,-2.025)@xmath110 ( 119,-2.025)@xmath110 ( 160,0)(30,120,160)23(127.5,8.5)@xmath110 ( 188.75,-19.485)(200,0)3 ( 200,0)(230,51.96 ) ( 200,0)(230,51.96)(230,51.96)(241.25,32.475 ) ( 248.75,19.485)(260,0 ) ( 260,0)(200,0 ) ( 260,0)(271.25,-19.485)3(230,51.96)(230,70.71)14 ( 236.5,31.2)@xmath110(245,17)@xmath110 ( 260,0)(30,140,180)23(233.5,17.5)@xmath110 ( 288.75,-19.485)(300,0)3 ( 300,0)(330,51.96 ) ( 300,0)(330,51.96)(330,51.96)(341.25,32.475 ) ( 348.75,19.485)(360,0 ) ( 360,0)(300,0 ) ( 360,0)(371.25,-19.485)3(330,51.96)(330,70.71)14 ( 336.5,31.2)@xmath110(345,17)@xmath110 ( 330,51.96)(30,240,280)23(332.25,19.5)@xmath110 ( 22,-28)(a ) ( 122,-28)(b ) ( 222,-28)(c ) ( 322,-28)(d ) ( 300,200)(-15,-25.98 ) ( -11.25,100.515)(0,120)3 ( 0,120)(30,171.96 ) ( 0,120)(30,171.96)(60,120)(37.5,120)(22.5,120)(0,120 ) ( 60,120)(71.25,100.515)3(30,171.96)(30,190.71)14 ( 33.3,117.975)@xmath110 ( 17.5,117.975)@xmath110 ( 30,171.96)(41.25,152.475 ) ( 48.75,139.485)(60,120 ) ( 36.5,151.2)@xmath110(45,137)@xmath110 ( 15,147.6)(48,120)1.68 ( 88.75,100.515)(100,120)3 ( 100,120)(130,171.96 ) ( 100,120)(130,171.96)(160,120)(137.5,120)(122.5,120)(100,120 ) ( 160,120)(171.25,100.515)3(130,171.96)(130,190.71)14 ( 133.3,117.975)@xmath110 ( 117.5,117.975)@xmath110 ( 130,171.96)(141.25,152.475 ) ( 148.75,139.485)(160,120 ) ( 136.5,151.2)@xmath110(145,137)@xmath110 ( 115,147.6)(154.375,128.8)1.68 ( 188.75,100.515)(200,120)3 ( 200,120)(230,171.96 ) ( 200,120)(230,171.96)(260,120)(237.5,120)(222.5,120)(200,120 ) ( 260,120)(271.25,100.515)3(230,171.96)(230,190.71)14 ( 233.3,117.975)@xmath110 ( 217.5,117.975)@xmath110 ( 230,171.96)(241.25,152.475 ) ( 248.75,139.485)(260,120 ) ( 236.5,151.2)@xmath110(245,137)@xmath110 ( 230,120)(15,0,180)1.68 ( -11.25,-19.485)(0,0)3 ( 0,0)(30,51.96 ) ( 0,0)(30,51.96)(60,0)(37.5,0)(22.5,0)(0,0 ) ( 60,0)(71.25,-19.485)3(30,51.96)(30,70.71)14 ( 33.3,-2.5)@xmath110 ( 18,-2.5)@xmath110 ( 30,51.96)(41.25,32.475 ) ( 48.75,19.485)(60,0 ) ( 36.5,31.2)@xmath110(45,17)@xmath110 ( 11.25,0)(54.375,9.675)1.68 ( 88.75,-19.485)(100,0)3 ( 100,0)(130,51.96 ) ( 100,0)(130,51.96)(160,0)(137.5,0)(122.5,0)(100,0 ) ( 160,0)(171.25,-19.485)3(130,51.96)(130,70.71)14 ( 133.3,-2.5)@xmath110 ( 117.5,-2.5)@xmath110 ( 130,51.96)(141.25,32.475 ) ( 148.75,19.485)(160,0 ) ( 136.5,31.2)@xmath110(145,17)@xmath110 ( 134.125,42.75)(145,0)1.68 ( 188.75,-19.485)(200,0)3 ( 200,0)(230,51.96 ) ( 200,0)(230,51.96)(260,0)(237.5,0)(222.5,0)(200,0 ) ( 260,0)(271.25,-19.485)3(230,51.96)(230,70.71)14 ( 233.3,-2.5)@xmath110 ( 217.5,-2.5)@xmath110 ( 230,51.96)(241.25,32.475 ) ( 248.75,19.485)(260,0 ) ( 236.5,31.2)@xmath110(245,17)@xmath110 ( 245,29.7)(15,126,293)1.68 ( 23,90)(a ) ( 125,90)(b ) ( 225,90)(c ) ( 23,-28)(d ) ( 125,-28)(e ) ( 225,-28)(f ) the vacuum average of non - local quark - gluon mixing operator @xmath147 is calculated to be @xmath148_{\beta\alpha}t^a_{ji},\end{aligned}\ ] ] where @xmath149 and @xmath150 are the abbreviations of @xmath151 and @xmath152 respectively . @xmath153 is the strong coupling . because we calculate up to the condensate of dimension - six operators , the external gluon field @xmath154 in fig.4 should be expanded up to the second term , which will contribute a dimension - six operator , @xmath155 where @xmath156 is the covariant derivative in the adjoint representation , @xmath157 . then another vacuum matrix element needed is @xcite @xmath158 we calculate these diagrams and find that the contributions of fig.4(c ) , ( d ) and fig.5(c ) , ( d ) vanish after double borel transformation in two variables @xmath60 and @xmath61 , because only one variable appearing in the denominator , for instance , @xmath159 . the borel transformation in @xmath60 will kill such terms . following the above method , after some tedious algebraic derivation with the software mathematica , we obtain the coefficients @xmath160 , @xmath161 , @xmath162 and @xmath163 needed in eq . ( [ formsr ] ) . they are listed in appendix . the borel parameters @xmath83 and @xmath84 are not physical parameters . the physical result should not depend on them if the operator product expansion can be calculated up to infinite order . however , ope has to be truncated to some finite orders in practice . therefore , borel parameters have to be selected in some windows " to get the best stability of the physical results . the requirement to select the stable windows " is : the borel parameters can not be too large , or , contributions of higher resonance and continuum states can not be effectively suppressed ; at the same time , they should not be too small , or , the truncated ope would fail because the series in ope generally depend on borel parameters in the denominator @xmath170 . we find the optimal stability with the requirements shown in table [ t1 ] and the thresholds @xmath76 , @xmath77 in the ranges @xmath171 , @xmath172 . the regions of borel parameters which satisfies the requirements of table [ t1 ] are shown in fig.[region ] in two - dimensional diagram of @xmath27 and @xmath28 . we find good stability of the form factors within these regions . .requirements to select borel parameters @xmath27 and @xmath28 for each form factors @xmath173 , @xmath174 , @xmath175 and @xmath176 [ cols="^,^,^,^ " , ] [ t4 ] we use the total decay width of @xmath23 meson @xmath177 @xcite to obtain the branching ratio of @xmath5 , our result is @xmath178 which is in good agreement with experimental data @xmath179 . we calculate the transition form factors for @xmath58 transition in the region @xmath180 in qcd sum rule , where no non - landau - type singularity occurs . then fit the result from qcd sum rule in this region of momentum transfer , and extrapolate it to the whole physical region in the decay @xmath181 . we treat the two borel parameters @xmath27 and @xmath28 as independent parameters , and select the allowed region for @xmath27 and @xmath28 by requiring that the higher resonance and continuum contributions in @xmath23 and @xmath35 channels are not large , at the same time requiring that the condensate of higher dimension operators do not contribute too much . we find good stability for the transition form factors @xmath1 , @xmath2 , @xmath3 and @xmath4 in the relevant two - dimensional regions of @xmath27 and @xmath28 . we obtain the results of the transition form factors @xmath1 , @xmath2 , @xmath3 and @xmath4 in these regions of @xmath27 and @xmath28 . our result of the ratios of these form factors @xmath182 and @xmath183 are well consistent with experimental data . we studied the process @xmath184 with the form factors calculated from qcd sum rule . for the transverse polarization state of the final @xmath35 meson , the rate of @xmath23 decaying to right - hand state is almost an order smaller than decaying to left - hand state . the ratio of @xmath8 and the branching ratio of @xmath184 are in good agreement with the experimental data within the error bars of both the present experimental data and theoretical calculation . borel transformed coefficients of perturbative and nonperturbative contributions to the transition form factors in eq . ( [ formsr ] ) are given here . the contributions of condensate of dimension - six operator @xmath185 are numerically negligible , whereas their expressions are more tedious , therefore we do not present all of them here . @xmath186 with @xmath187 & + s_1-s_2+q^2)m_c + m_s(-s_2 ^ 2 + 2m_s^2s_2+s_1s_2+q^2s_2+\lambda ) ] , \end{array } \eqno(a1)\ ] ] where @xmath188 . the lower integration limit @xmath189 is determined by the condition that all internal quarks are on their mass shell @xcite , @xmath190 @xmath191 & + q^2)m_s^4m_2 ^ 2-(m_1 ^ 2+m_2 ^ 2)m_s^6)m_1 ^ 4+m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^2m_s^2(3m_2 ^ 2\\[4 mm ] & -m_s^2)m_1 ^ 2+m_2 ^ 2m_cm_s(-3m_1 ^ 2m_2 ^ 4 + ( m_1 ^ 2+m_2 ^ 2+q^2)m_s^2m_2 ^ 2\\[4 mm ] & -(m_1 ^ 2+m_2 ^ 2)m_s^4)m_1 ^ 2 -m_2 ^ 4(m_1 ^ 2+m_2 ^ 2)m_c^3m_s^3 ] \times\langle \bar{s}s\rangle\ ; , \end{array } \eqno(a2)\ ] ] @xmath192 & -(3m_2 ^ 4 + 5m_1 ^ 2m_2 ^ 2)m_s^2)m_1 ^ 4-m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^2(3m_2 ^ 2-m_s^2)m_1 ^ 2\\[4 mm ] & + m_2 ^ 2m_cm_s(2m_2 ^ 4 - 2m_1 ^ 2m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^2)m_1 ^ 2+m_2 ^ 4(m_1 ^ 2+m_2 ^ 2)m_c^3m_s\\[4 mm ] & + q^2(m_1 ^ 4m_2 ^ 2(3m_2 ^ 2-m_s^2)-m_1 ^ 2m_2 ^ 4m_cm_s)]\times g\langle \bar{s}\sigma tg s \rangle\ ; , \end{array } \eqno(a3)\ ] ] @xmath193 & + ( m_1 ^ 2+m_2 ^ 2)m_c^5m_s^3m_2 ^ 4+m_1 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^4m_s^4m_2 ^ 2\\[4 mm ] & + m_1 ^ 2q^4m_s^3(m_sm_1 ^ 2+m_2 ^ 2m_c)m_2 ^ 2+m_1 ^ 2m_c^3m_s^3(-13m_2 ^ 4 + 2m_1 ^ 2m_2 ^ 2\\[4 mm ] & + ( m_1 ^ 2+m_2 ^ 2)m_s^2)m_2 ^ 2+m_1 ^ 4m_c^2(-54(-1+e^{m_s^2/m_2 ^ 2})m_1 ^ 2m_2 ^ 6\\[4 mm ] & + 54m_1 ^ 2m_s^2m_2 ^ 4-(m_1 ^ 2 + 10m_2 ^ 2)m_s^4m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^6)\\[4 mm ] & + q^2(54(-1+e^{m_s^2/m_2 ^ 2})m_2 ^ 6m_1 ^ 6 - 54m_2 ^ 4m_s^2m_1 ^ 6-(m_1 ^ 2+m_2 ^ 2)m_s^6m_1 ^ 4\\[4 mm ] & -m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_cm_s^5m_1 ^ 2+m_2 ^ 2(m_1 ^ 4 + 10m_2 ^ 2m_1 ^ 2-(2m_1 ^ 2\\[4 mm ] & + m_2 ^ 2)m_c^2)m_s^4m_1 ^ 2-m_2 ^ 2m_c(2m_1 ^ 4 - 13m_2 ^ 2m_1 ^ 2+(2m_1 ^ 2\\[4 mm ] & + m_2 ^ 2)m_c^2)m_s^3 ) ] \times g^2\langle \bar{s}s\rangle^2\ ; . \end{array } \eqno(a4)\ ] ] @xmath194 with @xmath195 & + s_2+q^2)m_sm_c^4 + 2s_2(-4s_1 ^ 2 + 8s_2s_1 - 4s_2 ^ 2 + 2q^4 - 6(s_1-s_2)m_s^2\\[4 mm ] & + \lambda+2q^2(3m_s^2+s_1+s_2))m_c^3 + 2s_2m_s[4s_1 ^ 2 - 8s_2s_1 + 4s_2 ^ 2 - 2q^4\\[4 mm ] & + 6(s_1-s_2)m_s^2 - 3\lambda-2q^2(3m_s^2+s_1+s_2)]m_c^2+[6(s_1\\[4 mm ] & -s_2)s_2m_s^4 + 2(4s_2 ^ 3 - 8s_1s_2 ^ 2 + 4s_1 ^ 2s_2 - 2\lambda s_2+s_1\lambda)m_s^2+(s_1\\[4 mm ] & -s_2)s_2(2s_1 ^ 2 - 4s_2s_1 + 2s_2 ^ 2-\lambda)-2s_2q^4(2m_s^2 + 2s_1\\[4 mm ] & + s_2)+q^2(-6s_2m_s^4 - 2(2s_2 ^ 2 + 2s_1s_2+\lambda)m_s^2+s_2(2s_1 ^ 2 - 6s_2s_1\\[4 mm ] & + 4s_2 ^ 2-\lambda))]m_c+m_s[2s_2 ^ 4 - 6s_1s_2 ^ 3 + 6s_1 ^ 2s_2 ^ 2 - 3\lambda s_2 ^ 2 + 6(s_2\\[4 mm ] & -s_1)m_s^4s_2-s_1 ^ 3s_2 + 3s_1\lambda s_2 + 2q^4(2m_s^2 + 2s_1+s_2)s_2+\lambda^2\\[4 mm ] & -2(4s_2 ^ 3 - 8s_1s_2 ^ 2 + 4s_1 ^ 2s_2 - 4\lambda s_2+s_1\lambda ) m_s^2+q^2(6s_2m_s^4 + 2(2s_2 ^ 2\\[4 mm ] & + 2s_1s_2+\lambda ) m_s^2+s_2(-2s_1 ^ 2 + 6s_2s_1 - 4s_2 ^ 2 + 3\lambda ) ) ] \}\ ; , \end{array } \eqno(a5)\ ] ] @xmath196 & + m_2 ^ 2)+q^2)m_s^4m_2 ^ 2-(m_1 ^ 2+m_2 ^ 2)m_s^6]m_1 ^ 4+m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^2m_s^2(3m_2 ^ 2\\[4 mm ] & -m_s^2)m_1 ^ 2-m_2 ^ 2m_cm_s(3m_1 ^ 2m_2 ^ 4+(m_1 ^ 2+m_2 ^ 2-q^2)m_s^2m_2 ^ 2\\[4 mm ] & + ( m_1 ^ 2+m_2 ^ 2)m_s^4)m_1 ^ 2-m_2 ^ 4(m_1 ^ 2+m_2 ^ 2)m_c^3m_s^3 \ } \times\langle \bar{s}s\rangle\ ; , \end{array } \eqno(a6)\ ] ] @xmath197 & + 5m_1 ^ 2m_2 ^ 2)m_s^2]m_1 ^ 4 + m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^2(3m_2 ^ 2-m_s^2)m_1 ^ 2-m_2 ^ 2m_cm_s\\[4 mm ] & [ 2(m_1 ^ 2 + 2m_2 ^ 2)m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^2 ] m_1 ^ 2-m_2 ^ 4(m_1 ^ 2+m_2 ^ 2)m_c^3m_s+q^2[m_2 ^ 2(m_s^2\\[4 mm ] & -3m_2 ^ 2)m_1 ^ 4 + m_2 ^ 4m_cm_sm_1 ^ 2]\}\times g\langle \bar{s}\sigma tg s \rangle\ ; , \end{array } \eqno(a8)\ ] ] @xmath198 & -2m_1 ^ 2)m_c^5m_s^3m_2 ^ 4 + m_1 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^4m_s^4m_2 ^ 2+m_1 ^ 2q^4m_s^3(m_1 ^ 2m_s\\[4 mm ] & -2m_2 ^ 2m_c)m_2 ^ 2 + m_1 ^ 2m_c^3m_s^3[-26m_2 ^ 4 + 4m_1 ^ 2m_2 ^ 2+(m_1 - 2m_2 ^ 2)m_s^2]\\[4 mm ] & m_2 ^ 2 + m_1 ^ 4m_c^2[-54(-1+e^{m_s^2/m_2 ^ 2})m_1 ^ 2m_2 ^ 6 + 54m_1 ^ 2m_s^2m_2 ^ 4 -4(m_1 ^ 2\\[4 mm ] & -2m_2 ^ 2)m_s^4m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^6 ] + q^2[54(-1+e^{m_s^2/m_2 ^ 2})m_1 ^ 6m_2 ^ 6\\[4 mm ] & -54m_2 ^ 4m_s^2m_1 ^ 6 -(m_1 ^ 2+m_2 ^ 2)m_s^6m_1 ^ 4-m_2 ^ 2(m_1 ^ 2 - 2m_2 ^ 2)m_cm_s^5m_1 ^ 2\\[4 mm ] & + m_2 ^ 2(4(m_1 ^ 4 - 2m_1 ^ 2m_2 ^ 2)-(2m_1 ^ 2+m_2 ^ 2)m_c^2)m_s^4m_1 ^ 2 + m_2 ^ 4m_c\\[4 mm ] & ( -4m_1 ^ 4 + 26m_2 ^ 2m_1 ^ 2+(4m_1 ^ 2 - 7m_2 ^ 2)m_c^2)m_s^3]\ } \times g^2\langle \bar{s}s\rangle^2\ ; . \end{array } \eqno(a9)\ ] ] @xmath199 with @xmath200 & + 3s_2-q^2)m_sm_c^4 + 2s_2[-4s_1 ^ 2 - 4s_2s_1 + 8s_2 ^ 2 + 2q^4 - 6(s_1 + 3s_2)\\[4 mm ] & m_s^2+\lambda + 2q^2(3m_s^2+s_1 - 5s_2)]m_c^3 + 2s_2m_s[4s_1 ^ 2 + 4s_2s_1 - 8s_2 ^ 2\\[4 mm ] & -2q^4 + 6(s_1 + 3s_2)m_s^2+\lambda -2q^2(3m_s^2+s_1 - 5s_2)]m_c^2 + [ 6s_2(s_1\\[4 mm ] & + 3s_2)m_s^4 + 2(s_1 + 2s_2)(-4s_2 ^ 2 + 4s_1s_2+\lambda ) m_s^2 + ( s_1-s_2)s_2(2s_1 ^ 2\\[4 mm ] & -2s_2 ^ 2-\lambda ) + 2s_2 q^4(-2m_s^2 - 2s_1+s_2 ) + q^2(-6s_2m_s^4 - 2(-10s_2 ^ 2\\[4 mm ] & + 2s_1s_2+\lambda ) m_s^2-s_2(-2s_1 ^ 2 - 10s_2s_1 + 4s_2 ^ 2+\lambda))]m_c -m_s[2s_2 ^ 4\\[4 mm ] & -2s_1s_2 ^ 3 - 2s_1 ^ 2s_2 ^ 2-\lambda s_2 ^ 2 + 6(s_1 + 3s_2)m_s^4s_2 + 2s_1 ^ 3s_2+s_1\lambda s_2\\[4 mm ] & + 2q^4(-2m_s^2 - 2s_1+s_2)s_2-\lambda^2 + 2(-8s_2 ^ 3 + 4s_1s_2 ^ 2 + 4s_1 ^ 2s_2\\[4 mm ] & + 4\lambda s_2+s_1\lambda ) m_s^2 + q^2(-6s_2m_s^4 - 2(-10s_2 ^ 2 + 2s_1s_2+\lambda ) m_s^2\\[4 mm ] & + s_2(2s_1 ^ 2 + 10s_2s_1 - 4s_2 ^ 2+\lambda ) ) ] \}\ ; , \end{array } \eqno(a10)\ ] ] @xmath201 & + m_2 ^ 2)+q^2)m_s^4m_2 ^ 2-(m_1 ^ 2+m_2 ^ 2)m_s^6]m_1 ^ 4+m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^2\\[4 mm ] & m_s^2(3m_2 ^ 2-m_s^2)m_1 ^ 2-m_2 ^ 2m_cm_s(3m_1 ^ 2m_2 ^ 4+(m_1 ^ 2 - 3m_2 ^ 2-q^2)m_s^2\\[4 mm ] & m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^4)m_1 ^ 2-m_2 ^ 4(m_1 ^ 2+m_2 ^ 2)m_c^3m_s^3 \ } \times\langle \bar{s}s\rangle\ ; , \end{array } \eqno(a11)\ ] ] @xmath202 & -m_2 ^ 4)m_s^2]m_1 ^ 4 + m_2 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^2(3m_2 ^ 2-m_s^2)m_1 ^ 2\\[4 mm ] & -m_2 ^ 4(m_1 ^ 2+m_2 ^ 2)m_c^3m_s - m_c[2m_1 ^ 4m_sm_2 ^ 4+m_1 ^ 2(m_1 ^ 2+m_2 ^ 2)m_s^3m_2 ^ 2 ] \\[4 mm ] & + q^2[m_2 ^ 2(m_s^2 - 3m_2 ^ 2)m_1 ^ 4 + m_2 ^ 4m_cm_sm_1 ^ 2]\}\times g\langle \bar{s}\sigma tg s \rangle\ ; , \end{array } \eqno(a13)\ ] ] @xmath203 & -2m_1 ^ 2)m_c^5m_s^3m_2 ^ 4 + m_1 ^ 2(m_1 ^ 2+m_2 ^ 2)m_c^4m_s^4m_2 ^ 2+m_1 ^ 2q^4m_s^3(m_1 ^ 2m_s\\[4 mm ] & -2m_2 ^ 2m_c)m_2 ^ 2 + m_1 ^ 2m_c^3m_s^3[-18m_2 ^ 4 + 4m_1 ^ 2m_2 ^ 2+(m_1 - 2m_2 ^ 2)m_s^2]\\[4 mm ] & m_2 ^ 2 + m_1 ^ 4m_c^2[-54(-1+e^{m_s^2/m_2 ^ 2})m_1 ^ 2m_2 ^ 6 + 54m_1 ^ 2m_s^2m_2 ^ 4 -4(m_1 ^ 2\\[4 mm ] & + 4m_2 ^ 2)m_s^4m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^6 ] + q^2[54(-1+e^{m_s^2/m_2 ^ 2})m_1 ^ 6m_2 ^ 6\\[4 mm ] & -54m_2 ^ 4m_s^2m_1 ^ 6 -(m_1 ^ 2+m_2 ^ 2)m_s^6m_1 ^ 4-m_2 ^ 2(m_1 ^ 2 - 2m_2 ^ 2)m_cm_s^5m_1 ^ 2\\[4 mm ] & + m_2 ^ 2(4m_1 ^ 2(m_1 ^ 2 + 4m_2 ^ 2)-(2m_1 ^ 2+m_2 ^ 2)m_c^2)m_s^4m_1 ^ 2 + m_2 ^ 4m_c\\[4 mm ] & ( -4m_1 ^ 4 + 18m_2 ^ 2m_1 ^ 2+(4m_1 ^ 2 - 7m_2 ^ 2)m_c^2)m_s^3]\ } \times g^2\langle \bar{s}s\rangle^2\ ; . \end{array } \eqno(a14)\ ] ] @xmath204 with @xmath205 & -s_2+q^2)m_c^3 + 2s_2m_s(2m - s^2 + s_1-s_2+q^2)m_c^2+[2s_2m_s^4\\[4 mm ] & + 2(-s_2 ^ 2+s_1s_2+\lambda ) m_s^2 + s_2\lambda + 2s_2q^2 ( m_s^2+s_1)]m_c - m_s[2s_2m_s^4\\[4 mm ] & + 2(-s_2 ^ 2+s_1s_2+\lambda ) m_s^2 -s_1\lambda + q^2(2s_2m_s^2 + 2s_1s_2+\lambda ) ] \ } \ ; , \end{array } \eqno(a15)\ ] ] @xmath206 & [ 3m_1 ^ 2m_2 -(m_1 ^ 2 + 2m_2 ^ 2)m_s^2]m_c^4 + [ 3m_1 ^ 4m_sm_2 ^ 6 - 9m_1 ^ 2(m_1 ^ 2+m_2 ^ 2)\\[4 mm ] & m_s^3m_2 ^ 4+(m_2 ^ 6 + 4m_1 ^ 2m_2 ^ 4 + 3m_1 ^ 4m_2 ^ 2)m_s^5 ] m_c^3+m_1 ^ 2[-6m_1 ^ 4m_2 ^ 6+(m_1 ^ 4\\[4 mm ] & + 4m_2 ^ 2m_1 ^ 2 + 3m_2 ^ 4)m_s^6-(6m_2 ^ 6 + 11m_1 ^ 2m_2 ^ 4 + 5m_1 ^ 4m_2 ^ 2)m_s^4\\[4 mm ] & + 3(3m_1 ^ 2m_2 ^ 6 + 2m_1 ^ 4m_2 ^ 4)m_s^2]m_c^2+m_1 ^ 2m_s[-15m_1 ^ 4m_2 ^ 6 + ( 2m_1 ^ 4\\[4 mm ] & + 3m_2 ^ 2m_1 ^ 2+m_2 ^ 4)m_s^6-(2m_2 ^ 6 + 13m_1 ^ 2m_2 ^ 4 + 11m_1 ^ 4m_2 ^ 2)m_s^4\\[4 mm ] & + ( 8m_1 ^ 2m_2 ^ 6 + 11m_1 ^ 4m_2 ^ 4)m_s^2]m_c+m_1 ^ 4m_s^2[3m_1 ^ 2m_2 ^ 6 - 4(m_1 ^ 2+m_2 ^ 2)\\[4 mm ] & m_s^4m_2 ^ 2+(m_1 ^ 2+m_2 ^ 2)m_s^6 + ( 2m_2 ^ 6 + 5m_1 ^ 2m_2 ^ 4)m_s^2]+m_1 ^ 2m_2 ^ 2q^4m_s^2\\[4 mm ] & [ ( m_s^2 - 3m_2 ^ 2)m_1 ^ 2+m_2 ^ 2m_cm_s]+ q^2[(6m_1 ^ 2m_2 ^ 6-(m_1 ^ 2 + 2m_2 ^ 2)m_s^6\\[4 mm ] & + ( 7m_2 ^ 4 + 5m_1 ^ 2m_2 ^ 2)m_s^4 - 3(m_2 ^ 6 + 2m_1 ^ 2m_2 ^ 4)m_s^2)m_1 ^ 4 + m_2 ^ 2m_c^2m_s^2\\[4 mm ] & ( 3m_2 ^ 2(2m_1 ^ 2+m_2 ^ 2)-(2m_1 ^ 2 + 3m_2 ^ 2)m_s^2)m_1 ^ 2+m_2 ^ 2m_cm_s ( -3m_1 ^ 2m_2 ^ 4\\[4 mm ] & -(3m_1 ^ 2 + 2m_2 ^ 2)m_s^4+(2m_2 ^ 4 + 9m_1 ^ 2m_2 ^ 2)m_s^2)m_1 ^ 2-m_2 ^ 4(2m_1 ^ 2+m_2 ^ 2)\\[4 mm ] & m_c^3m_s^3]\ } \times\langle \bar{s}s\rangle\ ; , \end{array } \eqno(a16)\ ] ] @xmath207 & -(m_1 ^ 2 + 2m_2 ^ 2)m_s^2]m_c^4 + m_2 ^ 2m_s[(3m_1 ^ 4 + 4m_2 ^ 2m_1 ^ 2+m_2 ^ 4)m_s^2\\[4 mm ] & -2m_1 ^ 2m_2 ^ 2(5m_1 ^ 2 + 3m_2 ^ 2)]m_c^3 + [ 2m_1 ^ 4(m_1 ^ 2 + 3m_2 ^ 2)m_2 ^ 4+(m_1 ^ 6\\[4 mm ] & + 4m_2 ^ 2m_1 ^ 4 + 3m_2 ^ 4m_1 ^ 2)m_s^4- 2(3m_2 ^ 2m_1 ^ 6 + 5m_2 ^ 4m_1 ^ 4)m_s^2]m_c^2+m_1 ^ 2m_s\\[4 mm ] & [ 4m_1 ^ 2(m_1 ^ 2 + 3m_2 ^ 2)m_2 ^ 4 + 2(m_1 ^ 4 + 3m_2 ^ 2m_1 ^ 2+m_2 ^ 4)m_s^4+(m_2 ^ 6\\[4 mm ] & -8m_1 ^ 2m_2 ^ 4 - 13m_1 ^ 4m_2 ^ 2)m_s^2]m_c + m_1 ^ 4[4m_1 ^ 2m_2 ^ 6+(m_1 ^ 2+m_2 ^ 2)\\[4 mm ] & m_s^2(2m_2 ^ 4+m_s^4)-(m_2 ^ 4 + 5m_1 ^ 2m_2 ^ 2)m_s^4]+q^4[m_2 ^ 2(m_s^2\\[4 mm ] & -3m_2 ^ 2)m_1 ^ 4+m_2 ^ 4m_cm_sm_1 ^ 2]-q^2[(2(m_1 ^ 2 + 3m_2 ^ 2)m_2 ^ 4 + ( m_1 ^ 2 + 2m_2 ^ 2)m_s^4\\[4 mm ] & -2(2m_2 ^ 4 + 3m_1 ^ 2m_2 ^ 2)m_s^2)m_1 ^ 4+m_2 ^ 2m_cm_s(m_2 ^ 4 -10m_1 ^ 2m_2 ^ 2\\[4 mm ] & + ( 3m_1 ^ 2 + 2m_2 ^ 2)m_s^2)m_1 ^ 2+m_2 ^ 2m_c^2((2m_1 ^ 2 + 3m_2 ^ 2)m_s^2\\[4 mm ] & -3(m_2 ^ 4 + 2m_1 ^ 2m_2 ^ 2))m_1 ^ 2+m_2 ^ 4(2m_1 ^ 2+m_2 ^ 2)m_c^3m_s ] \}\times g\langle \bar{s}\sigma tg s \rangle\ ; . \end{array } \eqno(a18)\ ] ] 99 m.a . shifman , a.i . vainshtein and v.i . zakharov , nucl . phys . b147 ( 1979 ) 385 , 448 . aliev , v.l . eletskii , and ya . i. kogan , sov . 40 ( 1984 ) 527 . ovchinnikov and v.a . slobodenyyuk , z. phys . c44 ( 1989 ) 433 ; v.n . baier and grozin , z. phys . c47 ( 1990 ) 669 . ovchinnikov , sov . j. nucl . phys . 50 ( 1989 ) 519 . p. ball , v.m . braun and h.g . dosch , phys . d44 ( 1991 ) 3567 . p. ball , phys . d48 ( 1993 ) 3190 . p. colangelo and f. de fazio , phys . b 520 ( 2001 ) 78 . k. kodama et al . ( fnal e653 collab . ) , phys . b309 ( 1993 ) 483 . frabetti et al . ( fnal e687 collab . ) , phys . b328 ( 1994 ) 187 . p. avery et al . ( cleo collab . ) b337 ( 1994 ) 405 . aitala et al . ( fnal e791 collab . ) , phys . b450 ( 1999 ) 294 . i. bediaga and m. nielsen , eprint hep - ph/0304193 . reinders , h. rubinstein and s. yazahi , phys . rep . 127 , no.1 ( 1985)1 - 97 . r.e.cutkosky , j. math . phys . 1 ( 1960 ) 429 . ioffe and a.v . smilga , nucl . b216 ( 1983 ) 373 . fock , phys . z. su 12 ( 1937 ) 404 ; j. schwinger , particles , sources , and fields , addison - wesley ( 1973 ) ; m.a . shifman , nucl . b173 ( 1980 ) 13 ; m.s . dubovikov and a.v . smilga , nucl . b185 ( 1981 ) 109 . smilga , yad . 35 ( 1982 ) 473 [ sov . j. nucl . 35 ( 1982 ) 271 . shifman , nucl . phys . b173 ( 1980 ) 13 . landau , nucl . ( 1959 ) 181 . s. narison , _ qcd spectral sum rules _ ( world scientific , singapore , 1989 ) . for a recent review article , see , p. colangelo and a. khodjamirian , in : at the frontier of particle physics / handbook of qcd , edited by m. shifman ( world scientific , singapore , 2001 ) , hep - ph/0010175 . particle data group , k. hagiwara et al . d66 ( 2002 ) 010001 .

in this paper , we present new _ a posteriori _ and _ a priori _ error bounds for the krylov subspace methods for computing for a given and , where is a large sparse non - hermitian matrix . the _ a priori _ error bounds relate the convergence to , ( the smallest and the largest eigenvalue of the hermitian part of ) and ( the largest eigenvalue in absolute value of the skew - hermitian part of ) , which define a rectangular region enclosing the field of values of . in particular , our bounds explain an observed superlinear convergence behavior where the error may first stagnate for certain iterations before it starts to converge . the special case that is skew - hermitian is also considered . numerical examples are given to demonstrate the theoretical bounds .

1603.07358

the problem of computing matrix exponentials arises in many theoretical and practical problems . numerous methods have been developed to efficiently compute @xmath7 or its product with a vector @xmath8 , where @xmath3 is an @xmath9 complex matrix and @xmath2 . we refer to the classical paper @xcite of moler and van loan for a survey of a general theory and numerical methods for matrix exponentials . for matrix exponential problems involving a large and sparse matrix @xmath3 , it is usually the product of the exponential with a vector that is of interest . this arises , for example , in solving the initial value problem ( @xcite ) @xmath10 see @xcite for some other applications . a large number of matrix exponential problems concern a _ positive definite _ @xmath3 ( i.e. @xmath11 is hermitian positive definite ) , which defines a stable dynamical system with a solution converging to a steady state . another important class of problems involve a skew - hermitian matrix @xmath3 ( i.e. @xmath12 with @xmath13 being hermitian ) , for which has a norm - conserving solution . such systems can be used to model a variety of physical problems where certain quantities such as energy are conserved . for example , a spectral method for solving the time - dependent schrdinger equation modeling n electrons leads to with a skew - hermitian matrix ; see @xcite . while we will study a general non - hermitian @xmath3 , we are particularly interested in these two important classes of problems , where stronger theoretical results can be derived . the krylov subspace methods are a powerful class of iterative algorithms for solving many large scale linear algebra problems . initially introduced by gallopoulos and saad @xcite , they have also become a popular method for approximating @xmath14 where @xmath15 is a fixed parameter typically representing a time step . for the ease of notation , we will assume throughout that @xmath16 . a comprehensive theory has been developed in the literature with error bounds demonstrating convergence of the krylov subspace methods and its relation to certain properties of the matrix . for example , earlier results in @xcite relate convergence of the krylov subspace methods to the norm of the matrix @xmath17 . more refined error bounds have later been derived , that provide sharper estimates of the errors by considering additional spectral information such as enclosing regions of the field of values of @xmath3 or positive definiteness of @xmath3 ; see @xcite and the references contained therein . for a real symmetric positive definite matrix @xmath3 , it has been shown in a recent work @xcite that the speed of convergence is also determined by the condition number of @xmath3 as in the conjugate gradient method . for positive definite matrices that are not necessarily hermitian , stronger convergence bounds have also been obtained in @xcite in terms of the field of values . however , most of these bounds are derived by assuming the field of values lying in a certain pre - defined region , which are not easy to apply or interpret . there is an inherited theoretical difficulty in quantitatively characterizing the influence on the convergence by the field of values , a two dimensional object . this issue also arises in the theory of the krylov subspace methods for solving linear systems . in this paper , we study the relation between the convergence of the krylov subspace methods and the field of values through its bounding rectangle @xmath18\times[-c , c]$ ] where @xmath19 , @xmath20 ( the smallest and the largest eigenvalue of the hermitian part of @xmath3 ) and @xmath21 ( the largest eigenvalue in absolute value of the skew - hermitian part of @xmath3 ) . with this approach , new _ a priori _ error bounds will be derived in terms of @xmath22 , @xmath23 and @xmath24 . simplified bounds will be presented for non - hermitian positive definite matrices and skew - hermitian matrices , which relate the speed of convergence to the size and the shape of the rectangular region . in particular , our bounds explain an interesting observed convergence behavior where the error may first stagnate for certain iterations before it starts to converge . numerical examples will be presented to demonstrate the behavior of the new error bounds . in developing our _ a priori _ error bounds , we also derive a new _ a posteriori _ error bound that is shown to provide a sharp and computable estimate of the error . the main technique used in deriving new _ a priori _ error bounds is the same as in the literature @xcite by constructing faber polynomial approximation of the exponential function in a region containing the field of values . the novelty in this work is to use the jacobi elliptic functions to construct a conformal mapping for the rectangular region that tightly encloses the fields of value and to show that this highly complicated mapping can be simplified to yield some simple final bounds . the paper is organized as follows . in section [ sec : preliminaries ] , we first present some preliminaries about the faber polynomial approximation and the jacobi elliptic functions . in section [ sec : aposteriori ] , we present a new _ a posteriori _ error bound , which relates the convergence to the decay properties of functions of banded matrices . to study this decay behavior , we construct a conformal mapping in section [ sec : conformal_mapping ] and present our new _ a priori _ error bound in section [ sec : apriori_nonhermitian ] . in section [ sec : apriori_skewhermitian ] , we apply the same idea on skew - hermitian matrices and derive simpler _ a priori _ bounds . numerical examples are presented in section [ sec : numerical_examples ] and some concluding remarks in section [ sec : concluding_remarks ] . in this section , we briefly discuss some related results in complex analysis that will be needed . faber polynomials extend the theory of power series to domains more general than a disk . it starts with the riemann mapping theorem ( * ? ? ? * theorem 1.2 ) that states that every connected domain in the extended complex plane whose boundary contains more than one point can be mapped conformally onto a disk with its center at the origin . let @xmath25 be the extended complex plane and @xmath26 be a bounded , closed continuum in the complex plane with boundary @xmath27 such that the complement of @xmath26 is simply connected in the extended plane and contains the point at @xmath28 . a continuum is a non - empty , compact and connected subset of @xmath29 . then there exists a function @xmath30 which maps the complement of @xmath26 conformally onto the exterior of a circle @xmath31 and satisfies the normalization conditions @xmath32 then , the function @xmath33 has a laurent expansion at infinity of the form @xmath34 moreover , given any integer @xmath35 , @xmath36^n$ ] has a laurent expansion of the form @xmath37^n = z^n+\alpha_{n-1}^{(n)}z^{n-1}+\cdots+\alpha_0^{(n)}+\frac{\alpha_{-1}^{(n)}}{z}+\cdots\ ] ] at infinity @xcite . then , we call the following polynomial containing non - negative powers of @xmath38 in the expansion @xmath39 the faber polynomials generated by @xmath26 . the faber polynomials can be used to approximate analytic functions on @xmath26 , essentially through the power series approximation of a transformed function on @xmath40 . let @xmath41 be the inverse of @xmath42 and let @xmath43 be the image under @xmath41 of the circle @xmath44 . we denote by @xmath45 the bounded region enclosed by @xmath43 . by ( * theorem 3.17 ) , every function @xmath46 analytic on @xmath47 can be represented on @xmath47 as a series of the faber polynomials @xmath48 with the coefficients @xmath49}{w^{n+1}}dw$ ] . the partial sum of the above series @xmath50 is a polynomial of degree at most @xmath51 that we can use to approximate @xmath46 on @xmath47 . the next theorem of @xcite presents some approximation bounds concerning @xmath52 . we first need to introduce the definition of total rotation of the boundary . for this , we assume @xmath26 is a closed jordan region , i.e. its boundary @xmath27 is rectifiable . then there exists a tangent vector that makes an angle @xmath53 with the positive real axis at almost all points @xmath54 . we say that @xmath27 has bounded total rotation @xmath55 if @xmath56 . then @xmath57 and the equality holds if @xmath26 is convex ; see @xcite . [ thm : razouk54 ] ( * ? ? ? * corollary 2.2 ) assume @xmath26 is a closed jordan region whose boundary @xmath27 has bounded total rotation @xmath55 . for any @xmath58 , let @xmath59 be an analytic function in @xmath45 . we have for any @xmath60 , @xmath61 where @xmath62 and @xmath63 denotes the uniform norm on @xmath45 . theorem [ thm : razouk54 ] is stated with @xmath43 defined from the conformal map @xmath42 satisfying the normalization condition . in the literature ( see @xcite for example ) , another normalization has also been used and may be more convenient in our application . we may consider a conformal map @xmath64 that maps the exterior of @xmath26 onto the exterior of the unit disk ( i.e. requiring @xmath65 rather than ) . the above theorem can be adapted to @xmath64 through a simple normalization transformation . namely , given @xmath64 , let @xmath66 and @xmath67 , where we assume @xmath68 is finite . then @xmath42 satisfies the normalization condition but now maps the exterior of @xmath26 onto the exterior of the disk @xmath69 . applying theorem [ thm : razouk54 ] to @xmath42 , holds for any @xmath58 . let @xmath70 . let @xmath43 be the inverse image under @xmath42 of the circle @xmath71 and @xmath72 be the inverse image under @xmath64 of the circle @xmath73 . it is easy to check that @xmath74 and then @xmath75 . thus , is reduced to @xmath76 where @xmath77 . namely , theorem [ thm : razouk54 ] holds verbatim for a conformal map that is normalized to map the exterior of @xmath26 onto the exterior of the unit disk . we note however that @xmath68 as defined in the two normalizations is invariant and is called logarithmic capacity of @xmath26 . in this subsection , we introduce the jacobi elliptic functions , which will be used to construct a conformal mapping in section [ sec : apriori_nonhermitian ] . more details about the jacobi elliptic functions can be found in @xcite . elliptic functions were first introduced as inverse functions of ( incomplete ) elliptic integrals . so before the introduction of the jacobi elliptic functions , we first state the definition and properties of elliptic integrals . given @xmath78 and a real parameter @xmath79 with @xmath80 , the ( incomplete ) jacobi elliptic integral of the first kind is defined as @xmath81 the ( incomplete ) jacobi elliptic integral of the second kind is defined as @xmath82 when @xmath83 , the corresponding integrals @xmath84 are called the complete jacobi elliptic integrals of the first kind and the second kind . let @xmath85 , the complementary parameter of @xmath79 . then , @xmath86 . for simplicity , we shall use the following notations . we now introduce the jacobi elliptic functions . there are a total of twelve jacobi elliptic functions in the family , but we will only discuss the basic three of them that will be used in this work . if @xmath88 where @xmath89 is the incomplete elliptic integral of the first kind defined in , three of the jacobi elliptic functions are defined as @xmath90 the notations @xmath91 , @xmath92 and @xmath93 indicate that @xmath94 , @xmath95 and @xmath96 are functions of two independent arguments : a complex argument @xmath97 and a real parameter @xmath98 . furthermore , for a fixed @xmath98 , @xmath99 , @xmath100 and @xmath101 are doubly periodical meromorphic functions defined on @xmath102 @xcite . in later sections , we will need some properties of the jacobi elliptic integrals and jacobi elliptic functions . we summarize them in the proposition below . for details , see @xcite , @xcite and @xcite . [ prop : jacobi ] 1 . @xmath103 and @xmath104 are positive - valued functions of @xmath79 . moreover , they are differentiable with respect to the parameter @xmath98 , and @xmath105 2 . * , p. 591 ) @xmath106=0\ ] ] 3 . * , p. 592 ) @xmath107 4 . @xmath94 , @xmath95 and @xmath96 satisfy @xmath108 5 . * table 16.2 , p. 570 ) @xmath94 , @xmath95 and @xmath96 are one - valued , doubly - periodic functions . for any @xmath109 , @xmath110 6 . * table 16.8 , p. 572 ) @xmath111 7 . * table 16.16 , p. 574 ) derivatives : @xmath112 8 . * , p. 575 ) let @xmath113 where @xmath114 and denote @xmath115 then @xmath116 we will also need to use the signs of the real and imaginary parts of @xmath117 , @xmath118 and @xmath119 when @xmath98 and @xmath102 is in the rectangular domain @xmath120\times[0,2ik']$ ] ( i.e. @xmath121 $ ] and @xmath122 $ ] ) . this is discussed in @xcite and we summarize it in table [ tab : sn ] , [ tab : cn ] and [ tab : dn ] for easy future references . .signs of @xmath123 [ cols="^,^,^",options="header " , ] in this section , we first introduce the arnoldi method for approximating @xmath124 and then discuss an _ a posteriori _ error bound . given @xmath125 and @xmath126 with @xmath127 , @xmath128 iterations of the arnoldi process generates an orthonormal basis @xmath129 for the krylov subspace @xmath130 by @xmath131 simultaneously , a @xmath128-by-@xmath128 upper hessenberg matrix @xmath132 $ ] is generated satisfying @xmath133 where @xmath134 $ ] and @xmath135 is the @xmath128-th coordinate vector . we note that @xmath136 we can approximate @xmath137 by its orthogonal projection on @xmath138 , @xmath139 , which is further approximated as @xmath140 we call @xmath141 the arnoldi approximation to @xmath142 in ; see @xcite . let @xmath143 be the field of values of @xmath3 and @xmath144 be the logarithmic norm of @xmath3 ( associated with the euclidean inner product ) . we also define @xmath145 . then we have @xmath146 where @xmath147 and @xmath148 denote the largest and the smallest eigenvalues respectively . in this notation , @xmath3 is positive definite if and only if @xmath149 . an important property associated with the logarithmic norm @xcite is that for @xmath150 , @xmath151 we now present a bound on the approximation error @xmath152 in terms of the @xmath153 entry of the matrix @xmath154 . [ thm : aposteriori_re ] let @xmath125 and @xmath2 with @xmath155 . let @xmath156 be the orthogonal matrix and @xmath157 be the upper hessenberg matrix generated by the arnoldi process for @xmath3 and @xmath158 satisfying . let @xmath159 be the arnoldi approximation to @xmath137 . then the approximation error satisfies @xmath160 where @xmath161 is the @xmath153 entry of the matrix @xmath154 and @xmath162 is defined in . first , we have @xmath163 and @xmath164 let @xmath165 . then @xmath166 note that @xmath167 . solving the initial value problem for @xmath168 , we have @xmath169 since @xmath170 in the integral , using , we have @xmath171 then the approximation error satisfies @xmath172 thus , if @xmath173 , we have @xmath174 . if @xmath175 , then @xmath176 this completes the proof . @xmath177 in the above bound is computable _ a posteriori _ for any given @xmath178 . being the @xmath153 entry of the matrix @xmath154 , it is expected to become small as @xmath128 increases because of a decay property associated with functions of a banded matrix ( see @xcite ) . this provides an understanding of the convergence of the error . indeed , in [ sec : apriori_nonhermitian ] , we shall extend the techniques introduced in @xcite to derive some sharp decay bounds on @xmath177 , which will result in some new _ a priori _ bounds . before we do that , we will need to construct some conformal mapping first in the next section . we also remark that the _ a posteriori _ bound in the theorem contains the integral of @xmath177 that is not directly computable . for practical error estimates , we can approximate it using a quadrature rule , say , the simpson s rule , by computing @xmath177 at some selected discrete points . this provides a fairly sharp _ a posteriori _ error estimates ; see the numerical examples in [ sec : numerical_examples ] . note that there are several _ a posteriori _ error estimates presented in @xcite derived from approximation of a different error expression , one of which is @xmath179 . in this section , we construct a conformal mapping which maps the exterior of a rectangle onto the exterior of a unit disk and discuss some of its properties . given a rectangle in @xmath180-plane whose vertices are @xmath181 and @xmath182 where @xmath183 and @xmath184 , we map the exterior of this rectangle conformally onto @xmath185 . this can be done in the following three steps . * step 1 : @xmath186 shifts the original rectangle to a new rectangle with vertices @xmath187 , where @xmath188 and @xmath189 . * step 2 : @xmath190 is defined through an auxiliary variable @xmath191 by @xmath192 where @xmath91 , @xmath92 and @xmath93 are jacobi elliptic functions and @xmath193 . the parameter @xmath79 is determined from @xmath194 by the equation @xmath195 here @xmath196 , @xmath197 , @xmath198 and @xmath199 are functions of @xmath79 or @xmath85 defined in . the existence and uniqueness of @xmath79 will be shown in lemma [ lem : unique ] below . it is shown in @xcite that @xmath200 conformally maps the exterior of the rectangle @xmath201\times[-\beta,\beta]$ ] to the upper half plane @xmath202 and that the range of @xmath191 is in the rectangle @xmath120\times[0,2ik']$ ] . * step 3 : @xmath203 maps @xmath202 onto @xmath204 . now let @xmath205 be the composition of the above three conformal mappings defined in , and . then @xmath206 maps the exterior of the rectangle @xmath18\times[-c , c]$ ] conformally onto the exterior of the unit circle . the rest of this section will present several results concerning @xmath206 that we will use in the next section , but first we give a proof of existence of a unique solution of that appears not readily available in the literature . [ lem : unique ] @xmath207 is an increasing function and @xmath208 is an decreasing function . for any @xmath209 , there exists a unique @xmath98 , as a function of @xmath210 , satisfying . let @xmath211 be a function of @xmath98 . then @xmath212 . by the definition of @xmath213 and @xmath214 , @xmath215 , @xmath216 , and then @xmath217 moreover , by , @xmath218=0,\ ] ] and therefore @xmath219 again by the definition of @xmath214 , @xmath220 . then @xmath221 by and , @xmath222 is differentiable in @xmath223 and @xmath224 so @xmath59 is an increasing function of @xmath79 over @xmath223 . now consider @xmath225 by and , @xmath226 is an increasing function of @xmath79 over @xmath223 with @xmath227 then for any @xmath209 , there exists a unique @xmath98 such that @xmath228 , i.e. , . the parameter @xmath79 determined by is defined by the aspect ratio @xmath229 ( or the shape ) of the rectangle @xmath18\times[-c , c]$ ] . for example , from the proof , @xmath230 if the rectangle is narrowly around the real axis , while @xmath231 if the rectangle is nearly a vertical line in the complex plane . when @xmath232 , the rectangle is a square . as in [ sec : preliminaries ] , we denote by @xmath233 in the @xmath180-plane the inverse image of the circle @xmath234 under @xmath206 for a given @xmath235 . we need to determine the minimum of @xmath236 in @xmath233 , i.e. the left most point of @xmath233 . first we prove a lemma about the jacobi elliptic functions , which is a direct result of proposition [ prop : jacobi ] . [ thm : same_sign ] for @xmath113 where @xmath237 and @xmath238 , @xmath239 by and , @xmath240 so , @xmath241 write @xmath242 . when @xmath237 , we have @xmath243 . so , @xmath244 by the definition of @xmath119 , for any @xmath114 , @xmath245 applying and to , we conclude that the imaginary part of @xmath118 and that of @xmath119 have the same sign . the following lemma shows that the minimum of @xmath236 in @xmath233 is attained at the inverse of @xmath246 . [ thm : min_re ] let @xmath247 be defined in . let @xmath248 be its inverse mapping and @xmath233 be the image of @xmath249 under @xmath250 . then @xmath251 by , @xmath252 recall the definition @xmath253 , the identities @xmath254 and @xmath255 , we have from that @xmath256 note that by and , we have @xmath257 and @xmath258 . then by , @xmath259 by , @xmath260 and then @xmath261 combining , , and , we have @xmath262 also implies @xmath263 on the other hand , by , @xmath264 so , @xmath265 and hence @xmath266 substituting this into , we have @xmath267 now let @xmath97 be on the circle of radius @xmath268 on the complex @xmath97-plane . then we can write @xmath269 where @xmath270 . hence @xmath271 treating @xmath272 as a function of @xmath273 , we have from and that @xmath274 so @xmath275 from and @xmath276 , we write @xmath93 as a function of @xmath273 , @xmath277 so @xmath278 when @xmath279 , and @xmath280 when @xmath281 $ ] . by lemma [ thm : same_sign ] , the imaginary part of @xmath92 always has the same sign as that of @xmath93 . thus , by , @xmath282 when @xmath279 , and @xmath283 when @xmath281 $ ] . the minimum value of @xmath236 is attained when @xmath284 , i.e. , @xmath246 . next , we find the explicit form for @xmath285 in lemma [ thm : min_re ] . [ thm : psi_r_negative ] let @xmath247 be the conformal mapping from the exterior of the rectangle @xmath18\times[-c , c]$ ] onto the exterior of the unit disk , as defined in , and let @xmath248 be its inverse . then for any @xmath235 , we have @xmath286 where the parameters @xmath79 is determined by and @xmath287 is the ratio in . recall that @xmath288 with @xmath289 , @xmath200 and @xmath290 the three conformal mappings defined in , and . let @xmath291 and @xmath41 be its inverse . then obviously @xmath292 the proof of this lemma consists of two parts . first , we prove that for any @xmath235 , @xmath293 by the same equation that was derived from and , @xmath294 in the map can be eliminated to define @xmath42 : @xmath295 through the auxiliary parameter @xmath191 as @xmath296 to compute @xmath297 , set @xmath298 above . then the corresponding @xmath191 satisfies @xmath299 by table [ tab : dn ] , @xmath300 is on the line segment connecting @xmath301 and @xmath302 . let @xmath303 where @xmath304 is on the line segment connecting @xmath301 and @xmath191 . by tables [ tab : sn ] , [ tab : cn ] and [ tab : dn ] , @xmath305 is purely imaginary with positive imaginary part , and @xmath306 and @xmath307 are both real and positive . then @xmath308 by and , @xmath309 then @xmath310 by , @xmath311 then @xmath312 thus , as @xmath304 moves along the positive imaginary axis from @xmath301 to @xmath191 , @xmath178 as defined by moves along the positive real axis from @xmath301 to @xmath313 . then @xmath314 this completes the proof of the first part . we next prove for any @xmath235 , @xmath315 let @xmath191 and @xmath316 be the auxiliary parameters in corresponding to @xmath268 and @xmath317 respectively . then @xmath318 by , @xmath319 . thus , using and , we get @xmath320\ } \\ \nonumber & = \alpha-\frac{i}{\lambda}\{-2i\cdot\lambda\alpha-[e(\sigma|m)-m_1\sigma]\ } \\ & = -\alpha+\frac{i}{\lambda}\{e(\sigma|m)-m_1\sigma\ } = -z(\sigma)=-\psi(r ) . \end{aligned}\ ] ] finally , applying @xmath321 to @xmath322 as in and noting that @xmath188 , is proved . finally , we show that @xmath206 can be normalized according to . [ thm : normalization ] let @xmath287 be the ratio in . we have @xmath323 first , by and @xmath324 , we have @xmath325 . applying it to , we have @xmath326 as @xmath327 , @xmath328 and @xmath329 ( see @xcite ) . since @xmath330 we have @xmath331 . applying it to , @xmath332 or @xmath333 as @xmath327 . then @xmath334 as @xmath327 . @xmath335 follows from lemma [ lem : unique ] . in this section , we derive new _ a priori _ error bounds for the arnoldi approximations of @xmath0 . we shall bound the error in terms of the following spectral information of @xmath3 : @xmath336 where @xmath337 ( @xmath338 ) are the eigenvalues of @xmath339 . these three numbers provide a region bounding @xmath340 , the field of values of @xmath3 , i.e. @xmath340 is contained in the rectangle @xmath18\times[-c , c]$ ] . we shall study the convergence of the arnoldi method through bounding @xmath341 ( the @xmath153 entry of @xmath154 ) in the _ a posteriori _ bound of [ sec : aposteriori ] as in @xcite . as mentioned before , analytic functions of banded matrices have a decay property , i.e. their entries decreases away from the main diagonal . sharp decay bounds were originally derived by benzi and golub @xcite for hermitian matrices ; see @xcite and the references contained therein for some further improvements . generalizations to the non - hermitian case , which is applicable to the hessenberg matrix @xmath157 here , have been obtained by benzi and razouk @xcite and benzi and boito @xcite . specifically , for non - hermitian matrices , the faber polynomial approximation and the conformal mappings on a circular region containing the field of value have been introduced in @xcite to bound the decay rate . here we will follow the same approach of @xcite , but we will use the conformal mapping that is constructed in [ sec : conformal_mapping ] so as to utilize a more precise region @xmath18\times[-c , c]$ ] that encloses the field of values . by using a smaller bounding region , a stronger approximation result and hence a stronger bound are obtained as follows . [ thm : bound_ht_re ] let @xmath157 be a @xmath128-by-@xmath128 upper hessenberg matrix and let @xmath342 be the @xmath153 entry of the matrix @xmath154 . let @xmath343 , @xmath344 and @xmath345 . then for any @xmath346 with @xmath347 , @xmath348 where @xmath349 , @xmath350 and the parameters @xmath79 is determined from @xmath351 , @xmath352 , @xmath353 by and @xmath287 is the ratio in . let @xmath247 be the conformal mapping from the exterior of the rectangle @xmath354\times[-c_k , c_k]$ ] onto the exterior of the unit disk , as defined in . for a fixed @xmath150 , let @xmath355 . since @xmath59 is an analytic function , it can be approximated by the partial sum @xmath356 of the series of faber polynomials generated by @xmath206 as defined in . let @xmath357 and consider @xmath233 , the inverse image under @xmath206 of the circle @xmath73 . applying theorem [ thm : razouk54 ] or , the approximation error in @xmath358 is bounded as @xmath359 where @xmath360 and we note that the total rotation around the rectangle is @xmath361 . since @xmath362 is a polynomial of degree @xmath363 , @xmath364_{k1}=e_k^t \pi_{k-2}(h_k)e_1=0 $ ] . then @xmath365_{k1}| = |[f(h_k)]_{k1}-[\pi_{k-2}(h_k)]_{k1}| \\ & \leq||f(h_k)-\pi_{k-2}(h_k)||_2 \\ & \leq q\cdot\max_{z\in w(h_k)}|f(z)-\pi_{k-2}(z)| , \end{aligned}\ ] ] where @xmath366 is the field of values of @xmath157 and the last inequality is by crouzeix s theorem @xcite . since @xmath367\times[-c_k , c_k]\subseteq c_r$ ] , we have @xmath368 now , the theorem follows from @xmath369 , where @xmath370 by lemma [ thm : min_re ] and lemma [ thm : psi_r_negative ] . we remark that @xmath349 is called crouzeix s constant and it is conjectured that it can be reduced to @xmath371 @xcite . combining the above theorem with theorem [ thm : aposteriori_re ] leads to the following _ a priori _ error bound in the following theorem . [ thm : apriori_re ] let @xmath125 and @xmath2 with @xmath155 and let @xmath159 be the arnoldi approximation to @xmath137 . then for any @xmath347 , the approximation error satisfies @xmath372 where @xmath349 , @xmath373 the parameters @xmath79 is determined by from @xmath374 of and @xmath287 is the ratio in . first note that @xmath375 for an orthogonal @xmath156 . then @xmath376\times[-c , c]\ ] ] now , theorem [ thm : bound_ht_re ] holds for @xmath342 , and indeed , from above and following the same proof , it holds with @xmath22 , @xmath23 , @xmath24 in place of @xmath351 , @xmath352 , @xmath353 . namely , @xmath377 with @xmath180 defined as in but from @xmath22 , @xmath23 , @xmath24 . now , using this bound in _ a posteriori _ error bound in theorem [ thm : aposteriori_re ] and noting that @xmath378 ( see ) , we have that , if @xmath379 , @xmath380 where we have used @xmath381 for any @xmath382 . if @xmath383 , the integration above gives @xmath384 and the final bound holds for this case as well . so the theorem is proved . for the rest of this section , we consider the case that @xmath3 is positive definite ( i.e. @xmath385 ) . in that case , the bound is simplified to @xmath386 bounding @xmath180 of using @xmath80 , we have @xmath387 this leads to a simple but obviously crude bound . in particular , the bound can be further simplified by setting the exponent @xmath388 to 0 , i.e. @xmath389 . we state these as the following corollary . under the the assumptions of theorem [ thm : apriori_re ] and that @xmath3 is positive definite ( i.e. @xmath385 ) , for any @xmath347 , the approximation error satisfies @xmath390 in particular , for @xmath389 , we have @xmath391 i.e. the error converges at least at the rate of @xmath392 . note that @xmath393 . since @xmath79 is a function of @xmath394 ( see lemma [ lem : unique ] ) and @xmath395 is the condition number of the hermitian part of @xmath3 , the bound relates the convergence to this condition number and the shape of the rectangle . more generally , we can find @xmath396 such that @xmath383 . then holds with this @xmath397 and the error converges at the rate @xmath397 . we call this @xmath397 the threshold convergence rate . however , this @xmath397 may not give the best bound possible among choices of @xmath346 . note that @xmath346 influences the error bound through two opposing actions of @xmath398 and @xmath399 . namely , choosing smaller @xmath346 results in a faster geometrically decreasing term @xmath398 , but @xmath399 may be much larger to result in an overall larger bound . so the best choice of @xmath346 should balance the two effects and will depend on @xmath128 . for example , smaller @xmath346 may be used for larger @xmath128 so that the more significant decrease in @xmath398 can offset the increase in @xmath399 . this suggest a superlinear convergence behavior where , as @xmath128 increases , the error is bounded with a smaller rate @xmath346 . in determining @xmath346 to be used in the bound , we consider the minimization at each step @xmath128 of @xmath400 taking derivative of @xmath197 with respect to @xmath346 and using @xmath401 we have @xmath402,\end{aligned}\ ] ] where @xmath403 . thus optimal @xmath404 can be found by solving @xmath405 note that a solution @xmath406 exists because the function in the equation is 1 when @xmath407 and @xmath408 when @xmath409 . finally , we discuss a special case , i.e. @xmath230 . [ thm : mto0 ] under the assumptions of theorem [ thm : apriori_re ] , and @xmath230 , the approximation error satisfies @xmath410 where @xmath411 and @xmath412 . @xmath413 and @xmath414 are both functions of @xmath79 and have the following expansions at @xmath415 ( * ? ? ? * , p. 591 ) @xmath416 then @xmath417 can be expanded at @xmath415 as @xmath418 since @xmath188 , @xmath419 then @xmath420 at the same time , for @xmath421 , @xmath422 so @xmath423 let @xmath396 be the unique solution of @xmath424 where the existence of @xmath397 and the uniqueness follow from the fact that the integral on the right is a function of @xmath346 monotonically decreasing from @xmath28 to @xmath301 for @xmath347 . using and , the equation is written as @xmath425 solving this , the solution @xmath397 with @xmath426 is @xmath427 using this @xmath397 in the bound , we have @xmath383 and the theorem is proved . note that @xmath79 is determined by @xmath210 . in particular , for @xmath230 , @xmath214 and @xmath213 have the expansions @xmath428 we also have the expansion of @xmath417 in . then @xmath429 so the above theorem applies to the case when @xmath430 is small or @xmath3 is nearly hermitian . in an earlier paper @xcite , it is shown that for a symmetric positive definite matrix @xmath3 , the approximation error satisfies @xmath431 where @xmath432 is the condition number of the matrix @xmath3 . this implies a conjugate gradient like convergence rate @xmath433 regardless of the norm of the matrix . then theorem [ thm : mto0 ] shows that the same conclusion holds if @xmath3 is nearly hermitian . in this section , we consider the special case that @xmath3 is skew - hermitian which , as discussed in the introduction , arises in some interesting applications . we write @xmath434 with @xmath13 being an hermitian matrix . in this case , the arnoldi algorithm is theoretically equivalent to the lanczos algorithm for @xmath13 . as we will see , the error bounds for computing @xmath435 is also significantly simplified . applying @xmath128 steps of the lanczos method to @xmath13 and @xmath436 with @xmath437 ( see @xcite ) , we obtain an orthonormal basis @xmath129 and a @xmath128-by-@xmath128 tridiagonal matrix @xmath438 such that @xmath439 where @xmath134 $ ] . this is equivalent to for the arnoldi algorithm for @xmath434 with @xmath440 and @xmath441 . then , the corresponding approximation of @xmath142 is @xmath442 which we call the lanczos approximation . then the same _ a posteriori _ error bound of theorem [ thm : aposteriori_re ] holds with @xmath443 and @xmath444 . namely , @xmath445 furthermore , slightly better bounds may be obtained by shifting the matrix . specifically , for any @xmath446 , we can consider the shifted matrix @xmath447 and correspondingly @xmath448 and @xmath449 . since @xmath450 , we can apply to @xmath447 to get @xmath451 where @xmath452 . thus @xmath453 we now bound @xmath177 as in the previous section to obtain the following _ a priori _ error bound . [ thm : apriori_im ] let @xmath454 be a skew - hermitian matrix and @xmath455 with @xmath155 . then , for any @xmath346 with @xmath347 , the error of the lanczos approximation @xmath456 satisfies @xmath457 where @xmath458 with @xmath459 and @xmath460 being the smallest and the largest eigenvalues of @xmath13 respectively . let @xmath461 and @xmath462 . we first bound @xmath444 as in theorem [ thm : bound_ht_re ] by constructing a conformal map and using the faber polynomial approximation . let @xmath463 where @xmath464 maps the exterior of @xmath465\}$ ] to the exterior of @xmath18 $ ] , @xmath466 maps the exterior of @xmath18 $ ] to the exterior of @xmath467 $ ] , @xmath468 maps the exterior of @xmath467 $ ] to @xmath469 . in the definition of @xmath290 , we choose the branch of @xmath470 such that @xmath471 . then @xmath42 maps the exterior of @xmath197 to the exterior of the unit circle @xmath472 with @xmath473 . construct the faber polynomials from this conformal map @xmath42 and the faber polynomial approximation @xmath474 of @xmath475 as defined in . let @xmath476 and let @xmath233 be the inverse image under @xmath42 of the circle @xmath73 . applying theorem [ thm : razouk54 ] or , the approximation error in @xmath358 is bounded as , @xmath477 where @xmath360 and we note that the total rotation of @xmath197 ( a line segment ) is @xmath361 . to find @xmath478 , for any @xmath479 , we write @xmath480 with @xmath481 where @xmath482 . then , it follows from the definition of @xmath42 that @xmath483 , \\ z_1 & = \frac{b - a}{2}z_2+\frac{b+a}{2}=\left[\frac{b - a}{4}\left(\frac{1}{q}+q\right)\sin\theta+\frac{b+a}{2}\right]-i\left[\frac{b - a}{4}\left(\frac{1}{q}-q\right)\cos\theta\right ] , \\ z & = iz_1=\frac{b - a}{4}\left(\frac{1}{q}-q\right)\cos\theta+i\left[\frac{b - a}{4}\left(\frac{1}{q}+q\right)\sin\theta+\frac{b+a}{2}\right ] . \end{aligned}\ ] ] thus @xmath484 now , let @xmath485 ( @xmath486 ) be the eigenvalues of @xmath487 . then @xmath488 . as in the proof of theorem [ thm : bound_ht_re ] , we have @xmath489_{k1}| = |[f(it_k)]_{k1}-[\pi_{k-2}(it_k)]_{k1}| \\ & \leq||f(it_k)-\pi_{k-2}(it_k)||_2 = \max_j |f(\lambda_j)-\pi_{k-2}(\lambda_j)| \\ & \leq\max_{z\in e}|f(z)-\pi_{k-2}(z)| \leq ||f-\pi_{k-2}||_\infty \\ & \leq\frac{2q^{k-1}}{1-q}e^{\frac{t(b - a)}{4}\left(\frac{1}{q}-q\right)}. \end{aligned}\ ] ] finally , using with @xmath490 , we have @xmath491 and hence @xmath492 where we have used @xmath493 for any @xmath494 . as before , we have an error bound for any given @xmath406 . using smaller @xmath346 results in a faster geometrically decreasing term @xmath398 , but @xmath495 is expected to be larger . so , again , we study the value of @xmath346 that minimizes the bound @xmath496 taking derivative of @xmath497 with respect to @xmath346 to get @xmath498.\ ] ] with @xmath499 as @xmath500 or @xmath501 , the optimal value @xmath502 that minimizes @xmath497 is given by the solution of the equation @xmath503 note that it can be shown that the above equation has a unique solution @xmath504 ( see @xcite for details ) . note that @xmath505 in @xmath497 is a well bounded term unless @xmath506 . for example , it is bounded by @xmath507 if @xmath508 . to quantitatively interpret the bound , we can consider minimization of @xmath509 which is essentially the same as @xmath497 unless @xmath510 . differentiate @xmath511 to get @xmath512.\ ] ] the discriminant of the quadratic @xmath513 is @xmath514 . so , if @xmath515 , @xmath516 is monotonically decreasing with the minimum occurring at @xmath517 . if @xmath518 , @xmath516 is minimized at @xmath519 . thus , the bound implies different convergence behavior at two stages of the lanczos iterations . 1 . when @xmath520 , there is essentially no decrease in the error bound . 2 . for @xmath518 , the error bounds for subsequent steps decrease at least at the rate of @xmath397 . the convergence behavior as implied from this theory is indeed what has been observed in the numerical examples ( see [ sec : numerical_examples ] ) , where the error initially stagnates for approximately @xmath521 steps and then begins to decrease superlinearly . thus our bound qualitatively explains this convergence property observed numerically . finally , we note that the convergence bound for skew - hermitian matrices have also been studied by hochbruch and lubich ( * ? ? ? * theorem 4 ) . it is proved there that for @xmath522 , @xmath523 interestingly , the range of validity of the bound coincides with the point of initial convergence as implied by our bound . it turns out that this bound can be implied from a special case of our error bound . for @xmath522 , . then our bound , simply using @xmath525 for the minimum , reduces to as follows : @xmath526 in this section , we present several numerical examples to demonstrate the error bounds obtained in this paper . all tests were carried out on a pc in matlab ( r2013b ) with the machine precision @xmath527 . the jacobi elliptic integrals that are needed for our bounds are computed using matlab built - in functions elliptick and elliptice . we will construct several testing matrices with different spectral distributions and compare the actual approximation error with the new _ a posterior _ error estimate and _ a priori _ bounds or . the integral in the _ a posterior _ error estimate is approximated using simpson s rule with @xmath507 subintervals on @xmath528 $ ] . we shall compare our bounds with the bounds by saad @xcite and where applicable with those of hochbruck and lubich @xcite as well . for example , if the matrices are positive semidefinite , we consider the following bound of saad ( * ? ? ? 2.2 ) : @xmath529 and the following bound of hochbruck and lubich ( * ? ? ? * theorem 2 ) : @xmath530 which holds for @xmath522 and with the assumption that the field of values @xmath340 is contained in the disk @xmath531 . _ example 1 . _ given an odd integer @xmath51 and a rectangle @xmath18\times[-c , c]$ ] in the complex plane where @xmath22 , @xmath23 and @xmath24 are all positive real numbers , let @xmath3 be the @xmath532 block diagonal matrix with the diagonal blocks being @xmath533 matrices @xmath534 for @xmath535 and @xmath536 , where @xmath537 , \ ; x_\ell= a+\frac{(\ell-1)(b - a)}{n-1 } \;\;\mbox { and } \;\ ; y_j = \frac{2jc}{n-1}.\ ] ] then , the eigenvalues of @xmath3 are @xmath538 with @xmath539 being the imaginary unit , which are the grid points of the @xmath540 lattice on @xmath18\times[-c , c]$ ] . clearly , @xmath3 is a normal matrix , so the field of values of @xmath3 is the convex hull of its eigenvalues , i.e. , the rectangle @xmath18\times[-c , c]$ ] . the primary purpose of this numerical test is to compare our _ a priori _ bound with hochbruch and lubich s bound . the latter is applicable when @xmath340 is contained in a disk @xmath531 . we therefore choose @xmath18\times[-c , c]$ ] to be the square @xmath541\times[-\frac{\sqrt{2}}{2},\frac{\sqrt{2}}{2}]$ ] which is enclosed in the circle @xmath542 and construct a matrix @xmath3 as above such that the eigenvalues of @xmath3 form a @xmath543 lattice in the square . we apply the arnoldi method to compute @xmath0 where @xmath158 is a random normalized vector and we use @xmath544 . in figure [ fig : ex1 ] , we plot against the iteration number the actual error @xmath152 in the solid line , the _ a posteriori _ error estimate in the @xmath545-line , our _ a priori _ bound in the dashed line , hochbruch and lubich s bound in the dotted line , and saad s bound in the x - line . note that hochbruch and lubich s bound is only valid for @xmath522 . in @xmath542 and @xmath544 . error ( solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) , and hochbruck and lubich s bound ( dotted ) . , title="fig : " ] in @xmath542 and @xmath544 . error ( solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) , and hochbruck and lubich s bound ( dotted ) . , title="fig : " ] in @xmath542 and @xmath544 . error ( solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) , and hochbruck and lubich s bound ( dotted ) . , title="fig : " ] in @xmath542 and @xmath544 . error ( solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) , and hochbruck and lubich s bound ( dotted ) . , title="fig : " ] we observe that when @xmath384 is relatively small , our new _ a priori _ bound is comparable to hochbruch and lubich s bound , but as @xmath384 increases , our bound improves significantly . in particular , for larger @xmath384 values , the error @xmath152 first stagnates for certain number of iterations before it starts to converge . our _ a priori _ bound nicely captures this behavior and the point where the convergence begins , while hochbruch and lubich s bound is pessimistic and is applicable to iterations long after the initial point of convergence . our _ a posteriori _ error estimate is sharp at the convergence stage for all tests . in the next example , we use the same construction as in example 1 , but consider the field of values contained in rectangles of different shape . this is to investigate the influence on the convergence rate by the shape of the rectangle through the parameter @xmath79 in . _ example 2 . _ for a given parameter @xmath98 , we determine the dimensions of the rectangle @xmath546 and @xmath547 by @xmath548 . we then construct a matrix as in example 1 whose field of values is contained in the rectangle @xmath549\times[-\beta,\beta]$ ] . we use @xmath550 whose corresponding values of @xmath194 are listed in figure [ fig : ex2 ] . note from section 3.3 that @xmath230 means that the matrix is close to being hermitian , and that @xmath231 means the matrix is close to being skew - hermitian with a real spectral shift . we apply the arnoldi method to compute @xmath0 for a random normalized vector @xmath158 and we use @xmath551 to give @xmath17 a moderate norm . in figure [ fig : ex2 ] we plot the error @xmath152 in the solid line , our _ a posteriori _ error estimate in @xmath545-line , our _ a priori _ bound in the dashed line and saad s bound in the x - line . ( top ) and @xmath552 ( bottom ) . error(solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] ( top ) and @xmath552 ( bottom ) . error(solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] ( top ) and @xmath552 ( bottom ) . error(solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] ( top ) and @xmath552 ( bottom ) . error(solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] figure [ fig : ex2 ] shows that the convergence is related to @xmath79 . for smaller @xmath79 when the eigenvalues lie close to the real axis , the convergence occurs at early iterations and at a faster rate . as @xmath79 increases to 1 , the convergence has an initial stagnation stage before the convergence occurs . again , this behavior is captured in our new _ a priori _ bound . our new bound also significantly improves saad s , which is based on the norm of the matrix only . our _ a posteriori _ error estimate is sharp for all tests . we further demonstrate our new bounds for non - positive definite matrices . we construct as in example 1 a matrix @xmath3 whose field of values is contained in the square @xmath553\times[-1,1]$ ] with @xmath554 and @xmath555 . we plot in figure [ fig : ex2_shift ] the actual error ( solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) and saad s bound ( x ) . we see that our bounds are still valid when @xmath3 is not positive definite . they also demonstrate the initial stagnation of convergence . however , the bound becomes more pessimistic for larger @xmath191 . . error ( solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , and saad s bound ( x ) . , title="fig : " ] . error ( solid ) , our _ a posteriori _ bound ( + ) , our _ a priori _ bound ( dashed ) , and saad s bound ( x ) . , title="fig : " ] in the next example , we consider matrices arising in the convection diffusion equation @xmath556 where @xmath557 ^ 2 $ ] . the finite - difference discretization in @xmath558 with a uniform mesh leads to an initial value problem and hence the problem of computing @xmath137 . _ example 3 . _ let @xmath559 be the finite - difference discretization of in a @xmath560 grid in @xmath561 ^ 2 $ ] scaled with @xmath562 so that @xmath563 . then @xmath3 is non - hermitian but positive definite . let @xmath158 be a random vector with @xmath127 and we compute the matrix exponential @xmath137 . we use various values of @xmath564 and apply the arnoldi method to @xmath3 and @xmath158 and the results are presented in figure [ fig : ex3 ] with @xmath152 in the solid line , our _ a posteriori _ error estimate in the @xmath545-line , our _ a priori _ bound in the dashed line and saad s bound in the x - line . . error(solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] . error(solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] . error(solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] . error(solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , saad s bound ( x ) . , title="fig : " ] we observe that for @xmath565 , our _ a priori _ bound is already a significant improvement on the classical bound by saad . for modestly large values of @xmath384 , saad s bound becomes very pessimistic due to the large norm of @xmath17 , while our _ a priori _ bound still follows the convergence curve of the error . for the case when @xmath566 ( @xmath567 ) or larger , our _ a priori _ bound also becomes very pessimistic . in all the cases , our _ a posteriori _ error estimate remains sharp . our final example concerns skew - hermitian matrices . _ example 4 . _ let @xmath13 be an @xmath9 diagonal matrix whose @xmath568-th diagonal entry is @xmath569 . let @xmath158 be a random @xmath570 normalized vector . then @xmath571 and the spectral gap @xmath572 is approximately 1 . we apply @xmath128 iterations of the lanczos method to compute @xmath573 . we will test @xmath574 with @xmath564 and the results are presented in figure [ fig : ex4 ] with @xmath152 in the solid line , our _ a posteriori _ error estimate in the @xmath545-line , our _ a priori _ bound in the dashed line , hochbruch and lubich s bound in the dotted line , and saad s bound in the x - line . diagonal matrix with @xmath575 . @xmath564 . error ( solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , hochbruch and lubich s bound ( dotted ) , and saad s bound ( x ) . , title="fig : " ] diagonal matrix with @xmath575 . @xmath564 . error ( solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , hochbruch and lubich s bound ( dotted ) , and saad s bound ( x ) . , title="fig : " ] diagonal matrix with @xmath575 . @xmath564 . error ( solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , hochbruch and lubich s bound ( dotted ) , and saad s bound ( x ) . , title="fig : " ] diagonal matrix with @xmath575 . @xmath564 . error ( solid ) , _ a posteriori _ bound ( + ) , _ a priori _ bound ( dashed ) , hochbruch and lubich s bound ( dotted ) , and saad s bound ( x ) . , title="fig : " ] we first observe that our bound only improves hochbruch and lubich s bound very slightly . it is significantly better than saad s bound when @xmath576 is large . in all cases , our and hochbruch and lubich s bound follow the actual error quite closely and our _ a posteriori _ error estimate is sharp . in addition , for larger @xmath384 , the error typically stagnates first for some iterations before it starts to converge . an analysis of our bound has shown that the convergence may be expected to start at @xmath577 . for @xmath564 , the corresponding @xmath128 is 1 , 5 , 10 and 25 , respectively . this basically matches the actual convergence curve in figure [ fig : ex4 ] , especially when @xmath384 is relatively large and more iterations are needed for the convergence . for the computation of @xmath0 with a non - hermitian matrix @xmath3 by the krylov subspace methods , we have presented an _ a posteriori _ error bound that provides a sharp estimate of the error . we have also derive new _ a priori _ error bounds based on the largest and the smallest eigenvalues of the hermitian and the skew - hermitian parts of @xmath3 . using this simple spectral information , our bounds capture convergence characteristics of the krylov subspace methods . they also explain often observed initial stagnation of the convergence curve . numerical comparisons with existing bounds also show that our new bounds may significantly improve the _ a priori _ bound by hochbruch and lubich @xcite that is based on a circular enclosing region of the field of values and the one by saad @xcite that is based on the norm . finally , it agrees with the bound @xcite for the symmetric positive definite case . the technique developed in this paper provides a new way to analyze convergence of the krylov subspace method for non - hermitian matrices through the bounding rectangle for the field of values . it may be extended to other linear algebra problems . for the future works , we plan to study convergence bounds for linear systems based on the hermitian and the skew - hermitian parts of @xmath3 , which may also add to the theory of the krylov subspace method for linear systems . g. dahlquist , _ stability and error bounds in the numerical integration of ordinary differential equations _ , almqvist & wiksells , uppsala , 1958 ; transactions of the royal institute of technology , stockholm , 1959 . x. guan , o. zatsarinny , k. bartschat , b.i . schneider , j. feist , and c.j . a general approach to few - cycle intense laser interactions with complex atoms _ , phys . a 76 ( 2007 ) , 053411 . m. ilic , i. w. turner , and v. anh , _ numerical solution of the fractional poisson equations using an adaptively preconditioned lanczos methods , _ journal of applied mathematics and stochastic analysis , vol . 2008 , article i d 104525 , 2008 . doi:10.1155/2008/104525 . h. kober , _ dictionary of conformal representations _ , dover publications inc . , 1957 . a. i. markushevich , _ theory of functions of a complex variable , vol . iii _ , revised english edition translated and edited by richard a. silverman , prenticd - hall inc . , englewood cliffs , n.j . , 1967 .

in the littlest higgs model with t - parity ( lht ) , top - partners ( ) are produced in pairs at the large hadron collider ( lhc ) . each top - partner decays into a top quark ( ) and the lightest-odd gauge partner . we demonstrate reconstruction of the system decaying hadronically , and measurement of the top - partner mass from the distribution . a top quark from a decay is polarized , and we discuss the effect of this polarization on the decay distributions . because the events consist of highly collinear jets which occasionally overlap each other , we compare distributions using different jet reconstruction algorithms ( snowmass cone , kt , cambridge , siscone ) . we find clustering algorithms are advantageous for studying top polarization effects .

0802.4142

in spite of the success of the standard model ( sm ) in explaining the particle interactions , there are yet two unsolved questions in the sm . one is the fine tuning problem and the other is the existence of the dark matter ( dm ) in our universe . the fine tuning problem is the question why the higgs is likely to be so light as expected from the lep data ( @xmath7 at the 95% confidence level @xcite ) , while naturally the higgs mass is the order of the cut off scale of the theory due to radiative corrections . some mechanism should protect the higgs mass , and interactions involving the higgs sector should be extended from the sm one . the existence of the dm is now established by the cosmological observations such as wmap , sdss and sn - ia @xcite . the sm should be extended to include the dm , which should be a neutral stable particle . moreover , provided that the dm is a thermal relic whose strength of the coupling to the sm particles of the order of the weak interaction , the mass can be a few 100 gev from a rough dimensional analysis . to solve the fine tuning problem , many models have been proposed . however , a so - called little hierarchy problem " @xcite arises in these models , once constraints from precision measurements are imposed . new operators arising from these models naively have the cut off scale @xmath8 tev to be consistent to the experimental data @xcite . some successful models solving the little hierarchy problem have parity structures , for example the mssm with r - parity @xcite , the little higgs model with t - parity ( lht ) @xcite , or the universal extra dimensional model ( ued ) with kk - parity @xcite . such a model which has a parity structure predicts the stable lightest parity odd particle as a candidate of the dm . the cut off scale of new operators in the model can be as low as a few tev @xcite , and the mass of the lightest parity odd particle can be on the order of a few 100 gev . the lhc is starting soon , and it is likely to discover new particles with masses up to a few tev @xcite . the models mentioned above are studied intensively by many authors . all of these models have partners of the sm colored particles which decay into the stable lightest parity odd particle through the parity conserving interactions . at the lhc , these partners are produced in pairs . the signal is multiple high @xmath9 jets and high @xmath9 leptons , each accompanied by missing transverse momentum @xmath10 . among the signals studied so far , the signals with high @xmath9 leptons are very promising @xcite , because the sm backgrounds are smaller . however , branching ratios of new particles into leptons strongly depend on model parameters . in addition , jets + leptons signals are often accompanied by undetectable neutrinos , which are also the source of missing momentum . they sometimes reduce the significance of the kinematical endpoints of the signal distributions for mass reconstructions . and even in the case that the lepton branching ratios are small , events with multiple jets and no lepton are enormously produced . in this paper , we focus on the top partner signal in the lht . the top partner pair productions occur with sizable cross section @xcite compared with stop pair productions in the mssm because the top partner is a fermion . the event has simple kinematics , @xmath11 where top quarks are highly boosted signals have been discussed in the rs1 model @xcite . however , it is different from the signal considered in this paper , because it is not accompanied by @xmath10 . ] . decay products from a boosted top are collimated and they are easily identified as a jet system with mass @xmath12 with high probability using hemisphere analysis @xcite . on the other hand , by imposing lepton veto and top tagging , background events from @xmath13jets can be reduced significantly . the process above was partly studied in ref . @xcite . in the paper , comphep @xcite and herwig6.5 @xcite are used for the event generation and acerdet1.0 @xcite is used for the detector simulation and jet reconstruction . acerdet1.0 implements the snowmass cone algorithm . however , the algorithm is not optimized in resolving the overlapping jets arising from boosted top quark correctly . recently , fastjet @xcite was released . infrared stable jet reconstruction algorithms ( kt @xcite , cambridge @xcite , siscone @xcite ) are implemented in the code , which is significantly faster than previous codes . we study improvements with these advanced jet reconstruction algorithms . in this study , we interface acerdet calorimeter information to fastjet and reanalyze the same process as in ref . @xcite to compare the results . we find the kt and cambridge algorithm have advantage to resolve overlapping jets . we also generate the signal events with underlying events using herwig6.5 + jimmy @xcite , and find the jet resolution with the kt algorithm is significantly affected by underlying events . this motivates us to show results with the cambridge algorithm mainly in this paper . we study the potential to measure the mass of the top partner using the reconstructed top candidates , although only the discovery potential is discussed in ref . one of the important variable is a @xmath5 @xcite . the @xmath5 is a function of two visible momenta , a missing transverse momentum and a test mass . in the case that the mass of the lightest @xmath2-odd particle ( @xmath14 ) is known , the endpoint of the @xmath5 distribution is equal to the top partner mass at @xmath15 . therefore we can measure the top partner mass @xmath16 using this distribution . we also generate the standard model backgrounds using alpgen @xcite+ herwig and conclude that they do not affect the endpoint of the @xmath5 distribution . we also discuss top polarization effects . a typical lht model predicts a top partner which decays dominantly into a right - handed top quark @xmath17 and a heavy photon @xmath3 . the polarization of tops can be measured through investigating decay distributions of tops . and we show that there are distinguishable difference between completely polarized case and non - polarized case in jet level analysis . this paper is organized as follows . in sec . [ lht ] , we explain our simulation setup for studying a top partner at the lhc . and we show how to reconstruct momenta of top quarks arising from top partner decays and how to measure the top partner mass @xmath16 using a reconstructed @xmath5 distribution . in sec . [ toppartner_label2c ] , we discuss differences among the jet reconstruction algorithms . in sec . [ toppolarization ] , we study top polarization effects . [ conclusion ] is devoted to the discussions and conclusions . in the following , we assume the top partner is the lightest in the fermion partners and decays exclusively to the lightest @xmath2-odd particle @xmath3 and a top . the top partner may be produced in pairs at the lhc and decays as , @xmath18 this process is similar to scalar top ( @xmath19 ) pair production process in the mssm . the production cross section of top partner is larger than that of scalar top in the case that the masses are the same , because top partner is a fermion . at the lhc , @xmath20 production cross section is 0.171 pb for @xmath21 gev . in order to identify this process , it is important to tag top quarks , and measure missing transverse momentum @xmath10 arising from escaping @xmath22 s . in the previous study @xcite , the events @xmath23 are generated by comphep @xcite , and top quark momenta are interfaced to herwig6.5 @xcite . in this paper , we also use comphep and herwig6.5 for signal events generation . our study is based on 8,550 signal events corresponding to @xmath24fb@xmath25 . the standard model background to the signal comes from the production of qcd , @xmath26 jets , @xmath27 jets , and @xmath28 jets events . in atlas study , it was shown that these four processes contribute to the background of the 0 lepton @xmath29 jets channel for susy search with approximately the same order of magnitude @xcite . among those , the qcd background arises due to the detector smearing and inefficiency , and we do not attempt to simulate it in this paper . even if qcd background is taken into account , it will not affect the results significantly because we require top mass cuts for the event selection . we will discuss this point later . the other processes contributes to the background due to hard @xmath30 produced from @xmath31 and @xmath32 decay . they are generated by alpgen+herwig in the paper . to reduce the computational time , we generate @xmath33 jets ( @xmath34 ) events corresponding to 5 fb@xmath25 with @xmath35 gev , @xmath36 jets ( @xmath34 ) corresponding to 10fb@xmath25 with @xmath37 gev , and @xmath38 jets ( @xmath39 ) corresponding to 12 fb@xmath25 with @xmath40 gev respectively . parton shower and matrix element matching are performed using mlm scheme provided by alpgen . we require @xmath41 , @xmath42gev and @xmath43 for parton level event generation before the matching . the @xmath44 followed by @xmath45 events become irreducible backgrounds , and we have generated the events for roughly @xmath46fb@xmath25 . we do not apply k - factor both for signal and background . we use acerdet1.0 for detector simulation , particle identification and jet reconstruction as in @xcite . in addition , we interfaced calorimeter information of acerdet1.0 to fastjet2.2beta @xcite so that we can compare different jet reconstruction algorithms simultaneously , and model the detector granularity . here the calorimeter information is the energy deposit @xmath47 to each cell @xmath48 centered at ( @xmath49 , @xmath50 ) with the size @xmath51 and @xmath52 ( 0.2 in the forward directions ) without smearing . they are interfaced as massless particles with momenta @xmath53 ) to fastjet . in this paper , we study jet distributions in the infrared stable algorithms , ( kt , cambridge and siscone ) , together in those for the snowmass cone algorithm provided by acerdet . to compare the four jet algorithms under the same conditions in sec . [ toppartner_label2 ] , we switch off the jet energy smearing . for the background , the smearing on @xmath10 might have the same impact on the estimation of the number of events after the cut , therefore we use the @xmath10 smeared by the acerdet . effects of jet energy smearing are discussed in the appendix [ smearing ] . @xmath54 & @xmath55 & @xmath56 & @xmath57 & @xmath58 both@xmath58 & @xmath58or@xmath59 & @xmath60 & @xmath61 @xmath62(signal ) & @xmath63&@xmath64&@xmath65&@xmath66&@xmath67&@xmath68&@xmath69&@xmath70@xmath13jets & @xmath71&@xmath72&@xmath73&@xmath74&@xmath75&@xmath76&@xmath77&@xmath78@xmath79 & @xmath80&@xmath81&@xmath82&@xmath83&@xmath84&@xmath85&@xmath86&@xmath86@xmath87jets & @xmath88&@xmath89&@xmath90&@xmath91&@xmath92&@xmath93&@xmath94&@xmath95@xmath96jets & @xmath97&@xmath98&@xmath99&@xmath100&@xmath101&@xmath102&@xmath103&@xmath95 we describe our cuts to select @xmath20 events . the summary of the numbers of the events after the cuts is shown in table [ toppartner_label3 ] . first , we impose our standard cuts for jet @xmath9 , @xmath104 and veto high @xmath9 isolated leptons , @xmath105 here , @xmath106 @xmath107 is a number of jets whose @xmath9 is larger than 50 ( 100 ) gev . @xmath54 is a number of isolated leptons ( @xmath108 ) with @xmath109 gev and @xmath110 . missing transverse energy is calculated using the energy deposit to the calorimeter and isolated leptons . it is calculated with smearing for the standard model background calculation . the lepton cut reduces @xmath111 jets and @xmath112 jets background , in which large @xmath104 is dominantly caused by neutrinos from leptonic @xmath32 decay . @xmath32 background still remains because @xmath32 can decay into @xmath113 . we applied a hemisphere analysis to find top candidates @xcite . each of high @xmath9 jets ( @xmath114 gev and @xmath115 ) is assigned to one of the two hemispheres which are defined as follows ; @xmath116 where @xmath117 @xmath118 @xmath119 to find hemispheres , we first take the highest @xmath9 jet momentum @xmath120 as @xmath121 and take the jet momentum @xmath122 which maximizes @xmath123 among all @xmath124 as @xmath125 . we group jets into hemisphere @xmath126 @xmath127 according to the eq.([distance ] ) . new @xmath128 s are then calculated from eq.([hemidef ] ) , and this process is repeated until the assignment converges . in this analysis , collinear objects tend to be assigned into the same hemisphere . top quarks from @xmath0 decays are highly boosted , then the decay products from the two top quarks are correctly grouped into different hemispheres with high probability . in this situation , the dependence on the definition of the distance eq.([hemidistancedef ] ) is weak . change the definition of the distance as @xmath129 causes negligible differences of the acceptance and our analyses in this paper . to assure the correct top reconstructions , we require both hemispheres transverse momenta are larger than a threshold , @xmath130 after imposing these cuts , distributions of the invariant masses of the hemisphere momenta ( @xmath131 ) for the @xmath20 and the standard model background events are shown in fig . [ toppartner_label5 ] . we can see peaks at the top mass both for the @xmath20 and @xmath13jets events in @xmath132 distributions ( fig . [ toppartner_label5]a , and [ toppartner_label5]c ) . on the other hand , such a peak is not seen in the @xmath133 distribution for @xmath13jets events ( fig . [ toppartner_label5]d ) . this is because at least one of the two tops should decay leptonicaly to give large @xmath104 . we also plot the distribution for @xmath134 jets and @xmath135 jets with dashed and dotted lines respectively . we do not see any structure in the hemisphere mass distributions . two dimensional scattering plots in @xmath136 vs. @xmath137 plane for the signal and @xmath13jets events are also shown in fig . [ toppartner_label6 ] , which show the clear difference between them . we can reduce the events from @xmath134 jets , @xmath112 jets and @xmath13 jets with hemisphere mass cuts . in table [ toppartner_label3 ] , the column @xmath138 shows the number of signal and background events after requiring @xmath139 gev @xmath140 gev . the number of events of @xmath13 jets ( @xmath134 jets , @xmath112 jets ) decreases by approximately @xmath141 after the cut . in the second hemisphere , only the signal distribution ( fig . 1b ) has a peak and background distributions ( fig . 1d ) are flat . we do not simulate qcd background in this paper . the magnitude of the qcd background for susy 0-lepton channel is approximately the same as that of the @xmath13 jets background @xcite . the hemisphere mass distribution of qcd background should be similar to that of @xmath134 jets or @xmath112 jets background . therefore the contribution of qcd background after the hemisphere mass cuts may be approximately the same as that of @xmath134 jets or @xmath112 jets background . the column both " shows the number of events after requiring the cut that both @xmath132 and @xmath133 are consistent with @xmath142 . the number of background becomes small by a factor of @xmath143 . however , the cut also reduces the signal events by a factor of @xmath144 . this is reasonable because we have the minimum jet energy cut for the hemisphere reconstruction ( @xmath145 gev ) , and some of top decay products may not contribute to the hemisphere momentum . additionally , in the case that a @xmath146-quark decays semi - leptonically , the hemisphere momentum and the invariant mass are also reduced . maybe the approach in ref . @xcite improves the mass resolution further . at this point , the background still dominate the signal , @xmath147 . if the sideband events can be used to estimate the background distribution , the significance of the signal events goes beyond 5 sigma . and b ) @xmath148 for the @xmath20 events for 50 fb@xmath25 , c ) @xmath149 and d ) @xmath148 for the events from @xmath13jets ( solid ) , @xmath112jets ( dotted ) and @xmath134jets ( dashed ) after the cut eq.([basiccut ] ) and eq.([ptcut ] ) . bg events are rescaled to 50 fb@xmath25.,title="fig : " ] and b ) @xmath148 for the @xmath20 events for 50 fb@xmath25 , c ) @xmath149 and d ) @xmath148 for the events from @xmath13jets ( solid ) , @xmath112jets ( dotted ) and @xmath134jets ( dashed ) after the cut eq.([basiccut ] ) and eq.([ptcut ] ) . bg events are rescaled to 50 fb@xmath25.,title="fig : " ] + and b ) @xmath148 for the @xmath20 events for 50 fb@xmath25 , c ) @xmath149 and d ) @xmath148 for the events from @xmath13jets ( solid ) , @xmath112jets ( dotted ) and @xmath134jets ( dashed ) after the cut eq.([basiccut ] ) and eq.([ptcut ] ) . bg events are rescaled to 50 fb@xmath25.,title="fig : " ] and b ) @xmath148 for the @xmath20 events for 50 fb@xmath25 , c ) @xmath149 and d ) @xmath148 for the events from @xmath13jets ( solid ) , @xmath112jets ( dotted ) and @xmath134jets ( dashed ) after the cut eq.([basiccut ] ) and eq.([ptcut ] ) . bg events are rescaled to 50 fb@xmath25.,title="fig : " ] ( vertical ) vs. @xmath137(horizontal ) distributions for @xmath20 ( the left figure ) and for @xmath4 ( the right figure ) . the units of axes are gev . , title="fig : " ] ( vertical ) vs. @xmath137(horizontal ) distributions for @xmath20 ( the left figure ) and for @xmath4 ( the right figure ) . the units of axes are gev . , title="fig : " ] to verify whether a momentum of a hemisphere whose mass is consistent with @xmath142 correctly matches a top momentum , we compare the @xmath121 with momentum of top partons @xmath150 . in fig . [ cbpartonmatch ] , we show the distributions of the @xmath151 and @xmath152 to see the difference between the two momenta . here , @xmath153 and @xmath154 is defined for one of the two top partons that gives smaller @xmath153 . the @xmath155 distribution has peak near 0 , and @xmath156 for most of the events . we conclude that a momentum of a hemisphere with @xmath157 may be considered as a momentum of a top partons . and @xmath153 for @xmath158.,title="fig : " ] and @xmath153 for @xmath158.,title="fig : " ] we now show that top partner mass ( @xmath16 ) can be measured using the endpoint of the distribution of the cambridge @xmath5 variable @xcite for @xmath4 + @xmath10 system if we know the ltp mass ( @xmath159 ) . first , only the signal distribution is considered and after that we will show that the background event does not contribute to the events near the endpoint and can be neglected for the determination of the endpoint . it turns to be the best discrimination between the signal and sm backgrounds . this variable is defined in the event @xmath160 , where @xmath161 and @xmath162 have the same masses @xmath163 , @xmath164 and @xmath146 are visible objects , and @xmath165 and @xmath166 are invisible particles with the same mass @xmath167 . in such a event , @xmath5 variable is defined as follows , @xmath168.\end{aligned}\ ] ] here , @xmath169 is an arbitrary chosen test mass and the transverse mass @xmath170 is defined as follows , @xmath171.\end{aligned}\ ] ] it is important that the following condition is satisfied in the case @xmath172 : @xmath173 thus @xmath174 can be extracted with measuring the upper endpoint of the @xmath175 distribution ( @xmath176 ) in the case that the true @xmath167 is known . in the case the true @xmath167 is not known , we can calculate @xmath177 for an arbitrary test mass @xmath169 . for each test mass @xmath169 , we can measure @xmath178 . the end point is expressed in terms of the following equation for the case that the masses of visible systems are the same ( @xmath179 ) and there is neither initial nor final state radiation @xcite , @xmath180 the endpoint contains the information on a combination of the relevant masses @xmath163 and @xmath167 . for a particle which undergoes more complicated decay process , @xmath179 can be various values . therefore one can obtain more than two independent information on the masses . practically , one can extract the true mass @xmath167 from a kink structure of the @xmath5 endpoint as a function of @xmath169 @xcite . for our case , visible particles are two top quarks , and invisible particles are two @xmath3 s . it is , therefore , not possible to determine top partner mass itself from the kink method because @xmath179 is always @xmath142 and not to be various values . we show the @xmath181 2-dimensional scattering plot in the fig [ toppartner_label4 ] . the dashed line is the line defined by eq.([mt2noisr ] ) substituted with the nominal values . the test mass dependence of the endpoint @xmath178 is well described by eq.([mt2noisr ] ) and no detectable kink structure can be seen . eventually , we can measure only a combination of masses ; @xmath182 in the case that a system of pair - produced particles has a net transverse momentum , the @xmath178 changes greater than the eq.([mt2noisr ] ) for all @xmath169 but for @xmath183 , therefore a kink structure might be seen @xcite . however , @xmath62 system generally has a small net @xmath184gev in average , a kink structure is not seen in our case even at parton level . then we can not measure the @xmath16 unless the @xmath159 is known for our case . the @xmath159 may be determined if productions of the other @xmath2-odd particles are observed . alternatively , if we assume the thermal relic density of @xmath3 is consistent with the dark matter density in our universe , @xmath159 is related to the higgs mass so that it is determined with two fold ambiguities @xcite . next , we show that @xmath16 can be measured using reconstructed tops at jet level assuming @xmath159 is known . ideally we may regard hemisphere momenta as top momenta if both hemisphere masses satisfy the condition @xmath185 . there is , however , not enough number of events left under these cuts as mentioned in the previous subsection . therefore in the following , we apply the cuts that one of the hemisphere masses @xmath186 satisfies @xmath139 gev@xmath187190 gev while the other satisfy @xmath46 gev @xmath188 gev ( the column or " in the table ) and regard the hemisphere momenta as top momenta . the endpoint of the @xmath5 distribution does not change under the relaxed cut or " . because @xmath5 is an increasing function of visible masses . additional sources of missing momentum ( such as neutrinos ) do not affect the endpoint either . it is easy to understand this as follows , a system of a ltp and the other sources of missing momentum can be regarded as an invisible pseudo - particle . the invisible pseudo - particle s invariant mass is always larger than the ltp mass ( @xmath189 ) , and @xmath190 is nevertheless smaller than @xmath174 because the system comes from @xmath191 pair production . on the other hand , @xmath177 is a monotonically increasing function of @xmath169 , therefore , @xmath192 is satisfied . scattering plot at the parton level . the dashed line is the line defined by the theoretical @xmath178 description with no transverse momentum limit , right ) the @xmath5 distributions for the nominal mass @xmath193 gev for the or `` ( solid line ) . the dashed line shows the contribution from the events survived after the both '' cut . the endpoint is 801.0 @xmath194 9.4 gev , and this value is consistent with @xmath21 gev . , title="fig : " ] 2dim . scattering plot at the parton level . the dashed line is the line defined by the theoretical @xmath178 description with no transverse momentum limit , right ) the @xmath5 distributions for the nominal mass @xmath193 gev for the or `` ( solid line ) . the dashed line shows the contribution from the events survived after the both '' cut . the endpoint is 801.0 @xmath194 9.4 gev , and this value is consistent with @xmath21 gev . , title="fig : " ] the @xmath5 distributions for the nominal value @xmath193 gev are shown in fig . [ toppartner_label4](right ) . the distribution after the or `` cut for the signal events are shown in the solid line . we fit the distribution near the endpoint by a linear function and obtain @xmath195 gev . this value is consistent with the nominal value @xmath21 gev . , we do not include the effects of jet energy smearing . in appendix [ smearing ] , we discuss the point . ] the fact supports validity of the relaxed cut ( or '' cut ) in determination of the @xmath5 endpoint . the contribution from the events survived after both " cut is shown in the dashed line and there are a few statistics . the distribution for the sm backgrounds is also shown in a dark histogram , and they have lower @xmath5 values . after imposing the cut @xmath196gev , the sm background is significantly reduced but the signal is not reduced as in table [ toppartner_label3 ] . moreover , after imposing the cut @xmath197gev , the sm background becomes negligible . therefore we can neglect them to fit the endpoint @xmath198gev in the present case . in a case that @xmath16 is lighter , for example @xmath199gev , a top quark momentum arising from a top partner is approximately 200gev ( @xmath200gev is assumed ) . it is boosted enough and the decay products distribute within @xmath201 @xcite , then the hemisphere analysis may work well . and the @xmath202 is ten times larger than the case of @xmath21gev @xcite . therefore it is possible to measure the endpoint by the method we proposed above can be discovered in the case of @xmath203gev . ] . we now study a dependence of the signal distributions on jet reconstructing algorithms . the reason to study different jet reconstruction algorithms is as follows . note that we need to study the jet system arising from a boosted top quark . in the rest frame of a top partner , a top quark momentum arising from top partner decay is expressed as @xmath204 therefore , typical @xmath9 of a top quark is above 300 gev . the jet angle separation is of the order of @xmath205 . if decay products of a top are aligned in the direction of the top momentum , the angle is even smaller . it is important to choose the algorithm that gives the best result in such a situation . four algorithms ( snowmass cone , kt , cambridge , siscone ) are used in the following analyses . we take two cone - type algorithms . the first one is snowmass cone",which is a simple algorithm implemented in acerdet1.0 . it defines a list of jets as follows , in this paper , we take cell momenta as massless particle momenta . as one can easily see from the algorithm , the highest @xmath9 jet in a region takes all activities within @xmath206 even if there are sub - dominant activities nearby @xmath207 . as we will see later , jet - parton energy matching is worse than the other algorithms . the second one is siscone @xcite . this algorithm is a seedless cone search algorithm . it defines a list of jets as follows , the reconstruction algorithm is infrared safe , because the reconstruction does not relay on the highest @xmath9 cell in a cell list . the number of reconstructed jets depends sensitively on the overlap parameter @xmath208 . if we set @xmath208 smaller , the algorithm tends to merge jets . for our choice @xmath209 , which is the default value , jets from a top quark tend to be merged and efficiency of resolving the three jets in a hemisphere is rather low compared with the other algorithms . the other category of jet finding algorithms is a clustering algorithm . a typical algorithm in this category , the kt algorithm @xcite is defined as follows : cambridge algorithm @xcite is similar to the kt algorithm but definition of @xmath210 and @xmath211 is modified as follows : in table [ toppartner_label8 ] , we show the numbers of signal events for the four jet algorithms after the same hemisphere mass cuts as in table [ toppartner_label3 ] . acerdet has an option to rescale jet energy , the results for the snowmass cone algorithm are given with jet calibration . the scale factor is determined so that an invariant mass distribution of the two jets from @xmath32 has the peak consistent with @xmath212 . @xmath213 & @xmath214 & @xmath59 both@xmath59 & @xmath59or@xmath59 & snowmass cone ( calibrated ) & 375 & 442 & 143 & 363 & @xmath215 gev kt & 420 & 415 & 135 & 411 & @xmath216 gev cambridge & 404 & 396 & 130 & 398 & @xmath217 gev siscone & 425 & 385 & 137 & 396 & @xmath218 gev we obtained 375 , 420 , 404 , 425 events after the @xmath219 cut for the snowmass cone , kt , cambridge , siscone respectively . [ cols="^,^ " , ] the distributions of @xmath136 for four algorithms are shown in fig . [ toppartner_label9 ] . the shaded regions denote the region satisfying @xmath220 . the kt , cambridge and siscone show nice resolutions in top mass . the peak for the snowmass cone algorithm is dull and has broad tail . this is because acerdet takes massless jets . this is rather an artificial difference as it is straight forward to define non - zero jet masses from calorimeter information . if such jet definitions are feasible at the lhc environment , the reconstruction efficiency may be increased significantly although we have not simulate the effect of mis - measurement of calorimeter energy . we regard the efficiency in the snowmass cone as a conservative estimate . fortunately , the endpoints of @xmath5 distributions are rather insensitive to the reconstruction algorithm . we find that they are @xmath215 gev ( snowmass cone ) , @xmath221 gev ( kt ) , @xmath217 gev ( cambridge ) , @xmath218 gev ( siscone ) with the statistical errors of the order of 10 gev . the difference among the algorithms is not essential at this point . distributions of hemisphere @xmath149 for @xmath20 events . the events with 150 gev@xmath222 190 gev are selected . ] [ partonmatchp ] shows deviation of a hemisphere momentum from a true top parton momentum @xmath223 for the four algorithms . we selected the signal events with 150 gev@xmath224 190 gev . the @xmath225 is mainly distributed within @xmath226% region for all algorithms . the @xmath153 for top parton and hemisphere momentum ( @xmath227 ) is mainly distributed less than 0.03 for all algorithms . these agreements justify regarding a hemisphere momentum as a top momentum . the peak position is larger than 0 by approximately 2% for the snowmass cone algorithm , because jet energy calibration by acerdet leads over - estimate of the jet energy for collimated jets . the jet energy calibration compensates the activities outside the jet cone , but for the collimated jets they are taken into account by the other jets . for the other algorithms , the peak position is less than 0 due to semi - leptonic decays of @xmath146-quark . distributions for the selected hemispheres ( 150 gev@xmath228 190 gev ) for the @xmath20 events . ] we now look into the matching between a @xmath146-parton and a @xmath146-jet in a selected hemisphere . the selected hemispheres satisfy the following conditions : 1 ) there are only three jets in the hemisphere and 150 gev@xmath229 gev , and 2 ) at least one jet pair satisfies @xmath230 gev and the other jet is @xmath146-tagged . here , we define a jet with @xmath231 as a @xmath146-jet ( @xmath232 . in fig . [ bpartonmatchp ] , the @xmath233 distributions are shown . the large tails found for @xmath234 come from a @xmath146-parton decaying semi - leptonically . in addition , the distribution for the snowmass cone algorithm shows a tail for @xmath235 . this tail arises because a locally highest @xmath9 jet takes over all energy in @xmath236 cone because the jet finding algorithm starts from the highest @xmath9 clusters . this feature can not be improved with minor modifications of the algorithm . for the siscone algorithm , the number of the three jet events is significantly small compared with the others because the algorithm actively merges overlapping jets . we will discuss this point in the next subsection . the algorithm is therefore not suitable for our analysis in the next section , in which we study the top spin dependence of @xmath146-jet distributions . in this subsection we compare clustering algorithms with the siscone in terms of the number of jets in a hemisphere . the numbers of jets inside a hemisphere with 150 gev@xmath237 190 gev are shown in the table [ nofj4 ] and [ nofj2 ] . since kt algorithm behaves similar to cambridge algorithm , only those for the cambridge and siscone are shown . the parameter @xmath238 for the clustering algorithms and for the siscone have different meanings and siscone has additional parameter @xmath208 as explained in sec.[toppartner_label2 ] . we investigate the distribution of the number of jets varying these parameters ( @xmath239 and @xmath240 for siscone ) . in order to study the top decay distribution , it is better to choose the parameters which give higher 3-jets acceptance . with the cambridge algorithm ( @xmath236 ) 511 events are classified into a group of 3-jets events , while with the siscone ( @xmath241 ) 324 events are classified into it although the total numbers of hemispheres with 150 gev @xmath228 190 gev are approximately the same . we can see @xmath242 or @xmath243 are optimal to enhance the number of 3-jets events with the siscone for our model point . the distribution of numbers of jet at those parameters are similar to that with cambridge ( @xmath236 ) . for such a small @xmath238 , however , some activities are missed outside a jet cone leading worse parton - jet matching , detector granularity might not be enough to resolve the jet ] . moreover , we found cambridge with @xmath244 also gives higher acceptance for 3-jets events than with @xmath236 . we do not find the parameter which improves the results for siscone over cambridge by changing @xmath238 and @xmath208 , therefore , we use clustering algorithms for the further analysis . appropriate @xmath238 should be used depending on top @xmath9 to protect unnecessary merging . sub - jet analysis based on clustering algorithms might be useful in such a case @xcite . we do not discuss these points any more because it is beyond the scope of this paper . @xmath245 1 jet & 9 & 103 & 58 & 41 2 jet & 244 & 429 & 413 & 362 3 jet & 511 & 295 & 324 & 362 4 jet or more & 36 & 5 & 15 & 23 total & 800 & 832 & 810 & 788 @xmath209 & @xmath245 & & @xmath246 & @xmath209 & @xmath245 1 jet & 2 & 51 & 16 & 12 & 1 & 10 & 2 & 2 2 jet & 138 & 335 & 294 & 235 & 47 & 135 & 106 & 84 3 jet & 574 & 419 & 430 & 462 & 467 & 477 & 440 & 430 4 jet or more & 85 & 36 & 55 & 71 & 136 & 107 & 122 & 115 total & 799 & 841 & 795 & 780 & 651 & 729 & 670 & 631 distributions in cambridge ( solid ) and kt ( dashed ) algorithms without ( left ) and with ( right ) underlying events . we take @xmath247.,title="fig : " ] without underlying events distributions in cambridge ( solid ) and kt ( dashed ) algorithms without ( left ) and with ( right ) underlying events . we take @xmath247.,title="fig : " ] with underlying events so far we have discussed the event distributions without underlying events . underlying events come from the soft parton interactions which occur with a hard collision , and whose nature at the lhc has large theoretical uncertainty . the top reconstruction efficiency may become worse with them , because the number of hit cells significantly increases with underlying events . we have generated the signal events with underlying events and multiple scattering using herwig6.5 + jimmy @xcite . in fig . [ withoutjimmy ] ( left ) , we show the distributions of @xmath136 for the kt ( dashed ) and cambridge ( solid ) algorithms without underlying events for @xmath236 , @xmath248 , @xmath249 , @xmath250 . the event selection cuts are the same as section [ toppartner_label1 ] . we can see that the locations of the peaks increase as increasing @xmath238 for both the kt and cambridge algorithms . the shapes of distributions are similar for all @xmath238 as top quarks are boosted enough so that the decay products are isolated from the other activities . however the kt algorithm tends to give higher invariant mass than the cambridge algorithm . in fig . [ withoutjimmy ] ( right ) , we show the same distributions with underlying events . we can see that the position of the peak for the kt algorithm is significantly larger than that for the cambridge algorithm in all @xmath238 values . even for @xmath236 , the peak position is larger than 175 gev for the kt algorithm . the situation becomes worse for the kt algorithm as @xmath238 increases . the reconstruction efficiency is reduced significantly for @xmath251 . this is because the kt algorithm over - collects soft activities which are far from the jet direction ( large @xmath252 ) due to the factor @xmath253 in the definition of the distance in eq.([ktdef ] ) , which is known as splash - in effects . on the other hand , the cambridge distance measure does not have the factor , therefore , it is not too sensitive to the existence of the underlying events . the effect of the underlying events can be safely neglected for @xmath254 . hence we take the cambridge algorithm in sec . [ lht ] and [ toppolarization ] . in this section , we consider top polarization effects . in the littlest higgs model with t - parity , the lagrangian relevant to a top partner decay is written as follows @xcite . @xmath255 where , @xmath256 a top partner @xmath0 decays dominantly into a top with @xmath257 if @xmath258 and @xmath259 , where @xmath260 is defined as the top helicity . it is the case in our model point , since @xmath261 . the amplitude is calculated in the appendix [ app_polarization ] , and we find @xmath262/[n(h_t=1/2)+n(h_t=-1/2)]\sim 0.85 $ ] . to simulate the top polarization effect we need to follow the decay cascade till the partons arising from top decay . instead of generating @xmath263 using comphep , we generate stop pair production @xmath264 followed by @xmath265 at a mssm model point using herwig and their decay products , which does not exist for @xmath266 . the correlation in principle appear in momentum distribution of @xmath1 and @xmath267 . however , the effect is rather small because @xmath0 is non - relativistic and also the system does not have enough kinematical constraints . ] . we take the mssm parameter that the other sparticles are heavy and the decay vertex of @xmath268 is approximately proportional to @xmath269 , so that herwig generates stop pair efficiently and they decay into approximately completely polarized top quark ( @xmath270 for our model point ) . herwig has an option to switch off polarization effects . for the mssm point , we do not find any distinguishable difference between the @xmath5 distributions with / without polarization effects . therefore the results shown in sec.[lht ] may be valid even for the lht because the spin correlation effects are not large . decay distributions of the top quark contains information on the interaction vertex @xcite . the amplitudes for @xmath271 are expressed as follows , @xmath272 the amplitudes for the other helicity combinations vanish . here , @xmath273 is the energy of the bottom quark in the rest frame of the decaying top . and @xmath274 and @xmath275 are the polar and azimuthal angles of the momentum of @xmath32 boson defined in the rest frame of the decaying top . the @xmath274 is measured from the top momentum direction and the @xmath276 is a helicity of @xmath32 . in fig.[distribution ] ( left ) , we show the parton level top decay distribution as the function of @xmath277 for the polarized and non - polarized cases , where @xmath278 is the angle between the @xmath146-quark momentum and the @xmath1-quark boost direction at the rest flame of the @xmath1-quark . we find evident difference between the two cases . in the region around @xmath279 , the emission of @xmath146 quark is suppressed because the amplitude proportional to @xmath280 is suppressed for @xmath257 . to see this at jet level , we study distributions of jets that are consistent to top decay products . some parton configuration is difficult to resolve at jet level . we analyse only the hemispheres with @xmath139 gev@xmath281 gev . we require that two of the three jets are consistent with those coming from @xmath32 , that is @xmath282gev , and the other jet is @xmath146-tagged . here , we regard a jet as a @xmath146-jet if the direction of the momentum is in a @xmath283 cone centered at a bottom parton momentum with @xmath284 gev . the analysis of the hemisphere which consists of 2 jets is given in appendix [ app_configuration ] . in fig.[distribution ] ( right ) we show the distribution of the angle between the b - jet momentum and the reconstructed top momentum @xmath285 . for the plot , we selected only the events with @xmath286gev . under the cut , standard model backgrounds are negligible as seen in sec.[lht ] . we use the measured hemisphere momentum to go back to the rest frame of the jet system . there is a distinguishable difference between polarized and non - polarized distributions . the ratio @xmath287 for the polarized case , while it is @xmath288 for the unpolarized one . a polarized top quark decays into a polarized @xmath32 @xmath289 . decay distribution of polarized @xmath32 ( @xmath290 ) can be calculated and the amplitudes are written as follows , @xmath291 here , @xmath292 and @xmath293 are the polar and the azimuth angles of the momentum of one of the jets from a @xmath32 decay to the @xmath32 momentum direction . these momenta are defined at the rest frame of the @xmath32 . a longitudinally polarized @xmath32 ( @xmath294 ) tends to decay transversely . on the other hand a transversely polarized @xmath32 ( @xmath295 ) tends to decay along a direction of the @xmath32 momentum . these differences may appear in the jet @xmath9 asymmetry @xmath296 , which defined as follows , @xmath297 jets from a @xmath298 decay tend to have @xmath299 while those from a @xmath300 give larger @xmath301 . this can be seen in fig.[asymmetry ] ( left ) . in this plot the @xmath302 distributions are shown for the events with @xmath303 . the ratio @xmath304 with @xmath305 is larger for polarized tops . therefore , @xmath296 for polarized top is distributed more around 0 than for non - polarized top . fig . [ asymmetry ] ( right ) shows @xmath296 distributions at jet level for @xmath306 . unfortunately , it is difficult to see the differences only by the shape of the distribution due to the limited statistics . in this paper we have studied reconstruction of top quarks arising from @xmath62 productions and its subsequent decay into a top and a stable gauge partner @xmath3 in the lht . we demonstrate the reconstruction of the top quarks through finding collinear jets whose invariant mass is consistent with @xmath142 using hemisphere analysis . main sm background processes are @xmath13jets , @xmath134jets and @xmath112jets productions , which can be reduced by imposing the cut on hemisphere momenta and @xmath5 variable . we also investigate the dependence on jet reconstructing algorithms . the cone algorithm used in the previous study @xcite is not optimal for the process . a top from @xmath0 is boosted , while the algorithm is designed so that the highest @xmath9 jets take all activities near the jet , mis - estimating the energy and the direction of the jets . an infrared safe version of the cone algorithms ( siscone ) also has some disadvantage for our case , because they tend to merge overlapping jets . we also study distributions with modern clustering algorithms ( kt and cambridge ) , which in general give better results than the cone algorithms . we also study effect of underlying events , and find the known tendency that the kt algorithm overestimates jet energies caused by collecting far and soft activities . whereas the reconstruction efficiency in the cambridge algorithm is not affected if @xmath254 . we also discuss top polarization effects . a top quark arising from a top partner decay is naturally polarized . this can be studied through looking into a distribution of the @xmath146-jet from @xmath1 decay especially the angle to a reconstructed top momentum in the rest frame of the top . we find that difference of the distributions between polarized and non - polarized top is still at detectable level with @xmath146-tagging for reasonable integrated luminosity ( @xmath30750 fb@xmath25 for @xmath21gev and @xmath200gev ) . these analyses are demonstrated using the cambridge algorithm , which shows good @xmath146-jet and @xmath146-parton matching . in many new physics scenarios , boosted gauge bosons and top quarks are produced at a significant rate . our study shows that choosing a right jet reconstruction algorithm or studying the dependence on them is important to reveal the physics behind the signal . this work is supported in part by the grant - in - aid for science research , ministry of education , culture , sports , science and technology , japan ( no.16081207 , 18340060 for m.m.n . ) . this work is also supported by world premier international research center initiative ( wpi program ) , mext , japan . in this paper , we show the distributions without jet energy smearing . in acerdet , there is an option to smear jet and missing transverse energies . the smearing is introduced for each jet energy in the snowmass cone algorithm and for a sum of the total transverse momentum ( the missing transverse momentum ) rather than for each calorimeter cell . we do not try to include smearing effects for the other three jet reconstruction algorithms ( kt , cambridge , siscone ) in this paper , because cells that jets consist of depend on the jet reconstruction algorithms , therefore comparison of smearing effects under the same condition is not easy . to obtain a rough idea on the signal and background distributions with smearing , we show @xmath5 distributions with / without smearing and jet energy calibration in the snowmass cone algorithm . the distribution near the end point is not significantly changed . the effect is rather small because we take the smearing based on atlas detector performance 50@xmath308% . which is less than 10 % for jet with @xmath309 30 gev . for the snowmass cone algorithm without / with smearing ( left / right figure respectively ) . the dashed line shows for the events with @xmath310 for both @xmath126 . the solid line shows for the events with @xmath311 . the endpoints are [email protected]/810.2@xmath19425 gev for nominal @xmath193 gev . ( @xmath312 gev ) . calibrated jets are used for both figures . dark histogram is the background distribution in the case that jet energy smearing is on . we do not show the background distribution without jet energy smearing in the right figure . , title="fig : " ] for the snowmass cone algorithm without / with smearing ( left / right figure respectively ) . the dashed line shows for the events with @xmath310 for both @xmath126 . the solid line shows for the events with @xmath311 . the endpoints are [email protected]/810.2@xmath19425 gev for nominal @xmath193 gev . ( @xmath312 gev ) . calibrated jets are used for both figures . dark histogram is the background distribution in the case that jet energy smearing is on . we do not show the background distribution without jet energy smearing in the right figure . , title="fig : " ] in this paper we took mass parameters @xmath313 gev , @xmath193 gev , and @xmath314 gev . the lagrangian relevant to our study is @xmath315 here , @xmath165 and @xmath166 is approximately expressed in terms of @xmath316 therefore @xmath317 at our model point . the amplitude of a top partner decay into a top with helicities @xmath318 can be calculated as follows , @xmath319,\end{aligned}\ ] ] where at the rest frame of @xmath0 , @xmath320 and , @xmath321 the @xmath322 is the wigner s function , @xmath323 the square of the amplitudes are therefore expressed for each helicity eigenstate as follows ; @xmath324 ^ 2 , & ( + , + , 0),\cr \gamma_{a}^2 \sin^2\frac{\theta}{2 } [ \beta_{a } ( \frac{s_\alpha + s_\beta } { 2}a_+^t + \frac{s_\alpha - s_\beta } { 2 } a_-^t ) - ( \frac{s_\alpha + s_\beta } { 2 } a_-^t + \frac{s_\alpha - s_\beta } { 2 } a_+^t ) ] ^2 , & ( + , -,0),\cr 2\sin^2 \frac{\theta}{2 } ( \frac{s_\alpha + s_\beta } { 2 } a_-^t - \frac{s_\alpha - s_\beta } { 2 } a_+^t ) ^2 , & ( -,-,+),\cr 2\cos^2 \frac{\theta}{2 } ( \frac{s_\alpha + s_\beta } { 2 } a_-^t - \frac{s_\alpha - s_\beta } { 2 } a_+^t ) ^2 , & ( -,+,-),\cr \gamma_{a}^2 \sin^2 \frac{\theta}{2 } [ \beta_{a } ( \frac{s_\alpha + s_\beta } { 2}a_+^t -\frac{s_\alpha - s_\beta } { 2 } a_-^t ) - ( \frac{s_\alpha + s_\beta } { 2 } a_-^t - \frac{s_\alpha - s_\beta } { 2 } a_+^t ) ] ^2 , & ( -,+,0),\cr \gamma_{a}^2 \cos^2 \frac{\theta}{2 } [ \beta_{a } ( \frac{s_\alpha + s_\beta } { 2}a_+^t -\frac{s_\alpha - s_\beta } { 2 } a_-^t ) + ( \frac{s_\alpha + s_\beta } { 2 } a_-^t - \frac{s_\alpha - s_\beta } { 2 } a_+^t ) ] ^2 , & ( -,-,0),\cr 0 & ( { \rm the\ others } ) . \cr \end{cases}\end{aligned}\ ] ] \ii . only two jets in a hemisphere . one of the jets has mass consistent with @xmath32(@xmath327 gev @xmath328 gev ) and the other is @xmath146-tagged . ( @xmath32 decay products are merged into a jet and @xmath146-jet is isolated ) \iii . only two jets in a hemisphere . none of the jets has mass consistent with @xmath32 mass and at least one of jets is @xmath146-tagged . ( one of the partons from @xmath32 decay and a @xmath146-parton are merged into a jet , and the other from @xmath32 is isolated . ) in fig . [ typehemi ] , we plot @xmath278 distribution of jets for each group , where @xmath278 is the angle between a @xmath146-parton and a @xmath1-parton momenta at the rest frame of the @xmath1-parton . the left figure is for non - polarized top quarks and the right figure is for polarized top quarks . for type iv events , the number of the events where a @xmath146-parton goes in the forward direction is strongly suppressed for polarized events compared with non - polarized events , while the events for @xmath329 is significantly enhanced . it is consistent with parton level distributions . note that the events near @xmath330 can not be accepted as type iv for both polarized and non - polarized cases because @xmath9 of the @xmath146-parton is too small , while the region is the most sensitive to polarization effects . the other distributions do not show clear dependence on the polarization . 99 r. barate _ [ lep working group for higgs boson searches ] , phys . b * 565 * , 61 ( 2003 ) [ arxiv : hep - ex/0306033 ] . d. n. spergel _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 148 * , 175 ( 2003 ) arxiv : astro - ph/0302209 . d. n. spergel _ et al . _ , arxiv : astro - ph/0603449 . d. g. york _ et al . _ [ sdss collaboration ] , astron . j. * 120 * , 1579 ( 2000 ) [ arxiv : astro - ph/0006396 ] . s. perlmutter _ et al . _ [ supernova cosmology project collaboration ] , astrophys . j. * 517 * , 565 ( 1999 ) [ arxiv : astro - ph/9812133 ] . a. g. riess _ et al . _ [ supernova search team collaboration ] , astron . j. * 116 * , 1009 ( 1998 ) [ arxiv : astro - ph/9805201 ] . r. barbieri and a. strumia , arxiv : hep - ph/0007265 . r. barbieri and a. strumia , phys . b * 462 * , 144 ( 1999 ) [ arxiv : hep - ph/9905281 ] . c. csaki , j. hubisz , g. d. kribs , p. meade and j. terning , phys . d * 67 * , 115002 ( 2003 ) [ arxiv : hep - ph/0211124 ] . h. p. nilles , phys . rept . * 110 * , 1 ( 1984 ) . h. e. haber and g. l. kane , phys . rept . * 117 * , 75 ( 1985 ) . s. p. martin , arxiv : hep - ph/9709356 . h. c. cheng and i. low , jhep * 0309 * , 051 ( 2003 ) [ arxiv : hep - ph/0308199 ] . h. c. cheng and i. low , jhep * 0408 * , 061 ( 2004 ) [ arxiv : hep - ph/0405243 ] . i. low , jhep * 0410 * , 067 ( 2004 ) [ arxiv : hep - ph/0409025 ] . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 429 * , 263 ( 1998 ) [ arxiv : hep - ph/9803315 ] . i. antoniadis , n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 436 * , 257 ( 1998 ) [ arxiv : hep - ph/9804398 ] . i. antoniadis , phys . b * 246 * , 377 ( 1990 ) . t. appelquist , h. c. cheng and b. a. dobrescu , phys . d * 64 * , 035002 ( 2001 ) [ arxiv : hep - ph/0012100 ] . j. hubisz and p. meade , phys . d * 71 * , 035016 ( 2005 ) [ arxiv : hep - ph/0411264 ] . altas technical design report , cern / lhcc-99 - 15 , atl - phys-98 - 131 ( 1998 ) . h. baer , x. tata and j. woodside , phys . d * 45 * , 142 ( 1992 ) . h. baer , c. h. chen , f. paige and x. tata , phys . d * 53 * , 6241 ( 1996 ) [ arxiv : hep - ph/9512383 ] . a. belyaev , c. r. chen , k. tobe and c. p. yuan , phys . d * 74 * , 115020 ( 2006 ) [ arxiv : hep - ph/0609179 ] . b. lillie , l. randall and l. t. wang , jhep * 0709 * , 074 ( 2007 ) [ arxiv : hep - ph/0701166 ] . f. moortgat and l. pape , cms physics tdr chapter 13.4 , p.400 - 403 , cern / lhcc/2006 - 021 ( 26 june 2006 ) . s. matsumoto , m. m. nojiri and d. nomura , phys . d * 75 * , 055006 ( 2007 ) [ arxiv : hep - ph/0612249 ] . a. pukhov _ et al . _ , arxiv : hep - ph/9908288 . g. corcella _ et al . _ , arxiv : hep - ph/0210213 . p. de jong , talk at susy 2008 , seoul p.19 , e. richter - was , arxiv : hep - ph/0207355 . m. cacciari , arxiv : hep - ph/0607071 . s. catani , y. l. dokshitzer , m. olsson , g. turnock and b. r. webber , phys . b * 269 * , 432 ( 1991 ) . s. catani , y. l. dokshitzer , m. h. seymour and b. r. webber , nucl . phys . b * 406 * , 187 ( 1993 ) . s. d. ellis and d. e. soper , phys . d * 48 * , 3160 ( 1993 ) [ arxiv : hep - ph/9305266 ] . m. cacciari and g. p. salam , phys . b * 641 * , 57 ( 2006 ) [ arxiv : hep - ph/0512210 ] . y. l. dokshitzer , g. d. leder , s. moretti and b. r. webber , jhep * 9708 * , 001 ( 1997 ) [ arxiv : hep - ph/9707323 ] . s. bentvelsen and i. meyer , eur . j. c * 4 * , 623 ( 1998 ) [ arxiv : hep - ph/9803322 ] . g. p. salam and g. soyez , jhep * 0705 * , 086 ( 2007 ) [ arxiv:0704.0292 [ hep - ph ] ] . j. m. butterworth , j. r. forshaw and m. h. seymour , z. phys . c * 72 * , 637 ( 1996 ) [ arxiv : hep - ph/9601371 ] . a. barr , c. lester and p. stephens , j. phys . g * 29 * , 2343 ( 2003 ) [ arxiv : hep - ph/0304226 ] . m. l. mangano , m. moretti , f. piccinini , r. pittau and a. d. polosa , jhep * 0307 * , 001 ( 2003 ) [ arxiv : hep - ph/0206293 ] . w. s. cho , k. choi , y. g. kim and c. b. park , arxiv:0711.4526 [ ] . a. j. barr , b. gripaios and c. g. lester , jhep * 0802 * , 014 ( 2008 ) [ arxiv:0711.4008 [ hep - ph ] ] . m. asano , s. matsumoto , n. okada and y. okada , phys . d * 75 * , 063506 ( 2007 ) [ arxiv : hep - ph/0602157 ] . g. brooijmans , atlas note : atl - com - phys-2008 - 001 t. han , r. mahbubani , d. g. e. walker and l. t. e. wang , arxiv:0803.3820 [ hep - ph ] . j. m. butterworth , a. r. davison , m. rubin and g. p. salam , arxiv:0802.2470 [ hep - ph ] . j. hubisz , p. meade , a. noble and m. perelstein , jhep * 0601 * , 135 ( 2006 ) [ arxiv : hep - ph/0506042 ] . g. l. kane , g. a. ladinsky and c. p. yuan , phys . d * 45 * , 124 ( 1992 ) .

we present the results of a multi lorentzian fit to the power spectra of two kilohertz qpo sources ; 4u 0614 + 09 and 4u 172834 . this work was triggered by recent results of a similar fit to the black hole candidates ( bhcs ) gx 3394 and cyg x1 by nowak in 2000 . we find that one to six lorentzians are needed to fit the power spectra of our two sources . the use of exactly the same fit function reveals that the timing behaviour of 4u 0614 + 09 and 4u 172834 is almost identical at luminosities which are about a factor 5 different . as the characteristic frequency of the lorentzians we use the frequency , , at which each component contributes most of its variance per log frequency as proposed by belloni , psaltis & van der klis in 2001 . when using instead of the centroid frequency of the lorentzian , the recently discovered hectohertz lorentzian is practically constant in frequency . we use our results to test the suggestions by , respectively , psaltis belloni and van der klis in 1999 and nowak in 2000 that the two lorentzians describing the high frequency end of the broad - band noise in bhcs in the low state can be identified with the kilohertz qpos in the neutron star low mass x ray binaries . the prediction for the neutron star sources is that if the two kilohertz qpos are present , then these two high frequency lorentzians should be absent from the broad band noise . we find , that when the two kilohertz qpos are clearly present , the low frequency part of the power spectrum is too complicated to draw immediate conclusions from the nature of the components detected in any one power spectrum . however , the relations we observe between the characteristic frequencies of the kilohertz qpos and the band limited noise , when compared to the corresponding relations in bhcs , hint towards the identification of the second highest frequency lorentzian in the bhcs with the lower kilohertz qpo . they do not confirm the identification of the highest frequency lorentzian with the upper kilohertz qpo .

astro-ph0107562

low mass x - ray binaries ( lmxbs ) can be divided into black hole candidates ( bhcs ) and neutron star lmxbs . the accretion process in these lmxbs can be studied through the timing properties of the associated x - ray emission ( see for an overview van der klis 2000 ) . the fourier power spectra of the neutron star lmxbs contain several timing features ; a power law red noise component in the lowest frequency range of the spectrum ( @xmath1 1 hz ) called very low frequency noise ( vlfn ) , a band limited noise component ( bln ) which is in most cases flat at low frequency and steepens to an approximate power law with an index of about 1 at higher frequency and several quasi periodic oscillations ( qpos ) at low ( @xmath2 hz ) and high frequencies . the high frequency qpos , at frequencies from a few hundred hz to more than 1000 hz are called kilohertz qpos . the correlation between the timing features at low frequency and the spectral properties led to a precise classification of the neutron star systems as z or atoll sources ( hasinger & van der klis 1989 ) . note , that the component that we call bln has been previously called `` high frequency noise '' in atoll sources ; this is presumably the same phenomenon that is called `` low frequency noise '' in z sources ( cf . van der klis 1995 ) . the timing features of the bhcs include among others a band limited noise component , a bump between 1 and 10 hz and several qpos . for the low frequency ( @xmath3 hz ) part of the power spectrum links between bhcs in the low state and neutron star lmxbs were suggested ( e.g. , van der klis 1994a , b ) , in particular concerning the band limited noise and some of the qpos . the improved sensitivity and the discovery of several new timing features with the rossi x ray timing explorer ( rxte ) has opened up new possibilities for connecting the timing properties of the neutron star lmxbs and the bhcs . in the atoll source 4u 0614 + 09 van straaten et al . ( 2000 , hereafter vs00 ) found that the relation between power density and break frequency of the band limited noise was similar to that established for bhcs in the low state . wijnands & van der klis ( 1999 , hereafter wk99 ) showed that the break frequency of the band limited noise and the centroid frequency of a low frequency qpo follow the same correlation for both the neutron star lmxbs and the bhcs . psaltis , belloni & van der klis ( 1999 , hereafter pbk99 ) made a systematic study of qpos and broad band noise components of the neutron star lmxbs and the bhcs . they suggested the identification of two features by plotting their frequencies versus each other . the bhcs high frequency bump between 1 and 10 hz was tentatively identified with the lower kilohertz qpo in the z and atoll sources and the bhcs low frequency qpo with the horizontal branch oscillations / low frequency qpo in the z / atoll sources . recently , nowak ( 2000 ) described the power spectra of the bhcs gx 3394 and cyg x1 with a fit function consisting of four lorentzian components and suggested the identification of the two highest frequency lorentzians , which together describe the high frequency end of the broad band noise , with the two kilohertz qpos . the lorentzian that is used to fit the low frequency end of the broad band noise has a centroid frequency fixed to zero . the additional lorentzian fits a bump at low frequencies . in the picture of nowak ( 2000 ) for the bhcs the first lorentzian corresponds to the broken power law of wk99 , the second lorentzian to the low frequency qpo of wk99 and pbk99 and the third lorentzian to the high frequency bump between 1 and 10 hz of pbk99 . this work forms the original motivation for the current paper . also triggered by nowak ( 2000 ) , belloni , psaltis & van der klis ( 2001 , hereafter bpk01 ) describe a paradigm for fitting the power spectra of low luminosity bursters and bhcs in their low / hard state . they describe the power spectra of these sources with a fit function consisting of four lorentzian components of which in practice three are zero centered . the suggestion of nowak ( 2000 ) might be tested by comparing his black hole results with neutron star sources , where both kilohertz qpos can be clearly identified and strong broad band noise is simultaneously present : the prediction for the neutron star sources is that if the two kilohertz qpos are present the two high frequency lorentzians should be absent from the broad band noise . in this work we test the multi lorentzian fit function of nowak ( 2000 ) and bpk01 on two of the atoll sources , namely 4u 172834 and 4u 0614 + 09 . the timing properties at low and high frequencies of 4u 172834 and 4u 0614 + 09 have previously been studied in connection with color diagrams by vs00 for 4u 0614 + 09 and di salvo et al . ( 2001 , hereafter ds01 ) . for both sources strong correlations were found between the frequencies of several components and the position of the source in the color diagram . we find that instead of four lorentzians one to six lorentzians are needed to fit the power spectra of 4u 172834 and 4u 0614 + 09 . we describe these different lorentzian components in 3 . in 4 we compare our results with those of nowak ( 2000 ) and bpk01 . in this work we analyse data from rxte s proportional counter array ( pca ; for instrument information see zhang et al . 1993 ) of the neutron star lmxbs 4u 172834 and 4u 0614 + 09 . for 4u 172834 we used the fourier power spectra 1 to 19 obtained by and described in ds01 . the power spectra were constructed by dividing the pca light curve into segments of 256 s and then binning the data in time before fourier transforming such that the nyquist frequency is always 2048 hz ; the normalization of leahy et al . ( 1983 ) was used . power spectra were combined based on the position in the color diagram , which is thought to be an indicator of mass accretion rate . the resulting power spectra were then converted to squared fractional rms . the poisson noise estimated between 1200 and 2048 hz ( where neither noise nor qpos are known to be present ) was subtracted before fitting the power spectra . for 4u 0614 + 09 we use a large data set previously studied by vs00 plus recent observations performed between 2000 september 1 and 4 and 2001 may 24 and 28 . in the september 2000 observations the source was at levels corresponding to the lowest mass accretion rate ( as inferred from colors and power spectral properties ) observed by vs00 . in the may 2001 observations the inferred mass accretion rate was at an even lower level . starting may 12 , 2000 , the propane layer on pcu0 , which functions as an anti coincidence shield for charged particles , was lost . this leads to a contamination of the data from electrons trapped in the earth s magnetosphere or from solar flare activity . for the september 2000 and may 2001 observations we therefore exclude all data with electron2 @xmath4 0.09 , where electron2 is the measured coincidence of events between the pcu propane layer and either of the two anodes in the first layer of pcu2 , the only detector that was switched on in all the observations . the electron2 screening led to a loss of about 12% of the data . the loss of the propane layer in pcu0 also leads to an increased background rate for pcu0 . we approximately take this into account by adding 30% of the pcu0 background rate to the total background rate . for the colors used above we have excluded pcu0 . in vs00 the 4u 0614 + 09 data were split into near continuous time intervals of approximately 2500 s which were called observations . for the present work , to improve statistics , we constructed representative intervals by adding up several observations that showed very similar power spectra in vs00 . as all characteristic frequencies in the power spectra are correlated with each other and with the position of the source in the hardness intensity diagram ( parametrized in vs00 by a variable @xmath5 ) we can select the data on one of these frequencies . the only power spectral feature that is present in all observations is the high frequency noise . however , at a @xmath6 ( the break frequency of the band limited noise , see for a definition vs00 ) of about 25 hz and higher the correlation with the other frequencies and with the position in hardness intensity diagram breaks down ( see figures 3 and 6 in vs00 ) . therefore , for a @xmath6 well below 25 hz we used @xmath6 to select the data and for a @xmath6 of about 25 hz and higher we used the centroid frequencies of the kilohertz qpos . to represent the highest mass accretion rate in 4u 0614 + 09 ( where the kilohertz qpos are mostly absent ) we add up the two observations at @xmath5 = 2.59 and @xmath5 = 2.64 in vs00 . as in the case of 4u 172834 , we number the intervals ( 19 ) in order of an increasing inferred mass accretion rate . we note that we could have selected the data based on @xmath5 value as was done for 4u 172834 ( see above ) , but because of the scatter in the correlation of @xmath5 with the frequencies of the power spectral features ( see figures 3 and 4 in vs00 ) , in 4u 0614 + 09 this leads to an artificial broadening of the power spectral features and in the case of the kilohertz qpos sometimes even to double peaks . for each interval of 4u 0614 + 09 we constructed a power spectrum in the same fashion as for 4u 172834 ( see above ) , but with a nyquist frequency of 4096 hz . an interval contains between 53 and 382 power spectra . we subtracted the poisson noise estimated between 2000 and 4000 hz ( as , different from 4u 172834 , qpos are known to be present between 1200 and 1400 hz ) before fitting the power spectra . we fitted the power spectra with a sum of lorentzian components . one to six lorentzian components were needed for a good fit . we only include those lorentzians in the fit whose significance based on the error in the power integrated from 0 to @xmath7 is above 3.0 @xmath8 . in addition to the lorentzians a power law component is used to fit the so called very low frequency noise ( vlfn ) . for the intervals where the kilohertz qpos have sufficiently high frequencies not to interfere with the low frequency features and vice versa , we fit the kilohertz qpos between 500 and 2048 hz and then fix the kilohertz qpo parameters when we fit the whole power spectra , similar to what was done in ds01 . this is for computational reasons only ; the results are the same as those obtained with all parameters free . we plot the power spectra and the fit functions in the power times frequency representation ( e.g. belloni et al . 1997 , nowak 2000 ) , where the power spectral density is multiplied with its fourier frequency . for a fit function consisting of many lorentzians this representation helps to visualize a characteristic frequency corresponding to each lorentzian component ( bpk01 ) namely , the frequency where each component contributes most of its variance per logarithmic frequency interval . this characteristic frequency , is not equal to the centroid frequency , @xmath9 , of the lorentzian but to @xmath10{\nu_0 ^ 2 + \delta^2}$ ] , where @xmath11 is the hwhm of the lorentzian . in this work we therefore use @xmath0 as characteristic frequency for broad features . for narrow features the characteristic frequency is nearly the same as @xmath9 . all power spectral features that we find in 4u 0614 + 09 and 4u 172834 become broader towards lower interval number ( see 3 ) . several features can be classified as qpos at higher characteristic frequencies but evolve into broad bumps as their frequencies become lower . therefore , to be consistent we use @xmath0 for all features . the actual lorentzian functions fitted were of the form : @xmath12 where @xmath13{1 + 4 q^2}$ ] , and @xmath14 ( the fractional rms integrated from 0 to @xmath7 ) , @xmath15 ( the quality factor , defined as @xmath16 ) and @xmath0 were the independent fit parameters . the two sources 4u 172834 and 4u 0614 + 09 yielded remarkably similar results . one to six lorentzian components were needed for a good fit . in addition to these lorentzians , intervals 1419 of 4u 172834 and 8 and 9 of 4u 0614 + 09 needed a power law component to fit the very low frequency noise ( vlfn ) . we show power spectra and fit functions in figure [ fig.powspec_1728 ] for 4u 172834 and in figure [ fig.powspec_0614 ] for 4u 0614 + 09 . the values of @xmath0 for all lorentzian components are listed in table [ tbl.numax ] , the values of @xmath15 are listed in table [ tbl.qvalues ] and the values of the integrated fractional rms ( over the full pca energy band ) are listed in table [ tbl.rms ] . the quoted errors in @xmath0 , @xmath15 and @xmath14 use @xmath17 = 1.0 . if we compare the @xmath18/dof values of our fits with the @xmath18/dof values of a broken power law fit ( for a description see ds01 ) to the same intervals there seems to be a clear statistical preference for the multi lorentzian function . for rather similar numbers of degrees of freedom ( between 131 and 145 , and usually higher in the multi lorentzian than in the broken power law fit ) the @xmath18 values of the multi lorentzian fits are more than 10 lower than of the broken power law fits in 15 out of 28 cases , whereas the inverse is true in only 4 of the 28 cases . the fits are generally better for 4u 0614 + 09 ( all with a @xmath18/dof below 1.9 ) than for 4u 172834 ( @xmath18/dof below 4.1 ) ; this could be due to the different methods used to select the data ( see 2 ) . allmost all fitted power spectral features ( described below ) were already identified in ds01 and vs00 . in ds01 and vs00 the bln component is described with a broken power law , whereas we use a zero - centered lorentzian . as in ds01 and vs00 we use lorentzians to describe two low frequency qpos ( when present ) , a broad high frequency feature ( @xmath19 hz ) and the two kilohertz qpos . we do not describe the behaviour of the vlfn component in this work ; for a discussion of this component in 4u 172834 and 4u 0614 + 09 see respectively ds01 and vs00 . in the current work , it only affects the power spectra below 0.11 hz . to confirm the identification of the components , we have plotted the @xmath0 of the lorentzians versus the @xmath0 of the lorentzian identified as the upper kilohertz qpo ( @xmath20 ) in figure [ fig.all_vs_upkilo ] . intervals 18 and 19 of 4u 172834 are not included in this plot as these intervals only showed one lorentzian . the grey symbols mark the 4u 172834 points , the black symbols the 4u 0614 + 09 points . in figure [ fig.all_vs_upkilo ] five correlations are present , of which the top three can be unambiguously identified with the lorentzians described in 3.23.4 . for the other two correlations the identification is more complicated ; see 3.1 . the points from interval 1 of 4u 0614 + 09 ( from recent observations in may 2001 ) are circled . the lorentzian at 233 hz of this interval can be identified based on frequency , @xmath15 value or fractional rms as either the upper kilohertz qpo or the hectohertz lorentzian . in tables [ tbl.numax ] , [ tbl.qvalues ] and [ tbl.rms ] we list this lorentzian as the upper kilohertz qpo . in figure [ fig.all_vs_upkilo ] and in all further figures we will use the parameters of this lorentzian both for the upper kilohertz qpo and for the hectohertz lorentzian . these points will be circled . the two sources show a remarkable similarity in their behaviour . this similarity is also shown in the behaviour of the @xmath15 values ( table [ tbl.qvalues ] ) and the fractional rms s of the different components ( table [ tbl.rms ] ) , although the fractional rms in 4u 0614 + 09 is generally higher by about a factor of 1.3 to 2 than that in 4u 172834 . note that the hydrogen column density , @xmath21 , is about a factor 10 higher for 4u 172834 ( e.g. schultz 1999 ) . this leads to the absorbtion of more low energy photons for 4u 172834 than for 4u 0614 + 09 and could increase the factor as most power spectral features are stronger at higher energies . in figures [ fig.rms_vs_upkilo ] and [ fig.q_vs_upkilo ] we plot the fractional rms and the @xmath15 value of the hectohertz lorentzian , the lower and upper kilohertz qpos ( see below for a description of these components ) versus @xmath20 . the zero centered lorentzian which is used to fit the band limited noise ( bln ) , is present in most of the intervals ( except interval 9 of 4u 0614 + 09 ) . as noted in 1 , previously the bln in the atoll sources was referred to as hfn ( high frequency noise ) in the literature , but as it has the lowest characteristic frequency of all components that name seems no longer appropriate . the characteristic frequency of this component ( @xmath22 ) , increases with interval number . this increase is halted at interval 13 for 4u 172834 and interval 5 for 4u 0614 + 09 . here an additional , non zero centered lorentzian component had to be included in the fit to the bln . as previously noted by ds01 , the characteristic frequency of this new component , @xmath23 , continues to follow the relations between the @xmath22 and the characteristic frequencies of the other components ( see fig . [ fig.all_vs_upkilo ] ) as well as the relation with interval number . we shall refer to this lorentzian as the very low frequency lorentzian . the very low frequency lorentzian is generally broader at lower frequencies ( see tables [ tbl.numax ] and [ tbl.qvalues ] ) except for interval 18 and 19 of 4u 172834 and 9 of 4u 0614 + 09 for which the identification as very low frequency lorentzian is ambiguous ( see below ) . note that in the broken power law description , the very low frequency lorentzian is usually not statistically required in 4u 0614 + 09 , whereas it is in 4u 172834 ( ds01 ) . in figure [ fig.upkilo_vs_bln ] we plot the fractional rms of the bln versus @xmath20 . in the top panel we plot the fractional rms of the zero centered lorentzian . in the middle panel we plot the fractional rms of the very low frequency lorentzian when it is present , and the fractional rms of the zero centered lorentzian when it is not . in the bottom panel finally we plot the fractional rms of the zero centered lorentzian and the very low frequency lorentzian summed together . in the top two panels the relation tends to jump when the very low frequency lorentzian appears , where this jump does not occur in the bottom plot . the relation is also smoother in the bottom plot . this behaviour together with the frequency behaviour of the zero centered lorentzian and the very low frequency lorentzian described above could hint at the zero centered lorentzian and the very low frequency lorentzian being two components that together fit one feature . it could however also be that the very low frequency lorentzian is present in the intervals where we do not detect it and biases our fractional rms and @xmath0 measurements for the zero centered lorentzian . this would then explain the fractional rms and @xmath0 connections between the zero centered lorentzian and the very low frequency lorentzian . we also investigated the possibility that the very low frequency lorentzian is a sub harmonic of the low frequency lorentzian . we found that in @xmath0 the deviations from a harmonic relation are 0.54.4 @xmath8 . however , note that in the @xmath0 representation a harmonic relation has no real physical meaning . in the @xmath24 representation , where a harmonic relation has physical meaning , the deviations from a harmonic relation between the very low frequency lorentzian and the low frequency lorentzian are 1.96.5 @xmath8 . when we require an harmonic relation between the centroid frequencies of the very low frequency lorentzian and the low frequency lorentzian in the fits to the power spectra where both features are simultaneously present ( 5 and 6 for 4u 0614 + 09 and 912 for 4u 172834 ) we find that this significantly worsens the fits ( f test probabilities from 0.05 to @xmath25 ) . note that for intervals 1 , 2 , 4 , 5 and 6 of 4u 172834 there appears to be some residual power near the frequency where the sub harmonic of the low frequency qpo should be . in intervals 18 and 19 of 4u 172834 and 9 of 4u 0614 + 09 there is only one lorentzian present . based on the relations of characteristic frequency , @xmath15 and rms fractional amplitude with interval number , this lorentzian can be identified as either the zero centered lorentzian or the very low frequency lorentzian . as fitting this lorentzian with a zero centered lorentzian provides bad fits , we have listed it as the very low frequency lorentzian in tables [ tbl.numax ] and [ tbl.qvalues ] , but this identification is uncertain . contrary to all other components , the characteristic frequency of this component decreases with interval number ( see tables [ tbl.numax ] and [ tbl.qvalues ] ) . the second lorentzian at low frequencies ( which we shall refer to as the low frequency lorentzian ) was present in intervals 112 for 4u 172834 and 16 for 4u 0614 + 09 ( see figures [ fig.powspec_1728 ] and [ fig.powspec_0614 ] ) . the characteristic frequency of this component , @xmath26 , also increases with interval number ( see table [ tbl.numax ] ) . in 4u 172834 @xmath26 ranges from 11.546.7 hz , in 4u 0614 + 09 from 2.044.6 hz . the low frequency lorentzian is too broad to be classified as a qpo ( @xmath27 ) at low frequencies ( @xmath28 25 hz ; see tables [ tbl.numax ] and [ tbl.qvalues ] ) and @xmath15 only exceeds 2 at higher frequencies . note that the low frequency lorentzian might be present in interval 6 of 4u 0614 + 09 ( @xmath26 = 60.5@xmath29 hz ) ; as the significance based on the error in the integrated power is only 2.4 @xmath8 we have not included this lorentzian in the fit . the fourth component , which we shall refer to as the hectohertz lorentzian , is fitted with a lorentzian with a characteristic frequency of the order of 100 hz . the hectohertz lorentzian is only absent in intervals 18 and 19 of 4u 172834 , at the highest inferred mass accretion rate . the lorentzian at 233 hz in interval 1 of 4u 0614 + 09 , at the lowest inferred mass accretion rate , can be identified as either the hectohertz lorentzian or the upper kilohertz qpo . the hectohertz lorentzian is broad at low interval numbers and becomes narrower at high interval numbers ( see table [ tbl.qvalues ] ) . this is a good example of how the characteristic frequency @xmath0 of a lorentzian is determined by @xmath9 and @xmath11 . in figure [ fig.hecto ] we plot @xmath0 , @xmath24 and @xmath11 versus the centroid frequency of the upper kilohertz qpo . as @xmath9 increases , @xmath11 falls such that @xmath0 is nearly constant . the fifth and sixth lorentzians were used to fit the pair of kilohertz qpos . this pair is present in intervals 1017 for 4u 172834 and 68 for 4u 0614 + 09 . in interval 1 of 4u 0614 + 09 there is a lorentzian present at @xmath30 hz . based on its frequency this lorentzian can be tentatively identified with the relation which pbk99 associate with the lower kilohertz qpo . however , this lorentzian is much broader and stronger than any of the `` normal '' lower kilohertz qpos in 4u 0614 + 09 and 4u 172834 ( see figures [ fig.rms_vs_upkilo ] and [ fig.q_vs_upkilo ] ) ; its broad shape is similar to the corresponding feature seen at similar frequencies in the low luminosity bursters ( bpk01 ) . in the other intervals ( except 18 and 19 for 4u 172834 ) there is only one kilohertz qpo present , which can be identified as the upper kilohertz qpo based on the correlations of its frequency with the frequency of the other features ( see fig . [ fig.all_vs_upkilo ] ) or interval number ( see ds01 ) . the lorentzian at 233 hz in interval 1 of 4u 0614 + 09 can be identified as either the hectohertz lorentzian or the upper kilohertz qpo . both kilohertz qpos are broader at lower frequencies ( see tables [ tbl.numax ] and [ tbl.qvalues ] ) . at frequencies below @xmath31 hz the upper kilohertz qpo becomes too broad to be classified as a real qpo ( @xmath27 ) . a way to compare the results of our multi lorentzian fit to the two atoll sources with previous fits of this function to other neutron star lmxbs and bhcs ( shirey 1998 , nowak 2000 , bpk01 ) is to plot the characteristic frequencies of the different lorentzians components versus the frequency of one lorentzian component which has already been identified for all these sources . there are three obvious candidates to plot everything against that have already been identified by wk99 and pbk99 ( see the introduction ) : the band limited noise component , the low frequency lorentzian and the lower kilohertz qpo . here we choose to present the results of comparing to the band limited noise component . if we use the characteristic frequency of the low frequency lorentzian instead , our conclusions do not change . if we use the characteristic frequency of the lower kilohertz qpo our points fall on the relations shown in bpk01 ; any differences between our conclusions and those of bpk01 are discussed below . the large advantage of using the band limited noise component is that in our data this component is present in all spectra , whereas the lower kilohertz qpo is only present in 11 ( perhaps 12 , see 3.4 ) and the low frequency lorentzian in 18 out of the 28 power spectra . also in the other sources the band limited noise component can almost always be clearly identified , where for the other peaks there can be ambiguity about which peak is which ( see , e.g. , homan et al . 2001 ) . a disadvantage of using the band limited noise component is that in both 4u 172834 and 4u 0614 + 09 at a certain point the frequency of the very low frequency lorentzian `` takes over '' from the frequency of the zero centered lorentzian ( see 3.1 ) . so , just plotting the characteristic frequencies of the different lorentzian components versus @xmath22 leads to a confusing picture where the correlations break down , i.e. at the intervals where both the zero centered lorentzian and the very low frequency lorentzian are present . this can be resolved by using @xmath23 as the characteristic frequency of the the band limited noise if the very low frequency lorentzian is present , and @xmath22 otherwise . we call the thus defined characteristic frequency of the band limited noise @xmath32 . although the physical connection between the zero centered lorentzian and the very low frequency lorentzian is uncertain ( see 3.1 ) , the smooth run of @xmath22 and @xmath23 with upper kilohertz qpo frequency visible in fig . [ fig.all_vs_upkilo ] indicates that for 4u 0614 + 09 and 4u 172834 @xmath32 could serve as horizontal coordinate to plot everything against as well as @xmath20 . we caution however , that due to the extensive data sets available for our two sources we are able to keep track of @xmath32 through the transformation from zero centered lorentzian to very low frequency lorentzian . with less data , confusion is likely . the correlation of @xmath32 with the other frequencies breaks down at interval 9 of 4u 0614 + 09 ( cf . however , as this is only one very extreme point this is not a real problem for our present purpose . we have excluded intervals 18 and 19 of 4u 172834 from the correlations , as for these intervals only one component , the very low frequency lorentzian , is present . but note that in these intervals also there is evidence of a break down in the correlations with @xmath32 ( 3.1 ) . in figure [ fig.comp ] we compare our results with those of nowak ( 2000 ; the bhcs gx 3394 , cyg x1 ) and bpk01 ( the low luminosity bursters 1e 17243045 , gs 182624 , slx 1735269 and the bhc xte j1118 + 480 ) by plotting all characteristic frequencies versus @xmath32 . the points for which we use @xmath22 as @xmath32 are black , those for which we use @xmath23 are grey . we have also included the results for cir x1 from shirey ( 1998 ) that were also used in pbk99 . that author also used a zero centered lorentzian to fit the broad band noise , and fitted two peaked features with lorentzians . the only difference with nowak ( 2000 ) and bpk01 is that the high frequency tail in shirey ( 1998 ) was fitted with a power law , whereas nowak ( 2000 ) and bpk01 use an additional lorentzian . the grey symbols in fig . [ fig.comp ] represent the results from shirey ( 1998 ) , nowak ( 2000 ) and bpk01 , the black symbols our results on 4u 172834 and 4u 0614 + 09 . for clarity figure [ fig.comp ] is split up in three panels . in the bottom panel we plot the characteristic frequency of the low frequency lorentzian ( @xmath26 ) , versus @xmath32 ( the wk99 relation ) . in this plot we have included the points from wk99 without the 4u 172834 and 4u 0614 + 09 points . however , note that wk99 use a broken power law instead of a zero centered lorentzian to fit the band limited noise . this will lead to systematic deviations in characteristic frequency ( see bpk01 ) . also note that for both 1e 17243045 and gs 182624 there are usually two low frequency peaks present ( see bpk01 ) ; a narrow peak and a broad component at a slightly higher frequency . if both these peaks are present we use the @xmath0 of the broad peak to compare with 4u 172834 and 4u 0614 + 09 based on the strong similarities of the broad peak with the low frequency lorentzian in these sources ( see also figure [ fig.cygx1_1e1724_4u0614 ] ) . most of our results on 4u 172834 and 4u 0614 + 09 coincide with the wk99 data . a few points lie slightly below the wk99 relation just as the points of gx 3394 , cyg x1 of nowak ( 2000 ) and 1e 17243045 , gs 182624 and slx 1735269 of bpk01 . this may be due to the systematic frequency deviations between a zero centered lorentzian and a broken power law ( see bpk01 ) . the xte j1118 + 480 points of bpk01 and the cir x1 points of shirey ( 1998 ) are well below the wk99 relation but may line up with each other . note that the qpo in cir x1 was previously identified as @xmath26 with the pbk99 relation ( see pbk99 ) . in the middle and top panels we plot the characteristic frequencies of the hectohertz lorentzian and both kilohertz qpos versus @xmath32 for 4u 172834 and 4u 0614 + 09 . we have fitted the correlations of the characteristic frequencies of the hectohertz lorentzian and the upper kilohertz qpo to @xmath32 with power laws in order to extrapolate these correlations to lower frequencies ( dashed lines in fig . [ fig.comp ] ) . for the lower kilohertz qpo the range in @xmath32 from these two sources alone is too small to draw strong conclusions from the extrapolation . for these fits we excluded interval 9 of 4u 0614 + 09 where the correlations between @xmath32 and the frequencies of the other lorentzians break down ( see fig . [ fig.all_vs_upkilo ] ) . we also exclude interval 1 of 4u 0614 + 09 from the fit ( although these points are shown in the figure ) as for this interval the identification of the hectohertz lorentzian and the lower and upper kilohertz qpo are uncertain ( see 3 ) . we use these power law fits only to get an indication for what the characteristic frequencies for the hectohertz lorentzian and the upper kilohertz qpo would be expected to be at low frequencies . the difference between the two panels is that in the middle panel we compare with the characteristic frequency of the lorentzian ( @xmath33 in bpk01 ) that pbk99 ( for the points from shirey 1998 ) , nowak ( 2000 ) and bpk01 identify with the lower kilohertz qpo and in the top panel with the lorentzian ( @xmath34 in bpk01 ) that nowak ( 2000 ) and bpk01 identify with the upper kilohertz qpo . we find that also when plotting versus @xmath32 the relation between @xmath33 and @xmath32 ( middle frame ) is similar to the relation of the lower kilohertz qpo with @xmath32 . also the @xmath30 hz lorentzian of interval 1 of 4u 0614 + 09 falls fairly close to the relation between @xmath33 and @xmath32 . however , more work needs to be done to connect both relations . concerning the top panel , nowak ( 2000 ) and bpk01 found that the bhc points ( gx 3394 , cyg x1 and xte j1118 + 480 ) as well as the slx 1735269 point followed the extension to low frequencies of the `` lower versus upper kilohertz qpo frequency '' of pbk99 . the points of 1e 17243045 and gs 182624 were above the relation . we find that the bhc points ( gx 3394 , cyg x1 and xte j1118 + 480 ) as well as the single point from slx 1735269 lie well below the extrapolated relation of @xmath32 with the characteristic frequency of the upper kilohertz qpo ; the points might be consistent with the extrapolation of the lower kilohertz qpo relation . we can therefore not identify the highest frequency lorentzian in these sources with the upper kilohertz qpo based on this relation . note again that the identification of @xmath32 might be wrong in sources for which only litle data has been analyzed ( see above ) . the points of gs 182624 are well above the relation and only the points of 1e 17243045 fall in the right range . however , the 1e 17243045 points also fall in the right range to be identified with the hectohertz lorentzian . the same conclusion can be drawn for the 233 hz lorentzian of interval 1 of 4u 0614 + 09 . interval 1 of 4u 0614 + 09 shows a @xmath32 similar to that of the cyg x1 observation from nowak ( 2000 ) and to that of observations of the low luminosity bursters 1e 17243045 and gs 182624 from bpk01 . in figure [ fig.cygx1_1e1724_4u0614 ] we plot the power spectra of 4u 0614 + 09 , 1e 17243045 and cyg x1 that have a @xmath32 of about 0.3 hz . for 1e 17243045 we use the same data that made up interval e of bpk01 and for cyg x1 we use the rxte observations on 1996 october 22 , the same as used by nowak ( 2000 ) . we constructed the power spectra in the same fashion as for 4u 172834 and 4u 0614 + 09 ( see 2 ) . the power spectra of 4u 0614 + 09 and 1e 17243045 are almost identical at both low and high frequencies . the power spectrum of cyg x1 is only similar to those of the neutron stars at low frequencies . the two high frequency peaks are both weaker relative to the low frequency peaks and about a factor of 4 lower in characteristic frequency than those in 4u 0614 + 09 and 1e 17243045 . an explanation for this might be that the characteristic frequencies of the high frequency peaks in the spectrum scale with the mass of the central object and that the low frequency peaks are independent of this mass . the use of exactly the same fit function for both 4u 172834 and 4u 0614 + 09 shows the similar timing behaviour of these two atoll sources ( see 3 ) . especially the frequencies and the @xmath15 values are almost identical . this similarity is quite remarkable , as the x ray luminosity is about a factor 5 larger for 4u 172834 ( ford et al . 2000 ) . note that the luminosities for 4u 172834 and 4u 0614 + 09 in ford et al . ( 2000 ) are based on distance to these sources calculated from type i x ray bursts . the luminosity is thought to reach the eddington limit in the so called radius expansion bursts ( lewin , van paradijs & taam 1993 ) . for 4u 172834 a radius expansion burst was used ( foster et al . 1986 ) but for 4u 0614 + 09 no radius expansion bursts have been observed , therefore only an upper limit on the distance was derived by assuming that the flux is less than the eddington limit ( brandt et al . . so the luminosity in ford et al . ( 2000 ) for 4u 0614 + 09 is also an upper limit . the fractional rms in 4u 172834 is systematically lower than that in 4u 0614 + 09 , which may be related to its higher luminosity . note that the data are inconsistent with a model in which the kilohertz qpo amplitudes are the same and the higher luminosity of 4u 172834 is caused by an additional source of x rays unrelated to the kilohertz qpos ; then the rms values in 4u 172834 would be expected to be a factor of 5 lower , while they are only lower by about a factor of 1.3 to 2 . another stunning difference between 4u 172834 and 4u 0614 + 09 is that in the whole rxte data set on 4u 0614 + 09 obtained until now ( @xmath35 ks ) no type i x ray bursts have been observed , whereas 4u 172834 is a well known burster that shows type i x ray bursts in all parts of the color color diagram ( van straaten et al . 2001 , franco 2001 ) . note that although no type i x ray bursts have been observed for 4u 0614 + 09 with rxte , two x ray bursts from this source have been previously detected . one with the oso8 satellite ( swank et al . 1978 ) and one with the granat satellite ( brandt et al . 1992 ) . the hectohertz lorentzian has been identified in several other neutron star lmxbs ( sax j1808.4 - 3658 , 4u 170544 ; wijnands & van der klis 1998 ) . sunyaev & revnivtsev ( 2000 ) use a sample of 9 bhcs and 9 neutron star lmxbs ( including sax j1808.4 - 3658 , 4u 170544 , 4u 0614 + 09 and 4u 172834 ) to show that neutron star systems show significant broad band noise above several hundred hz where bhc do not . the hectohertz lorentzian is one of the contributers to this excess of power in sax j1808.4 - 3658 , 4u 170544 , 4u 0614 + 09 and 4u 172834 and could be in the other sampled neutron star lmxbs . the hectohertz lorentzians could be related to the @xmath36 hz qpo in the bhc grs 1915 + 105 ( morgan , remillard & greiner 1997 ) and the @xmath37 hz qpo in the bhc gro j165540 ( remillard et al . 1999 ) . these oscillations also have stable frequencies and fall in the same frequency range as the hectohertz lorentzians we observe . however , these qpos are much narrower ( @xmath38 for the @xmath39 hz qpo and @xmath40 for the @xmath36 hz qpo ) and have much weaker rms fractional amplitudes ( @xmath411 % ) . models to explain these stable qpos in the bhcs all invoke strong gravity near the central black hole ( see for a review cui , chen & zhang 2000 ) . recently , fragile , mathews & wilson ( 2001 ) made a tentative identification of the 9 hz qpo in the bhc gro j165540 ( remillard et al . 1999 ) with the orbital frequency at the bardeen petterson transition radius ( e.g. bardeen & petterson 1975 ) . based on the similarities between the hectohertz lorentzian in the neutron star lmxbs and the 9 hz qpo in gro j165540 they also suggest this identification for the hectohertz lorentzian . for 4u 0614 + 09 and 4u 172834 we can confirm that in indeed several properties of the hectohertz lorentzian are quite similar to those of the 9 hz qpo in gro j165540 . as in the case of the 9 hz qpo , the hectohertz lorentzian always has @xmath42 and for both features the characteristic frequency is nearly constant whilst the frequencies of other components change . as noted in 1 , if the two highest frequency qpos in the bhcs could be identified with the lower and upper kilohertz qpo in the atoll sources ( as suggested by respectively pbk99 and nowak 2000 ) , one would predict that in 4u 172834 and 4u 0614 + 09 the two high frequency lorentzians should be absent from the broad band noise if the two kilohertz qpos are present . our results show a more complicated picture ; when the two kilohertz qpos are clearly present , then in addition to the zero centered lorentzian and the low frequency lorentzian , two more lorentzian components are present , the very low frequency lorentzian and the hectohertz lorentzian . these lorentzians however have quite a special character . the characteristic frequency of the very low frequency lorentzian is strongly connected with the characteristic frequency of the zero centered lorentzian and might be part of the bln ( see 3.1 ) , which is not clear for the two highest frequency lorentzians describing the broad band noise in the bhcs . the hectohertz lorentzian is almost constant in frequency with a frequency of about 150 hz . the highest frequency lorentzian describing the broad band noise in the bhcs might also be constant in frequency at about 30 hz , but there is currently not enough data available for the bhcs to confirm this . the factor of @xmath43 difference in frequency might then be caused by a scaling with the mass of the compact object which is about a factor of 5 lower for the neutron stars . the power spectra of the low luminosity bursters and bhcs studied in nowak ( 2000 ) and bpk01 showed four lorentzian components three of which in practice are zero centered . the atoll sources 4u 172834 and 4u 0614 + 09 also show four lorentzian components at low inferred mass accretion rates ( intervals 18 of 4u 172834 and 14 of 4u 0614 + 09 ) . so can we identify all four peaks in these intervals of 4u 172834 and 4u 0614 + 09 with the four peaks in the bhcs and the low luminosity bursters ? the first two lorentzian components in the low luminosity bursters , bhcs and the atoll sources 4u 172834 and 4u 0614 + 09 may have the same physical origin as they also show a similar relation between their respective characteristic frequencies ( the wk99 relation , see fig . [ fig.comp ] , bottom panel ) . interval 1 of 4u 0614 + 09 is almost identical to that of the low luminosity bursters . however , because interval 1 of 4u 0614 + 09 is an extreme point in the current dataset available for 4u 0614 + 09 , its two highest frequency lorentzians can not be linked directly with the two highest frequency lorentzians of intervals 24 . more observations of 4u 0614 + 09 with a @xmath32 between 0.3 and 1.4 hz are needed to connect these points . the third lorentzian of the bhcs and low luminosity bursters however , behaves completely differently from the third lorentzian in 4u 172834 and 4u 0614 + 09 ( except for interval 1 ) . where the characteristic frequency of the third lorentzian in 4u 172834 and 4u 0614 + 09 ( i.e. the hectohertz lorentzian ) is almost constant at about 150 hz , the third lorentzian in the bhcs and low luminosity bursters varies over several orders of magnitude ( fig . [ fig.comp ] , middle panel ) . instead , the pbk99 relation as well as the relation between the characteristic frequency of the third lorentzian in the bhcs and low luminosity bursters with @xmath32 suggest the identification of this lorentzian with the lower kilohertz qpo that is present in intervals 1017 of 4u 172834 and intervals 68 of 4u 0614 + 09 . in the middle panel of fig . [ fig.comp ] we have illustrated this with a dotted line that indicates a power law fit to the characteristic frequencies of the lower kilohertz qpo of 4u 172834/4u 0614 + 09 and the third lorentzian of the low luminosity bursters to @xmath32 . based on the similarity between the relation between the characteristic frequencies of the third and the fourth lorentzians in the bhcs gx 3394 and cyg x1 and the relation between the characteristic frequencies of the lower and upper kilohertz qpos , nowak ( 2000 ) suggested the identification of the fourth lorentzian with the upper kilohertz qpo . indeed the upper kilohertz qpo is also the fourth peak in the intervals at lower inferred mass accretion rates , where 4u 172834 and 4u 0614 + 09 mimic the bhcs most . however , the relation between the characteristic frequency of the fourth lorentzian and @xmath32 does not confirm this identification ( see 4 ) . for the low luminosity bursters 1e 17243045 and gs 182624 studied in bpk01 the identification of the fourth peak with the upper kilohertz qpo is even harder ( see 4 ) . but keep in mind that due to the extensive data sets available for our two sources we are able to keep track of @xmath32 through the transformation from zero centered lorentzian to very low frequency lorentzian . with less data , confusion might occur . * the use of exactly the same fit function for 4u 0614 + 09 and 4u 172834 shows that these two atoll sources show a remarkably similar timing behaviour at luminosities that differ by a factor of 5 . * interval 1 of 4u 0614 + 09 shows a remarkably similar power spectrum to that of the low luminosity bursters . unfortunately , this is an extreme point in the current dataset available for 4u 0614 + 09 and therefore the two highest frequency lorentzians can not be linked directly with the high frequency lorentzians of the other intervals . more observations of 4u 0614 + 09 with a @xmath32 between 0.3 and 1.4 hz are needed to complete the picture . * when the two kilohertz qpos are clearly present , the low frequency part of the power spectrum is too complicated to draw immediate conclusions from the nature of the components detected in any one power spectrum . the relation of characteristic frequency of respectively the two high frequency lorentzians with the characteristic frequency of the band limited noise ( @xmath32 ) hints towards the identification of the second highest frequency lorentzian in the bhcs with the lower kilohertz qpo ( as suggested by pbk99 ) but can not confirm the identification of the highest frequency lorentzian with the upper kilohertz qpo ( as suggested by nowak 2000 ) . * when using @xmath0 instead of the centroid frequency , the recently discovered hectohertz lorentzians are almost stable in frequency . * all lorentzian components in 4u 0614 + 09 and 4u 172834 become broader as their frequency decreases . this might explain why many lorentzians in other sources are zero centered when present at low frequencies . * for 4u 0614 + 09 the use of the multi lorentzian fit function clearly shows the presence of the band limited noise qpo , where in the broken power law description , the very low frequency lorentzian is usually not statistically required . the similarity between the very low frequency lorentzian in 4u 0614 + 09 and 4u 172834 , where the very low frequency lorentzian is statistically required in both descriptions , confirms the identification of the very low frequency lorentzian in 4u 0614 + 09 . this work was supported by nwo spinoza grant 080 to e.p.j . van den heuvel , by the netherlands organization for scientific research ( nwo ) , and by the netherlands research school for astronomy ( nova ) . this research has made use of data obtained through the high energy astrophysics science archive research center online service , provided by the nasa / goddard space flight center . we would like to thank bob shirey for providing us with a table of the cir x1 data . + 1 & [email protected] & & [email protected] & 178@xmath4435 & & 399@xmath4413 + 2 & [email protected] & & [email protected] & 173@xmath4422 & & 416@xmath4412 + 3 & [email protected] & & [email protected] & 203@xmath4428 & & 497@xmath4414 + 4 & [email protected] & & [email protected] & [email protected] & & [email protected] + 5 & [email protected] & & [email protected] & [email protected] & & [email protected] + 6 & [email protected] & & [email protected] & 198@xmath4413 & & [email protected] + 7 & [email protected] & & [email protected] & 175@xmath4414 & & [email protected] + 8 & [email protected] & & [email protected] & [email protected] & & [email protected] + 9 & [email protected] & [email protected] & [email protected] & [email protected] & & [email protected] + 10 & [email protected] & [email protected] & [email protected] & [email protected] & 513@xmath4418 & [email protected] + 11 & [email protected] & [email protected] & [email protected] & [email protected] & 561@xmath4411 & [email protected] + 12 & [email protected] & [email protected] & [email protected] & [email protected] & 604@xmath4414 & [email protected] + 13 & [email protected] & [email protected] & & [email protected] & 680@xmath4410 & [email protected] + 14 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 15 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 16 & [email protected] & [email protected] & & [email protected] & [email protected] & 1134@xmath4412 + 17 & 12.9@xmath45 & [email protected] & & 224@xmath4427 & [email protected] & 1161@xmath4416 + 18 & & [email protected]@xmath46 & & & & + 19 & & [email protected]@xmath46 & & & & + + 1 & [email protected] & & [email protected] & & [email protected]@xmath46 & [email protected]@xmath46 + 2 & [email protected] & & [email protected] & 157@xmath4412 & & [email protected] + 3 & [email protected] & & [email protected] & 153@xmath4412 & & 481@xmath4415 + 4 & [email protected] & & [email protected] & [email protected] & & [email protected] + 5 & 7.0@xmath47 & [email protected] & [email protected] & [email protected] & & [email protected] + 6 & [email protected] & [email protected] & 44.4@xmath48 & [email protected] & 517@xmath4412 & [email protected] + 7 & 11.4@xmath49 & [email protected] & & [email protected] & [email protected] & [email protected] + 8 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 9 & & [email protected]@xmath46 & & [email protected] & & [email protected] + + 1 & & [email protected] & [email protected] & & 1.66@xmath50 + 2 & & [email protected] & [email protected] & & [email protected] + 3 & & [email protected] & [email protected] & & [email protected] + 4 & & [email protected] & [email protected] & & [email protected] + 5 & & [email protected] & [email protected] & & [email protected] + 6 & & [email protected] & [email protected] & & [email protected] + 7 & & [email protected] & [email protected] & & [email protected] + 8 & & [email protected] & [email protected] & & [email protected] + 9 & 0.90@xmath51 & [email protected] & [email protected] & & [email protected] + 10 & [email protected] & [email protected] & [email protected] & 5.4@xmath52 & [email protected] + 11 & 1.64@xmath53 & 3.8@xmath54 & [email protected] & 6.5@xmath55 & [email protected] + 12 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 13 & [email protected] & & [email protected] & [email protected] & [email protected] + 14 & [email protected] & & [email protected] & [email protected] & [email protected] + 15 & [email protected] & & [email protected] & [email protected] & [email protected] + 16 & [email protected] & & [email protected] & [email protected] & [email protected] + 17 & [email protected] & & [email protected] & [email protected] & 10.7@xmath56 + 18 & [email protected]@xmath46 & & & & + 19 & [email protected]@xmath46 & & & & + + 1 & & [email protected] & & [email protected]@xmath46 & [email protected]@xmath46 + 2 & & [email protected] & [email protected] & & [email protected] + 3 & & [email protected] & [email protected] & & [email protected] + 4 & & [email protected] & [email protected] & & [email protected] + 5 & [email protected] & 6.2@xmath57 & [email protected] & & [email protected] + 6 & [email protected] & 6@xmath58 & [email protected] & 5.9@xmath59 & [email protected] + 7 & [email protected] & & [email protected] & 13@xmath60 & [email protected] + 8 & [email protected] & & [email protected] & [email protected] & [email protected] + 9 & [email protected]@xmath46 & & [email protected] & & [email protected] + + 1 & [email protected] & & [email protected] & [email protected] & & [email protected] + 2 & [email protected] & & [email protected] & [email protected] & & [email protected] + 3 & [email protected] & & [email protected] & [email protected] & & [email protected] + 4 & [email protected] & & [email protected] & [email protected] & & [email protected] + 5 & [email protected] & & [email protected] & [email protected] & & [email protected] + 6 & [email protected] & & [email protected] & [email protected] & & [email protected] + 7 & [email protected] & & 5.8@xmath61 & [email protected] & & [email protected] + 8 & [email protected] & & [email protected] & [email protected] & & [email protected] + 9 & [email protected] & 6.2@xmath62 & [email protected] & 12.23@xmath63 & & [email protected] + 10 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 11 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 12 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + 13 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 14 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 15 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 16 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 17 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 18 & & [email protected]@xmath46 & & & & + 19 & & [email protected]@xmath46 & & & & + + 1 & [email protected] & & [email protected] & & [email protected]@xmath46 & [email protected]@xmath46 + 2 & [email protected] & & [email protected] & [email protected] & & [email protected] + 3 & [email protected] & & [email protected] & [email protected] & & [email protected] + 4 & [email protected] & & [email protected] & [email protected] & & [email protected] + 5 & 7.8@xmath64 & [email protected] & 3.57@xmath65 & [email protected] & & [email protected] + 6 & 8.5@xmath66 & [email protected] & 3.3@xmath67 & [email protected] & [email protected] & [email protected] + 7 & 7.8@xmath68 & [email protected] & & [email protected] & 9.45@xmath69 & [email protected] + 8 & [email protected] & [email protected] & & [email protected] & [email protected] & [email protected] + 9 & & [email protected]@xmath46 & & [email protected] & & [email protected] +

we consider in what sense quantum tunnelling is associated with non - classical probabilistic behaviour . we use the wigner function quasi - probability description of quantum states . we give a definition of tunnelling that allows us to say whether in a given scenario there is tunnelling or not . we prove that this can only happen if either the wigner function is negative and/or a certain measurement operator which we call the tunnelling rate operator has a negative wigner function .

1607.01764

quantum systems can ` tunnel ' through barriers , a wave - like behaviour that is a prototypical of non - classical behaviour . tunnelling plays a crucial role in a range of systems , including the theory of radioactivity , in which tunnelling is used to explain how radioactivity emerges from a nucleus despite the surrounding coulomb barrier @xcite ; and more recent applications include adiabatic quantum computing and annealing @xcite , and quantum electronics in nano - devices @xcite . we here try to clarify our understanding of in what sense tunnelling is non - classical , in particular to what extent it is associated with non - classical probabilistic behaviour . this is in order to understand tunnelling better , and we hope that this understanding will help in clarifying to what extent tunnelling allows for non - classical information processing . for example , the crucial component of information processing is the mathematical structure of the probabilistic system , rather than the physical parameters involved . we find that tunnelling is indeed associated with non - classical probabilistic behaviour . in particular , we show that the wigner function of the state , a commonly - used and powerful phase space quasi - probability representation of quantum states , and/or a certain operator , which we termed in this article as tunnelling rate operator , has to contain negative values at some phase space points to be able to tunnel . here we list two examples : 1 . a gaussian wave packet inbound on a square potential has a positive wigner function @xcite . however , once it hits the potential , it loses its gaussian nature and this forces it to have a negative wigner function while part of it tunnels ; 2 . a ground state of a simple harmonic oscillator , which has a positive wigner function , has some probability of being found inside the binding potential region . the tunnelling rate operator we will define in this article contains negative values in this case . to be able to derive our main statement we needed to define clearly what tunnelling is and this definition is also one of the contributions of this article . finally we note that wigner function as a real - vector representation of quantum states , fits into the generalised probabilistic framework , and therefore allows for extension to post - quantum theories . to our knowledge , this has not been realised before . lastly , we discuss the possibility of studying tunnelling in post - quantum theories . we proceed as follows . firstly we give a brief technical introduction to tunnelling and wigner functions . then we give new results , our definition of tunnelling and our main theorem . we discuss the interpretation of the theorem . finally we show that phase space quantum mechanics could be fitted into the generalised probabilistic framework , and demonstrate possible road maps into studying of the phenomenon of tunnelling in post - quantum theories . we now give a technical introduction . * quantum tunnelling * refers to the phenomenon where a quantum system with insufficient energy penetrates and passes through a potential barrier , which defies the laws of classical mechanics @xcite . it is usually demonstrated mathematically by solving the one spatial - dimension schrdinger equation with a rectangular potential barrier , such that tunnelling @xcite@xcite@xcite : [ thm : sd ] * standard definition of tunnelling . * for a system in a rectangular potential barrier with the form @xmath0 an energy eigenstate with definite energy @xmath1 is a tunnelling state iff @xmath2 and the probability of finding the state in the region @xmath3 is non - zero . similar behaviours had been studied substantially in quantum systems in other potentials , such as series of rectangular barriers @xcite and double wells @xcite , where there is finite probability of locating an energy eigenstate in classically forbidden region . however , it is difficult to conceive an analogous definition for quantum systems in a superposition of energy eigenstates or in a more complicated potential , because it is less clear of what the corresponding classically forbidden regions are for both cases . in order to provide a more general definition of quantum tunnelling and to understand this phenomenon in terms of non - classical probabilistic behaviour , we have chosen the * phase space formulation of quantum mechanics * as the fundamental framework . in this formulation , the state of a quantum system is described by a quasi - probability distribution , and observables are replaced by ordinary c - number functions in phase space . mathematically , a quantum state described by a vector in the hilbert space formulation @xmath4 can be transformed to a real function _ wigner function _ @xmath5 in the phase space formulation as @xmath6 such function satisfies the normalisation condition @xmath7 . however , a wigner function , in general , is not everywhere positive , and therefore can not be considered as a legitimate joint probability distribution in phase space . it was demonstrated that a wigner function of a pure continuous - variable state is positive if and only if the state is gaussian , which is known as _ hudson s theorem _ @xcite . a more general approach to describe states and operators @xmath8 as phase space functions @xmath9 is via _ weyl transformation _ : @xmath10 and the inverse transformation , _ wigner transformation _ , is given by the formula : @xmath11}\,d\alpha\,d\beta\,dx\,dp.\end{gathered}\ ] ] it is clear , from and , the wigner function of a pure state is the weyl transformation of a pure density operator @xmath12 . with the fact that a general density operator represented by @xmath13 , any mixed state in the phase space representation is described by a convex sum of pure wigner functions @xmath14 , where @xmath15 are the pure wigner functions and @xmath16 . another point to note about the transition between hilbert space formulation to phase space formulation of quantum mechanics is the preservation of non - commutative nature of operators via the introduction of _ star product _ @xmath17 , @xmath18,\ ] ] such that the mapping can be written as @xmath19 where @xmath20 is the weyl transformation of the corresponding operator @xmath21 . in the phase space formulation , the probability of a measurement outcome @xmath22 corresponding to @xmath23 ( where @xmath24 ) for a quantum state @xmath4 and a corresponding wigner function @xmath25 is @xmath26 where @xmath27 is the wigner function of the state @xmath28 . in other words , this probability is simply the integral of the product of wigner functions corresponding to the measurement outcome and the quantum state itself up to a normalisation constant @xmath29 . this result is analogous to its classical counterpart described by a distribution function @xmath30 , where the probability of a dynamic variable @xmath31 is given by @xmath32 f(x , p)\,dx\,dp . , \ ] ] it should be clear that the probability corresponding to a measurement of a dynamic variable @xmath31 is the inner product of the state of the system @xmath30 , and a function that describes the measurement of probability of certain dynamic variable , which is known as an _ effect _ @xmath33 . for instance , in the examples above , the effects for calculating the probability @xmath22 , @xmath34 are @xmath35 $ ] and @xmath36 for classical and quantum cases respectively . from the brief discussions above , it is shown that the mathematical structures of classical and quantum theories in phase space are similar to one another . such similarities are anticipated by the fact that classical probability theories and finite - dimensional quantum theories can be described by a unified framework known as the * generalised probabilistic theories * ( gpts ) @xcite . a finite - dimensional probabilistic theory under the framework of gpts has three major components : + + * 1 . preparation . * a state of a finite - dimensional system is represented by a real vector from a _ state space _ , which is defined by the convex set of extremal states , or _ pure states _ , and the _ null state _ , within a finite - dimensional vector space . such mathematical object completely determines the characteristics of the state of a physical system . + + * 2 . transformation*. the evolution of any state vector is represented by a linear map that maps any state vector into another within the state space . + + * 3 . measurement*. the probability of a particular measurement outcome for any state is described by the inner product between the state vector and an _ effect vector _ corresponding to such measurement . the inner product calculated must be ranging between 0 and 1 as a valid probability @xcite . the set of valid effects is known as the _ effect space _ , which is constructed as the convex sum of the set of _ pure effects _ and _ null effect_. under the gpt framework , it is possible to construct probabilistic theories other than classical and quantum theories by varying its state space , the effect space and the set of allowed transformations , hence providing a way of generalising existing theories into _ post - quantum theories_. it was shown via these constructed post - quantum theories that certain quantum phenomena , such as non - unique decomposition of mixed states into pure states , and no - cloning theorem @xcite , which were thought to be novel to quantum systems , were actually generic properties of generalised probabilistic theories . however , these studies are limited to finite - dimensional systems and therefore can not be immediately applied to the study of quantum tunnelling , which is a wave - mechanical quantum phenomenon . therefore if it is possible to rewrite infinite - dimensional quantum mechanics in the framework of gpts , we could determine whether tunnelling is unique to quantum theories , and examine such phenomenon in post - quantum theories . it is clear from the previous section that is too restrictive to describe the entire class of tunnelling behaviour . a general quantitative definition of tunnelling is therefore required to be : 1 . mathematically well - defined ; 2 . able to recover as a special case ; 3 . able to reflect the non - classicality of the behaviour ; 4 . able to supply a quantitative criteria of tunnelling for systems with general potentials and states without definite energy . in order to construct such a definition of tunnelling , we starts with the law of conservation of energy for a classical particle in a potential @xmath37 : 1 . classical kinetic energy @xmath38 is positive , due to the reality of momentum @xmath39 . by conservation of energy , for a particle with energy @xmath40 , @xmath41 this implies that a particle with energy @xmath40 is not allowed in the region @xmath42 classically , where @xmath43 is a real variable representing position . denote this region as @xmath44 , as the classically forbidden region for states with energy @xmath40 . 2 . for two energies @xmath45 and @xmath46 , such that @xmath47 , their respective classically forbidden regions satisfy @xmath48 , since @xmath49 3 . classically , only a particle with energy @xmath50 is allowed to be in @xmath44 , since : 1 . for @xmath51 , @xmath44 is classically forbidden region ; 2 . for @xmath52 , by ( 2 ) , @xmath53 is also classically forbidden . with the concluding statement , and with the general principle that tunnelling is a phenomenon that violates this classical constraint , we formulate the definition of tunnelling to be : [ def : qm - gd ] * general definition of tunnelling . * for a state in a potential given by @xmath37 , it is tunnelling if and only if there exists some energy @xmath40 , such that the probability of locating the state in region where @xmath54 is greater than that of measuring the state to have energy @xmath55 , or mathematically , @xmath56 notice that this definition does not require the state in question to be quantum or classical , and therefore allows the definition to be applied as a condition of general phenomenon on any physical systems , provided they can be described by energy and position as physical parameters . another advantage of using such general definition is its capabilities of extending to other non - classical behaviour , such as reflection over barrier as discussed in appendix a. it is also shown , in appendix b , that the conventional definition could be recovered as a special case of such general definition . the necessary and sufficient condition for tunnelling , or equivalently the general definition , are based on cumulative probabilities @xmath57 and @xmath58 . in the phase space formulation , they are given by the inner product of the functions representing the difference of the effects corresponding to the two probabilities , and the wigner function representing the state . the general definition of tunnelling applied in this picture implies : [ thm : pstns ] * necessary and sufficient conditions of tunnelling in phase space . * a state represented by a distribution function @xmath30 in phase space is tunnelling if and only if there exists some @xmath40 : @xmath59(x , p ) f(x , p)\,dx\,dp > 0,\end{gathered}\ ] ] where @xmath60 and @xmath61 are the effects corresponding to the probability measurement of outcome @xmath62 and @xmath57 respectively . this statement is basically a rewritten form of the general definition . as by the definition of the effects @xmath33 as stated in the theorem , @xmath63 by , the necessary and sufficient condition of tunnelling is therefore @xmath64 : @xmath65(x , p ) f(x , p)\,dx\,dp > 0,\end{gathered}\ ] ] which is the theorem to be proved . the theorem therefore shows that tunnelling is related to the distribution function and @xmath66 , and the latter is denoted as the _ tunnelling rate operator _ at @xmath40 . two important consequences of are : [ col : pstn ] a state is non - tunnelling if , for every energy @xmath40 , the tunnelling rate operator , and the distribution function are non - negative everywhere in the phase space . an equivalent way of expressing the necessary and sufficient conditions of tunnelling is to specify the necessary and sufficient conditions of a non - tunnelling state , which is for all @xmath40 : @xmath67 f(x , p ) \,dx \,dp & \leq 0 \\ \iint \left[\mathcal{e}_{e > e^ * } - \mathcal{e}_{\{x|v(x)>e^*\}}\right ] f(x , p ) \,dx \,dp & \geq 0.\end{aligned}\ ] ] it is clear that if both @xmath68 and @xmath30 is non - negative over all of phase space , then this condition will hold . notice here this theorem only provides a sufficient but not necessary condition for the case of non - tunnelling , as it is not the case where @xmath69 and @xmath30 must be positive for the non - tunnelling condition to be satisfied . another way of rephrasing the corollary , or indeed as a direct consequence of the main theorem , is : [ col : pst ] if a state is tunnelling , then either tunnelling rate operator at some @xmath40 , or the distribution function , contain negativities . the result can be directly inferred from , by a simple reshuffling of the inequality that defines tunnelling : for some energy @xmath40 , @xmath70(x , p ) f(x , p)\,dx\,dp > 0 , \\ [ \mathcal{e}_{e > e^ * } -\mathcal{e}_{\{x|v(x ) > e^*\}}](x , p ) f(x , p)\,dx\,dp < 0.\end{aligned}\ ] ] since it is impossible for the statement above to hold if any of the two functions are positive over the entire phase space , then either @xmath68 or @xmath30 must contain negativities . the last statement demonstrates the main claim of this paper : under a phase space framework , tunnelling implies negativities in the distribution representing the state , and/or the tunnelling rate operator at some energy levels . in the case of quantum systems , tunnelling implies negativities in the wigner function of the state and/or the tunnelling rate operator . the results of the previous section could immediately be applied to the study of generic behaviours of classical and quantum theories . one important result is that classical phase space mechanics do not allow tunnelling , because as shown in appendix c , both the phase space distributions representing a state and the tunnelling rate operator does not contain non - negativities , and by , classical states can not tunnel . however , in quantum theories , the non - negativity conditions of the functions are relaxed . this can be attributed to the following two elements of quantum phase space theory : + + * 1 . wigner function as quasi - probability distribution . * a quantum state in phase space is represented by wigner function , which generally contains negative values in general . + + * 2 . deformation of effect * @xmath71 generally speaking , the effect @xmath72 is identical for both classical and quantum case , while the effect @xmath61 is altered in the quantum case . both phenomena can be explained by the alteration in the eigenvalue equations for both position and hamiltonian operator represented in the phase space picture , as will be shown here . generally , for any dynamical variables @xmath73 , for both classical and quantum systems , their eigenstate with respect to the corresponding algebra has definite value @xmath74 if one conducts a measurement of @xmath75 on such system . consider first the dynamical variable @xmath43 as the position . the eigenvalue equation in classical systems with eigenstate @xmath76 and eigenvalue @xmath77 is @xmath78 and this equation has an obvious solution of @xmath79 , where @xmath80 is an arbitrary continuous positive real function which is normalised to unity @xmath81 . if the state in question is of such a form , the measurement of position on such state must give @xmath82 . at the same time , such function could be used as an effect to determine what is the probability of a certain state to have a position measurement of @xmath82 , as the inner product of this delta function against any state would give the proportion of the state with @xmath82 , i.e. the corresponding probability . therefore , classically , @xmath83 . in the case of quantum systems , the product of any operators replaced by star - product of functions as a result of deformation quantisation . therefore , the eigenvalue problem is mapped to an equation with position eigenstate @xmath84 and eigenvalue @xmath77 : @xmath85 one simple way of solving the problem is to switch back to the hilbert space picture , where the equivalent problem is @xmath86 which has a well - known solution of @xmath87 . the wigner function for such a state is given by @xmath88 which is again of the form of a dirac delta function in position space after integrating away the momentum dependence . by the duality of effect and state in quantum theory , the effect for @xmath82 is therefore @xmath89 . therefore , since both the classical and quantum position eigenstates are of the form of a dirac delta function @xmath90 , the effect corresponding to measurement in position space are therefore identical , which implies @xmath72 is of the same form for both classical and quantum calculations . however , this is not the case for @xmath61 . the energy eigenstate @xmath91 with energy @xmath1 for a classical system satisfies @xmath92 which gives @xmath93 $ ] . the quantum energy eigenvalue equation in phase space for an energy eigenstate @xmath94 and energy @xmath1 , however , is altered to be @xmath95 here , the energy eigenstate as a weyl map of @xmath96 is no longer of the form @xmath97 $ ] , as the introduction of star product implies the functional dependence of energy eigenstate is no longer purely on the functional form of the dynamical variable . this can be seen by the bopp shift representation of star product , where the functional dependence of the energy eigenstate depends also on the position and momentum derivatives of the hamiltonian function . therefore , the energy eigenstate @xmath61 is not identical to its classical counterpart . with the energy effect @xmath98 , and the fact that wigner functions are generally not completely positive , @xmath99 generally contains negativity . therefore , the function @xmath100 generally contains negativity , given that @xmath72 is non - negative . this demonstrates that how this operator can violate the classical case by containing negativities . a special class of states in quantum mechanics is gaussian states , which has non - negative wigner functions in phase space representation . by , a statement for gaussian states could be made as : [ cor : ptg ] a gaussian state can tunnel only if the function @xmath100 contains negativities for some @xmath40 . by , if a state is tunnelling , then either @xmath100 or @xmath5 contains negativities for some energy @xmath40 . by hudson s theorem @xcite , the wigner function representing gaussian state is non - negative over all of phase space . therefore , if a gaussian state is tunnelling , then @xmath100 must contain negativities for some energy @xmath40 . this corollary serves two purposes . first of all , since a gaussian state is represented by a positive wigner function over phase space , it is often considered to be a valid joint - probability distribution and as the ` least non - classical ' state @xcite . this corollary serves as a reminder that despite it is true that negativity in wigner functions as distribution functions is a novel feature in phase space quantum theory , a completely positive wigner function can still exhibit non - classical behaviours , which , in this specific case , is due to the deformation in effects as discussed in the last section . it has been demonstrated that the sub - theory of gaussian quantum mechanics can be constructed by imposing certain epistemic restriction on classical phase space mechanics @xcite , in which only gaussian states , measurements and operations are considered . this reinforces the conclusion that the tunnelling rate operator corresponding to the measurements in question is the culprit for a gaussian state to exhibit tunnelling , for otherwise , the system would simply be classical and can not tunnel . secondly , the discussion of tunnelling in gaussian states allows for simplified examples of the application of the previous results , as the tunnelling rate operator @xmath100 determines the tunnelling behaviour of these gaussian states . consider the following two examples of gaussian states : + + * 1 . quantum tunnelling of wave packets . * a particle in quantum mechanics is often described by a wave packet , which has a localised waveform as a superposition of energy eigenstates @xcite . in particular , gaussian wave packets are often used in the study of quantum tunnelling and general quantum mechanics problem @xcite . by studying gaussian wave packets in rectangular potential barrier , it also introduces dynamical aspects into the problem of tunnelling , as the position probability distribution of a quantum system now changes over time due to the relative phase differences between the different components of its energy eigenstates . a gaussian wave packet that centres at position @xmath82 and momentum @xmath101 , with uncertainty in position as @xmath102 , has the form @xmath103{$-$}}{\frac{(x - x_0)^2}{4\sigma_x^2}}\right ] } \exp{\left(\frac{ip_0 x}{\hbar}\right)},\ ] ] which has a gaussian distribution over position space , @xmath104{$-$}}{\frac{(x - x_0)^2}{2\sigma_x^2}}\right]}.\ ] ] , title="fig : " ] , title="fig : " ] it is possible to solve for the dynamics of the gaussian wave packet as the superposition of time - dependent energy eigenstates of the rectangular potential barrier , or via numerical simulation of schrdinger equation . an example of such a simulation is shown in , where a wave packet , with initial gaussian shape and average energy lower than the potential height , passes through the barrier , and the gaussian nature of the wave packet is destroyed . ( red ) and @xmath105 ( orange ) for a gaussian wave packet tunnelling through a barrier as a function of time @xmath106 . at times that @xmath107 , the state is considered to be tunnelling . [ fig : gwpt ] ] the fact that there is no definite energy for such gaussian wave packet creates difficulties in applying the standard definition of tunnelling to this case . however , using our general definition , it could be demonstrated that such a state is indeed tunnelling at certain times during the propagation of the wave packet , as @xmath108 as shown in . the detailed calculations could be found in appendix d. + + * 2 . simple harmonic oscillator . * a simple harmonic oscillator has a potential of the form @xmath109 where @xmath110 is the mass of the particle , and @xmath74 is the _ angular frequency _ of the oscillator . a simple harmonic oscillator is one of the most well - studied potentials in physics , and has many nice features that are exploited in both classical and quantum theories . a particle as a classical simple harmonic oscillator will carry out sinusoidal oscillatory motion with @xcite @xmath111 where the amplitude @xmath112 and the phase @xmath113 depends on the initial conditions of the particle . by conservation of energy , a particle with energy @xmath1 is only allowed to be in the region @xmath114 , for otherwise the particle would have negative kinetic energy . in the quantum case , in order to solve for the energy eigenstates of the hamiltonian @xmath115 one can use the _ ladder operator _ method by defining the _ creation operator _ @xmath116 and _ annihilation operator _ @xmath117 as @xcite@xcite@xcite @xmath118 which gives a commutation relation @xmath119 = 1 $ ] . with the two operators , the hamiltonian can be rewritten as @xmath120 one can identify the operators @xmath121 as the _ number operator _ @xmath122 , and label the energy eigenstates as @xmath123 with energy @xmath124 and the operations of the annihilation and creation operators on the energy eigenstates are @xmath125 , title="fig : " ] , title="fig : " ] in this example , the wave function of interest is the ground state of quantum harmonic oscillator , @xmath126 , which can be solved by the equation @xmath127 . solving the equation would give the solution : @xmath128{$-$}}{\frac{m \omega x^2}{2\hbar}}\right)},\ ] ] which has a gaussian waveform . since this energy eigenstate has energy @xmath129 , and the gaussian wave function has non - zero amplitudes over the entire position space , there is non - zero probability of finding the ground state in classically forbidden region where @xmath130 , hence @xmath131 , satisfying the criteria of tunnelling . this phenomenon also applies to excited states , which can be clearly seen in . to mathematically develop the statement about whether the ground state is indeed tunnelling from the viewpoint of the phase space , one could refer to and consider the operator @xmath132 at @xmath133 . it can be shown that , in appendix e , that such an operator for the quantum case is calculated to be @xmath134(x , p ) \nonumber \\ = & \ , \begin{cases } -2 \exp \left(-\frac{m\omega x^2}{\hbar } \right)\exp\left(-\frac{p^2}{m\omega\hbar}\right ) & \text{for } v(x ) > e_0 , \\ 1 - 2\exp \left(-\frac{m\omega x^2}{\hbar}\right ) \exp \left(-\frac{p^2}{m\omega\hbar}\right ) & \text{otherwise , } \end{cases}\end{aligned}\ ] ] which contains negativities , indicating the ground state of a quantum harmonic oscillator can tunnel . to demonstrate such a state is indeed tunnelling , one could calculate the inner product between this tunnelling rate operator and the wigner function of the ground state , which is indeed negative as shown in appendix e. by , the ground state of a quantum harmonic oscillator is a tunnelling state . in order to demonstrate that phase space quantum theory can be structured under the generalised probabilistic theory framework , it is necessary to recover the mathematical objects corresponding to the three components of a gpt : + + * 1 . preparation . * a state of a one - spatial dimensional quantum state can be described by a wigner function @xmath5 , which is essentially the weyl transformation of a density operator representing that quantum state in hilbert space formulation . however , there may be difficulties in designating such state as a real vector , which is demanded by the framework of gpt , for mathematical manipulations may not be well - behaved for objects in infinite - dimensional spaces @xcite . nonetheless , real functions @xmath30 in the phase space can be considered as vectors in a vector space , with the vector addition as addition of functions and scalar multiplication as multiplying a scalar @xcite , in which the set of valid phase space functions , or the state space , is a subset . in addition , one can define an inner product on such vector space such that , for any two functions @xmath135 and @xmath136 : @xmath137 such that a wigner function corresponding to a pure state is square - integrable , @xmath138 which implies that the function converges and the inner product exists for these states as a sign of well - behaved theory . moreover , for general normalised states that is represented by @xmath139 with @xmath16 and @xmath140 , the wigner representation is @xmath141 where @xmath15 are wigner functions corresponding to the pure states @xmath142 . in this case , since @xmath16 , these wigner functions corresponds to the normalised states , and therefore any normalised is a convex sum of the pure states @xmath143 . therefore , it is demonstrated that the wigner functions corresponding to the pure states are the extremal states of the gpt , and the set of normalised states is the convex hull of such extremal states . therefore , these normalised mixed states are square - integrable as well , as @xmath144 ^ 2\,dx\,dp \nonumber \\ = & \,\sum_{i , j } \lambda_i \lambda_j \iint w_i(x , p ) w_j(x , p ) \,dx \,dp \nonumber \\ \leq&\ , \sum_i \frac{\lambda_i^2}{2\pi\hbar } + \sum_{i \neq j } \frac{\lambda_i\lambda_j}{2\pi\hbar } \nonumber \\ = & \,\frac{1}{2 \pi \hbar } \left(\sum_i \lambda_i\right)^2 = \frac{1}{2 \pi \hbar } , \end{aligned}\ ] ] which implies @xmath145 is the maximum of the square - integral of any wigner function . in fact , such result is related to the purity measure @xmath146 . this measure can be mapped to @xmath147 where @xmath148 , with @xmath149 . this implies the square integral , or the inner product of the state on itself , or geometrically the square of the length of the state vector , is linearly related to purity . an interesting comparison to be made here is that the length of a vector in bloch sphere corresponds also to the purity measure of a qubit , and that in quantum theory of qubit , such length is also related to uncertainty of a state . sub - normalised states can be generated by rescaling a normalised wigner function with some non - negative constant @xmath150 , and therefore in general , any state can be expressed as @xmath151 , \end{aligned}\ ] ] as the convex sum of extremal states and the null vector . therefore , the states of phase space formulation concur with the requirements of the gpt framework . there is , however , one caveat : the conventional wigner representation of a position eigenstate is linearly related to a dirac delta function @xmath90 . this state is not well - behaved in the sense that the square - integral of the function diverges , causing problems in defining the inner product of such vector space as the integral of two phase space functions . nonetheless , it should be noted that the same problem arises in the hilbert space formulation of quantum theory , where the wave function of a position eigenstate is not considered to be a state in hilbert space and therefore does not correspond to a physical state @xcite . therefore , the introduction of such states in the gpt framework does not lead to additional non - physical features other than the ones inherent in the conventional quantum theory @xcite . as a side remark , a possible method to bypass this problem by discretising the wigner function into an @xmath152 grid in phase space , in which each grid at @xmath153 represents the quasi - probability of locating the state between some range @xmath154 and @xmath155 , i.e. @xmath156 . even though it would require @xmath157 , it would regularise the dirac delta functions @xmath90 into having the value @xmath158 at each grid where @xmath159 , instead of infinities . since these quasi - probabilities are real numbers , at @xmath157 a scaled wigner function is recovered , the gpt formulation can be applied to such object as representation of a quantum state . needless to say , this process would require a detailed formulation of discrete wigner functions and phase space quantum mechanics . + + * 2 . transformation . * the set of valid transformations is the set of functions with star - product as the weyl transform of transformations in hilbert space formulation . an example of such transformation is the unitary evolution of state . the weyl transformation of such evolution is given by @xmath160 where @xmath161 is the weyl transform of the unitary operator @xmath162 . it can be demonstrated that complex conjugate transposition of an operator in hilbert space formulation is mapped to a complex conjugation of the phase space function , as @xmath163{$-$}}\infty}^\infty e^{2ipy/\hbar } { \langle x+y|u^\dagger|x - y|\,\rangle}dy \\ = & \ , { \scalebox{0.75}[1.0]{$-$}}2\int_{\infty}^{{\scalebox{0.75}[1.0]{$-$}}\infty } e^{{\scalebox{0.75}[1.0]{$-$}}2ipy/\hbar } { \langle x - y|u^\dagger|x+y|\,\rangle}dy \\ = & \ , \left(2 \int_{{\scalebox{0.75}[1.0]{$-$}}\infty}^\infty e^{2ipy/\hbar } { \langle x+y|u|x - y|\,\rangle}dy\right)^ * \\ = & \ , u^*(x , p ) . \end{aligned}\ ] ] therefore , the unitary condition for a reversible transformation is expressed as @xmath164 it is also interesting to note that unitary transformations preserve purity by @xmath165 ^ 2\right ) = \operatorname{tr}(\hat{u}\hat{\rho}^2\hat{u}^\dagger ) = \operatorname{tr}(\hat{u}^\dagger \hat{u } \hat{\rho^2 } = \operatorname{tr}(\hat{\rho}^2 ) . \end{aligned}\ ] ] in the phase space formulation , this is equivalent to the case where the unitary transformations preserve the length of the state vector . in general , unitary transformations preserve the inner products between two states , and therefore in the phase space picture such transformations resembles an orthogonal transformation , despite the dimension of the state space is infinite . this feature is analogous to the isomorphism between @xmath166 and @xmath167 in the bloch sphere representation of quantum theory of qubits . in order to demonstrate that transformations in phase space formulation of quantum theory concurs with that of a gpt , it is necessary to demonstrate that such transformations are linear . since it is the case in hilbert space formulation that a transformation , represented by a cptp map as @xmath168 , must be linear , by weyl transformation , @xmath169 where @xmath170 is the operator in phase space representation , and @xmath15 are wigner functions corresponding to density operators @xmath171 this can also be seen from the construction of star - product , where the fact that the operators @xmath172 and @xmath173 are linear implies that the star - product as a expansion of these operators is linear as well , and therefore is consistent with the formulation of transformation in a gpt . it is also relatively straightforward to demonstrate that such transformations map any valid state into another by directly converting such result from the hilbert space formulation via weyl transformation . since it is the case that each state in hilbert space formulation can be mapped to a wigner function as a valid state , and that each cptp map is a transformation that maps one valid state into another , it must be the case that these transformations in phase space formulation map any state into valid states . + + * 3 . measurement . * the set of valid effects in the phase space formulation is the set of functions as the weyl transformation of _ positive operator valued measurement _ , or _ povms _ , in hilbert space formulation . in particular , the set of pure effects @xmath174 are the phase space functions corresponding to projective measurement in the form @xmath175 , such that @xmath176 which is isomorphic to the set of pure states . with the povms as probabilistic mixture of projection operators , any mixed effects can be expanded as a convex sum of pure effects and null effect . from this result , the self - duality of state space and effect space , a property of quantum theory , is recovered . notice that under the weyl transformation , a probability of some event corresponding to @xmath177 occurring is given by the inner product of the effect @xmath174 and the state @xmath178 , @xmath179 here , the probabilities satisfy @xmath180 . these projectors corresponding to a measurement summing to identity by the completeness equation , and the set of orthogonal eigenstates is mapped to a set of orthogonal functions in phase space as shown in appendix f. therefore , for some projection operators @xmath181 corresponding to multiple outcomes , @xmath182 . hence , for general states and general effects , the inner product has the upper bound @xmath183 by cauchy - schwarz inequality and the condition @xmath184 and @xmath185 . such inner product is also non - negative , as @xmath186 and that the trace of pure states and pure effects must be non - negative . by mapping these results into the phase space formulation , these effects satisfy the requirement of the gpt framework as to give valid probability values under inner product with any states in state space . to study the phenomenon of tunnelling in post - quantum theories , one will have to devise a method of consistently extending the existing state space in phase space picture into inclusion of non - physical states . extension of state space is relatively easy with wigner function representation , as by varying the values of the wigner function at some phase space points , it is possible to generate a non - physical state . the difficulty , however , lies in construction of the corresponding effect space , and the physical interpretation of these post - quantum states . while a complete post - quantum theory is not devised in this article , some preliminary work on gaussian states is carried out as a precursor towards an eventual development of a post - quantum phase space theory . a gaussian bivariate distribution in phase space @xmath187 has the form @xcite @xmath188{$-$}}\frac{1}{2 } ( \vec{x } - \vec{\mu})^t \gamma^{-1 } ( \vec{x}-\vec{\mu})\right],\ ] ] where @xmath189 is a vector of coordinates , @xmath190 is a vector of mean values of coordinates , and @xmath191 is the covariance matrix , where @xmath192 is the covariance of @xmath193-th and @xmath194-th coordinates . a possible way of generalising the existing state space is to include gaussian distributions that violates the purity condition , such that @xmath195 the main advantage of considering only gaussian distributions as post - quantum states is that it is a proper joint - probability distribution in phase space , and therefore such an object gives proper probabilities when one conducts measurement in position or momentum space . another advantage is that such a distribution is positive over all phase space , and therefore , by , any analysis of tunnelling on these post - quantum states are dependent only on the effects @xmath61 and @xmath60 , provided that the effects remain valid under such an extension of state space . it is interesting to note the physical meaning behind extension of purity . one way of understanding such an extension is to calculate the variances in both position and momentum for @xmath187 . since it is the property of a gaussian bivariate distribution to yield a gaussian distribution in one coordinate after integrating the distribution over the other coordinate , the variances in @xmath43 and @xmath39 are simply @xmath196 and @xmath197 . for a state with given purity @xmath198 , @xmath199{$-$}}(\vec{x } - \vec{\mu})^t \gamma^{-1 } ( \vec{x } - \vec{\mu})\right]\,dx\,dp \nonumber \\ & = \frac{2\pi\hbar}{4 \pi \sqrt{\sigma_x^2\sigma_p^2-\gamma_{xp}^2 } } \geq \frac{\hbar}{2\sqrt{\sigma_x^2\sigma_p^2 } } \nonumber \\ \sigma_x \sigma_p & \geq \frac{\hbar}{2\mu}.\end{aligned}\ ] ] since @xmath200 . the final result of the calculation closely resembles the uncertainty principle . in fact , if one substitutes @xmath201 as the condition of pure state , the uncertainty principle is recovered . therefore , by relaxing the purity relation to states that has @xmath202 , the lower bound of the uncertainties @xmath203 reaches below the lower bound allowed by quantum theory , and therefore violates the quantum uncertainty principle . this matches the intuition on an extreme case where a dirac delta function @xmath204 represents a distribution with perfect information of both position and momentum , is a state with infinite purity @xmath198 since @xmath205 , which maximally violates the uncertainty principle . it should be noted that certain studies had related purity and uncertainty before @xcite ; yet the results are limited to purity being between @xmath148 . nonetheless , the practical implication of this result is that one can simply change the purity condition and generate post - quantum states according to by altering the covariant matrix @xmath191 in . another way of interpreting the violation of purity condition is to look at the hilbert space formulation of quantum mechanics . for a general state @xmath5 such that it is real , it is mapped by wigner transformation to an operator @xmath206 such that @xmath207}\,d\alpha\,d\beta\,dx\,dp \right]^\dagger \nonumber \\ = & \ , \frac{1}{(2\pi)^2 } \iiiint w^*(x , p ) e^{{\scalebox{0.75}[1.0]{$-$}}i[\alpha(\hat{x } - x ) + \beta ( \hat{p } - p)]}\,d\alpha\,d\beta\,dx\,dp \nonumber \\ = & \ , \frac{1}{(2\pi)^2 } \iiiint w(x , p ) e^{i[({\scalebox{0.75}[1.0]{$-$}}\alpha)(\hat{x } - x ) + ( { \scalebox{0.75}[1.0]{$-$}}\beta)(\hat{p } - p ) ] } \nonumber \\ & \,d({\scalebox{0.75}[1.0]{$-$}}\alpha)\,d({\scalebox{0.75}[1.0]{$-$}}\beta)\,dx\,dp \nonumber \\ = & \ , \hat{\rho}.\end{aligned}\ ] ] in other words , the reality condition of a wigner function is mapped to the hermiticity of the density operator . since it is possible to find an eigen - decomposition for any hermitian operators @xcite , one can write a density operator as @xmath208 where @xmath209 are eigenvalues corresponding to @xmath210 as @xmath193-th eigenvector . in this representation , by the orthogonality of eigenvectors , purity is simply @xmath211 therefore , for a general normalised post - quantum state with purity @xmath202 represented by a real wigner function , it can be mapped to a density operator in its eigenbasis @xmath13 , such that @xmath212 assume that all @xmath213 . by some algebraic manipulation , @xmath214 therefore , it is impossible to satisfy the both and together with the assumptions @xmath213 . by reductio ad absurdum , it is necessary that the density operator corresponding to post - quantum states with purity greater than unity to have negative eigenvalues . in other words , in the hilbert space formulation , these post - quantum states violates quantum theory by introducing non - positive definite operators as states . this serves as a warning of altering the state space without correspondingly changing effect space , as this would introduce observable non - physical probabilities beyond the conventional range between 0 and 1 . with the procedure ultimately related to introducing negativities into the density operator , there is an obvious problem with the extension towards post - quantum theory . it can no longer be conceived that the set of measurements is invariant under such alteration , for to do so is to allow negative probabilities when one conduct an inner product of an effect corresponding to the eigenstate with negative eigenvalue and the state . therefore , the set of allowed measurements must shrink accordingly . despite it is true that by the construction of , the effect @xmath215 is still valid , it is not so apparent that @xmath61 remains valid . if in certain states that the set of effects corresponding to energy measurement is invalid , then one must find another set of effects corresponding to a new energy measurement , which creates great difficulties in interpreting energy as a physical and observable quantity . nonetheless , a qualitative argument could be given here regarding the status of tunnelling as a phenomenon in post - quantum scenarios . consider the ground state of the quantum harmonic oscillator : if one varies the state by shrinking @xmath216 and @xmath217 simultaneously , and therefore violating the uncertainty principle and purity condition by altering the covariance matrix @xmath191 in , then while the ground state energy effect vector would shrink correspondingly , the variation is continuous and therefore the inner product that specifies the tunnelling rate would still retains negativity in the close vicinity of the quantum case . therefore , it seems that tunnelling can be a generic property of post - quantum theories and is not unique to quantum theory . however , to fully study the phenomenon of tunnelling in post - quantum theories rigorously , it is necessary to construct a systematic theory that describes the effects on the effect space by alteration of the state space . we have shown that tunnelling necessitates a negative wigner function of the state and/or a tunnelling rate operator at some energies as we have defined . this links tunnelling with non - classical probabilistic behaviour ( negative quasi - probabilities ) in a concrete manner . we also argued that our approach can be used to investigate tunnelling in generalised probabilistic theories , showing the wigner function representation fits into that framework . a very intriguing question for future research is how these results relate to recent studies that suggest negative wigner function cite ( which is also argue to be equivalent to contextuality cite ) is the source of the putative power of quantum computation @xcite . we are grateful for discussions with dan browne , jonathan halliwell , benjamin yadin , andrew garner , vlatko vedral , and dominic branford . we acknowledge funding from the eu collaborative project thermiq ( grant agreement no . 618074 ) , wolfson college , university of oxford and the london institute for mathematical sciences . 99 m. razaby , _ quantum theory of tunneling _ singapore : world scientific publishing , 2014 . m. arndt , t. juffmann , and v. vedral , _ quantum physics meets biology _ , hsfp . j. * 3(6 ) * , pp . 386 - 400 ( 2009 ) . a. ghosh , and s. mukherjee , _ quantum annealing and computation : a brief documentary note _ , arxiv e - prints ( 2013 ) , 1310.1339v4 . a. j. shields , _ quantum electronics : new light on quantum tunnelling _ , nat . photonics , * 6 * , pp . 348 - 349 ( 2012 ) . r. l. hudson , rep . * 6 - 2 * , 249 - 252 ( 1974 ) . c. cohen - tannoudji , b. diu , and f. laloe , _ quantum mechanics , vol . 1_. new york , ny : john wiley & sons , 1991 . d. j. griffins , _ introduction to quantum mechanics _ , 2nd ed . upper saddle river , nj : prentice hall , 1995 . r. shankar , _ principles of quantum mechanics _ , 2nd ed . new york , ny : plenum press , 1994 . c. l. roy , and a. khan , _ a study of tunneling through multibarrier systems _ , phys . ( b ) , * 176(1 ) * , pp . 101 - 108 ( 1993 ) . v. jelic , and f. marsiglio , _ the double well potential in quantum mechanics : a simple , numerically exact formulation _ , eur . * 33 * , pp . 1651 - 1666 ( 2012 ) . c. k. zachos , d. b. fairlie , and t. l. curtright , _ quantum mechanics in phase space : an overview with selected papers_. singapore : world scientific ( 2005 ) . p. janotta , and h. hinrichsen , _ generalized probability theories : what determines the structure of quantum theory ? _ , arxiv e - print ( 2014 ) , quant - ph/1402.6562v3 . l. hardy , _ quantum theory from five reasonable axioms _ , arxiv e - prints ( 2008 ) , quant - ph/0101012v4 . j. barrett , _ information processing in generalized probabilistic theories _ , arxiv e - print ( 2006 ) , quant - ph/0508211v3 . a. mari , k. kieling , b. melholt nielsen , e. s. polzik , and j. eisert , _ directly estimating non - classicality _ , phys . lett . * 106 * , 010403 ( 2011 ) s. d. barlett , t. rudolph , and r. w. spekkens , _ reconstruction of gaussian quantum mechanics from liouville mechanics with an epistemic restriction _ , arxiv e - print ( 2012 ) , quant - ph/1111.5057v2 . t. e. hartman , _ tunneling of a wave packet _ , j. appl . * 33(12 ) * , pp . 3427 - 3433 ( 1962 ) . d. kleppner , and r. j. kolenkow , _ an introduction to mechanics _ , 2nd ed . cambridge , uk : cambridge university press ( 2013 ) . s. hassini , _ mathematical physics - a modern introduction to its foundations _ , 2nd ed . switzerland : springer international publishing ( 2013 ) . v. v. dodonov , _ purity- and entropy - bounded uncertainty relations for mixed quantum states _ , j. opt . b : quantum semiclass . opt . * 4 * ( 2002 ) , pp . 98 - 108 . m. howard , j. wallman , v. veitch , and j. emerson , _ contextuality supplies the ` magic ' for quantum computation _ , nature * 510 * , pp . 351 - 355 ( 2014 ) . the general definition could be thought of as a case of non - classical behaviour in position space . a corresponding non - classical behaviour known as reflection over barrier , could be considered under the same framework as tunnelling in momentum space . following the same analysis , the classically forbidden region for a particle with energy @xmath40 is @xmath218 , denoted as @xmath219 , where @xmath39 is a real variable representing momentum . such constraint also applies to states with energy @xmath55 , or that @xmath220 . therefore , one can formulate the definition of reflection over barrier as : [ def : qm - rb ] * general definition of reflection over barrier . * for a state in a potential given by @xmath37 , it is reflecting over barrier if and only if there exists some energy @xmath40 , such that the probability of locating the state in region where @xmath221 is greater than that of measuring the state to have energy @xmath52 , or mathematically , @xmath222 it should be noted that the two definitions , and , has similar structure , and this provides an example of constructing a mathematical formulation of various non - classical quantum processes : starting from certain classical relations between physical quantities , such as position @xmath43 and energy @xmath1 in , one could define some classically forbidden region @xmath44 which relates the two quantities , and any states that violates this relation are considered to be a non - classical state . the general definition can be applied to an energy eigenstate solution with energy @xmath223 of a rectangular potential barrier with the form with potential height @xmath224 and length @xmath225 and provide a condition definition of tunnelling in such case . mathematically , the two probabilities in the general definition becomes @xmath226 since , by , a state is tunnelling if and only if there exists some energy @xmath40 such that @xmath227 , the general definition is equivalent to : [ def : qm - ed ] * equivalent general definition of tunnelling for energy eigenstates in rectangular potential barrier . * an energy eigenstate with energy @xmath223 in a rectangular potential barrier of the form is tunnelling if and only if @xmath228 and the probability of finding the state in the region @xmath229 is non - zero . while this equivalent statement has much similarities with the conventional definition , a crucial distinction is that the convention definition provides a less strict condition of having non - zero probability of locating the state in @xmath3 . this does illustrate an important distinction between the rationale behind the formulation of the general and conventional definition . the reason why the conventional definition includes the region @xmath230 to be classically forbidden region for a state with @xmath228 is that it assumes the energy eigenstate to be incoming from @xmath231{$-$}}{\infty}$ ] , and therefore can not classically pass through the barrier . on the other hand , the classically forbidden region of general definition does not take the initial condition of the state into account , so in this sense it can be equally applied to cases where the state is incoming from @xmath231{$-$}}{\infty}$ ] and @xmath232 . in this sense , the general definition can be applied to a broader classes of states . undeniably , however , the equivalent definition provides a stricter definition of tunnelling , and hence is not immediately an equivalent statement with the conventional definition . however , one can justify that , specifically in the quantum case , the two definitions are equivalent with the inclusion of the continuity condition of wave functions . with this condition , a quantum energy eigenstate can only have non - zero probability in the region @xmath230 if there is non - zero probability distribution of locating the state at @xmath229 , which implies the condition @xmath233 , assuming the state is incoming from @xmath231{$-$}}{\infty}$ ] . therefore , the general definition and the continuity condition of wave function leads to the recovery of the conventional definition . with the framework of tunnelling in phase space stated in and , it is straightforward , then , to demonstrate that it is impossible for classical systems to tunnel in the following manner : [ thm : pstc ] * impossibility of tunnelling in classical systems . * a classical system , specified with the hamiltonian @xmath234 and the distribution function @xmath30 , can not tunnel . for a general classical system , the effect corresponding to probability @xmath62 is @xmath235 and similarly , the effect corresponding to probability @xmath57 is @xmath236 however , since the set @xmath237 , as @xmath238 therefore @xmath54 implies @xmath239 for a classical system , which suggests that the function @xmath240 ( x , p ) \geq 0 $ ] . also , since @xmath30 in this scenario is a joint probability function , @xmath30 must be non - negative over all phase space . therefore , by , a classical system can not tunnel . in some sense , this proof is anticipated by the design of the general definition , as part of the original intentions of constructing such a definition . however , this exercise is still valuable , because the previous discussion in the main text is largely based on classical particles rather than phase space ensembles . another important point is that this proof illustrates is the dual nature of tunnelling in phase space , as tunnelling does not only depend on condition on the state , i.e. the distribution function , but also the difference in effects @xmath241 . in particular , both functions have to be non - negative for a state to not tunnel . this concurs with the discussion on , where a gaussian state has a positive wigner function representation , yet in certain scenarios such states could indeed tunnel . while it is rather difficult to obtain the energy distribution of a wave packet due to the piecewise nature of the energy eigenstate wave functions , there are certain features of the problem that simplifies the analysis in principle . firstly , despite the time evolution of the state , the cumulative probability distribution @xmath57 is invariant . this is due to the fact that a general state can be considered as a superposition of energy eigenstates @xmath242 , and under the unitary operator @xmath243{$-$}}i \hat{h } t/\hbar]$ ] , @xmath244{$-$}}i\hat{h}t/\hbar } { |\psi\rangle } & = \sum_i c_i e^{{\scalebox{0.75}[1.0]{$-$}}i \hat{h}t/\hbar } { |e_i\rangle } \nonumber \\ & = \sum_i c_i e^{{\scalebox{0.75}[1.0]{$-$}}ie_it/\hbar } { |e_i\rangle}.\end{aligned}\ ] ] therefore , the probability @xmath57 is the sum of the norm squared amplitude corresponding to energy eigenstates with energy greater than @xmath40 , or @xmath245 which is independent of time . hence , the dynamical nature of the wave packet only changes the probability @xmath62 . secondly , given that for a rectangular potential barrier , @xmath246 is a constant value between @xmath247 as shown in , it can then be demonstrated that the state is tunnelling if and only if @xmath248 , by demonstrating that @xmath248 is equivalent to the condition @xmath249 , i.e. the condition of a tunnelling state : 1 . @xmath250 assume it is the case that @xmath249 and @xmath251 . construct @xmath252 according to , @xmath253 where in the region @xmath247 , @xmath254 , and in the region @xmath255 , @xmath256 . therefore , for all possible @xmath40 , @xmath257 , which contradicts with the premise @xmath258 . therefore , by reductio ad absurdum , @xmath259 . 2 . @xmath260 since @xmath261 is simply @xmath262 , @xmath263 implies @xmath249 . what the previous exercise shows is that it suffices to use two probabilities , @xmath105 and @xmath264 of a state in rectangular potential barrier to determine whether the state is tunnelling or not , which could be applied to our example of gaussian state tunnelling through a rectangular barrier . using the approximation that the energy eigenstates are roughly free momentum eigenstates , one can conduct a fourier transform on and obtain the gaussian wave packet has a probability distribution over momentum space as @xmath265{$-$}}\frac{2\sigma_x(p - p_0)^2}{\hbar}\right]},\ ] ] which is also gaussian , as expected from the property of fourier transform of gaussian probability distributions . therefore , the cumulative energy probability is given by @xmath266{$-$}}\infty}^{{\scalebox{0.75}[1.0]{$-$}}\sqrt{2me^*}}p(p)\,dp + \int_{\sqrt{2me^*}}^\infty p(p)\,dp \\ & = 1 - \frac{1}{2}\left[\operatorname{erf}\left(\frac{\sqrt{2}\sigma_x}{\hbar}(\sqrt{2me^*}-p_0)\right)+\operatorname{erf}\left(\frac{\sqrt{2}\sigma_x}{\hbar}(\sqrt{2me^*}+p_0)\right)\right],\end{aligned}\ ] ] where @xmath267 is _ error function_. while it is difficult to find the close form of the probability @xmath264 , via some numerical simulation , it is possible to obtain this probability as a function of time , as shown as . regardless , this analysis demonstrates how the general definition of tunnelling can be used to determine whether a general state as a superposition of energy eigenstates is tunnelling or not , as well as to assign a quantitative value to how such a state violates the classicality constraints . it therefore provides evidence on how the general definition satisfies the criteria of providing quantitative criteria of tunnelling for systems with states without definite energy . for a quantum harmonic oscillator with the form , the ground state wave function @xmath268 is given by . such wave function can be mapped to a wigner function @xmath84 by weyl transformation to be @xmath269{$-$}}\frac{m\omega x^2}{\hbar}\right)\exp\left({\scalebox{0.75}[1.0]{$-$}}\frac{p^2}{m\omega\hbar}\right).\ ] ] notice here that it can be clearly seen how such gaussian state saturates the uncertainty principle , as @xmath270 . also , such a wigner function , as a map of a gaussian state , is a bivariate gaussian distribution which is positive over all phase space . this example is hence a direct verification of hudson s theorem . as stated in the main text , the analysis of tunnelling for gaussian states lies predominantly on the effects . first of all , consider the effect @xmath72 . by the condition @xmath54 , the region corresponding to each @xmath40 can be found by latexmath:[\ ] ] which is simply the fourier transform in the position coordinate , with @xmath298 and @xmath299 as two independent parameters .

we study switching between period - two states of an underdamped quantum oscillator modulated at nearly twice its natural frequency . for all temperatures and parameter values switching occurs via quantum activation : it is determined by diffusion over oscillator quasienergy , provided the relaxation rate exceeds the rate of interstate tunneling . the diffusion has quantum origin and accompanies relaxation to the stable state . we find the semiclassical distribution over quasienergy . for , where the system has detailed balance , this distribution differs from the distribution for ; the distribution is also destroyed by small dephasing of the oscillator . the characteristic quantum activation energy of switching displays a typical dependence on temperature and scaling behavior near the bifurcation point where period doubling occurs .

cond-mat0602288

switching between coexisting stable states underlies many phenomena in physics , from diffusion in solids to protein folding . for classical systems in thermal equilibrium switching is often described by the activation law , with the switching probability being , where @xmath2 is the activation energy . as temperature is decreased , quantum fluctuations become more and more important , and below a certain crossover temperature switching occurs via tunneling @xcite . the behavior of systems away from thermal equilibrium is far more complicated . still , for classical systems switching is often described by an activation type law , with the temperature replaced by the characteristic intensity of the noise that leads to fluctuations @xcite . quantum nonequilibrium systems can also switch via tunneling between classically accessible regions of their phase space @xcite . in addition to classical activation and quantum tunneling , nonequilibrium systems have another somewhat counterintuitive mechanism of transitions between stable states . we call this mechanism quantum activation and study it in the present paper . it describes escape from a metastable state due to quantum fluctuations that accompany relaxation of the system @xcite . these fluctuations lead to diffusion away from the metastable state and , ultimately , to transitions over the classical `` barrier '' , that is , the boundary of the basin of attraction to the metastable state . we study quantum activation for periodically modulated systems . switching mechanisms for such systems are shown schematically in fig . [ fig : d ] , where @xmath3 is the effective potential of the system in the rotating frame . this figure describes , in particular , a nonlinear oscillator studied in the present paper , which displays period doubling when its frequency is periodically modulated in time . the energy of periodically modulated systems is not conserved . instead they are characterized by quasienergy @xmath4 . it gives the change of the wave function @xmath5 when time is incremented by the modulation period @xmath6 , @xmath7 , and is defined modulo @xmath8 fr fr . of a parametrically modulated oscillator . sketched are scaled period - two quasienergy levels ( see sec . [ sec : dynamics ] ) in the neglect of interwell tunneling . the minima of @xmath9 correspond to classically stable states of period two motion . the arrows indicate relaxation , diffusion over quasienergy away from the minima , and interwell tunneling . the effective hamiltonian @xmath10 is defined by eq . ( [ eq : g ] ) , and @xmath11 ; the figure refers to @xmath12 in eq . ( [ eq : g]).,width=288 ] coupling to a thermal reservoir leads to transitions between the states of the system . for @xmath0 the transitions are accompanied by the creation of excitations in the thermal reservoir . the energy of the system decreases in each transition . however , the quasienergy may decrease or increase , albeit with different probabilities @xmath13 and @xmath14 . this effect has quantum origin . it is due to the functions @xmath15 being superpositions of the eigenfunctions @xmath16 of the energy operator of the system in the absence of modulation . therefore bath - induced transitions down in energy @xmath17 lead to transitions @xmath18 with @xmath19 . the values of @xmath13 , @xmath14 for the latter transitions are determined by the appropriate overlap integrals and depend on @xmath20 . more probable transitions determine in which direction , with respect to quasienergy , the system will most likely move . such motion corresponds to relaxation over quasienergy . [ fig : d ] refers to the case @xmath21 . in this case relaxation corresponds to quasienergy decrease . the minima of @xmath3 are the classical stable states . however , quantum transitions in which quasienergy increases have a nonzero probability even for @xmath0 . because this probability is less than @xmath14 , such transitions lead to diffusion over quasienergy @xmath4 away from the minima of @xmath3 . in turn , the diffusion leads to a finite - width distribution over @xmath4 and ultimately to activated - type overbarrier transitions between the wells in fig . [ fig : d ] . in fact , discussed in this paper and sketched in fig . [ fig : d ] are period - two quasienergy states , with quasienergy @xmath4 defined by the condition @xmath22 . they are more convenient for the present problem ; their relation to the standard quasienergy states is explained in sec . [ sec : dynamics ] . of interest for the problem of switching is the semiclassical situation where the basins of attraction to the metastable states ( the wells in fig . [ fig : d ] ) have a large number @xmath23 of localized states . in this case the rate of tunneling decay is exponentially small . the activation rate is also exponentially small since it is determined by the ratio of transition probabilities @xmath24 raised to the power @xmath23 . both the tunneling and activation exponents are @xmath25 for the situation sketched in fig . [ fig : d ] . indeed , the tunneling exponent is given by the action , in the units of @xmath26 , for classical motion in the inverted effective potential @xmath27 from one maximum of @xmath27 to the other . it is easy to see that this action is of order @xmath23 , unless @xmath3 has a special form . therefore for @xmath28 the activation exponent is either much larger or much smaller than the tunneling exponent . in this paper we develop a theory of the statistical distribution and consider switching of a parametrically modulated quantum oscillator . we show that , irrespective of temperature , the activation exponent is smaller than the tunneling exponent . therefore switching always occurs via activation , not tunneling , as long as the relaxation rate exceeds the tunneling rate . we study a nonlinear oscillator with frequency modulated at nearly twice the natural frequency @xmath29 . when the modulation is sufficiently strong , the oscillator has two stable vibrational states with periods @xmath30 . these states differ in phase by @xmath31 but otherwise are identical @xcite . they correspond to the minima of the effective potential in fig . [ fig : d ] . the lowest quantized states in fig . [ fig : d ] are squeezed . squeezing in a parametric oscillator has attracted interest in many areas , from quantum optics @xcite to phonons @xcite , microcantilevers @xcite and electrons and ions in penning traps @xcite . recent progress in systems based on josephson junctions and nanoelectromechanical systems @xcite makes it possible to study squeezed states in a well controlled and versatile environment . both classical and quantum fluctuations can be investigated and the nature of switching between the states can be explored . the results can be further used in quantum measurements for quantum computing , as in the case of switching between coexisting states of a resonantly driven oscillator @xcite . the probability distribution and interstate transitions of a parametrically modulated oscillator have attracted considerable attention . much theoretical work has been done for models where fluctuations satisfy the detailed balance condition either in the classical limit @xcite or for @xmath0 @xcite . generally this condition does not hold in systems away from equilibrium . in particular a classical nonlinear parametric oscillator does not have detailed balance . switching of such an oscillator was studied experimentally for electrons in penning traps @xcite . the measured switching rate @xcite agreed quantitatively with the theory @xcite . of the distribution over scaled period - two quasienergy @xmath9 of a parametrically modulated oscillator for different oscillator planck numbers @xmath32 in the limit of a large number of intrawell quasienergy states . the figure refers to @xmath33 . , width=288 ] a quantum parametric oscillator also does not have detailed balance in the general case . the results presented below show that breaking the special condition where detailed balance holds leads to a sharp change of the statistical distribution and the switching rate . this change occurs already for an infinitesimally small deviation from detailed balance , in the semiclassical limit . the fragility of the detailed balance solution is previewed in fig . [ fig : fr ] . this figure shows the effective inverse temperature of the intrawell distribution over the scaled period - two quasienergy @xmath9 . the function @xmath34 is the exponent of the distribution @xmath35 , where @xmath36 is the population of an @xmath37-th state and @xmath38 is the scaled planck constant defined in eq . ( [ eq : l ] ) below . the effective inverse temperature @xmath39 depends on @xmath9 and differs from the inverse temperature of the bath @xmath40 . the @xmath41 result for @xmath42 is obtained from the solution with detailed balance ( here @xmath43^{-1}$ ] is the planck number of the oscillator ) . it is seen from fig . [ fig : fr ] that the @xmath44 ( @xmath45 ) limit of the solution without detailed balance does not go over into the @xmath0 result . the effective quantum activation energy @xmath46 , which gives the exponent of the switching rate @xmath47 , is shown in fig . [ fig : acte2 ] . it is equal to @xmath48 , where @xmath49 and @xmath50 are , respectively , the values of @xmath3 at the top and the minimum of the wells in fig . [ fig : d ] . the @xmath51 detailed balance value for @xmath52 strongly differs from the @xmath53 value in a broad range of the control parameter @xmath54 that characterizes the detuning of the modulation frequency from @xmath55 , see eq . ( [ eq : mu ] ) below . it is also seen from fig . [ fig : acte2 ] that the quantum activation energy @xmath52 decreases with increasing temperature . ultimately when the planck number of the oscillator becomes large , @xmath56 , we have @xmath57 , the law of thermal activation , and the results coincide with the results @xcite obtained by a different method . , where @xmath58 , eq . ( [ eq : l ] ) , is the effective planck constant . the scaled switching exponent @xmath52 is plotted as a function of the single scaled parameter @xmath54 that controls oscillator dynamics , eq . ( [ eq : mu ] ) , for different values of the oscillator planck number @xmath32 . , width=288 ] in sec . [ sec : dynamics ] we obtain an effective hamiltonian that describes the oscillator dynamics in the rotating frame . we describe the classical dynamics and calculate semiclassical matrix elements of the coordinate and momentum . in sec . [ sec : balance_equation ] we derive the balance equation for the distribution over quasienergy levels . this distribution is found in the eikonal approximation in sec . [ sec : distribution ] . its explicit form is obtained in several physically important limiting cases in sec . [ sec : limiting_cases ] . in sec . [ sec : switching_exponent ] the probability of interstate switching due to quantum activation is analyzed . in sec . [ sec : fragility ] it is shown that the distribution obtained for @xmath59 , where the system has detailed balance , is fragile : in the limit of a large number of intrawell states it differs from the distribution for @xmath1 . it is also dramatically changed by even weak dephasing due to external noise . in sec . [ sec : tunneling ] we discuss tunneling between the coexisting states of period two vibrations and show that interstate switching occurs via quantum activation rather than tunneling , if the relaxation rate exceeds the tunneling rate . in sec . [ sec : conclusions ] we summarize the model , the approximations , and the major results . we will study quantum fluctuations and interstate switching using an important model of a bistable system , a parametric oscillator . the hamiltonian of the oscillator has a simple form @xmath60 we will assume that the modulation frequency @xmath61 is close to twice the frequency of small amplitude vibrations @xmath29 , and that the driving force @xmath62 is not large , so that the oscillator nonlinearity remains small , @xmath63 here , @xmath64 is the mean squared oscillator displacement ; in what follows for concreteness we set @xmath65 . we will change to the rotating frame using the canonical transformation @xmath66 , where @xmath67 and @xmath68 are the raising and lowering operators , . it is convenient to introduce the dimensionless coordinate q and momentum p , @xmath69\ , , \\ u^{\dag}(t)p u(t)=-c\frac{\omega_f}{2}\left[p\sin(\omega_f t/2)+q\cos(\omega_f t/2)\right ] , \nonumber\end{aligned}\ ] ] where @xmath70 . the commutation relation between @xmath71 and @xmath72 has the form @xmath73=-i\lambda\ , , \quad \lambda=3\gamma\hbar / f\omega_f\ , .\ ] ] the dimensionless parameter @xmath58 will play the role of @xmath26 in the quantum dynamics in the rotating frame . this dynamics is determined by the hamiltonian @xmath74 with @xmath75 here we used the rotating wave approximation and disregarded fast oscillating terms @xmath76 . as a result @xmath77 is independent of time . in the `` slow '' dimensionless time @xmath78 the schrdinger equation has the form @xmath79 . the effective hamiltonian @xmath80 , eq . ( [ eq : g ] ) , depends on one parameter @xmath81 for @xmath82 , the function @xmath10 has two minima . they are located at @xmath83 , @xmath84 , and @xmath85 . for @xmath86 the minima are separated by a saddle at @xmath87 , as shown in fig . [ fig : g ] . when friction is taken into account , the minima become stable states of the parametrically excited vibrations in the classical limit @xcite [ we note that the function @xmath88 in ref . is similar to @xmath10 , but has opposite sign ] . , eq . ( [ eq : g ] ) , for @xmath89 . the minima of @xmath10 correspond to the stable states of period two vibrations.,width=288 ] the function @xmath10 is symmetric , @xmath90 . this is a consequence of the time translation symmetry . the sign change @xmath91 corresponds to the shift in time by the modulation period @xmath92 , see eq . ( [ eq : udu ] ) . in contrast to the standard hamiltonian of a nonrelativistic particle , the function @xmath10 does not have the form of a sum of kinetic and potential energies , and moreover , it is quartic in momentum p. this structure results from switching to the rotating frame . it leads to a significant modification of quantum dynamics , in particular tunneling , compared to the conventional case of a particle in a static potential . the quasienergies of the parametric oscillator @xmath93 are determined by the eigenvalues @xmath94 of the operator @xmath95 , with an accuracy to small corrections @xmath96 . they can be found from the schrdinger equation @xmath97 the exact eigenfunctions @xmath98 are either symmetric or antisymmetric in @xmath72 . the eigenstates @xmath98 give the floquet states of the parametric oscillator @xmath7 . from the form of the unitary operator @xmath99 it is clear that @xmath100 , with @xmath101 " for symmetric and @xmath102 " for antisymmetric states @xmath98 . therefore we have @xmath103 for symmetric and @xmath104 for antisymmetric states . it is convenient to define period - two quasienergy by the condition @xmath105 . in this case the relation @xmath106 holds for both symmetric and antisymmetric states . at the same time , there is simple one to one correspondence with the standard quasienergy states . period - two quasienergy states are particularly convenient for describing intrawell states @xmath107 for small tunneling between the wells of @xmath10 . such states are combinations of symmetric and antisymmetric states @xmath98 . for this reason we use period - two quasienergies throughout the paper . we will study the intrawell states @xmath107 in the semiclassical approximation . the classical equations of motion @xmath108 can be explicitly solved in terms of the jacobi elliptic functions . the solution is given in appendix a. there are two classical trajectories for each @xmath109 , one per well of @xmath10 in fig . [ fig : g ] . they are inversely symmetrical in phase space and double periodic in time , @xmath110 with one real period @xmath111 and one complex period @xmath112 . this means that @xmath113 , where @xmath114 , and @xmath115 here @xmath116 is the complete elliptic integral of the first kind , @xmath117 , and @xmath118 , @xmath119 is the parameter @xcite . the real part of @xmath120 can be positive or negative ; for @xmath121 , the period @xmath122 has not only imaginary , but also a nonzero real part . the vibration frequency is @xmath123 . it monotonically decreases with increasing @xmath9 in the range where @xmath10 has two wells , @xmath124 , and @xmath125 for @xmath126 , i.e. for @xmath9 approaching the saddle - point value . the periodicity of @xmath127 , @xmath128 in imaginary time turns out to be instrumental in calculating matrix elements of the operators @xmath71 , @xmath72 on the intrawell wave functions @xmath107 as discussed in the next section . of central interest to us will be the case where the number of states with @xmath129 , that is the number of states inside the wells of @xmath10 in fig . [ fig : g ] , is large . formally , this case corresponds to the limit of a small effective planck constant @xmath58 . for @xmath130 the wave function @xmath131 in the classically accessible region can be written in the form @xmath132+{\rm c.c.}\ , , \\ s_n(q)=\int_{q_{n}}^{q}p(q',g_n)dq'\ , .\nonumber\end{aligned}\ ] ] here , @xmath133 is the normalization constant , @xmath134 is the momentum for a given @xmath72 as determined by the equation @xmath135 , @xmath136 is the classical turning point , @xmath137 , and the derivative @xmath138 is calculated for @xmath139 . we parameterize the classical trajectories ( [ eq : dqpdt ] ) by their phase @xmath140 , which we count ( modulo @xmath141 ) from its value at @xmath136 . with this parametrization we have @xmath142 for @xmath143 ; on a classical trajectory , @xmath72 is an even and @xmath71 is an odd function of @xmath144 . with eqs . ( [ eq : psi ] ) and ( [ eq : ds ] ) , we can write the semiclassical matrix element of the lowering operator @xmath145 as a fourier component of @xmath146 $ ] over the phase @xmath140 @xcite , @xmath147 to lowest order in @xmath58 . as we discuss below , these matrix elements determine relaxation of the oscillator . we will be interested in calculating @xmath148 for @xmath109 . then if we neglect interwell tunneling , we have two sets of wave functions @xmath131 , one for each well . the interwell matrix elements of the operator @xmath149 are exponentially small . the intrawell matrix elements are the same for the both wells except for the overall sign , because @xmath72 and @xmath71 in different wells have opposite signs on the trajectories with the same @xmath9 . we will consider the matrix elements @xmath150 for the states in the right well @xmath151 . and momentum @xmath71 . the left and right panels refer , respectively , to the negative and positive real part of the parameter of the elliptic functions @xmath120 , eq . ( [ eq : m ] ) [ the imaginary part of @xmath120 is infinitesimally small ] . both @xmath72 and @xmath71 have poles at @xmath152 , but @xmath153 has a pole only for @xmath154 . the increment of phase by @xmath155 leads to a transition between the trajectories with a given @xmath9 in different wells of @xmath10.,width=288 ] the integral ( [ eq : a ] ) can be evaluated using the double periodicity of the functions @xmath72 , @xmath71 . it is clear that @xmath156 for any complex @xmath144 . in addition , from eq . ( [ eq : pe ] ) for @xmath109 we have @xmath157 , where @xmath158 is a complex number , which is determined by the ratio of the periods of @xmath72 and @xmath71 . the shift in time by @xmath159 corresponds to a transition from @xmath160 in one well to @xmath160 in the other well . we can now write the matrix element @xmath148 as @xmath161^{-1}\oint_c d\phi \,e^{-im\phi}a(\phi;g)\ , , \ ] ] where integration is done along the contour @xmath162 in fig . [ fig : cl ] . it is explained in appendix [ ap : cm ] that each of the functions @xmath163 , @xmath164 has two poles inside the contour @xmath162 @xcite . therefore it is easy to calculate the contour integral ( [ eq : cint ] ) . the result has a simple form @xmath165 with @xmath166 given by the equation @xmath167 = \frac{1}{2}{\textrm{im}}\,\phi_0 . \nonumber\end{aligned}\ ] ] an important feature of the matrix elements @xmath148 is their exponential decay for large @xmath168 . from eqs . ( [ eq : am ] ) and ( [ eq : ts ] ) @xmath169\ , , \quad m\gg 1\ , , \\ a_m(g)\propto\exp[-|m|{\textrm{im}}\,\phi_*]\ , , \quad -m\gg 1\ , .\nonumber\end{aligned}\ ] ] we note that the decay is asymetric with respect to the sign of @xmath170 . this leads to important features of the probability distribution of the oscillator . coupling of the oscillator to a thermal reservoir leads to its relaxation . we will first consider the simplest type of relaxation . it arises from coupling linear in the oscillator coordinate @xmath171 and corresponds to decay processes in which the oscillator makes transitions between neighboring energy levels , with energy @xmath172 transferred to or absorbed from the reservoir . we will assume that the oscillator nonlinearity is not strong and that the detuning @xmath173 of the modulation frequency is small , whereas the density of states of the reservoir weighted with interaction is smooth near @xmath29 . then the quantum kinetic equation for the oscillator density matrix @xmath174 in the rotating frame has the form @xmath175-\hat{\eta}\rho\ , , \\ \hat{\eta}\rho=\eta\left[(\bar{n}+1)({\hat{a}^{\dag}}{\hat{a}}\rho-2{\hat{a}}\rho{\hat{a}^{\dag}}+ \rho{\hat{a}^{\dag}}{\hat{a}})\right.\nonumber\\ \left.+\bar{n}({\hat{a}}{\hat{a}^{\dag}}\rho-2{\hat{a}^{\dag}}\rho{\hat{a}}+\rho{\hat{a}}{\hat{a}^{\dag}})\right]\ , , \nonumber\end{aligned}\ ] ] where @xmath176 is the dimensionless relaxation constant and @xmath177^{-1}$ ] is the planck number . we will assume that relaxation is slow so that the broadening of quasienergy levels is much smaller than the distance between them , @xmath178 . then off - diagonal matrix elements of @xmath174 on the states @xmath131 are small . we note that , at the same time , off - diagonal matrix elements of @xmath174 on the fock states of the oscillator @xmath179 do not have to be small . to the lowest order in @xmath180 relaxation of the diagonal matrix elements @xmath181 , is described by the balance equation @xmath182 with dimensionless transition probabilities @xmath183 it follows from eqs . ( [ eq : am ] ) and ( [ eq : wmn ] ) that , even for @xmath0 , the oscillator can make transitions to states with both higher and lower @xmath9 , with probabilities @xmath184 with @xmath185 and @xmath186 , respectively . this is in spite the fact that transitions between the fock states of the oscillator for @xmath0 are only of the type @xmath187 . the explicit expression for the matrix elements ( [ eq : am ] ) makes it possible to show that the probability of a transition to a lower level of @xmath188 is larger than the probability of a transition to a higher level , that is , @xmath189 for @xmath185 . therefore the oscillator is more likely to move down to the bottom of the initially occupied well in fig . [ fig : g ] . this agrees with the classical limit in which the stable states of an underdamped parametric oscillator are at the minima of @xmath10 . however , along with the drift down in the scaled period - two quasienergy @xmath9 , even for @xmath0 there is also diffusion over quasienergy away from the minima of @xmath10 , due to nonzero transition probabilities @xmath190 with @xmath185 . equation ( [ eq : be ] ) must be slightly modified in the presence of tunneling between the states with @xmath129 . the modification is standard . one has to take into account that the matrix elements of @xmath174 depend not only on the number @xmath37 of the period - two quasienergy level inside the well , but also on the index @xmath191 which takes on two values @xmath192 , that specify the wells of @xmath10 . these values can be associated with the eigenvalues of the pseudospin operator @xmath193 . the matrix elements of the operator @xmath194 on the wave functions of different wells are exponentially small and can be disregarded . the operator describing interwell tunneling can be written in the pseudospin representation as @xmath195 $ ] , where @xmath196 is the tunneling splitting of the states in different wells . we assume that the tunneling splitting is much smaller than @xmath176 . then after a transient time @xmath197 , there is formed a quasistationary distribution over the states @xmath131 inside each of the wells of @xmath10 in fig . [ fig : g ] . this distribution can be found from eq . ( [ eq : be ] ) using the wave function @xmath131 calculated in the neglect of tunneling . interwell transitions occur over much longer time . the stationary intrawell probability distribution can be easily found from eqs . ( [ eq : be ] ) and ( [ eq : wmn ] ) if the number of levels with @xmath129 is small . much more interesting is the situation where this number is large . it corresponds to the limit of small @xmath58 . we will be interested primarily in the quasistationary distribution over the states in the well in which the system was initially prepared . it is formed over time @xmath198 and is determined by setting the right hand side of eq . ( [ eq : be ] ) equal to zero . the resulting equation describes also the stationary distribution in both wells , which is formed over a much longer time given by the reciprocal rate of interwell transitions . for @xmath199 we can use the wentzel - kramers - brillouin ( wkb ) approximation both to calculate the matrix elements in @xmath184 ( [ eq : wmn ] ) and to solve the balance equation . the solution should be sought in the eikonal form @xmath200\ , , \quad r_n = r(g_n).\ ] ] it follows from eq . ( [ eq : ae ] ) that the transition probabilities @xmath201 rapidly decay for large @xmath168 . therefore in the balance equation ( [ eq : be ] ) we can set @xmath202\ , , \\ & & r'(g)=dr / dg.\nonumber\end{aligned}\ ] ] the corrections to the exponent in eq . ( [ eq : rp ] ) of order @xmath203 , @xmath204 are small for @xmath130 ( prime indicates differentiation over g ) . we note that @xmath205 is not small in the quantum regime and the exponential in ( [ eq : rp ] ) will not be expanded in a series in @xmath206 . from eq . ( [ eq : be ] ) , the function @xmath207 is determined by the polynomial equation @xmath208.\nonumber\end{aligned}\ ] ] here the sum goes over positive and negative @xmath170 . in obtaining eq . ( [ eq : bep ] ) we have used the relation @xmath209 for @xmath210 , which is the consequence of the wkb approximation for the matrix elements : @xmath211/dg\ll 1 $ ] for @xmath130 . equation ( [ eq : bep ] ) has a trivial solution @xmath212 , which is unphysical . because all coefficients @xmath201 are positive and @xmath213 for @xmath214 , one can show that the polynomial in the left hand side of eq . ( [ eq : bep ] ) has one extremum in the interval @xmath215 . since its derivative for @xmath212 is negative and it goes to @xmath216 for @xmath217 , it has one root in this interval . this root gives the value @xmath218 . since @xmath219 , we have @xmath220 . in turn @xmath207 gives @xmath221 and thus the distribution @xmath36 . in obtaining @xmath221 from @xmath207 , in the spirit of the eikonal approximation , one should set @xmath222 , where @xmath223 is the minimal quasienergy . of the distribution over scaled period - two quasienergy @xmath9 for two values of the control parameter @xmath54 and for different oscillator planck numbers @xmath32 . for @xmath224 the function @xmath225 only weakly depends on @xmath9 inside the well , @xmath226.,width=288 ] it is seen from eq . ( [ eq : rho ] ) that @xmath227 has the meaning of the effective inverse temperature of the distribution over period - two quasienergy . the function @xmath207 can be calculated for different values of the control parameter @xmath54 and different oscillator planck numbers @xmath228 by solving eq . ( [ eq : bep ] ) numerically . the numerical calculation is simplified by the exponential decay of the coefficients @xmath229 with @xmath168 . the results are shown in fig . [ fig : dsdg ] . the function @xmath230 smoothly varies with @xmath9 in the whole range @xmath231 of intrawell values of @xmath9 , except for very small @xmath228 . numerical results on the logarithm of the distribution @xmath232 obtained from eq . ( [ eq : bep ] ) for @xmath233 are compared in fig [ fig : num](a ) with the results of the full numerical solution of the balance equation ( [ eq : be ] ) . in this latter calculation we did not use the wkb approximation to find the transition probabilities @xmath190 . instead they were obtained by solving numerically the schrdinger equation @xmath234 and by calculating @xmath190 as the appropriately weighted matrix elements @xmath235 , eq . ( [ eq : wmn ] ) . in the calculation we took into account that , for small @xmath58 , the levels of @xmath188 form tunnel - split doublets . the tunneling splitting is small compared to the distance between intrawell states @xmath236 . the exact eigenfunctions of the operator @xmath237 are well approximated by symmetric and antisymmetric combinations of the intrawell wave functions . this allows one to restore the intrawell wave functions from the full numerical solution and to calculate the matrix elements @xmath190 . by construction , such numerical approach gives @xmath221 only for the values of @xmath9 that correspond to quasienergy levels @xmath238 . it is seen from fig . [ fig : num](a ) that the two methods give extremely close results for small @xmath58 ( see also below ) . over period - two quasienergy @xmath9 ( solid lines ) with the results obtained by direct calculation of the transition probabilities followed by numerical solution of the balance equation ( squares).,width=288 ] the effective inverse temperature @xmath230 and the distribution @xmath36 can be found in several limiting cases . we start with the vicinity of the bottom of the wells of @xmath10 . here classical vibrations of @xmath71 , @xmath72 are nearly harmonic . to leading order in @xmath239 we have @xmath240 for @xmath241 , and therefore the transition probabilities @xmath242 . then eq . ( [ eq : bep ] ) becomes a quadratic equation for @xmath243 , giving @xmath244 the inverse effective temperature @xmath245 as given by eq . ( [ eq : rpa1 ] ) monotonically decreases with the increasing planck number @xmath228 , i.e. , with increasing bath temperature @xmath246 . we note that @xmath245 smoothly varies with @xmath228 for low temperatures @xmath247 , except for small @xmath248 . we have @xmath249 for @xmath250 . for @xmath12 , on the other hand , we have @xmath251 , i.e. the effective inverse temperature for @xmath50 is simply @xmath252 . we now consider the parameter ranges where @xmath207 can be found for all @xmath9 . for @xmath56 the transition probabilities become nearly symmetric , @xmath253 . as a result , the effective inverse temperature @xmath254 becomes small , and the ratio @xmath255 can be expanded in @xmath256 . this gives @xmath257 it is shown in appendix [ ap : c ] that eq . ( [ eq : cw ] ) can be written in the simple form @xmath258 where the integration is performed over the area @xmath259 of the phase plane ( q , p ) encircled by the classical trajectory @xmath127 , @xmath128 with given @xmath9 . from eq . ( [ eq : rc ] ) , the effective inverse temperature @xmath254 is @xmath260^{-1}$ ] . in the high - temperature limit @xmath261 . therefore @xmath254 is @xmath252 and does not contain @xmath26 , as expected . ( [ eq : rc ] ) in this limit coincides with the expression for the distribution obtained in ref . using a completely different method . the function @xmath207 may be expected to have a simple form for @xmath54 close to the bifurcation value @xmath262 where the two stable states of the oscillator merge together and @xmath10 becomes single well . this is a consequence of the universality that characterizes the dynamics near bifurcation points and is related to the slowing down of motion and the onset of a soft mode . the situation we are considering here is not the standard situation of the classical theory where the problem is reduced to fluctuations of the soft mode . the oscillator is not too close to the bifurcation point , its motion is not overdamped and the interlevel distance exceeds the level broadening . still as we show @xmath221 has a simple form . the motion slowing down for @xmath54 approaching @xmath263 leads to the decrease of the vibration frequencies . from eq . ( [ eq : period ] ) , @xmath264^{1/2}\ ] ] for @xmath265 . therefore one may expect that @xmath266 becomes small , and we can again expand @xmath255 in @xmath267 , as in eq . ( [ eq : cw ] ) . one can justify this expansion more formally by noticing that the transition probabilities @xmath268 are nearly symmetric near the bifurcation point @xmath269 . in the present case the latter inequality is a consequence of the relation @xmath270 in eq . ( [ eq : am ] ) , which leads to latexmath:[$\bigl|\left|a_{-m}(g)\right|-\left|a_{m}(g)\right|\bigl|\ll and @xmath166 can be obtained from eqs . ( [ eq : m ] ) and ( [ eq : ts ] ) , which show that the right - hand side of eq . ( [ eq : ts ] ) is @xmath272 for @xmath273 . since @xmath274 @xcite , we have from eqs . ( [ eq : period ] ) and ( [ eq : ts ] ) @xmath275 . it follows from the above arguments that near the bifurcation point @xmath207 is given by eq . ( [ eq : rc ] ) for arbitrary planck number @xmath228 , i.e. , for arbitrary temperature . since @xmath276 for @xmath265 , we obtain from eq . ( [ eq : rc ] ) a simple explicit expression @xmath277 eq . ( [ eq : rc0 ] ) agrees with eq . ( [ eq : rpa1 ] ) near @xmath278 in the limit @xmath265 . it shows that the inverse effective temperature @xmath254 is independent of @xmath9 . it monotonically decreases with increasing temperature @xmath246 . the function @xmath207 can be also obtained for @xmath41 . this is a consequence of detailed balance that emerges in this case @xcite . the detailed balance condition is usually a consequence of time reversibility , which does not characterize the dynamics of a periodically modulated oscillator . so , in the present case detailed balance comes from a special relation between the parameters for @xmath0 . detailed balance means that transitions back and forth between any two states are balanced . it is met if the ratio of the probabilities of direct transitions between two states is equal to the ratio of probabilities of transitions via an intermediate state @xmath279 one can see from eqs . ( [ eq : am ] ) and ( [ eq : wmn ] ) that the condition ( [ eq : detailed_balance_condition ] ) is indeed met for @xmath280 . therefore the balance equation has a solution @xmath281 , which immediately gives @xmath282 the function @xmath207 is plotted in fig . [ fig : randt ] for two values of the control parameter @xmath54 . it displays different behavior depending on the sign of @xmath54 . for @xmath283 the parameter of the elliptic function @xmath120 , eq . ( [ eq : m ] ) , is negative for all @xmath109 . therefore the periods @xmath284 of @xmath128 , @xmath127 , eq . ( [ eq : period ] ) , are smooth functions of @xmath9 , except near @xmath285 , where @xmath286 . this divergence of @xmath287 is seen in fig . [ fig : randt ] . the function @xmath207 remains finite near @xmath278 , eq . ( [ eq : rpa1 ] ) . and the imaginary part of the period @xmath122 of the oscillator coordinate and momentum @xmath288 for @xmath0 . for @xmath289 both @xmath206 and @xmath290 have a logarithmic singularity.,width=288 ] for @xmath291 , on the other hand , the function @xmath207 has a singularity . its location @xmath292 is determined by the condition @xmath293 , which gives @xmath294 . for small @xmath295 we have @xmath296 , whereas @xmath166 , eq . ( [ eq : ts ] ) remains finite for @xmath297 . therefore @xmath207 diverges logarithmically at @xmath292 , as seen from fig . [ fig : randt](b ) . physically the divergence of the inverse temperature is related to the structure of the transition probabilities @xmath190 for @xmath280 . for @xmath298 we have @xmath299 for @xmath214 , see eqs . ( [ eq : am ] ) and ( [ eq : wmn ] ) . this means that , for @xmath238 close to @xmath292 , diffusion towards larger @xmath9 slows down . the slowing down leads to a logarithmic singularity of @xmath207 . the eikonal approximation is inapplicable for @xmath9 close to @xmath292 . however , the width of the range of @xmath9 where this happens is small , @xmath300 , as follows from the discussion below eq . ( [ eq : rp ] ) . in addition , there are corrections to the balance equation due to off - diagonal terms in the full kinetic equation . these corrections give extra terms @xmath301 in the transition probabilities @xmath190 . the analysis of these corrections as well as features of @xmath302 that are not described by the eikonal approximation is beyond the scope of this paper because , as we show , these features are fragile . the function @xmath303 obtained by integrating eq . ( [ eq : r0 ] ) is compared in fig . [ fig : num](b ) with the result of the numerical solution of the balance equation ( [ eq : be ] ) with numerically calculated transition probabilities @xmath190 . the semiclassical and numerical results are in excellent agreement . we checked that the agreement persists for different values of the control parameter @xmath54 and for almost all @xmath304 , except for a few extremely narrow resonant bands of @xmath58 . quantum diffusion over quasienergy described by eq . ( [ eq : be ] ) leads to switching between the classically stable states of the oscillator at the minima of @xmath10 in fig . [ fig : g ] . the switching rate @xmath305 is determined by the probability to reach the top of the barrier of @xmath10 , that is by the distribution @xmath36 for such @xmath37 that @xmath306 . to logarithmic accuracy @xmath307 where @xmath207 is given by eq . ( [ eq : bep ] ) . the parameter @xmath308 is of the order of the relaxation rate @xmath176 due to coupling to a thermal bath . the quantity @xmath52 plays the role of the activation energy of escape . the activation is due to quantum fluctuations that accompany relaxation of the oscillator , and we call it quantum activation energy . as we show in sec . [ sec : tunneling ] , @xmath52 is smaller than the tunneling exponent for tunneling between the minima of @xmath10 . therefore if the relaxation rate @xmath176 exceeds the tunneling rate , switching between the stable states occurs via quantum activation . quantum activation energy @xmath52 obtained by solving eq . ( [ eq : bep ] ) numerically for @xmath206 is plotted in fig . [ fig : acte2 ] . it depends on the control parameter @xmath54 and the planck number @xmath228 , and it monotonically increases with increasing @xmath54 and decreasing @xmath228 . close to the bifurcation point @xmath262 , i.e. , for @xmath309 , it displays scaling behavior with @xmath310 . from eq . ( [ eq : rc0 ] ) @xmath311 the scaling exponent @xmath312 coincides with the scaling exponent near the bifurcation point of a classical parametric oscillator where the oscillator motion is still underdamped in the rotating frame , i.e. , @xmath310 is not to small @xcite . it can be seen from the results of refs . that the exponent @xmath312 also describes scaling of the activation energy of escape due to classical fluctuations closer to the pitchfork bifurcation point where the motion is necessarily overdamped . as a function of the scaled frequency detuning @xmath54 . the transition rate is @xmath313 . with increasing @xmath32 , the value of @xmath52 multiplied by @xmath314 quickly approaches the classical limit @xmath315 shown by the dashed line . in this limit the ratio @xmath316 is @xmath252 and does not contain @xmath26 . , width=240 ] in the classical limit @xmath317 we have from eq . ( [ eq : rc ] ) @xmath318 , i.e. , the switching rate @xmath319 $ ] , with temperature independent @xmath320 being the standard activation energy . the quantity @xmath321 in the classical limit as obtained from eq . ( [ eq : rc ] ) , is shown with the dashed line in fig . [ fig : t_scaled ] . it is seen that @xmath52 quickly approaches the classical limit with increasing planck number @xmath228 , so that even for @xmath322 the difference between @xmath321 and its classical limit is @xmath323 . effectively it means that , for small @xmath247 , the exponent in the probability of quantum activation can be approximated by the classical exponent for activated switching in which one should replace @xmath324 it turns out that the expression for the distribution ( [ eq : rho ] ) and ( [ eq : r0 ] ) found from the detailed balance condition for @xmath0 does not generally apply even for infinitesimally small but nonzero temperature , in the semiclassical limit . this is a consequence of this solution being of singular nature , in some sense . a periodically modulated oscillator should not have detailed balance , because the underlying time reversibility is broken ; the detailed balance condition ( [ eq : detailed_balance_condition ] ) is satisfied just for one value of @xmath32 and when other relaxation mechanisms are disregarded . formally , in a broad range of @xmath54 the correction @xmath325 to the @xmath0 solution diverges . the divergence can be seen from eqs . ( [ eq : ae ] ) , ( [ eq : wmn ] ) , ( [ eq : r0 ] ) . the transition probabilities @xmath326 have terms @xmath327 which vary with @xmath170 as @xmath328 $ ] for @xmath329 . at the same time , the @xmath0 solution ( [ eq : r0 ] ) gives @xmath330= \exp\left[2\,m\;{\textrm{im}}(\phi_0 - 2\phi_*)\right]$ ] . therefore for the series ( [ eq : bep ] ) with the term @xmath331 in @xmath326 to converge we have to have @xmath332 the condition ( [ eq : con ] ) is met at the bottom of the wells of @xmath10 and also close to the bifurcation values of the control parameter , @xmath309 . we found that , with increasing @xmath54 , the condition ( [ eq : con ] ) is broken first for @xmath9 approaching the barrier top , @xmath126 . in this region @xmath333 . a somewhat tedious calculation based on the properties of the elliptic functions shows that the condition ( [ eq : con ] ) is violated when @xmath334 , and for @xmath335 we have @xmath336 for @xmath126 . the increase in @xmath54 leads to an increase of the range of @xmath9 where eq . ( [ eq : con ] ) does not apply . the detailed balance distribution is inapplicable for @xmath44 in this range . to find the distribution for small planck number @xmath228 in the range where @xmath337 we seek the solution of the balance equation ( [ eq : bep ] ) in the form @xmath338\ , , \,\epsilon\ll 1\ , .\ ] ] this solution does not give diverging terms for @xmath339 . the terms @xmath340 $ ] in eq . ( [ eq : bep ] ) are @xmath341 for @xmath342 , and their sum over @xmath170 is @xmath343 . using the explicit expression for the transition probabilities we obtain that , to leading order in @xmath228 @xmath344\sinh\left(m\;{\textrm{im}}\,\phi_*\right)/ \left|\cos\left(m\;{\textrm{im}}\,\phi_0/2\right)\right|^2\right]^{-1}.\ ] ] here again , the sum goes over positive and negative @xmath170 . in obtaining eq . ( [ eq : eps ] ) we used the general expression ( [ eq : am ] ) for the coefficients @xmath148 in @xmath201 . in fact , the asymptotic expressions ( [ eq : ae ] ) for @xmath148 allow one to calculate the sum over @xmath170 in eq . ( [ eq : eps ] ) explicitly and the result provides a very good approximation for @xmath4 . equations ( [ eq : rapp ] ) and ( [ eq : eps ] ) give the effective inverse temperature @xmath254 and therefore the distribution @xmath36 as a whole for @xmath247 . it is clear that the @xmath345 limit is completely different from the detailed balance solution ( [ eq : r0 ] ) for @xmath280 , that is , the transition to the @xmath41 regime is nonanalytic . the difference between the @xmath280 and the @xmath345 solution is seen in figs . [ fig : fr ] and [ fig : acte2 ] . it follows from eq . ( [ eq : eps ] ) that the perturbation theory diverges for the value of @xmath9 where @xmath346 . at such @xmath9 the @xmath345 solution ( [ eq : rapp ] ) coincides with the @xmath280 solution ( [ eq : r0 ] ) . for smaller @xmath9 the condition ( [ eq : con ] ) is satisfied and the @xmath0 solution for @xmath207 applies ; the corrections to this solution are @xmath347 . the derivative of the effective inverse temperature @xmath230 over @xmath9 is discontinuous at the crossover between @xmath345 and @xmath280 solutions , as seen in fig . [ fig : fr ] . figure [ fig : fr ] illustrates also the smearing of the singularity of @xmath207 due to terms @xmath348 in @xmath184 , which is described by the numerical solution of the balance equation ( [ eq : bep ] ) . away from the crossover the analytical solutions provide a good approximation to numerical results . dephasing plays an important role in the dynamics of quantum systems . it comes from fluctuations of the transition frequency due to external noise or to coupling to a thermal reservoir . a simple mechanism is quasielastic scattering of excitations of the reservoir off the quantum system . since the scattering amplitude depends on the state of the system , the scattering leads to diffusion of the phase difference of different states . for an oscillator , dephasing has been carefully studied , both microscopically and phenomenologically , see refs . and papers cited therein . it leads to an extra term @xmath349 in the quantum kinetic equation ( [ eq : master ] ) , with @xmath350\right],\ ] ] where @xmath351 is the dimensionless dephasing rate . if both @xmath352 and @xmath176 are small compared to @xmath353 , populations of the steady states @xmath36 are described by the balance equation ( [ eq : be ] ) in which one should replace @xmath354 , with @xmath355 it follows from eq . ( [ eq : ae ] ) that , for large @xmath356 , we have @xmath357 $ ] , that is , the transition probability exponentially decays with increasing @xmath356 , and the exponent is determined by @xmath358 . even slow dephasing is sufficient for making the detailed balance condition inapplicable . mathematically , the effect of slow dephasing is similar to the effect of nonzero temperature . if the condition ( [ eq : con ] ) is violated , the sum @xmath359 with @xmath207 given by the detailed balance solution ( [ eq : r0 ] ) diverges . the correct distribution for the appropriate @xmath9 and @xmath54 is given by eq . ( [ eq : rapp ] ) . the parameter @xmath360 is given by eq . ( [ eq : eps ] ) in which @xmath228 is replaced , @xmath361\right|^2.\end{aligned}\ ] ] it is seen from eq . ( [ eq : eph ] ) and figs . [ fig : fr ] and [ fig : acte2 ] that , for low temperatures , even weak dephasing , @xmath362 , leads to a very strong change of the probability distribution . the fragility of the detailed balance solution discussed in this section is a semiclassical effect . it occurs if the number of states in each well @xmath363 is large . formally we need @xmath364\gtrsim 1 $ ] , with @xmath365 . in other words , the detailed balance solution is fragile provided @xmath58 is sufficiently small . a full numerical solution of the balance equation confirms that , when @xmath58 is no longer small , the distribution for small @xmath366 is close to that described by the @xmath367 . the oscillator localized initially in one of the wells of the effective hamiltonian @xmath10 can switch to another well via tunneling . for small @xmath58 tunneling can be described in the wkb approximation . we will first find the tunneling exponent assuming that the oscillator is in the lowest intrawell state and show that it exceeds the quantum activation exponent @xmath316 . we will then use standard arguments to show that this is true independent of the initially occupied intrawell state , for all temperatures . if we now go back to the original problem of switching between stable states of a periodically modulated oscillator , we see that switching via tunneling can be observed only where the relaxation rate is smaller than the tunneling rate , that is the prefactor in the rate of quantum activation is very small . such experiment requires preparing the system in one of the wells . tunneling between period two states of a strongly modulated strongly nonlinear system has been demonstrated in atomic optics @xcite . for a weakly nonlinear oscillator , tunneling splitting between the lowest states of the hamiltonian @xmath188 for @xmath12 was found in ref . @xcite . the effective hamiltonian @xmath10 is quartic in the momentum @xmath71 . therefore the semiclassical momentum @xmath368 as given by the equation @xmath369 has 4 rather than 2 branches . this leads to new features of tunneling compared to the standard picture for one - dimensional systems with a time - independent hamiltonian quadratic in @xmath71 . for concreteness , we will consider tunneling from the left well , which is located at @xmath370 . the rate of tunneling from the bottom of the well of @xmath10 in fig . [ fig : g ] is determined by the wkb wave function with @xmath371 . this wave function is particularly simple for @xmath372 . in the region @xmath373 and not too close to @xmath374 it has the form @xmath375,\nonumber\\ & & s_0(q)=\int\nolimits_{q_{l0}}^q\,p_-(q')\,dq',\end{aligned}\ ] ] where @xmath376.\ ] ] for @xmath283 the wave function @xmath377 monotonically decays with increasing @xmath72 . the exponent for interwell tunneling is @xmath378 , with @xmath379 ( we use that @xmath380 is the position of the right well ) , @xmath381 a more interesting situation arises in the case @xmath382 . here , inside the classically forbidden region @xmath383 decay of the wave function is accompanied by oscillations . we leave the analysis of this behavior for a separate paper . here we only note that the tunneling exponent is still given by eq . ( [ eq : s_tun ] ) . in the tunneling probability as a function of the parameter @xmath384 ( solid line ) . also shown for comparison is the quantum activation energy @xmath52 for @xmath385 and @xmath250.,width=288 ] if the level splitting due to tunneling is small compared to their broadening due to relaxation , the tunneling probability is quadratic in the tunneling amplitude , @xmath386 the action @xmath387 as a function of @xmath54 is plotted in fig . [ fig : tunneling ] . it is seen from this figure that the tunneling exponent exceeds the quantum activation exponent @xmath316 for all values of the control parameter @xmath54 . this indicates that , as mentioned above , it is exponentially more probable to switch between the classically stable states of the oscillator via activation than via tunneling from the ground intrawell state , for not too small relaxation rate . in the same limit where the relaxation rate exceeds the tunneling rate , we can consider the effect of tunneling from excited intrawell states of the hamiltonian @xmath188 . the analysis is similar to that for systems in thermal equilibrium @xcite . over the relaxation time there is formed a quasiequilibrium distribution over the states inside the initially occupied well of @xmath10 . as before , we assume that relaxational broadening of the levels @xmath238 is small compared to the level spacing . the probability of tunneling from a given state @xmath37 is determined by its occupation @xmath36 . the overall switching probability is given by a sum of tunneling probabilities from individual intrawell states @xmath388 where @xmath389 is a prefactor that smoothly depends on @xmath37 , and @xmath390 is the imaginary part of the action of a classical particle with hamiltonian @xmath10 and energy @xmath238 , which moves in complex time from the turning point @xmath83 in one well of @xmath10 to the turning point in the other well . it follows from the analysis in appendix a and eq . ( [ eq : pe ] ) , that the duration of such motion is @xmath391 . therefore @xmath392 . taking into account that @xmath393 , we see that the derivative of the overall exponent in eq . ( [ eq : multilevel_w ] ) over @xmath238 is @xmath394 . it was shown in secs . [ sec : distribution ] and [ sec : limiting_cases ] that this derivative is always positive , cf . [ fig : randt ] . therefore the exponent monotonically increases ( decreases in absolute value ) with increasing @xmath9 . this shows that , with overwhelming probability , switching occurs via overbarrier transitions , i.e. , the switching mechanism is quantum activation . we emphasize that this result is independent of the bath temperature . in this paper we studied switching between the states of period two vibrations of a parametrically modulated nonlinear oscillator and the distribution over period - two quasienergy levels . we considered a semiclassical case where the wells of the scaled oscillator hamiltonian in the rotating frame @xmath10 , fig . [ fig : g ] , contain many levels . the distance between the levels is small compared to @xmath395 but much larger than the tunneling splitting . we assumed that the oscillator is underdamped , so that the interlevel distance exceeds their width . at the same time , of primary interest was the case where this width exceeds the tunneling splitting . we considered relaxation due to coupling to a thermal bath , which is linear in the oscillator coordinate , as well as dephasing from random noise that modulates the oscillator frequency . the problem of the distribution over intrawell states was reduced to a balance equation . the coefficients in this equation were obtained explicitly in the wkb approximation , using the analytical properties of the solution of the classical equations of motion . the balance equation was then solved in the eikonal approximation . the eikonal solution was confirmed by a full numerical solution of the balance equation , which did not use the wkb approximation for transition matrix elements . we found that the distribution over period - two quasienergy has a form of the boltzmann distribution with effective temperature that depends on the quasienergy . this temperature remains nonzero even where the temperature of the thermal bath @xmath1 . it is determined by diffusion over quasienergy , which accompanies relaxation and has quantum origin : it is due to the floquet wave functions being combinations of the fock wave functions of the oscillator . unexpectedly , we found that the quasienergy distribution for @xmath0 , where the system has detailed balance , is fragile . it differs significantly from the solution for @xmath1 , for a large number of intrawell states . the @xmath0 solution is also destroyed by even small dephasing . the probability of switching between period two states @xmath47 is determined by the occupation of the states near the barrier top of the effective hamiltonian @xmath10 . we calculated the effective quantum activation energy @xmath52 which gives @xmath396 . both @xmath316 and the exponent of the tunneling probability are proportional to the reciprocal scaled planck constant @xmath397 . however , for all parameter values and all bath temperatures , @xmath316 is smaller than the tunneling exponent . therefore in the case where intrawell relaxation is faster than interwell tunneling , switching occurs via quantum activation . in the limit where fluctuations of the oscillator are classical , @xmath398 , we have @xmath399 , and @xmath47 is described by the standard activation law . down to small planck numbers @xmath400 the quantum activation energy @xmath52 is reasonably well described by the classical expression even for small @xmath401 provided @xmath402 is replaced by @xmath403 , with @xmath32 being the planck number of the oscillator . the inapplicability of this description for small @xmath32 indicates , however , that classical and quantum fluctuations do not simply add up . the replacement @xmath404 becomes exact for all @xmath32 close to the bifurcational value of the control parameter @xmath405 where the period - two states first emerge . in this range @xmath52 scales with the distance to the bifurcation point as @xmath406 . the results on switching rate are accessible to direct experimental studies in currently studied nano- and microsystems , in particular in systems based on josephson junctions , including those used for highly sensitive quantum measurements . we gratefully acknowledge a discussion with m. devoret . this work was partly supported by the nsf through grant no . the hamiltonian @xmath10 ( [ eq : g ] ) is quartic in @xmath71 and @xmath72 . this makes it possible to solve classical equations of motion ( [ eq : dqpdt ] ) . trajectories with given @xmath9 lie on the cross - section @xmath369 of the surface @xmath10 in fig . [ fig : g ] . we will be interested only in the intrawell trajectories , in which case @xmath407 . the trajectories in different wells are inversely symmetrical and can be obtained by the transformation @xmath408 , @xmath409 . their time dependence can be expressed in terms of the jacobi elliptic functions @xcite . for trajectories in the right well in fig . [ fig : g ] , where @xmath151 , we have @xmath410 here @xmath411 the parameter of the elliptic functions @xmath412 is given by eq . ( [ eq : m ] ) . for @xmath283 the function @xmath413 monotonically decreases with increasing @xmath9 from @xmath293 for @xmath414 to @xmath415 for @xmath416 . for @xmath291 the function @xmath413 becomes nonmonotonic . it first increases from @xmath293 with increasing @xmath9 , but than decreases , goes trough @xmath293 for @xmath417 , and goes to @xmath216 for @xmath126 . the jacobi functions @xmath418 , @xmath419 and @xmath420 for @xmath421 are equal to @xmath422 , @xmath423 , and @xmath424 with @xmath425 and @xmath426 @xcite . the double periodicity of the functions @xmath127 , @xmath128 discussed in sec . [ sec : gspec ] is a consequence of the double periodicity of elliptic functions . the expressions for the periods @xmath427 , eq . ( [ eq : period ] ) , follow from eq . ( [ eq : qpt ] ) . the trajectories in the left well of @xmath10 in fig . [ fig : g ] , where @xmath428 , can be written in the form @xmath429 this expression shows how to make a transition from one well to another by moving in complex time , which simplifies the analysis of oscillator tunneling . we note that the function @xmath430 in the expressions for @xmath431 ( [ eq : qpt ] ) has periods @xmath111 , @xmath432 @xcite as a function of @xmath433 . it s period parallelogram is shown in fig . [ fig : cl ] . in this parallelogram @xmath430 takes on any value twice . therefore both @xmath127 and @xmath128 have two poles located at @xmath434 . the values of @xmath435 are given by the equation @xmath436 the positions of the poles are shown in fig . [ fig : cl ] for @xmath437 and @xmath421 , respectively , with @xmath438 . for concreteness we choose @xmath439 using the relations between the jacobi elliptic functions @xcite we find that , near the pole at @xmath440 , @xmath441 whereas @xmath442 is not singular at @xmath443 . ( [ eq : pole_a ] ) was used to obtain the explicit form of the matrix element of the operator @xmath145 in sec . in this appendix we calculate the effective inverse temperature @xmath207 in the limit of large oscillator planck number @xmath444 . the explicit form of the coefficients in eq . ( [ eq : cw ] ) for @xmath206 follows from the general expression ( [ eq : wmn ] ) for the transition rates @xmath184 , @xmath445 the semiclassical matrix elements @xmath446 are given by eq . ( [ eq : a ] ) . they are fourier components of the function @xmath153 on the classical trajectory with given @xmath447 . using the completeness condition @xmath448 we can rewrite @xmath449 with @xmath450 and @xmath451 . the integrals over @xmath144 can be written as contour integrals over @xmath452 , @xmath453 along the trajectories with given @xmath9 . the contour integrals can be further simplified using the stokes theorem . this gives @xmath454 where the functions @xmath455 and @xmath456 are given by eq . ( [ eq : rc ] ) .

in this paper , we present photometry for young star clusters in m31 , which are selected from caldwell et al . these star clusters have been observed as part of the beijing arizona taiwan connecticut ( batc ) multicolor sky survey from 1995 february to 2008 march . the batc images including these star clusters are taken with 15 intermediate - band filters covering 300010000 . combined with photometry in the _ galex _ far- and near - ultraviolet , broad - band , sdss , and infrared of two micron all sky survey , we obtain their accurate spectral energy distributions ( seds ) from . we derive these star clusters ages and masses by comparing their seds with stellar population synthesis models . our results are in good agreement with previous determinations . the mean value of age and mass of young clusters ( gyr ) is about 385 myr and , respectively . there are two distinct peaks in the age distribution , a highest peak at age 60 myr and a secondary peak around 250 myr , while the mass distribution shows a single peak around . a few young star clusters have two - body relaxation times greater than their ages , indicating that those clusters have not been well dynamically relaxed and therefore have not established the thermal equilibrium . there are several regions showing aggregations of young star clusters around the 10 kpc ring and the outer ring , indicating that the distribution of the young star clusters is well correlated with m31 s star - forming regions . the young massive star clusters ( age myr and mass ) show apparent concentration around the ring splitting region , suggesting a recent passage of a satellite galaxy ( m32 ) through m31 disk .

1208.6419

star clusters are considered as important tracers for understanding the formation and evolution of their host galaxies @xcite . star cluster systems have been traditionally separated into two populations globular clusters and open clusters ( gcs and ocs)on their ages , masses , metallicities , and positions . however , more recent studies have discovered that the distinction between gcs and ocs becomes increasingly blurred ( see * ? ? ? * for details ) . @xcite listed photometric colors and magnitudes for star clusters in magellanic clouds ( mcs ) and the fornax dwarf system and divided them into two groups . they found that star clusters in blue group have central condensation properties similar to those of the red group , which were considered as gcs , however they could not be identified with the galactic ocs . @xcite termed 23 clusters in the large magellanic cloud ( lmc)differing from gcs in their relative youth and ocs in their richness and shape as `` young populous clusters '' , which were called `` young massive clusters '' ( ymcs ) or `` blue luminous compact clusters '' ( blccs ) by @xcite . actually , the blue integrated colors for a cluster may be influenced by several factors , such as poor metallicity ( the luminosity of the horizontal branch ) , young age ( the position of the main - sequence turnoff stars ) , and some exotic stellar populations ( e.g. , blue stragglers , wolf - rayet stars ) . however , several studies ( e.g. , * ? ? ? * ; * ? ? ? * ) have reached similar conclusions that the exceedingly blue colors of blccs are a direct consequence of their young ages ( see * ? ? ? * for details ) . m31 is the largest galaxy in the local group , and has a large number of star clusters , including young clusters having been studied by many authors . @xcite listed 11 objects in m31 classified as blue clusters using the uv colors , most of which have been proved to be young clusters @xcite , except for b133 and b145 , which were stated as a star and an old gc @xcite , respectively . @xcite derived ages and masses for a large sample of young clusters , and found that these star clusters are less than 2 gyr old , and most of them have ages between @xmath10 and @xmath11 yr and masses ranging from @xmath12 to @xmath13 . these authors also stated that the young star clusters in m31 show a range of structures , most of which have low concentrations typical of ocs in the milky way ( mw ) , however , there are a few with high concentrations similar to the mw gcs . @xcite carried out a survey of compact star clusters in the southwest part of m31 , and suggested a rich intermediate - mass star cluster population in m31 , with a typical age range of 30 myr @xmath14 3 gyr , peaking at @xmath6 70 myr . in order to ascertain the properties of the blccs , @xcite performed an image survey for 20 blccs lying in the disk of m31 using the wide field and planetary camera-2 ( wfpc2 ) on the _ hubble space telescope _ ( _ hst _ ) . in addition , another key aim of this _ hst _ survey was to determine the fraction of contamination of blccs by asterisms , since @xcite suggested that a large fraction of the putative blccs may in fact be just asterisms . @xcite presented the resulting @xmath15 images of six very young or young star clusters in m31 observed with the keck laser guide star adaptive optics system , and indicated that the four youngest out of these six objects are asterisms . however , @xcite presented a conclusion that these four objects are true clusters based on spectra . the _ hst _ images @xcite showed that nineteen of the twenty surveyed candidates are real star clusters , and one ( nb67 ) is a bright star . @xcite measured surface brightness profiles for 23 bright young star clusters using images from the wfpc2 , including the sample clusters of @xcite , and derived the structural properties by fitting the surface brightness profiles to several structural models . the authors stated that the sample young clusters are expected to dissolve within a few gyr and will not survive to become old gcs , and that young star clusters in m31 and mcs follow the same fundamental plane relations as old gcs of m31 , mcs , the mw and ngc 5128 , regardless of their host galaxy environments . @xcite presented a m31 stellar cluster catalog utilizing the panchromatic hubble andromeda treasury survey data , which will cover @xmath16 of m31 disk with multiple filters and allow the identification of thousands of star clusters . the large population of young star clusters reflect a high efficiency of cluster formation , possibly triggered by a current interaction event between m31 and its satellite galaxy @xcite , suggesting that young star clusters should be associated with the star - forming ( sf ) regions of m31 . @xcite found that young clusters ( @xmath17 2 gyr ) are spatially coincident with m31 s disk , including the 10 kpc ring and the outer ring @xcite . although these authors also found the young star clusters in the halo of m31 , all of the clusters outside of the optical disk of m31 are old , globular clusters ( see * ? ? ? * for details ) . @xcite stated that most of young star clusters kinematics have the thin , rotating disk component ( see also * ? ? ? * ) . the young star clusters distribution has a distinct peak around @xmath18 kpc from the center in m31 disk , and some young star clusters show concentration around the 10 kpc ring splitting regions near m32 and most of them have systematically younger ages ( @xmath19 myr ) . @xcite also stated that the young star clusters show a spatial distribution similar to ob stars , uv sf regions , and dust , all of which are important tracers of disk structures . several criteria were developed for selecting young clusters from the integrated spectrum and colors . @xcite comprehensively studied the properties of 67 very blue and likely ymcs in m31 selected according to their color @xmath20 $ ] and/or the strength of @xmath21 spectral index ( @xmath22 ) . @xcite presented a catalog of m31 gcs based on images from the sdss and the wide field camera on the united kingdom infrared telescope and selected a population of young clusters with a definition of @xmath23 $ ] . @xcite published a catalog of m31 young clusters ( @xmath24 gyr ) and supported the selection criteria @xmath25 $ ] and @xmath26 $ ] @xcite . these criterions may play important roles in distinguishing young from old clusters for those whose ages can not be derived accurately . the formation and disruption of young star clusters represent a latter - day example of the hierarchical formation of galaxies @xcite . motivated by that , we decided to describe some basic properties of young star clusters in m31 , such as positions , distributions of ages and masses , correlations of the ages and masses with structure parameters , which may provide important information about the processes involved in their formation and disruption . in this paper , we will provide photometry of a set of young star clusters in m31 using images obtained with the beijing arizona taiwan connecticut ( batc ) multicolor sky survey telescope . by comparing the observed seds with the galev simple stellar population ( ssp ) models , we derive their ages and masses . this paper is organized as follows . in section 2 we present the batc observations of the sample clusters , the relevant data - processing steps , and the _ galex _ ( fuv and nuv ) , optical broad - band , sdss @xmath1 and 2mass nir data that are subsequently used in our analysis . in section 3 we derive ages and masses of the sample clusters . a discussion on the sample young clusters ( @xmath17 2 gyr ) will be given in section 4 . finally , we will summarize our results in section 5 . the sample of star clusters in this paper is selected from @xcite , who presented a series of studies of m31 young and old clusters , respectively . we selected 178 young clusters given in @xcite , and fortunately , all the young clusters have been observed with 15 intermediate - band filters of the batc photometric system . however , there are 42 clusters for which we can not obtain accurate photometric measurements with different reasons as following , a ) some clusters have one or more nearby very bright objects ; b ) some clusters are very close to other objects ; c ) some clusters are very faint and the signal - to - noise ratio ( snr ) is low ; d ) some clusters are superimposed onto a bright background ; e ) some clusters are superimposed onto a strongly variable background . in addition , there are several remarkable clusters with `` adhered '' @xcite objects in our images , such as m088 and its neighbor m089 , and g099 and c037-g099x ( see also * ? ? ? in a previous paper , @xcite presented the seds in 15 intermediate - band filters of the batc photometric system for one ymc vdb0-b195d and determined its age and mass by comparing its seds with the theoretical evolutionary population synthesis models . thus , here we will analyze the multicolor photometric properties of the remaining 135 clusters . the m31 field is part of a galaxy calibration program of batc multicolor sky survey . the batc program uses the 60/90 cm schmidt telescope at the xinglong station of the national astronomical observatories , chinese academy of sciences ( naoc ) . this system includes 15 intermediate - band filters , covering a range of wavelength from 3000 to 10000 ( see * ? ? ? * for details ) . before february 2006 , a ford aerospace @xmath27 thick ccd camera was applied , which has a pixel size of 15 @xmath28 and a field of view of @xmath29 , resulting in a resolution of @xmath30 . after february 2006 , a new @xmath31 ccd with a pixel size of 12 @xmath32 m was used , with a resolution of @xmath33 pixel@xmath34 @xcite . we obtained 143.9 hours of imaging of the m31 field covering about 6 square degrees , consisting of 447 images , through the set of 15 filters in five observing runs from 1995 to 2008 , spanning 13 years ( see * ? ? ? * ; * ? ? ? * for details ) . figure 1 shows the spatial distribution of the sample clusters and the m31 fields observed with the batc multicolor system , in which a box only indicates a field view of @xmath35 @xmath36 @xmath35 of the thick ccd camera . all the sample star clusters are indicated with black dots , with high - accuracy coordinates from @xcite , which are based on the images from the local group galaxies survey @xcite and the digitized sky survey . we processed all the ccd images to apply standard procedures including bias subtraction and flat - fielding using an automatic data reduction software named pipeline i , developed for the batc multicolor sky survey of the ccd images . batc magnitudes are defined and obtained in a similar way as for the spectrophotometric ab magnitude system @xcite . in order to improve the image quality , multiple images of the same filter were combined to one , on which the magnitudes of the sample star clusters were determined . the absolute flux of the combined images in the central field of m31 ( m31 - 1 in figure 1 ) was calibrated using observations of standard stars , while the absolute flux of the combined images of the m31 - 2 to m31 - 7 fields was calibrated based on secondary standard transformations using the m31 - 1 field ( see * ? ? ? * for details ) . we performed standard aperture photometry of our sample objects using the phot routine in daophot @xcite . to ensure that we adopted the most appropriate photometric radius that included all light from the objects , we used 9 different aperture sizes ( with radii of @xmath37 pixels for the old ccd , while for the new ccd , the radii were given with pixels corresponding to the same arcsecs on the old ccd ) to determine the magnitude . we also checked the aperture radii carefully on the images by visual examination in order not to include the light from extraneous objects . the local sky background was measured in an annulus with an inner radius of @xmath38 pixel and a width of 4.0 pixels for the old ccd , and with an inner radius of @xmath38 pixel and a width of 5.0 pixels for the new ccd , respectively . there are 40 clusters in this paper which are in common with our series of previous papers @xcite , and the photometric data of these 40 clusters were also derived by these studies . we found that most photometric data obtained here are in good agreement with those obtained by previous studies . we checked the images of those clusters with photometric discrepancy , and found most of them were loaded near the bulge ( b091 ) or in some disk regions with bright or variable background ( e.g. , b210 , m020 , m023 ) . the different choice of the aperture for photometry and the annulus for background caused the discrepancy . the seds for the sample clusters in m31 are listed in table 1 . columns ( 1 ) gives the cluster names . columns ( 2 ) to ( 16 ) present the magnitudes in the 15 batc passbands . the @xmath39 magnitude uncertainties from daophot are listed for each object on the second line for the corresponding passbands . for some objects , the magnitudes in some filters could not be obtained because of low snr . we should remind that magnitudes with an uncertainty larger than 0.3 will not be used in the following analysis , although they are listed in table 1 . columns ( 17 ) is the photometric aperture adopted in this paper . as our series of papers has pointed out , accurate and numerous photometric points can derive accurate ages of star clusters @xcite . @xcite stated that the uv photometry is powerful for age estimation of young stellar populations , and the combination of uv photometry with optical observations enables one to break the age - metallicity degeneracy . @xcite and @xcite showed that the age - metallicity degeneracy can be partially broken by adding nir photometry to optical colors ( see * ? ? ? * and references therein ) . several previous studies @xcite have provided magnitudes for star clusters in different passbands , which will be used to estimate ages of sample star clusters in this paper . the latest revised bologna catalogue of m31 gcs and candidates ( hereafter rbc v.4 ) @xcite includes _ galex _ ( fuv and nuv ) fluxes from @xcite , optical broad - band , 2mass nir magnitudes for 2045 objects . for @xmath0 magnitudes given in rbc v.4 , the relevant photometric uncertainties are not listed . therefore , we adopted the original @xmath0 measurements of @xcite as our preferred reference , including their published photometric errors . for the remaining objects , the @xmath0 magnitudes from rbc v.4 were adopted , with the photometric uncertainties set following @xcite , i.e. , @xmath40 mag in @xmath41 and @xmath42 mag in @xmath43 @xcite . in rbc v.4 , the 2mass @xmath2 magnitudes were transformed to cit photometric system @xcite . however , we needed the original 2mass @xmath2 data to compare the observed seds with the ssp models , so we reversed the transformation using the equations given by @xcite . there were no magnitude errors for @xmath44 bands in rbc v.4 , and we obtained them by comparing the photometric data with figure 2 of @xcite , in which the photometric error was shown as a function of magnitude for stars brighter than their observational completeness limits @xcite . in addition , since rbc v.4 provided @xmath2 magnitudes only for a small number of sample star clusters , we adopted @xmath2 magnitudes from the 2mass-6x - psc catalog , with 6 times the normal exposure of 7.2 s on most fields of m31 @xcite . there are 3 kinds of magnitudes given by 2mass-6x - psc catalog , the `` default '' magnitude , the @xmath45 aperture and @xmath46 aperture magnitudes . we found that the @xmath45 aperture magnitudes agree well with the magnitudes in rbc v.4 , while the other 2 kinds of magnitudes have large discrepancy with magnitudes in rbc v.4 . however , for the @xmath45 aperture magnitudes , when @xmath2 magnitudes are fainter than @xmath47 mag , the dispersions are considerable . in this paper , we preferentially adopted 2mass @xmath2 magnitudes in rbc v.4 . for the remaining star clusters , we adopted the @xmath45 aperture magnitudes in the 2mass-6x - psc catalog when the magnitudes brighter than @xmath48 mag . @xcite performed sdss @xmath1 photometry for 1595 m31 clusters and cluster candidates using the program sextractor on drift scan images of m31 obtained by the sdss 2.5-m telescope , which are on the ab photometric system ( see * ? ? ? * for details ) . we found that there was a magnitude offset ( @xmath49 ) between sdss and @xmath0 for 35 objects by transforming between the @xmath1 and @xmath0 bands following the transformations from @xcite . because the sdss @xmath1 magnitudes provided a more homogeneous set of photometric measurements , we adopted the sdss magnitudes and abandoned @xmath0 magnitudes for these objects in the following analysis . these 35 objects are flagged with small `` a '' in column 1 of table 2 . @xcite presented a catalog of 700 confirmed star clusters in m31 , providing the most extensive and updated uv integrated photometry on the ab photometric system based on _ galex _ imaging , superseding the uv photometry published by @xcite , which were included by rbc v.4 . therefore , we used the magnitudes of the fuv and nuv from @xcite as the uv photometry in our following sed fitting process . we listed the _ galex _ , optical broad - band , sdss @xmath1 , and 2mass nir photometry of the sample clusters in table 2 ( columns 2 to 16 ) , where the photometric errors are listed for each object on the second line for the corresponding passbands . as we discussed above , the magnitudes with an uncertainty larger than 0.3 will not be used in the following analysis . to check our photometry , we transformed the batc intermediate - band system to the broad - band system using the relationships between these two systems derived by @xcite : @xmath50 @xmath51 @xmath52-band photometry can be derived from the batc @xmath53 , and @xmath54 bands , while @xmath55-band magnitude can be obtained from the batc @xmath56 , and @xmath57 bands . figure 2 shows a comparison of the @xmath52 and @xmath55 photometry of our m31 sample objects with previous measurements from @xcite ( circles ) and @xcite ( triangles ) . there are several objects with larger offsets ( @xmath58 ) , shown with black solid marks in figure 2 ( m045 in the top panel ; b200d , sk036a , and sk068a in the bottom panel ) . the snrs of m045 , b200d , and sk036a are low , and sk068a is superimposed onto a bright background , thus we can not derive accurate photometries for these four star clusters . the mean @xmath52 and @xmath55 magnitude differences in the sense of this paper minus others are @xmath59 mag and @xmath60 , i.e. , there is no system offset between our magnitudes and previous determinations . we required independently determined reddening values to estimate ages of the sample clusters robustly and accurately . here we used @xcite and @xcite as our reference . @xcite derived reddening values from three ways : 1 ) mean reddening values from available literature @xcite ; 2 ) median reddening values of star clusters located within an annulus at each 2 kpc radius from the center of m31 , for these star clusters there were no available reddening values in the literature ; 3 ) for star clusters at distances larger than 22 kpc from the center of m31 , the foreground reddening value of @xmath61 was adopted . because all of our sample clusters have a projected galactocentric radius smaller than 22 kpc , the reddening values for them were derived from the first two methods . the reddening value of cluster lgs04131.1_404612 was not given by @xcite , so we adopted the value @xmath62 from @xcite . its reddening uncertainty was simply adopted half of the reddening value , i.e. , @xmath63 . we noticed that , for star cluster b449 , the reddening value determined by @xcite is very different with the value determined by @xcite ( @xmath64 ) . we treated both age and reddening value as free parameters , and determined the reddening value to be @xmath65 which was in good agreement with @xcite . so , in this paper , we adopted @xmath61 in @xcite . the reddening uncertainty for b449 was adopted to be 0.07 . column 4 of table 4 lists the reddening values adopted for the sample clusters , while column 5 lists the methods for deriving the reddening values ( @xmath66 and @xmath67 indicate that the reddening values were obtained by the first and second method in kang et al . 2012 , respectively ; @xmath68 indicates that the reddening values were from caldwell et al . 2009 , 2011 , only for lgs04131.1_404612 and b449 ) . in order to determine the ages and masses of the sample star clusters , we compared their seds with theoretical stellar population synthesis ( sps ) models . the ssp models of galev ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) were adopted @xcite in this paper , which are based on the padova isochrones ( with the most recent versions using the updated bertelli 1994 isochrones , including the thermally - pulsing asymptotic giant - branch [ tp - agb ] phase ) , and a @xcite stellar initial mass function ( imf ) with a lower - mass limit of @xmath69 and the upper - mass limit between 50 and 70 @xmath70 depending on metallicity . the full set of models span the wavelength range from 91 to 160 @xmath32 m . these models cover ages from @xmath71 to @xmath72 yr , with an age resolution of 4 myr for ages up to 2.35 gyr , and 20 myr for greater ages . the galev ssp models include five initial metallicities , @xmath73 ( solar metallicity ) , and 0.05 . since our observational data were integrated luminosities through our set of filters , we convolved the galev ssp seds with the _ galex _ fuv and nuv , broad - band @xmath0 , sdss @xmath1 , batc , and 2mass @xmath2 filter response curves to obtain synthetic ultraviolet , optical , and nir photometry for comparison . the synthetic @xmath74 filter magnitude in the ab magnitude system can be computed as @xmath75 where @xmath76 is the theoretical sed and @xmath77 is the response curve of the @xmath74 filter of the corresponding photometric systems . here , @xmath76 varies with age and metallicity . we used a @xmath78 minimization test to determine which galev ssp models are most compatible with the observed seds , following @xmath79 ^ 2}{\sigma_{i}^{2}}},\ ] ] where @xmath80 is the integrated magnitude in the @xmath74 filter of a theoretical ssp at metallicity @xmath81 and age @xmath82 , @xmath83 is the number of the filters used for fitting , @xmath84 represents the intrinsic integrated magnitude in the same filter and @xmath85 here , @xmath86 is the observational uncertainty , and @xmath87 is the uncertainty associated with the model itself , for the @xmath74 filter . @xcite estimated the uncertainty associated with the term @xmath87 by comparing the colors obtained from different stellar evolutionary tracks and spectral libraries . following @xcite and @xcite , we adopted @xmath88 mag in this paper . @xmath89 is the uncertainty in the reddening value , and @xmath90 , where @xmath91 is taken from @xcite , @xmath92 , and @xmath93 is the uncertainty of the distance modulus , which is always 0.07 from @xmath94 mag @xcite . @xcite determined ages for 20 possible ymcs in m31 with metallicity as a free parameter of their fit . their results showed that most ymcs in m31 were best fitted with solar metallicity model . @xcite claimed that it seemed likely the young star clusters have supersolar abundances . in this paper , the galev models of solar metallicity ( @xmath95 ) were used to fit the intrinsic seds for all the sample clusters here . as an example , we presented the fitting for some sample clusters in figure 3 . during the fitting , we found that , for a small number of clusters , some photometric data can not be fitted with any ssp models . we therefore did not use these deviating photometric data points to obtain the best fits . these deviating photometric data points are the @xmath96 magnitude of m070 , @xmath97 magnitude of m040 , @xmath98 magnitude of v133 , @xmath99 magnitude of m082 and m101 , @xmath83 and @xmath100 magnitudes of b195 , @xmath101 magnitude of b118d , and @xmath102 and @xmath103 magnitudes of m091 . we also noticed that , for some star clusters ( b091 , b305 , b319 , b392 , b458 , b480 , b484 , and dao69 ) , the photometries in @xmath2 bands show obvious offsets from _ fuv and nuv bands ( see also * ? ? ? * ) . considering that the uv photometry is powerful for age estimation of young stellar populations @xcite , we would adopt the _ galex _ fuv and nuv photometries in the sed fitting , and abandon the @xmath2 magnitudes for these clusters . the masses of the sample star clusters were determined sequentially . the galev models provide absolute magnitudes in 77 filters for ssps of @xmath104 , including 66 filters of the _ hst _ , johnson @xmath0 @xcite , cousins @xmath105 @xcite , and @xmath106 @xcite systems . the difference between the intrinsic and model absolute magnitudes provides a direct measurement of the cluster mass , in units of @xmath104 ( see * ? ? ? * for details ) . we transformed the 2mass @xmath2 magnitudes to the photometric system of @xcite using the equations given by @xcite , and estimated masses of the clusters using magnitudes in all of the @xmath0 and @xmath2 bands . the masses of clusters obtained based on the magnitudes in different filters were different , therefore , we averaged them as the final cluster mass . the masses of 22 clusters were not derived , because the @xmath107 magnitudes can not be used : ( 1 ) there are no @xmath107 magnitudes ( lgs04131.1_404612 ) ; ( 2 ) there are no @xmath2 magnitudes and the @xmath0 magnitudes were abandoned because of the discrepancy with the sdss @xmath1 magnitudes ( e.g. , b195 ) ; ( 3 ) the @xmath0 magnitudes were abandoned because of the discrepancy with the sdss @xmath1 magnitudes and the @xmath2 magnitudes were abandoned because of the discrepancy with the _ galex _ uv magnitudes ( e.g. , b319 ) . the ages and masses of the sample clusters obtained in this paper are listed in table 4 . in this paper , we determined ages and masses for 135 star clusters by comparing their multicolor photometries with theoretical sps models . these star clusters were from @xcite , who presented a series of studies of m31 young and old clusters , respectively . as discussed in section 2.3 , there are 40 clusters in this paper which are in common with our series of previous papers , and the ages of 27 clusters were also derived by our series of studies @xcite . in the study of @xcite , the ssp models of bruzual & charlot ( g. bruzual & charlot 1996 , unpublished ) were used ; in the studies of @xcite and @xcite , the ssp models of @xcite were used . in addition , in these three studies , we used only the batc photometries in 13 passbands , and did not include uv data , which is powerful tool for age estimation of young stellar populations ( see * ? ? ? * and reference therein ) . so , we re - estimated the ages for these young star clusters with more photometric date including uv data , and with the same ssp models as @xcite and @xcite used . in the studies of @xcite and @xcite , we used the metallicities obtained by @xcite and @xcite when estimating the ages of star clusters . the metallicities in @xcite and @xcite were determined from the lick indices which were calibrated from the galactic old gcs . however , @xcite claimed that young star clusters are probably not so metal - poor as deduced from the metallicities obtained by @xcite , and concluded that @xmath108band line strength tends to underestimate [ fe / h ] values in @xcite by more than 1 dex ( see also * ? ? ? * ) . as a result , most of ages obtained in @xcite and @xcite are older than those obtained in this paper . so , we re - estimated the ages for the young star clusters in @xcite and @xcite with the solar metallicity ssp models . in addition , nine clusters ( b476 , bh11 , m026 , m040 , m045 , m053 , m057 , m058 , and m070 ) were estimated to be older than 2 gyr in this paper , which are considered to be old star clusters ( e.g. , * ? ? ? * ) . since this paper focused on young clusters , we would not consider these nine old clusters in the following analysis . however , we pointed out that the ages of six of these nine star clusters were estimated to be younger than 2 gyr by previous studies : the age of b476 was estimated to be 1.2 gyr by @xcite ( however , 7.1 gyr by fan 2010 ) ; the age of bh11 was estimated to be 1.6 gyr by @xcite ; the age of m040 was estimated to be @xmath109 myr by @xcite and @xmath110 myr by @xcite ; the age of m053 was estimated to be 1 gyr by @xcite and @xmath111 myr by @xcite ; the age of m058 was estimated to be @xmath112 myr by @xcite ; the age of m070 was estimated to be 1.2 gyr by @xcite and @xmath113 myr by @xcite . the ages of these nine clusters derived by previous papers and here are listed in table 3 . there are three clusters with ages of 4 myr , which is the lowest age limit of galev models . clusters khm31 - 37 and v133 were estimated slightly older by other authors @xcite , while b196d was estimated slightly younger ( @xmath114 myr ) by @xcite . in figure 4 , the estimated ages for the young ( @xmath4 gyr ) star clusters in this paper were compared with those from previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the star clusters with ages of 4 myr obtained in this paper are drawn with open squares in figure 4 . there are five clusters in common between @xcite and this paper . we can see that the ages of @xcite are in good agreement with ours . @xcite estimated ages of star clusters located in the southern disk of m31 with @xmath0 sed - fitting . there is an obvious offset between their estimated ages and ours , which is caused by some large scatters . if the three clusters with ages greater than 3 gyr given by @xcite , which are drawn with arrows in figure 4 , are not included , the systematic offset can be reduced to be @xmath115 gyr . @xcite estimated ages of star clusters in m31 with multi - band ( @xmath107 ) sed - fitting , and their results agree well with ours , with a small offset ( @xmath116 gyr ) . @xcite determined ages of 20 possible ymcs in m31 by comparing the observed color magnitude diagrams and the isochrones of different metallicities and ages of @xcite , and estimated masses of these clusters based on the maraston s ssp models of solar metallicity and @xcite and @xcite imfs and the ir magnitudes in the 2mass-6x - psc catalog . in general , the ages obtained in this paper are in good agreement with the determinations by @xcite , with a small deviation ( @xmath117 gyr ) . there is a small systematic offset ( @xmath118 gyr ) between @xcite and this paper , i.e. the ages obtained by @xcite are larger than the ages obtained here . @xcite also found the systematic offset as the ages obtained by @xcite are larger than the ages obtained by @xcite . @xcite suggested that this offset is caused by the super - solar metallicity models ( @xmath119 ) adopted by @xcite when they determined the ages of star clusters . @xcite derived ages for young clusters by fitting the multi - band photometry with model seds , and their results are in good agreement with ours . in figure 5 , we compared the masses of clusters obtained in this paper with those from previous studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the masses of two ( khm31 - 37 and v133 ) of the three clusters with ages of 4 myr , which are drawn with open squares in figure 5 , were derived to be lower than @xmath120 . the masses estimated in this paper are in good agreement with those estimated by @xcite , @xcite , @xcite , and @xcite . there is an obvious offset between @xcite and this paper , which is mainly caused by some scatters . the cluster with the largest discrepancy is b335 , with a mass estimate of @xmath121 by @xcite , and @xmath122 in this paper . when b335 is excluded , the offset can be reduced to be ( @xmath123 ) . the masses estimated by @xcite are slightly less than those estimated here , with an offset of @xmath124 . figure 6 shows the number , ages and masses of young star clusters ( @xmath4 gyr ) as a function of projected radius from the center of m31 , adopted at @xmath125 and @xmath126 ( j2000.0 ) following @xcite and @xcite . in the top panel , the histogram for the radial distribution of young star clusters shows clearly two peaks at @xmath127 kpc and @xmath128 kpc , while in the middle panel and bottom panel , wide age and mass distributions can be seen in these two peak regions . @xcite presented the radial distribution of clusters against the distance from the center of m31 , and found that the young clusters show two peaks around @xmath18 kpc and @xmath129 kpc . they also found that the uv sf regions show two distinct peaks : a main peak at @xmath130 kpc and a secondary peak around 11 kpc . in addition , a small peak at @xmath131 kpc in the distribution of ages of uv sf regions against the projected radius ( see figure 19 of * ? ? ? * ) can be clearly found . we argued that the peak at @xmath127 kpc obtained in this paper should be associated with the peak at @xmath131 kpc for the uv sf regions , while the peak at @xmath128 kpc obtained in this paper correlate with the well - known 10 kpc ring @xcite . figure 7 displays the spatial distribution and radial distribution of the m31 young clusters with different age bins : ( a ) @xmath132 gyr ; ( b ) 0.1 gyr @xmath133 0.4 gyr ; ( c ) 0.4 gyr @xmath133 1 gyr ; ( d ) 1 gyr @xmath133 2 gyr . in the top panel , young star clusters in different age ranges are drawn with different marks . the inner , solid ellipse and the dashed contour represent the 10 kpc ring and the outer ring from @xcite based on infrared observations with the multiband imaging photometer for spitzer ( mips ) instrument on the _ spitzer space telescope _ , respectively . the 10 kpc ring was drawn with a center offset from the m31 nucleus by [ @xmath134 , @xmath135 @xcite with a radius of 44 arcmin ( 10 kpc ) . there are several regions drawn with open rectangles which show aggregations of young star clusters to different extents . however , we should point out that these aggregations of young star clusters may be caused by the projection effect because of the inclination of m31 disk . @xcite noted two clumps of young clusters , both of which are located in one rectangle ( @xmath136 kpc @xmath137 @xmath138 kpc and @xmath139 kpc @xmath140 0 kpc ) . the star clusters in this study are spatially coincident with the disk and the rings , indicating that the distribution of the young star clusters correlates with the galaxy s sf regions , which is consistent with previous studies @xcite . in the bottom panel , the number of young star clusters in different age bins as a function of projected radial distance from the m31 center was shown . we can see that clusters younger than @xmath141 gyr show most obvious aggregation around the 10 kpc ring . @xcite ran a number of numerical simulations of the m31m32 and m31-ngc 205 interactions , and assumed a passage of m32 through the disk of m31 occurring 20 myr ago , resulting in a burst of star formation that propagates outward through the disk . @xcite suggested that m32 and m31 had an almost head - on collision about 210 myr ago , and m32 passed through m31 disk again about 110 myr ago ( see figure 2 of * ? ? ? * ) , which induced two off - center rings an inner ring with projected dimensions of @xmath6 1.5 kpc and the 10 kpc ring . both of the simulations recurred the 10 kpc ring and the observed split . we divided our sample star clusters younger than 300 myr into six groups , and showed the spatial distribution for each group in figure 8 . we can see that only star clusters with ages 50 myr @xmath14 100 myr appear around the 10 kpc ring and the ring splitting region ( @xmath142 kpc @xmath137 @xmath143 kpc and @xmath144 kpc @xmath140 @xmath145 kpc ) @xcite , indicating that 1 ) the 10 kpc ring may begin to form about 100 myr ago ; 2 ) m32 passed through the southern part of m31 disk around 100 myr and in turn resulted in the split in the form of a hole . this appears to be consistent with the prediction by @xcite of a second passage of m32 about 110 myr ago . after the second passage , star clusters formed around the split for a long period , since there are a number of star clusters around the split with ages younger than 50 myr . @xcite reported that the star formation rate ( sfr ) of the m31 disk would be elevated greatly and quickly after an encounter event , and it would finally drop when the interstellar medium is depleted and disrupted . however , from figure 8 , we can not find evidence of radial trend of star cluster ages ( see also * ? ? ? * ; * ? ? ? figure 9 shows the spatial and radial distribution of the m31 young star clusters with different mass bins : ( a ) @xmath146 ; ( b ) @xmath147 ; ( c ) @xmath148 . in the top panel , clusters of these three groups are drawn with different marks . the bottom panel presents the number of young clusters in different mass bins as a function of projected radial distance from m31 center , and it shows that young clusters more massive than @xmath7 are most concentrated nearby the 10 kpc ring . figure 10 plots the distribution of estimated ages and masses for the young star clusters . a prominent correlation can be seen that mass increases with age . there are two distinct peaks in the age histogram : a highest peak at age @xmath6 60 myr ( @xmath149 ) and a secondary peak around 250 myr ( @xmath150 ) . the mass distribution of the young star clusters show a single peak around @xmath7 . the mean values of age and mass of young clusters are about 385 myr and @xmath5 , slightly higher than the values presented by @xcite , which are 300 myr and @xmath7 , respectively . most of our young clusters have masses ranging from @xmath151 to @xmath152 , which are more massive than ocs in the solar neighborhood @xcite , but less massive than typical gcs in the mw @xcite . the lack of young clusters more massive than @xmath153 is also noted by @xcite and @xcite , possibly caused by a low - average sfr of m31 @xcite or hidden by dust clouds in the disk due to the inclination angle of m31 @xcite . @xcite have listed three phases for the evolution of a young star cluster : 1 ) the first few myr , during which the star formation activity is still proceeding and the star cluster is rich in gas ; 2 ) a subsequent period after the first supernovae ( some 3 myr after formation ) , in which a young cluster is experiencing a serious loss of gas and dust , and stellar mass loss plays an important role in the cluster evolution ; 3 ) a later stage that stellar dynamical processes dominate the cluster evolution . the dividing line between phase 2 and phase 3 may be anywhere between 100 myr and 1 gyr , and most of our young clusters are experiencing the phase 2 or phase 3 . @xcite presented that after 5 gyr , both mass and galactic location are important evolutionary parameters for gcs . @xcite discussed the destructive effects of encounters of clusters with giant molecular clouds ( gmcs ) , and presented that the disruption time for a star cluster varies directly with the cluster density and is about 200 myr for a mean density of 1 @xmath154 . @xcite also reported that two - body relaxation is effective at destroying low - mass clusters and this may account for the scarcity of low - mass older clusters . actually , the two - body relaxation and the encounters with gmcs are also important processes that lead to young cluster disruption ( see * ? ? ? * and references therein ) , while @xcite presented that mass loss due to stellar evolution is the most important process in the young cluster dissolution . it is evident that star cluster mass is one key parameter in the star cluster evolution . @xcite derived an empirical relation between the disruption time and the initial mass of star clusters in the solar neighborhood , small magellanic cloud ( smc ) , m51 , and m33 . @xcite determined a disruption time of 1.3 gyr for a @xmath7 cluster in the solar neighborhood , while @xcite reported that most of m31 young clusters would be destroyed in the next gyr or so , and only some massive and dense ones may survive for a longer time . several features are shown in figure 10 : 1 ) there is an obvious gap in the age distribution around 100 myr . 2 ) there are few clusters older than 400 myr ( @xmath155 ) with mass lower than @xmath7 . although many low - mass clusters can be easy to disrupt , this gap may be caused by a selection effect . in fact , @xcite found that the completeness of m31 ground - based sample drops precipitously at @xmath156 ( @xmath157 ) , which is about @xmath158 . 3 ) there are few clusters more massive than @xmath153 , which may be caused by a low - average sfr of m31 or the hidden by dust clouds in the m31 disk as discussed above . 4 ) there is a gap of clusters with very low masses ( @xmath159 ) and younger than 30 myr ( @xmath160 ) . these clusters may be too faint to be sample objects of @xcite , indicating that our sample is not complete in these age and mass ranges ( see also * ? ? ? figure 11 shows the age distribution in different mass intervals ( top panel ) and mass function in different age intervals ( bottom panel ) . the histograms are derived using a 0.4-dex bin width with different starting values . these distributions contain information about the formation and disruption history of star clusters @xcite , however , the interpretation of the empirical distributions of clusters depends strongly on how incompleteness affects the sample @xcite . in the top panel , we can see an obvious gap before 40 myr ( @xmath161 ) , which is caused by a selection effect . the age distribution of the clusters does not declines monotonically , with an apparent bend around 200 myr ( @xmath162 ) . we argued that this bend near 200 myr may be explained as a burst of cluster formation , possibly caused by a current interaction event between m31 and its satellite galaxy , such as the collision between m31 and m32 about 210 myr ago suggested by @xcite . the two decline trends starting from 40 myr and 200 myr reflect a rapid disruption of clusters . @xcite noted a peak of the cluster age distribution at 70 myr , and suggested an enhanced cluster formation episode at that epoch . in the bottom panel , the gap in the number of clusters in the low - mass regions ( @xmath163 ) is apparently due to a sample incompleteness , but not physical @xcite . the initial mass function for star clusters should be slightly steeper @xcite than what is shown here because of the short lifetimes of low - mass clusters . recently , @xcite compared the observed age distributions and mass functions of star clusters in the mw , mcs , m83 , m51 , and antennae , and found that these distributions of clusters are similar in different galaxies . however , due to the incompleteness of our cluster sample , partly due to the exclusion of clusters that can not derive accurate photometry , we would not give any empirical formulas of the distributions for age and mass . in this section , we will discuss the correlations of ages and masses with structure parameters , which are derived by king - model @xcite fits for clusters in m31 @xcite . because the sample clusters are younger than 2 gyr , the structure parameters obtained from the bluer filters are preferred ( see * ? ? ? * in detail ) . there are four clusters ( b315 , b319 , b368 , and b374 ) which have been studied twice by @xcite and @xcite , and we would use the new results in @xcite . figure 12 shows structure parameters as a function of age for young clusters in this paper . some correlations can be seen , the concentration @xmath164 , defined as @xmath165 , decreases with age . the trend is largely driven by clusters b342 and b368 , both of which have large @xmath164 values ( 3.98 for b342 and 3.87 for b368 ) . both the scale radius @xmath166 and projected core radius @xmath167 increase with age . clusters b342 and b368 have very small @xmath166 and @xmath167 values and are drawn with arrows in figure 12 ( the values of @xmath166 and @xmath167 are @xmath6 0.014 pc for b342 , while are @xmath6 0.011 pc for b368 ) . @xcite and @xcite discussed the trend for core radius against age , and argued that this trend may represent real evolution in the structure of clusters as they grow old , partially explained by the effect of mass segregation @xcite , or dynamical effects such as heating by black hole ( bh ) binaries @xcite . @xcite also demonstrated that neither large differences in primordial binary fraction nor a tidal heating due to differences in the cluster orbits could account for the observed trend . the best - fit central surface brightness @xmath168 shows a decreasing trend with age , and @xcite argued that this trend may be likely due to the fading of stellar population and the increase of core radius @xmath167 with age . we also see that the central mass density @xmath169 decreases with age , although the scatters are great . @xcite presented that the central mass density shows very little trend with age for both the m31 young clusters and young clusters in the mcs . there is no obvious correlation between @xmath170 , the two - body relaxation time at the model - projected half - mass radius , and age . the dashed line represents the region that @xmath170 equal to age . it can be seen that most clusters ( except for dao38 and m091 ) have ages less than @xmath170 , indicating that these young clusters have not been well dynamically relaxed . because two - body encounters can transfer energy between individual stars and then impel the system to establish thermal equilibrium @xcite , we argue that these young clusters have not established thermal equilibrium . figure 13 shows structure parameters as a function of mass for the sample young clusters . the concentration @xmath164 increases with mass , although the trend is much weak . @xcite presented @xmath164 plotted against mass for clusters in mcs , and found that there was no correlation between @xmath164 and mass . actually , we found that all the clusters in @xcite have @xmath164 less than 2.5 , much smaller than the largest value in our sample ( @xmath171 ) . if we do not include the two clusters b342 and b368 , the correlation for @xmath164 with mass nearly disappear . both @xmath166 and @xmath167 increase with mass , however , the trend is largely weaken by cluster b327 , which has very small values of @xmath166 and @xmath167 , but larger than those of b342 and b368 which are drawn with arrows in figure 13 . both the central surface brightness @xmath168 and central mass density @xmath169 decrease weakly with mass , while no obvious correlation between @xmath170 and mass can be seen . we checked the surface brightness profiles of b327 , b342 , and b368 displayed in @xcite , which have very small @xmath166 and @xmath167 and very large @xmath168 and @xmath169 , and found that these core profiles are cuspy . @xcite concluded that the cores of these clusters did not appear to be resolved in the _ hst_/wfpc2 images and the structural parameters for these clusters would be uncertain if the central cluster luminosity is dominated by only a few bright stars . however , if these cuspy core profiles are true integrated properties , which may be better fitted by a power - law structure model ( e.g. , * ? ? ? * ) , the three clusters may have been post core - collapse ( see * ? ? ? * in detail ) . ymcs are often related to the violent sf episodes triggered by galaxy collisions , mergers , and close encounters @xcite . however , based on a sample of 21 nearby spirals , @xcite found that ymcs can exist in a wide variety of host galaxy environments , including quiescent galaxies , and that there is no correlation between the morphological type of the galaxies and their contents of ymcs . ymcs are dense aggregates of young stars , which are also expected to be the nurseries for many unusual objects , including exotic stars , binaries , and bhs @xcite . many studies @xcite that focused on m31 ymcs have derived remarkable achievements in understanding their stellar populations , structure parameters , and dynamical properties . there are 13 ymcs in our cluster sample with a definition of age @xmath8 myr and mass @xmath9 @xcite . figure 14 shows the spatial distribution of the 13 ymcs , while different sizes of the open circles indicate ymcs in different mass ranges . the rectangle between the 10 kpc ring and the outer ring represents the split in the southern part of the m31 disk , and the two black filled triangles represent m32 and ngc 205 . it is not surprising to see that most of the ymcs gather around the split , indicating that there has been a high - level star formation activity , which is consistent with previous studies @xcite . in this paper , we determined the ages and masses for a sample of m31 star clusters by comparing the multicolor photometry with theoretical sps models . multicolor photometric data are from the _ galex _ fuv and nuv , broadband @xmath0 , sdss @xmath1 , 15 intermediate - band filters of batc , and 2mass @xmath2 , which constitute the seds covering @xmath3 . we made a discussion on the spatial distribution , distribution of ages and masses , correlations of ages and masses with structure parameters for the sample young clusters ( @xmath4 gyr ) . the mean value of age and mass of young clusters is about 385 myr and @xmath5 , respectively . there are two distinct peaks in the age distribution , a highest peak at age @xmath6 60 myr and a secondary peak around 250 myr , while the mass distribution shows a single peak around @xmath7 . there are several regions showing aggregations of young clusters around the 10 kpc ring and the outer ring , indicating that the distribution of the young clusters correlates well with m31 s sf regions . the ages and masses show apparent correlations with some structure parameters . we also found the correlation between core radius @xmath167 and age , which has been studied by many authors . a few young clusters have the two - body relaxation times @xmath170 greater than their ages , indicating that they have not been well dynamically relaxed . we argued that these young clusters have not established the thermal equilibrium . the ymcs ( age @xmath8 myr and mass @xmath9 ) show obvious aggregation around the split in the southern part of the m31 disk , suggesting a high efficiency of star formation , possibly triggered by a recent passage of a satellite galaxy ( m32 ) through m31 disk . we would like to thank the anonymous referee for providing rapid and thoughtful report that helped improve the original manuscript greatly . this work was supported by the chinese national natural science foundation grant nos . 10873016 , 10633020 , 11073032 , and 11003021 , and by the national basic research program of china ( 973 program ) no . 2007cb815403 . lcccccccccccccccc object & @xmath96 & @xmath97 & @xmath164 & @xmath172 & @xmath54 & @xmath173 & @xmath174 & @xmath57 & @xmath175 & @xmath176 & @xmath177 & @xmath99 & @xmath83 & @xmath178 & @xmath100 & @xmath179 + & ( mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag ) & ( @xmath180 ) + b006d - d036 & 20.07 & 19.46 & 18.99 & 18.76 & 18.75 & 18.61 & 18.51 & 18.44 & 18.32 & ... & 18.38 & 18.29 & ... & 18.10 & ... & 5.0 + & 0.142 & 0.121 & 0.049 & 0.056 & 0.066 & 0.066 & 0.059 & 0.063 & 0.074 & ... & 0.108 & 0.115 & ... & 0.130 & ... & + b010d & 20.89 & 20.03 & 19.79 & 19.28 & 19.23 & 18.99 & 18.96 & 18.78 & 18.68 & 18.63 & 18.54 & 18.57 & 18.45 & 18.37 & 18.11 & 4.2 + & 0.254 & 0.161 & 0.045 & 0.038 & 0.039 & 0.034 & 0.048 & 0.040 & 0.045 & 0.052 & 0.072 & 0.059 & 0.100 & 0.078 & 0.124 & + b012d - d039 & ... & 20.14 & 19.46 & 19.19 & 19.14 & 18.83 & 18.85 & 18.73 & 18.55 & 18.47 & 18.36 & 18.34 & 18.25 & 17.98 & 18.01 & 5.0 + & ... & 0.261 & 0.058 & 0.087 & 0.093 & 0.080 & 0.113 & 0.114 & 0.111 & 0.119 & 0.110 & 0.117 & 0.139 & 0.105 & 0.167 & + b014d & ... & 19.80 & 19.64 & 19.40 & 19.21 & 19.04 & 18.79 & 18.73 & 18.59 & 18.43 & 18.33 & 18.16 & 18.41 & 18.02 & 17.89 & 5.0 + & ... & 0.158 & 0.066 & 0.099 & 0.080 & 0.074 & 0.067 & 0.070 & 0.072 & 0.074 & 0.093 & 0.067 & 0.128 & 0.097 & 0.148 & + b015d - d041 & ... & 19.51 & 19.26 & 18.92 & 18.84 & 18.61 & 18.46 & 18.34 & 18.03 & 17.98 & 18.04 & 17.86 & 17.91 & 17.65 & 17.65 & 5.0 + & ... & 0.135 & 0.051 & 0.067 & 0.057 & 0.053 & 0.059 & 0.058 & 0.052 & 0.061 & 0.077 & 0.060 & 0.098 & 0.068 & 0.107 & + b017d & 20.22 & 19.44 & 18.98 & 18.51 & 18.32 & 18.09 & 17.97 & 17.87 & 17.68 & 17.53 & 17.41 & 17.22 & ... & 17.14 & 17.32 & 6.7 + & 0.216 & 0.154 & 0.042 & 0.051 & 0.046 & 0.042 & 0.055 & 0.048 & 0.057 & 0.059 & 0.075 & 0.062 & ... & 0.099 & 0.157 & + b018-g071 & 19.81 & 18.78 & 18.32 & 17.93 & 17.88 & 17.80 & 17.45 & 17.39 & 17.20 & 17.13 & 16.84 & 16.95 & 16.79 & 16.71 & 16.86 & 5.8 + & 0.274 & 0.052 & 0.061 & 0.083 & 0.048 & 0.037 & 0.037 & 0.033 & 0.026 & 0.041 & 0.045 & 0.034 & 0.048 & 0.081 & 0.125 & + b019d & ... & 19.81 & 20.18 & 19.90 & 19.48 & 19.23 & 18.96 & 18.79 & 18.51 & 18.44 & 18.18 & 18.39 & 18.15 & 17.84 & 17.95 & 5.0 + & ... & 0.176 & 0.139 & 0.179 & 0.140 & 0.128 & 0.124 & 0.111 & 0.108 & 0.101 & 0.099 & 0.103 & 0.130 & 0.103 & 0.163 & + b035d & 21.08 & 19.74 & 19.14 & 18.91 & 18.67 & 18.56 & 18.41 & 18.35 & 18.13 & 18.10 & 18.05 & 18.01 & 18.18 & 17.94 & 18.08 & 5.0 + & 0.399 & 0.192 & 0.032 & 0.047 & 0.040 & 0.040 & 0.048 & 0.046 & 0.053 & 0.054 & 0.081 & 0.060 & 0.117 & 0.120 & 0.177 & + b040-g102 & 18.68 & 17.73 & 17.64 & 17.65 & 17.54 & 17.48 & 17.37 & 17.23 & 17.18 & 17.18 & 16.94 & 17.16 & 16.98 & 16.93 & 16.90 & 5.8 + & 0.097 & 0.020 & 0.030 & 0.056 & 0.036 & 0.027 & 0.037 & 0.027 & 0.019 & 0.038 & 0.055 & 0.040 & 0.054 & 0.112 & 0.124 & + lccccccccccccccc object & fuv & nuv & @xmath41 & @xmath52 & @xmath55 & @xmath181 & @xmath182 & @xmath183 & @xmath174 & @xmath184 & @xmath175 & @xmath185 & @xmath101 & @xmath102 & @xmath103 + & ( mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag)&(mag ) + b006d - d036 & 21.50 & 20.87 & ... & ... & 18.69 & ... & ... & 19.72 & 18.44 & 18.30 & 18.10 & 17.99 & ... & ... & ... + & 0.100 & 0.060 & ... & ... & 0.050 & ... & ... & 0.114 & 0.069 & 0.077 & 0.090 & 0.134 & ... & ... & ... + b010d & ... & ... & ... & ... & 18.92 & ... & ... & 20.80 & 19.11 & 18.62 & 18.40 & 18.27 & ... & ... & ... + & ... & ... & ... & ... & 0.050 & ... & ... & 0.224 & 0.114 & 0.120 & 0.127 & 0.171 & ... & ... & ... + b012d - d039 & ... & ... & ... & ... & 19.05 & ... & ... & 20.60 & 19.03 & 18.48 & 18.13 & 17.81 & ... & ... & ... + & ... & ... & ... & ... & 0.050 & ... & ... & 0.187 & 0.052 & 0.068 & 0.085 & 0.123 & ... & ... & ... + b014d & ... & 22.81 & ... & ... & 18.53 & ... & ... & 20.54 & 19.05 & 18.49 & 18.14 & ... & ... & 15.79 & 15.72 + & ... & 0.380 & ... & ... & 0.050 & ... & ... & 0.131 & 0.044 & 0.057 & 0.072 & ... & ... & 0.133 & 0.224 + b015d - d041 & ... & ... & ... & ... & 18.70 & ... & ... & 20.39 & 18.70 & 18.01 & 17.71 & 17.42 & ... & 15.92 & 15.72 + & ... & ... & ... & ... & 0.050 & ... & ... & 0.183 & 0.091 & 0.098 & 0.108 & 0.131 & ... & 0.265 & 0.151 + b017d & ... & ... & ... & ... & 18.23 & ... & ... & 20.05 & 18.32 & 17.77 & 17.39 & 17.17 & 15.21 & ... & ... + & ... & ... & ... & ... & 0.050 & ... & ... & 0.132 & 0.048 & 0.057 & 0.066 & 0.088 & 0.094 & ... & ... + b018-g071 & ... & 22.19 & 18.47 & 18.25 & 17.53 & 17.00 & 16.38 & 19.39 & 17.85 & 17.22 & 16.82 & 16.63 & 15.46 & 14.78 & 14.60 + & ... & 0.140 & 0.080 & 0.050 & 0.050 & 0.050 & 0.050 & 0.088 & 0.059 & 0.063 & 0.067 & 0.076 & 0.094 & 0.086 & 0.112 + b019d & ... & ... & ... & ... & 18.93 & ... & ... & 21.14 & 19.38 & 18.61 & 18.11 & 17.59 & 16.07 & 15.73 & 14.79 + & ... & ... & ... & ... & 0.050 & ... & ... & 0.246 & 0.043 & 0.064 & 0.082 & 0.110 & 0.112 & 0.112 & 0.112 + b035d & 22.37 & 22.95 & ... & ... & 18.48 & ... & ... & 20.05 & 18.68 & 18.27 & 17.91 & 17.58 & 15.99 & 15.35 & 15.18 + & 0.340 & 0.310 & ... & ... & 0.050 & ... & ... & 0.162 & 0.111 & 0.121 & 0.135 & 0.164 & 0.094 & 0.112 & 0.130 + b040-g102 & 19.84 & 19.30 & 17.42 & 17.69 & 17.30 & 16.92 & 16.63 & 18.32 & 17.38 & 17.19 & 17.05 & 16.98 & 16.07 & 14.96 & 15.51 + & 0.030 & 0.020 & 0.046 & 0.022 & 0.010 & 0.022 & 0.022 & 0.048 & 0.035 & 0.039 & 0.045 & 0.059 & 0.112 & 0.086 & 0.130 + lcc object & log ( age)@xmath186 & log ( age)@xmath187 + & ( yr ) & ( yr ) + b476-d074 & @xmath188 & @xmath189 + bh11 & @xmath190 & @xmath191 + m026 & @xmath192 & @xmath193 + m040 & @xmath194 & @xmath195 + m045 & ... & @xmath196 + m053 & @xmath197 & @xmath198 + m057 & ... & @xmath199 + m058 & @xmath200 & @xmath201 + m070 & @xmath202 & @xmath203 + lcccc object & log ( age ) & log ( mass ) & @xmath204 & flag@xmath186 + & ( yr ) & ( @xmath70 ) & & + b006d - d036 & @xmath205 & @xmath206 & @xmath207 & 1 + b010d & @xmath208 & @xmath209 & @xmath210 & 1 + b012d - d039 & @xmath211 & @xmath212 & @xmath213 & 1 + b014d & @xmath214 & @xmath215 & @xmath216 & 1 + b015d - d041 & @xmath217 & @xmath218 & @xmath219 & 1 + b017d & @xmath220 & @xmath221 & @xmath222 & 2 + b018-g071 & @xmath223 & @xmath224 & @xmath225 & 1 + b019d & @xmath226 & @xmath227 & @xmath228 & 1 + b035d & @xmath229 & @xmath230 & @xmath210 & 1 + b040-g102 & @xmath231 & @xmath232 & @xmath210 & 1 +

redshift - space distortions ( rsd ) offer an attractive method to measure the growth of cosmic structure on large scales , and combining with the measurement of the cosmic expansion history , it can be used as cosmological tests of gravity . with the advent of future galaxy redshift surveys aiming at precisely measuring the rsd , an accurate modeling of rsd going beyond linear theory is a critical issue in order to detect or disprove small deviations from general relativity ( gr ) . while several improved models of rsd have been recently proposed based on the perturbation theory ( pt ) , the framework of these models heavily relies on gr . here , we put forward a new pt prescription for rsd in general modified gravity models . as a specific application , we present theoretical predictions of the redshift - space power spectra in gravity model , and compare them with-body simulations . using the pt template that takes into account the effects of both modifications of gravity and rsd properly , we successfully recover the fiducial model parameter in-body simulations in an unbiased way . on the other hand , we found it difficult to detect the scale dependence of the growth rate in a model - independent way based on gr templates .

1309.6783

redshift - space distortions ( rsd ) of galaxy clustering , which appear as systematic effects in determining the redshift of each galaxy via spectroscopic measurements and manifestly break statistical isotropy @xcite , are now recognized as a sensitive probe of the growth of structure . combining the distance measurement of galaxies using the baryon acoustic oscillations as standard ruler ( e.g. , @xcite ) , rsd offers a unique opportunity to test the theory of gravity on cosmological scales ( e.g. , @xcite ) and help us obtain a deeper understanding of the current accelerating expansion of the universe . this is indeed one of the main goal of on - going and up - coming galaxy surveys such as the baryon oscillation spectroscopic survey of sloan digital sky survey iii , the wigglez survey , the subaru measurement of imaging and redshifts , the dark energy survey , the bigboss project , and the esa / euclid survey , which will provide precision measurements of the power spectrum ( or correlation function ) . the late - time cosmic acceleration , first discovered by the observations of distant type ia supernovae @xcite , may be the result of a dark energy which can be realized in the presence of dynamical scalar field , or it may indicate the breakdown of general relativity ( gr ) on cosmological scales . the latter case requires a consistent model of gravity that explains the accelerating expansion on large scales with the modification of gravity , while neatly evading the stringent constraints on the deviation from gr at solar system scales ( e.g. , @xcite ) . in this respect , the large - scale structure offers the best opportunity to distinguish between modified gravity and dark energy models in gr , and the measurement of rsd is a powerful tool to probe gravity . given the high - precision measurements of rsd in the near future , accurate theoretical templates of the redshift - space power spectrum or correlation function is highly demanded in order to detect a small deviation of gravity from gr . this is indeed now active research subject , and there are many studies to accurately model rsd . the rsd measurement is basically made at the scales close to the linear regime of gravitational evolution , but the nonlinearity arising both from the gravity and the rsd is known to play a crucial role . moreover , due to the non - gaussian nature of rsd @xcite , the applicable range of linear theory prediction is fairly narrower than that in real space . thus , beyond the linear scales , a sophisticated treatment is required for reliable theoretical predictions with a wider applicable range . recently , based on the perturbation theory of large - scale structure , several improved models of rsd have been proposed @xcite . these models properly account for the non - gaussian nature of rsd , and are tested against @xmath1-body simulations , successfully describing redshift - space power spectrum and/or correlation function at weakly nonlinear regime . applying these models to real observations , constraints on the growth of structure have been also obtained ( e.g. , @xcite ) . however , it should be noted that the proposed models of rsd have been so far tested only in the case of gr . further , beyond linear theory , the template of rsd is computed with the perturbation theory under the assumption that gravity is described by gr . thus , the observational constraints derived from the pt - based template can be only used as a consistency test with gr , and a care must be taken in addressing the constraint on a specific model of modified gravity . the aim of the present paper is to examine these issues based on an improved model of rsd developed by ref . @xcite . the power spectrum expression of this model is similar to the one proposed by ref . @xcite and the so - called streaming model frequently used in the literature ( e.g. , @xcite ) , but it includes two important pt corrections as a result of the low-@xmath2 expansion . although the model also includes a phenomenological term to account for the finger - of - god damping arising from the small - scale physics , combining the recently developed resummed pt , it successfully describes not only the matter but also the halo power spectra in @xmath1-body simulations @xcite . it is shown that the model can be used as a theoretical template to simultaneously constrain the parameters associated with the cosmic expansion and the structure growth in an unbiased manner , and applying it to the luminous red galaxy sample of sloan digital sky survey data release 7 , a robust contraint is obtained @xcite . here , extending these previous works in gr , we put forward a prescription to compute the redshift - space power spectrum in modified gravity models . as a specific example , we explicitly compute the redshift - space power spectrum in @xmath0 gravity model , as one of the representative modified theory of gravity @xcite . the theoretical prediction based on the standard pt calculation is compared with the results of @xmath1-body simulations , and a good agreement is found . then , we will discuss the potential impact of the precision modeling of rsd on the model - independent test of gr and/or constraint on modified gravity models . we will show that a tight and unbiased constraint on modified gravity models is achieved only with an improved pt model of rsd in which the effect of modified gravity is properly taken into account in the pt calculation . with the improved pt template , testing gr will be made possible in a model - independent way , but we argue that a quantitative characterization of the small deviation from gr generally requires a prior knowledge of modified gravity models . the paper is organized as follows . in sec . [ sec : rsd_model ] , we begin by briefly reviewing the model of rsd proposed by ref . @xcite . employing the standard pt calculation , we then give a prescription on how to compute the redshift - space power spectrum in modified gravity models . in appendix [ sec : basiceqs_mg ] , we summarize the basic formalism of the standard pt in a general context of modified gravity models , and explicitly give expressions for the second - order pt kernels used to compute the higher - order corrections of rsd . in sec . [ sec : rsd_frmodel ] , as one of the representative models of modified gravity , we consider the @xmath0 gravity model , and quantitatively compare the pt predictions in redshift space with @xmath1-body simulations . based on this , in sec . [ sec : implications ] , a potential impact of the precision pt model of rsd is discussed , particularly focusing on a precision constraint on the model parameter of modified gravity , and model - independent analysis of detecting or characterizing a small deviation from gr . finally , sec . [ sec : conclusion ] is devoted to the summary and conclusion . we begin by writing the exact expression for redshift - space power spectrum . let us denote the density and velocity fields by @xmath3 and @xmath4 . owing to the distant - observer approximation , which is usually valid for the observation of distant galaxies of our interest , one can write ( e.g. , @xcite ) @xmath5 where @xmath6 denotes the separation in real space and @xmath7 indicates an ensemble average . in the above expression , the @xmath8-axis is taken as an observer s line - of - sight direction , and we define the directional cosine @xmath9 by @xmath10 . further , we defined @xmath11 , and @xmath12 for the line - of - sight component of the velocity field . note that the above expression has been derived without invoking the dynamical information for velocity and density fields , i.e. , the euler equation and/or continuity equations . thus , eq . ( [ eq : pkred_exact ] ) does hold even in modified gravity models . ( [ eq : pkred_exact ] ) can be re - expressed in terms of the cumulants . then , the term in the bracket is factorized into two terms , each of which includes the exponential factor ( e.g. , see eq.(6 ) of ref . @xcite for explicit expression ) . among these , the overall factor , expressed as @xmath13 with @xmath14 being the cumulant , is responsible for the suppression of the power spectrum arising mostly from the virialized random and coherent motion on small scales . the effect of this is known to be partly non - perturbative , and seems difficult to treat petrubatively . since it has been shown to mainly change the broadband shape of the power spectrum , we may phenomenologically characterize it with a general functional form @xmath15 with @xmath16 being a scale - independent constant . on the other hand , the remaining factor includes the term leading to the kaiser effect in the linear regime @xcite , and is likely to affect the structure of power spectrum on large - scales . although there also appears the exponential factor @xmath17 in each term of this factor , these contributions should be small as long as we consider the large scales , and the perturbative treatment may be applied . with the proposition given above , ref . @xcite applied the low-@xmath2 expansion , keeping the overall prefactor as general functional form @xmath18 . the resultant power spectrum expression at one - loop order becomes @xmath19\ , \nonumber\\ & \quad\times\,\bigl\ { p_{\rm kaiser}(k,\mu)+a(k,\mu)+b(k,\mu ) \bigr\ } , \label{eq : tns_model}\end{aligned}\ ] ] which we hereafter call tns model . owing to the single - stream approximation in which the dynamics of large - scale structure is described by the density @xmath3 and velocity divergence @xmath20 , the quantities @xmath21 , @xmath22 , and @xmath23 are explicitly written as @xmath24 where @xmath25 , @xmath26 , and @xmath27 respectively denote auto - power spectra of density and velocity divergence , and their cross power spectrum . the function @xmath28 is cross bispectrum defined by @xmath29 note that in deriving eq . ( [ eq : tns_model ] ) , we do not assume any gravity model . although the expression ( [ eq : tns_model ] ) has been originally derived based on the consideration in gr , as long as the deviation from gr is small , eq . ( [ eq : tns_model ] ) can apply to any model of modified gravity . as we mentioned in sec . [ sec : intro ] , the main characteristic of the model given in eq . ( [ eq : tns_model ] ) is the two additional terms @xmath22 and @xmath23 , which represent the higher - order coupling between velocity and density fields . it has been shown in previous studies in gr that these two terms enhance the power spectrum amplitude over the scales of baryon acoustic oscillations , and moderately but notably change the acoustic structure imprinted in the power spectrum @xcite . as a result , the model ( [ eq : tns_model ] ) successfully describes both the matter and halo power spectra of @xmath1-body simulations at weakly nonlinear scales @xcite . these features are expected to hold qualitatively even in the modified theory of gravity , but quantitative aspect of the rsd would generally differ from that of gr , which we will study in detail . to compute the redshift - space power spectrum beyond linear theory , we apply the pt treatment of gravitational evolution , and calculate each term in eq . ( [ eq : tns_model ] ) in the quasilinear regime . while the power spectrum calculation in the case of gr has been made possible with resummed pt scheme up to the two - loop order ( e.g. , @xcite in redshift space , and @xcite in real space ) and the applicable range of the pt prediction has become wider , we here work with the standard pt calculation at one - loop order for the predictions in modified theory of gravity . although the standard pt treatment in gr is known to produce ill - behaved pt expansion that lacks good convergence properties ( e.g. , @xcite ) , using the standard pt as theoretical template , we can still get a fruitful cosmological constraint at the quasilinear scales ( e.g. , @xcite ) . in the present paper , we use the standard pt formalism developed by ref . @xcite , which is suited to deal with a wide class of modified gravity models . in what follows , we separately give a prescription on how to compute the power spectrum corrections in eq . ( [ eq : tns_model ] ) . the term @xmath21 in eq . ( [ eq : tns_model ] ) is the leading - order contribution to the redshift - space power spectrum . in the large - scale limit where the linear theory prediction is safely applied , we have @xmath30 , and eq . ( [ eq : kaiser ] ) is reduced to the kaiser formula , @xmath31 @xcite , where @xmath32 is the linear growth rate defined by @xmath33 with @xmath34 being the linear growth factor . beyond the linear theory , a simple relation between density and velocity divergence fields no longer hold , and we need to separately evaluate the three power spectra , @xmath25 , @xmath27 and @xmath26 , especially in the modified gravity models @xcite . in contrast to the gr , one crucial point in the modified gravity models is that a new scalar degree of freedom , sometimes referred to as the scalaron , arises and modifies the force law . in the presence of an extra scalar field , even though the conservation law of energy momentum tensor remains unchanged , the poisson equation is inevitably modified and is coupled to the field equation for scalaron . in particular , in successful modified gravity models that have a mechanism to recover gr on small scales , the scalaron generally acquires nonlinear interaction terms , and they play an important role to recover gr on small scales . thus , we need to properly take into account such a nonlinear interaction of the scalaron , and consistently solve the evolution equations of density and velocity divergence . in ref . @xcite , we have developed a formalism to calculate the nonlinear power spectrum in a wide class of modified gravity models , including @xmath0 gravity and dvali - gabadadze - porratti ( dgp ) braneworld @xcite models . the formalism perturbatively treats the effect of nonlinear scalaron , and employing the standard pt technique , we have explicitly computed the power spectrum of density field , @xmath25 , at one - loop order , which reproduces the @xmath1-body results at quasi - linear scales . in what follows , we adopt this formalism to perturbatively compute auto- and cross - power spectra of density and velocity divergence . the basic equations for perturbations are briefly summarized in appendix [ sec : basiceqs_mg ] . to compute the one - loop power spectra , specifically in the @xmath0 gravity model presented below ( sec . [ sec : rsd_frmodel ] ) , the analytic calculations starting naively with the basic equations in appendix [ sec : basiceqs_mg ] is technically difficult in practice , because the perturbation equations can not be separately treated in time and scales . instead of solving the equations for @xmath3 and @xmath35 in appendix [ sec : basiceqs_mg ] , we will numerically solve the evolution equations for the power spectra @xmath36 called the closure equation in ref . @xcite [ eqs . ( 4.3)(4.4)(4.5 ) ] , which has been derived by truncating an infinite chain of the moment equations at one - loop order . implementation and technical detail of the numerical scheme to solve the closure equations are presented in ref . @xcite ( see also appendix a of ref . @xcite ) . next consider the @xmath22 term . the expression given in eq . ( [ eq : a_term ] ) can be rewritten with a more convenient form suited for numerical integration . introducing the doublet with @xmath32 being the linear growth rate , @xmath37 ( e.g. , @xcite ) . ] , @xmath38 , we define the bispectrum @xmath39 : @xmath40 in terms of this , the three - dimensional integral is reduced to the sum of the two - dimensional integrals , and the final form of the @xmath22 term becomes @xcite @xmath41 with @xmath42 and @xmath43 . the non - vanishing components of @xmath44 and @xmath45 are exactly the same as those presented in ref . @xcite ( see sec . iii - b2 ) . since the @xmath22 term appears as a next - to - leading order correction , the tree - level calculation of the bispectrum is sufficient for a consistent calculation of redshift - space power spectrum at one - loop order . expanding the doublet @xmath46 as @xmath47 , and assuming the gaussian initial condition , the bispectrum at the tree - level order becomes @xmath48 where the functions @xmath49 are the symmetrized standard pt kernel of the @xmath50-th order perturbative solutions : @xmath51 with @xmath52 . the function @xmath53 is the initial density field , for which we assume gaussian statistics . the statistical property of @xmath53 is characterized by the power spectrum : @xmath54 the remaining task in computing the @xmath22 term is to evaluate the pt kernels up to the second order , which can be done analytically . this is also the case with the @xmath0 gravity model given below , although some numerical works are involved . in appendix [ sec : basiceqs_mg ] , based on the basic equations , we derive the explicit functional form of the pt kernels @xmath55 and @xmath56 , and summarize the procedure to compute these kernels . the explicit expression for the pt kernels in @xmath0 gravity and dgp models is also presented . similar to the @xmath22 term , the @xmath23 term given in eq . ( [ eq : b_term ] ) is reduced to the sum of two - dimensional integrals . from refs . @xcite , the resultant expression becomes @xmath57 here , the coefficients @xmath58 are the same as presented in appendix b of ref . @xcite . ) exactly coincides with eq . ( a4 ) of ref . @xcite , but the definition of power spectra @xmath36 is somewhat different . ] for the one - loop order of the redshift - space power spectrum , the linear - order power spectra are sufficient to evaluate eq . ( [ eq : b_term2 ] ) , and we have @xmath59 with the linear pt kernel @xmath55 . in this section , as an illustrative example showing the rsd beyond linear scales in modified gravity model , we compute the redshift - space power spectrum in the @xmath0 gravity model , and compare the pt prediction with results of @xmath1-body simulations . the @xmath0 gravity is one of the representative gravity models , and it has a mechanism to recover gr on small scales . generalizing the einstein - hilbert action to include an arbitrary function of the scalar curvature @xmath60 , the model is given by : @xmath61 , \label{eq : action_fr}\ ] ] where @xmath62 and @xmath63 is the lagrangian of the ordinary matter . this theory is equivalent to the brans - dicke theory with parameter @xmath64 , but there is a non - trivial potential . this can be seen from the trace of modified einstein equations : @xmath65 where @xmath66 and @xmath67 is a laplacian operator and we assumed the matter dominated universe . we can identify @xmath68 as the scalaron , i.e. , extra scalar field , and its perturbations are defined as @xmath69 where the bar indicates that the quantity is evaluated on the background universe . here , we consider the cases with @xmath70 and @xmath71 . these conditions are necessary to have the background close to cold dark matter ( @xmath72cdm ) cosmology . then the perturbations for scalaron satisfy @xmath73 note that this is nothing but the equation for the bd scalar perturbations with @xmath64 . in what follows , we consider the specific function @xmath0 of the form @xmath74 where @xmath22 is a constant with dimensions of length squared @xcite . if we take the limit @xmath75 , we obtain @xmath76 and cosmological constant does not appear . for high curvature @xmath77 , on the other hand , @xmath0 can be expanded as @xmath78 where @xmath79 is determined by @xmath22 . the quantity @xmath80 is the background curvature today , and we defined @xmath81 ( see e.g. , @xcite for recent cosmological constraints on @xmath82 ) . in the setup of @xmath1-body simulations and pt calculation below , we will take @xmath83 , and assume that the background expansion just follows the @xmath72cdm history with the same @xmath79 . we use the subset of cosmological @xmath1-body simulations presented in ref . the data set of @xmath1-body simulations were created by the @xmath1-body code for modified gravity models , ecosmog @xcite , which is a modification of the mesh - based n - body code , ramses @xcite . with cubic boxes of side length @xmath84gpc and @xmath85 particles , the initial conditions were generated at redshift @xmath86 using mpgrafic[multiblock footnote omitted ] , according to the linear matter power spectrum determined by the cosmological parameters : @xmath87 , @xmath88 , @xmath89 , @xmath90 , @xmath91 , @xmath92 . indicated in ref . @xcite is a typo . ] in the analysis presented below , we mainly consider gr and @xmath0 gravity with @xmath93 ( labeled as @xmath72cdm and f4 in ref . @xcite ) , and focus on the output results at @xmath94 . in ref . @xcite , with @xmath95 independent realizations , the matter power spectra are measured in redshift space , applying the distant - observer approximation . adopting the line - of - sight direction perpendicular to each side of the simulation box , the density field is assigned on @xmath96 grids with cloud - in - cell interpolation , and the power spectra are measured for three different line - of - sight directions . here , in comparison with the pt model , we use the power spectra averaged over all line - of - sight directions and realizations . with this treatment , the measured power spectra tend to be smooth and the outliers disappear , while the estimation of the error covariance is rather complex because of the duplicated power spectrum measurement with same realizations . in what follows , for the analysis of the fitting and parameter estimation , we assume a hypothetical galaxy survey of the volume @xmath97gpc@xmath98 , and consider the statistical error limited by the cosmic variance . unless otherwise stated , the error bars of the @xmath1-body results indicate the 1-@xmath99 error of this hypothetical survey computed with the linear power spectrum ( see appendix c of ref . @xcite ) . ] finally , in addition to the redshift - space power spectra , we also compare the real - space power spectra of the density and velocity fields in ref . @xcite , which are used to estimate the valid range of pt predictions . term ( left ) and @xmath23 term ( right ) . the plotted results are at @xmath94 , and they are multiplied by @xmath100 just for illustrative purpose . the @xmath22 and @xmath23 terms are respectively expanded as @xmath101 and @xmath102 , and we here plot the scale - dependent coefficients @xmath103 and @xmath104 . the dotted lines are the results in gr , while the solid lines are those in @xmath0 gravity model with @xmath105 . those results are computed with the same linear power spectrum ( see text ) . for reference , the power spectra @xmath106 , @xmath107 , and @xmath108 are also shown in black , blue and red lines for gr case . [ fig : pkcorr_a_b],width=321 ] before presenting the redshift - space power spectrum , we first separately compute the contribution of each term in the power spectrum expression ( [ eq : tns_model ] ) , and compare it with @xmath1-body simulation . [ fig : pkcorr_a_b ] shows the results of standard pt calculation at one - loop order at @xmath94 in the gr ( dotted ) and f4 ( @xmath0 gravity with @xmath109 , solid ) cases , with the same cosmological parameters as adopted in @xmath1-body simulations . the @xmath22 ( left ) and @xmath23 ( right ) terms are plotted together with the auto- and cross - power spectra of density and velocity divergence fields , multiplied by @xmath100 . according to eqs . ( [ eq : a_term_simplified ] ) and ( [ eq : b_term2 ] ) , the @xmath22 and @xmath23 terms are expanded as @xmath101 and @xmath102 , and we here plot the scale - dependent coefficients @xmath103 and @xmath104 ( @xmath110 : magenta , @xmath111 : cyan , @xmath112 : green , @xmath113 : yellow ) . overall , the resultant amplitude of power spectrum corrections in f4 is rather larger than that in gr . the acoustic signature are clearly seen in both gr and f4 cases not only for the real - space quantities but also the rsd correction ( i.e. , @xmath22 term ) . the reason for a larger amplitude in @xmath0 gravity is mainly attributed to the scale - dependent enhancement of the linear growth factor on small scales in @xmath0 gravity , and with a large value of @xmath109 , the mechanism to recover gr is still inefficient at quasi - linear scales . this implies that the redshift - space power spectrum can be quite different between gr and f4 , and the rsd corrections ( i.e. , @xmath22 and @xmath23 terms ) would play an important role . for gr ( left ) and @xmath0 gravity with @xmath109 . top panels shows the power spectra multiplied by the cube of wavenumber , i.e. , @xmath114 , while bottom panels present the ratio of the power spectra to the linear theory predictions , @xmath115 . [ fig : pk_dd_dt_tt],width=321 ] to see the domain of applicability of the standard pt calculation , we next plot in fig . [ fig : pk_dd_dt_tt ] the real - space power spectra @xmath25 , @xmath27 , and @xmath26 from @xmath1-body simulations , and compare those with the pt results . according to a phenomenological rule calibrated with @xmath1-body simulations @xcite , the standard pt power spectrum at one - loop order is expected to agree with @xmath1-body simulations at @xmath116mpc@xmath117 with an accuracy of @xmath118 level . although the simulation results show somewhat noisy behavior , the pt predictions seem to work well at least at the scales indicated by the empirical rule , where the deviation from linear theory is around @xmath119 in both gr and f4 . the result suggests that even in the presence of a substantial difference in the linear growth , the nonlinear gravitational growth itself does not change so much between gr and modified gravity models . this would be probably true as long as the mechanism to recover gr is still inefficient at quasi - linear scales . keeping in mind the applicability of pt calculation , we now focus on the redshift - space power spectrum , and compare the pt calculations with @xmath1-body simulations . in fig . [ fig : pkred ] , top panels show the monopole ( @xmath120 ) and quadrupole ( @xmath121 ) moments of power spectrum multiplied by @xmath100 , while bottom panels present the ratio of monopole and quadrupole spectra to the linear theory prediction taking only account of the kaiser effect . the multipole power spectrum @xmath122 is defined by @xmath123 with @xmath124 being the legendre polynomials . the pt results based on an improved model of rsd [ i.e. , tns model , eq . ( [ eq : tns_model ] ) ] are depicted as solid lines , while the results without correction terms are also shown in dashed lines . in both cases , we adopt the gaussian damping function in computing pt predictions : @xmath125.\end{aligned}\ ] ] here , the velocity dispersion @xmath16 is a free parameter and is determined by fitting the model predictions to the @xmath1-body results of monopole and quadrupole spectra up to @xmath126mpc@xmath117 ( indicated by vertical arrows ) , corresponding to the valid range of pt . note that we also examined the lorentzian form , but the choice of the damping function did not change the results as long as we consider the applicable range of standard pt one - loop . for gr ( left ) and @xmath0 with @xmath109 ( right ) . top panels show the monopole and quadrupole power spectra multiplied by @xmath100 , while bottom panels present the ratio of power spectra to linear theory predictions , @xmath127 . solid and dashed lines respectively show the pt results based on the tns model [ eq . ( [ eq : tns_model ] ) ] with and without @xmath22 and @xmath23 terms . in each panel , vertical arrow indicates the maximum wavenumber used to estimate @xmath16 . [ fig : pkred],width=321 ] gravity to that of gr , @xmath128 . the results at @xmath94 is shown . solid and dashed lines are the pt predictions based on the tns model with and without @xmath22 and @xmath23 terms , while dotted lines are the linear theory predictions . the vertical arrow indicates the maximum wavenumber used to estimate @xmath16 . [ fig : qmratio],width=321 ] fig . [ fig : pkred ] shows that the model ( [ eq : tns_model ] ) successfully describes the @xmath1-body results of rsd in both gr and @xmath0 model . although the applicable range of standard pt one - loop is limited , the @xmath22 and @xmath23 terms still play an important role . in the presence of these terms , the acoustic signature of redshift - space power spectrum tends to be smeared compared to the real - space power spectrum , and this indeed improves the agreement with @xmath1-body simulations . in the panels of fig . [ fig : pkred ] , we show the reduced chi - squared statistic defined by @xmath129 ^ 2}{[\delta p^{\rm(s)}_\ell(k_i)]^2 } , \label{eq : chi2_red}\end{aligned}\ ] ] with the quantity @xmath130 being the number of degrees of freedom . here , the statistical error @xmath131 is estimated from the cosmic variance error assuming the survey volume @xmath132gpc@xmath98 . the number of fourier bins in the above summation can be inferred from the maximum wavenumber shown in fig . [ fig : pkred ] , depicted as vertical arrows . the resultant @xmath133 taking account of the @xmath22 and @xmath23 terms ( i.e. , tns model ) are clearly lower than those ignoring the corrections . gravity with @xmath134 ( middle ) and @xmath135 ( right ) . the best - fit parameter @xmath136 and the resultant values of reduced chi - squared are respectively shown in top and bottom panels . for comparison , linear theory prediction of the velocity dispersion , @xmath137 , is also depicted as solid lines in top panels . in each panel , filled and open triangles indicate the pt results based on the tns model with and without @xmath22 and @xmath23 terms , respectively . note that the maximum wavenumber in the fitting , @xmath138 , is set to @xmath139 and @xmath140mpc@xmath117 at @xmath141 and @xmath142 . [ fig : fitting_sigmav2],width=332 ] to show the quantitative difference of rsd between gr and @xmath0 gravity , fig . [ fig : qmratio ] shows the ratio of quadrupole - to - monopole ratio in f4 to that in gr , i.e. , @xmath143 . note that the errorbars of the @xmath1-body simulation shown in the panel are not the cosmic variance error , but are estimated from the @xmath1-body data of the @xmath95 realizations for a particular line - of - sight direction . the linear theory predicts a slight enhancement of the ratio , while the actual @xmath1-body result rather shows a noticeable reduction at small scales . this basically comes from a stronger suppression of the power spectra in @xmath0 gravity , as shown in fig . [ fig : pkred ] ( see bottom panel ) . fig . [ fig : fitting_sigmav2 ] summarizes the fitting results of the parameter @xmath16 together with the resultant reduced chi - squared . at @xmath94 , the fitted value of the velocity dispersion is relatively large in f4 by @xmath144 . neglecting the correction terms , the relative difference of @xmath16 between @xmath0 model and gr is more prominent ( @xmath145 ) , although the values themselves are even smaller than those taking account of the @xmath22 and @xmath23 terms . as a result , the pt prediction ignoring the corrections exhibits a strong damping behavior in fig . [ fig : qmratio ] , and tends to deviate from @xmath1-body simulations at small scales . by contrast , the prediction with @xmath22 and @xmath23 terms ( i.e. , tns model ) faithfully traces the @xmath1-body trend beyond the applicable range of the standard pt . finally , while we mainly presented the results at @xmath94 , we briefly comment on other cases at @xmath141 , where we also examined the f5 case ( @xmath0 gravity with @xmath146 ) . all the results are summarized in fig . [ fig : fitting_sigmav2 ] . at @xmath141 , the nonlinear clustering is strongly developed , and the applicable range of standard pt one - loop is quite limited . nevertheless , with a limited fitting range of @xmath147mpc@xmath117 , the pt results show an excellent performance with @xmath148 , and the prediction including the @xmath22 and @xmath23 terms gives a better agreement with @xmath1-body simulations . [ fig : fitting_sigmav2 ] shows that the fitted value of velocity dispersion in @xmath0 gravity is generally larger than that in gr , roughly consistent with the one estimated with linear theory ( solid lines ) . this suggests that a stronger damping of the power spectrum amplitude may be a good indicator for modified gravity , as pointed out by ref . @xcite ( see also ref . note , however , that the actual value of @xmath16 depends on the underlying model of rsd . further , our observable is not dark matter but galaxy distribution , which does not faithfully trace the dark matter distribution . a careful study is needed , and we leave this issue to future work . as function of the maximum wavenumber @xmath138 used for mcmc analysis . assuming the cosmic variance limited survey of the volume @xmath149gpc@xmath98 , we fit the pt template to the @xmath1-body simulation of the f4 run at @xmath94 , and derive the best - fit values and 1-@xmath99 statistical error of @xmath150 , allowing the parameter @xmath16 to be free . filled circles are the results based on the tns model [ eq . ( [ eq : tns_model ] ) ] in @xmath0 gravity , while filled triangles are the cases ignoring the @xmath22 and @xmath23 terms . open circles represent the results similar to filled circles , but the corrections @xmath22 and @xmath23 are calculated in gr . for comparison , crosses are the results ignoring not only the @xmath22 and @xmath23 terms but also the damping function @xmath18 . [ fig : fr0_mcmc],width=340 ] having confirmed that the pt model of rsd works well at quasi - linear scales , in this section , we discuss the potential impact of the pt template on the rsd measurement at quasilinear scales . in testing gravity with rsd , a primary goal would be to clarify whether gr really holds on cosmological scales or not . in this respect , the measurement of the linear growth rate , @xmath151 , provides an important clue , and with the gr - based pt template , we may look for a possible deviation of @xmath32 from gr prediction . one important property in a large class of modified gravity models , including @xmath0 gravity , is the scale dependence of @xmath32 . thus , a detection of scale - dependent @xmath32 immediately implies the deviation of gravity from gr . the crucial question is how well one can detect or characterize such scale dependence in a model - independent manner . alternatively , we may consider some specific gravity models , and try to directly constrain the models themselves . in this case , the linear growth rate @xmath32 might not be an appropriate indicator to characterize a possible deviation from gr . rather , one tries to directly constrain the model parameter of modified gravity ( e.g. , @xmath150 in the case of @xmath0 gravity ) . then , with the prior assumption of the specific gravity models , the question is how well we can accurately constrain the model parameter based on the pt template of rsd in an un - biased way . below , we will separately consider these two issues , and examine the parameter estimation analysis . note that we will adopt below @xmath152 of eq . ( [ eq : chi2_red ] ) to estimate the goodness - of - fit . strictly speaking , this is not entirely correct , because the nonlinear gravitational evolution induces the non - gaussian contribution , which produces non - vanishing power spectrum covariances between diffrent fourier modes . however , it is shown in the gr case that as long as we consider the quasi - linear scales at moderately high redshift , the off - diagonal componets are small enough , and the diagonal components can be approximately described by the simple gaussian contribution , leading to a negligible influence on the parameter etstimation ( e.g. , @xcite ) . we thus expect that the same would be true in our case of the @xmath0 gravity model that is close to gr , and our simple treatment with gaussian error contribution would be validated at quasi - linear scales . let us first consider the model - dependent analysis to constrain the model parameter of modified gravity , assuming the @xmath0 gravity with @xmath134 as our fiducial gravity model . for specific functional form with eq . ( [ eq : model_approx ] ) [ or eq . ( [ eq : model ] ) ] , the parameter @xmath153 is the only parameter characterizing a deviation of gravity from gr . thus , the test of gravity is made possible with constraining the model parameter @xmath153 by fitting the theoretical template to the data set of redshift - space power spectrum . here , as a simple demonstration , we ignore the effect of galaxy bias , and allowing @xmath153 to flow , we fit the pt template to the @xmath1-body data at @xmath94 . [ fig : fr0_mcmc ] shows the results of parameter estimation based on the markov chain monte carlo ( mcmc ) technique . assuming the hypothetical survey limited by the cosmic variance error with the survey volume @xmath149gpc@xmath98 , the best - fit value of @xmath153 and the 1-@xmath99 statistical error are derived , and are plotted ( top ) as function of maximum wavenumber , @xmath138 , together with the reduced chi - squared statistic @xmath154 ( bottom ) , where @xmath138 represents the range of the wavenumber used for parameter estimation . note here that the number of free parameters is two , i.e. , @xmath153 and @xmath16 . accordingly , the derived constraint is rather tight , and a slight discrepancy between the template and data can lead to a biased estimation of the @xmath153 . [ fig : fr0_mcmc ] shows that only the improved model of rsd computed in @xmath0 gravity ( filled circles ) recovers the fiducial @xmath153 out to @xmath155mpc@xmath117 , corresponding to the applicable range of standard pt one - loop . a slight change of the pt template , depicted as open circles and filled triangles , leads to a biased estimation of the model parameter . ignoring the damping function @xmath18 ( crosses ) further adds a large systematic error this is even true at @xmath156mpc@xmath117 . left panel of fig . [ fig:2d_mcmc ] shows the representative result of the two - dimensional constraints on @xmath153 and @xmath16 taken from fig . [ fig : fr0_mcmc ] , where we fix the maximum wavenumber to @xmath155mpc@xmath117 . the meaning of color types are the same as in fig . [ fig : fr0_mcmc ] , and in each error contour , inner and outer contours respectively represent the @xmath142-@xmath99 ( @xmath157 c.l . ) and @xmath158-@xmath99 ( @xmath159 c.l . ) constraints . overall , the degeneracy between @xmath153 and @xmath16 is weak , and the result suggests that at the scales accessible by pt template , the model parameter @xmath153 can be constrained down to @xmath160 from future rsd measurements . note , however , that this is only true when we properly take account of the effect of modified gravity in computing the pt template . most of the analysis in the literature considered the effect of modified gravity only in the linear growth rate @xmath32 and incorporated it into the gr - based template to constrain the model parameter @xmath153 using the measurements of rsd ( e.g. , @xcite for recent works ) . the right panel of fig . [ fig:2d_mcmc ] indeed demonstrates such a case . that is , we adopt the gr - based pt template in which the effect of modified gravity is only incorporated in the linear growth rate @xmath32 . in gr , the velocity - divergence field @xmath35 is known to be factorized as @xmath161 , where @xmath162 is perturbatively expanded as @xmath163^n\,\widetilde{\theta}_n({\mbox{\boldmath$k$}})$ ] . as a result , at a given redshift , the pt template of the redshift - space power spectrum is described as the function of @xmath2 , @xmath9 and @xmath32 , i.e. , @xmath164 . since the growth rate @xmath32 controlls the strength of rsd , we naively expect that simply incorporating the scale - dependent @xmath32 in modified gravity into the pt template allows us to faithfully constrain the model parameter @xmath153 . however , this actually leads to a biased estimation of the model parameter @xmath153 , as shown in the contour with orange color of fig . [ fig:2d_mcmc ] . the reason for the large systematic bias is ascribed to the fact that the modification of gravity not only alters the linear growth rate but also affects the shape of the real - space power spectra because of the scale - dependent growth , as clearly shown in fig . [ fig : pk_dd_dt_tt ] . thus , for an unbiased estimation of @xmath153 , we need to additionally incorporate the effect of _ gravity bias _ , that accounts for the relative difference of the clustering amplitude between gr and @xmath0 gravity , into the pt template . the contour with magenta color is the results taking account of this gravity bias , simply assuming the following relation : @xmath165 where @xmath166 is the density field in @xmath1-body simulation , whilst @xmath167 is the density field for the pt calculation . the function @xmath168 characterizes the scale - dependent growth relative to the gr prediction , and we adopt here the functional form similar to those frequently used to model the galaxy bias ( e.g. , @xcite ) . allowing the parameters @xmath169 and @xmath170 to float , the result marginally reproduces the fiducial value of @xmath153 , and the goodness - of - fit quantified by @xmath154 is improved . with the increased number of free parameters , however , constraining power is significantly reduced , and the size of error contour indeed becomes large ( c.f . left panel of fig . [ fig:2d_mcmc ] ) . this proves that the heterogeneous pt template is insufficient to tightly constrain the model parameter of modified gravity , and a full pt modeling taking proper account of the modified gravity is required for unlocking the full power of precision rsd measurement . to spatially vary in three wavenumber bins . adopting the gravity bias prescription given in eq ( [ eq : gravity_bias ] ) and fixing the maximum wavenumber to @xmath155mpc@xmath117 , we derive the constraint on @xmath32 in each wavenumber bin . the vertical errorbars indicate the @xmath142-@xmath99 error assuming the cosmic - variance limited survey of @xmath149gpc@xmath98 , and the dotted and solid lines respectively represent the linear theory prediction and its binned average in the underlying @xmath0 gravity model . [ fig : mcmc2 ] , width=340 ] consider next the model - independent test of gr , and discuss how well we can characterize or detect the scale dependence of the linear growth rate , @xmath32 . here , for illustrative purpose , we examine the two simple cases . one is to divide the power spectrum data into several wavenumber bins , and in each bin , we try to estimate @xmath32 to see a possible deviation from spatially homogeneous @xmath32 . the other case is to assume a specific functional form of @xmath32 , and to constrain its parameters . in both cases , similar to the analysis shown in right panel of fig . [ fig:2d_mcmc ] , we adopt the gr - based pt template with an improved model of rsd ( i.e. , tns model ) , and take account of the gravity bias in eq . ( [ eq : gravity_bias ] ) . we then fit the template to the monopole and quadrupole power spectra at @xmath94 measured from @xmath1-body simulations of @xmath0 gravity with @xmath134 . [ fig : mcmc2 ] shows the result of mcmc analysis for the binned linear growth rate , where we set @xmath155mpc@xmath117 , and divide the power spectrum data into three equal bins . dotted and solid lines represents the linear growth rate of the @xmath0 gravity with and without binning , while the vertical errorbars of the binned results indicate the @xmath142-@xmath99 statistical uncertainty derived from the mcmc analysis , marginalized over other nuisance parameters . note that number of free parameters is @xmath95 . the best - fit value of @xmath32 in each bin is close to the fiducial value , but slightly away from linear theory prediction except for the central bin . as a result , the errorbars share almost the same value of @xmath32 , and no notable trend of the scale - dependent growth is found from the binned estimate of @xmath32 . [ fig : mcmc2_fk_f0_float ] examines the other case , in which we assume a specific functional form of @xmath32 given below : @xmath171 . \label{eq : f_approx}\end{aligned}\ ] ] allowing the parameters @xmath172 , @xmath173 and @xmath174 to float , we perform the mcmc analysis . again , the number of free parameters is @xmath95 , and we set @xmath155mpc@xmath117 . note that for the scales of our interest , eq . ( [ eq : f_approx ] ) is shown to accurately describe the scale - dependent linear growth rate of @xmath0 gravity , and fitting directly eq . ( [ eq : f_approx ] ) to the linear theory prediction at @xmath94 , we obtain @xmath175 , @xmath176 , and @xmath177mpc@xmath117 . the left panel of fig . [ fig : mcmc2_fk_f0_float ] shows the two - dimensional projected errors on the parameters , @xmath172 , @xmath173 , and @xmath178 . the mcmc analysis of the rsd measurement favors non - zero values of these parameters , strongly indicating a deviation from spatially constant @xmath32 . however , a closer look at two - dimensional contours reveals a substantial difference in @xmath173 between the best - fit result and the directly fitted value , @xmath179 . as a result , the mcmc result is unable to reproduce the underlying scale - dependent linear growth rate . right panel of fig . [ fig : mcmc2_fk_f0_float ] shows the constraint on the scale - dependence of @xmath32 . based on the best - fit values and the associated @xmath142-@xmath99 errors shown in left panel of fig . [ fig : mcmc2_fk_f0_float ] , the best - fit curve is plotted in red solid line , and its @xmath142-@xmath99 statistical uncertainty is shown in red shaded region . the scale - dependence inferred from the mcmc result is rather stronger than that of the linear theory prediction ( black dashed ) . these two examples imply that the model - independent detection and characterization of the scale - dependent @xmath32 are generally difficult , and the results are rather sensitive to the choice of parameterized form of the linear growth rate . this is presumably because each of the parameters characterizing the scale - dependent @xmath32 can not be determined locally , but rather it must be estimated with a wide range of wavenumber . then the parameters tend to be highly correlated with each other , leading to a biased estimation . in this respect , a sophisticated treatment with principal component analysis may provide a way to robustly detect a scale - dependent @xmath32 . in this paper , we studied how well we can clarify the nature of gravity at large scales with redshift - space distortions ( rsd ) , especially focusing on the quasilinear regime of the gravitational evolution . while most of previous works have been done with the theoretical template assuming gr as underlying gravity theory , we here developed a new perturbation theory ( pt ) prescription for rsd in the general context of the modified gravity models . extending our previous works on the improved model of rsd proposed by ref . @xcite , we applied the standard pt framework by ref . @xcite , which has been formulated to deal with a wide class of modified gravity models , to the computation of the redshift - space power spectrum . as a specific application , in this paper , we consider the @xmath0 gravity model , and compared the pt prediction of rsd with results of @xmath1-body simulations . albeit the limited applicable range of the standard pt , the pt results successfully describe the @xmath1-body simulations , and the predicted monopole and quadrupole spectra quantitatively agree with @xmath1-body results . then , we next considered how well we can characterize and/or constrain the deviation of gravity from gr . one obvious approach is to first assume a specific modified theory of gravity as an underlying gravity model and constrain their model parameters . using the pt as a theoretical template , we performed the parameter estimation analysis , and checked if the theoretical template correctly recovers the fiducial value of the model parameter in the @xmath1-body simulations . adopting the improved model of rsd [ tns model , eq . ( [ eq : tns_model ] ) ] , a full pt template calculated in the modified gravity model was found to reproduce the correct model parameter , while a slight deficit in the pt template led to a biased parameter estimation . as another approach , we have also examined the model - independent analysis , and based on the pt template calculated in gr , we tried to characterize the scale - dependent linear growth rate from monopole and quadrupole power spectra . without assuming any modified gravity model , the parameterization of the scale - dependent linear growth rate @xmath32 is necessary , and the parameters characterizing @xmath32 are highly correlated in general . our simple two examples suggest that the results are highly sensitive to the choice of parameterization , and it is generally difficult to characterize the scale - dependence of @xmath32 in an unbiased manner unless employing some sophisticated methods such as principal component analysis . throughout the paper , we have worked with the standard pt , but the standard pt is known to have a bad convergence property . while we can still get a fruitful constraint on modified gravity models , resummed pt schemes with a wide applicable range are highly desirable to improve the observational constraint . a development of improved pt template in redshift space is an important future direction ( see @xcite for recent attempt ) . another important issue is the application of the present prescription to the real measurement of rsd . with full pt implementation of the theoretical template , a tight cosmological constraint is expected to be obtained in a robust and unbiased way . in doing this , however , a proper account of the galaxy bias would be crucial . although the present paper mainly focused on the matter power spectrum , the galaxy bias would be also affected by the modification of gravity , and this may produce a non - trivial scale - dependent shape of the observed power spectrum . in fact , the @xmath1-body study of the halo clustering properties has revealed that the halo bias in @xmath0 gravity is systematically lower than that in gr ( e.g. , @xcite ) . since even the velocity dispersion and clustering amplitude of the dark matter distribution in @xmath0 gravity differ from those in gr ( see figs . [ fig : pk_dd_dt_tt ] and [ fig : pkred ] in sec . [ sec : rsd_frmodel ] ) , coupled with the nonlinear gravity , this could impose a non - trivial trend in the halo / galaxy bias ( see e.g. , @xcite for recent study on the abundance and clustering of halos and galaxies ) . hence , a careful study of the halo / galaxy bias is necessary together with extensive tests with @xmath1-body mock catalogs toward an unbiased test of gravity . we are grateful to elise jennings for providing us the power spectrum data of @xmath1-body simulations . we also thank baojiu li and gong - bo zhao for allowing us to use the simulation data . this work is supported in part by a grant - in - aid for scientific research from the japan society for the promotion of science ( no . 23740186 for t.h and no . 24540257 for a.t ) . k.k is supported by stfc grant st / k0090x/1 , the european research council and the leverhulme trust . t.h . acknowledges a support from mext hpci strategic program . in this appendix , after briefly reviewing the formalism developed in ref . @xcite , we derive general expressions for the second - order pt kernel @xmath56 , as well as the linear growth factor @xmath55 in modified gravity models . in sec . [ subsec : evolution_eqs ] , we begin by reviewing the framework to treat the evolution of matter fluctuations in modified gravity models . we then develop the perturbation theory and derive the pt kernels up to the second - order in sec . [ subsec : pt_kernel ] . the explicit expressions for pt kernels are given in specific modified gravity models , i.e. , @xmath0 gravity and dvali - gabadadze - porratti ( dgp ) models . let us first consider the matter sector . apart from the force law of gravity , the basic equations governing the evolution of matter sector is basically described by the conservation of energy momentum tensor , which would remain unchanged even in the modified gravity model . hence , under the single - stream approximation , the matter fluctuations are treated as pressureless fluid flow , whose evolution equations are the continuity and euler equations : @xmath180=0 , \label{eq : eq_continuity}\\ & \frac{\partial { \mbox{\boldmath$v$}}}{\partial t } + h\,{\mbox{\boldmath$v$}}+ \frac{1}{a}({\mbox{\boldmath$v$}}\cdot\nabla)\cdot{\mbox{\boldmath$v$}}=-\frac{1}{a}\nabla\psi , \label{eq : eq_euler}\end{aligned}\ ] ] where @xmath181 is the newton potential . on the other hand , for the gravity sector , there may appear a new scalar degree of freedom referred to as the scalaron , which results in a large - distance modification to the gravity . on large scales , the scalaron @xmath182 mediates the scalar force , and behaves like the brans - dicke scalar field without potential and self - interactions , while it should acquire some interaction terms on small scales , which play an important role to recover gr . indeed , for several known mechanisms such as chameleon and vainshtain mechanisms ( e.g. , @xcite ) , the nonlinear interaction terms naturally arise and eventually become dominant , leading to a recovery of gr . as a result , even on subhorizon scales , the poisson equation is modified , and is coupled to the field equation for scalaron @xmath182 with self - interaction term : @xmath183 with @xmath62 and @xmath184 being the brans - dicke parameter . here , we have used the quasi - static approximation and neglected the time derivatives of the perturbed quantities compared with the spatial derivatives . this treatment is always valid as long as we consider the evolution of matter fluctuations inside the hubble horizon . the function @xmath185 represents the nonlinear self - interaction , and it can be expanded as @xmath186 in fourier space , eqs . ( [ eq : eq_continuity])([eq : eom_scalaron ] ) can be reduced to a more compact form . assuming the irrotationality of fluid quantities , the velocity field is expressed in terms of velocity divergence , @xmath20 . then , we have @xcite , @xmath187 with the mode - coupling kernels , @xmath188 and @xmath189 given by @xmath190 in the above , the function @xmath191 characterizes the deviation of the newton constant from gr , while the quantity @xmath192 is originated from the non - linear interactions of the scalaron , which is responsible for the recovery of gr at small scales . the functional form of these are obtained from the poisson equation and field equation for scalaron , and the expressions relevant for the second - order perturbations are respectively given by @xcite : @xmath193 here , in deriving the last expression , we perturbatively solve the scalaron field @xmath182 in terms of @xmath3 using eqs . ( [ eq : eom_scalaron ] ) and ( [ eq : i_expansion ] ) . the explicit functional form of @xmath191 or @xmath194 , and @xmath195 depends on actual modified gravity models , which will be specified later . let us now perturbatively solve the evolution equations ( [ eq : eom1 ] ) and ( [ eq : eom2 ] ) . consider first the linear - order solutions . ignoring all the non - linear terms in eqs . ( [ eq : eom1 ] ) and ( [ eq : eom2 ] ) , we obtain @xmath196 the solutions @xmath197 and @xmath198 can be formally expressed as @xmath199 where the function @xmath200 is the initial density field [ see eq . ( [ eq : initial_pk ] ) ] . the function @xmath201 is the linear growth factor , and satisfies the following evolution equation : @xmath202 accordingly , the first - order pt kernels are @xmath203 @xmath204\,\delta_0({\mbox{\boldmath$k$}}_1)\delta_0({\mbox{\boldmath$k$}}_2 ) , \label{eq : eom_2nd_pt}\\ & \theta^{(2)}({\mbox{\boldmath$k$}})=-\frac{1}{h}\dot{\delta}^{(2)}({\mbox{\boldmath$k$}})+\frac{1}{h } \int\frac{d^3{\mbox{\boldmath$k$}}_1d^3{\mbox{\boldmath$k$}}_2}{(2\pi)^3}\,\delta_d({\mbox{\boldmath$k$}}-{\mbox{\boldmath$k$}}_{12})\ , \alpha({\mbox{\boldmath$k$}}_1,{\mbox{\boldmath$k$}}_2)\,\dot{d}_1\,d_2 \,\delta_0({\mbox{\boldmath$k$}}_1)\,\delta_0({\mbox{\boldmath$k$}}_2)\end{aligned}\ ] ] @xmath208\,\delta_0({\mbox{\boldmath$k$}}_1)\,\delta_0({\mbox{\boldmath$k$}}_2 ) , \label{eq:2nd_order_solutions1 } \\ & \theta^{(2)}({\mbox{\boldmath$k$}};t ) = \frac{1}{h}\int\frac{d^3{\mbox{\boldmath$k$}}_1d^32{\mbox{\boldmath$k$}}_2}{(2\pi)^3 } \delta_d({\mbox{\boldmath$k$}}-{\mbox{\boldmath$k$}}_{12 } ) \nonumber\\ & \qquad\qquad\qquad\times \left[\frac{1}{2}\left\{\left(\dot{d}_1d_2-\dot{d}^{(2)}_{1,2}\right)\alpha_{1,2}+ \left(\dot{d}_2d_1-\dot{d}^{(2)}_{2,1}\right)\alpha_{2,1}\right\ } -\dot{e}^{(2)}_{1,2}\,\beta_{1,2}-\dot{f}^{(2)}_{1,2}\right]\ , \delta_0({\mbox{\boldmath$k$}}_1)\,\delta_0({\mbox{\boldmath$k$}}_2 ) . \label{eq:2nd_order_solutions2}\end{aligned}\ ] ] thus , the _ symmetrized _ second - order pt kernels @xmath56 are respectively given by @xmath209 . \label{eq : f2_velocity}\end{aligned}\ ] ] here , the functions @xmath210 , @xmath211 , and @xmath212 are the second - order growth functions , whose governing equations are @xmath213 with the operator @xmath214 given by @xmath215 below , we will present a more explicit expression for evolution equations of the growth functions in @xmath0 gravity and dgp models . in @xmath0 gravity models described in eq . ( [ eq : action_fr ] ) , the scalaron @xmath182 is identified with @xmath216 [ eq . ( [ eq : scalaron_fr ] ) ] , and it behaves like the brans - dicke scalar with @xmath64 and the nonlinear interaction @xmath217 . then , the functions @xmath191 and @xmath195 are generally given by @xcite @xmath218 where we define @xmath219 and @xmath220 . in @xmath0 gravity , all the second - order growth functions @xmath221 , @xmath222 , and @xmath223 as well as the linear growth factor @xmath201 are scale - dependent , and no simplified expressions are obtained without invoking any approximations . we numerically solve evolution equations below : with the linear operator being @xmath225 here , the function @xmath221 is asymmetric with respect to the change of the arguments , i.e. , @xmath226 . the second - order growth functions are generally given as the function of @xmath227 , @xmath228 , and @xmath229 . note that in deriving the evolution equations above , we did not specify the functional form of @xmath0 , and the functions , @xmath230 and @xmath231 , still remains unspecified . to solve the equations , in this paper , we consider the specific function given in eq . ( [ eq : model ] ) , and then the functions @xmath230 and @xmath231 are expressed in terms of the background quantities . as another example of modified gravity model , we consider the dvali - gabadadze - porratti ( dgp ) braneworld model @xcite . in dgp models , the brans - dicke parameter of the scalaron becomes time - dependent , and is given by @xcite @xmath232 with @xmath233 , which represents two distinct branches of the background solutions ( @xmath234 : self - accelerating branch , @xmath235 normal branch ) . the parameter @xmath236 is the crossover scale which characterizes the ratio of @xmath237 newton constant to @xmath238 newton constant . in this model , the nonlinear interaction of the scalaron comes from the vainshtein mechanism . as a result , the functions @xmath191 and @xmath195 are respectively given by @xcite @xmath239 then , the second - order growth functions @xmath221 , @xmath222 , and @xmath223 as well as the linear growth factor @xmath201 become all independent of scale . to further get a simplified expression , we may employ the einstein - de sitter approximation . in this approximation , the non - linear growth functions in the higher - order pt solutions are first obtained assuming the einstein - de sitter background , and they are expressed in terms of the scale factor . then , simply replacing the scale factor with the linear growth factor @xmath240 , we obtain an approximate description of the non - linear growth functions : @xmath241 here , the evolution equation for linear growth factor is given by @xmath242 substituting eq . ( [ eq : d2_e2_eds ] ) into eqs . ( [ eq : f2_density ] ) and ( [ eq : f2_velocity ] ) , we obtain the approximate expressions for the symmetrized pt kernels : @xmath243 with @xmath244 . in the above , we rewrite the second - order growth function @xmath223 as @xmath245 , with @xmath246 being the scale - independent function satisfying the following evolution equation : @xmath247 eqs . ( [ eq : f2_density_dgp ] ) and ( [ eq : f2_velocity_dgp ] ) coincide with the results in ( b5 ) and ( b6 ) of ref . the first term of the right - hand - side in each kernel is exactly the same kernel as found in gr , while the second term is originated from the non - linear interactions of the scalar degree of freedom .

in this paper , dynamical systems theory and bifurcation theory are applied to investigate the rich dynamical behaviours observed in three simple disease models . the 2- and 3-dimensional models we investigate have arisen in previous investigations of epidemiology , in - host disease , and autoimmunity . these closely related models display interesting dynamical behaviors including bistability , recurrence , and regular oscillations , each of which has possible clinical or public health implications . in this contribution we elucidate the key role of backward bifurcation in the parameter regimes leading to the behaviors of interest . we demonstrate that backward bifurcation facilitates the appearance of hopf bifurcations , and the varied dynamical behaviors are then determined by the properties of the hopf bifurcation(s ) , including their location and direction . a maple program developed earlier is implemented to determine the stability of limit cycles bifurcating from the hopf bifurcation . numerical simulations are presented to illustrate phenomena of interest such as bistability , recurrence and oscillation . we also discuss the physical motivations for the models and the clinical implications of the resulting dynamics .

1504.05260

convex incidence rate , backward bifurcation , supercritical and subcritical hopf bifurcations , recurrence , bistability , disease model in the mathematical modelling of epidemic diseases , the fate of the disease can be predicted through the uninfected and infected equilibria and their stability . the basic reproduction number , @xmath1 , represents the average number of new infectives introduced into an otherwise disease - free system by a single infective , and is usually chosen as the bifurcation parameter . if the model involves a forward bifurcation , the uninfected equilibrium is in general globally asymptotically stable @xcite , characterized by @xmath2 , and infection fails to invade in this parameter regime . the threshold @xmath3 defines a bifurcation ( or critical ) point , and when @xmath4 , a stable infected equilibrium emerges . this simple exchange of stability implies that complex dynamics will not typically occur in forward bifurcation . in contrast , backward bifurcation describes a scenario in which a turning point of the infected equilibrium exists in a region where all state variables are positive , and @xmath5 . this induces multiple infected equilibria , disrupting the global stability of the uninfected equilibrium . multiple stable states ( e.g. , bistability ) may likewise appear in @xcite , and yu et al . ( submitted for publication ) . instead of converging globally to the uninfected equilibrium when @xmath2 , the solution may approach an infected equilibrium , depending on initial conditions . in practice , the phenomenon of backward bifurcation gives rise to new challenges in disease control , since reducing @xmath1 such that @xmath2 is not sufficient to eliminate the disease @xcite . instead , @xmath1 needs to be reduced past the critical value given by the turning point @xcite , since the result in yu et al . ( submitted for publication ) shows that the uninfected equilibrium in backward bifurcation is globally stable if @xmath1 is smaller than the turning point . furthermore , an infective outbreak or catastrophe may occur if @xmath1 increases and crosses unity , while the upper branch of the infected equilibrium remains stable @xcite . in addition , oscillation or even recurrent phenomena may occur if uninfected and infected equilibria coexist in a parameter range , and both are unstable @xcite . @xcite predicted oscillations arising from backward bifurcation , and @xcite pointed out that the unstable infected equilibrium `` commonly arises from hopf bifurcation '' , but did not demonstrate oscillations . several mechanisms leading to backward bifurcation have been proposed , such as partially effective vaccination programs @xcite , educational influence on infectives behavior @xcite , the interaction among multi - group models @xcite and multiple stages of infection @xcite . in this study , we will investigate the emergence of backward bifurcation in three simple disease models which have arisen in the study of epidemiology , in - host disease and autoimmunity . in each case , we find that backward bifurcation facilitates the emergence of hopf bifurcation(s ) , and hopf bifurcation in turn underlies a range of complex and clinically relevant dynamical behaviors . a central theme in our investigation is the role of the incidence rate in the epidemiological and in - host disease models . the incidence rate describes the speed at which an infection spreads ; it denotes the rate at which susceptibles become infectives . under the assumptions of mass action , incidence is written as the product of the infection force and the number of susceptibles . for example , if @xmath6 and @xmath7 denote the susceptible and infective population size respectively , a bilinear incidence rate , @xmath8 ( where @xmath9 is a positive constant ) , is linear in each of the state variables : @xmath6 and @xmath7 . the possibility of saturation effects @xcite has motivated the modification of the incidence rate from bilinear to nonlinear . saturation occurs when the number of susceptible contacts per infective drops off as the proportion of infectives increases . a nonlinear incidence rate , therefore , typically increases sublinearly with respect to the growth of the infective population , and may finally reach an upper bound . the development of nonlinear incidence was first investigated in the form @xmath10 , where @xmath9 , @xmath11 , and @xmath12 are positive constants @xcite . other forms of nonlinear incidence have also been analysed , such as @xmath13 @xcite , and @xmath14 @xcite . since the nonlinear incidence functions described above were often developed to incorporate saturation effects , these functions are typically concave at realistic parameter values . @xcite used this feature to derive general results for disease models with concave incidence . they proved that standard epidemiological models with concave incidence functions will have globally asymptotically stable uninfected and infected equilibria for @xmath2 and @xmath4 , respectively . more specifically , denoting the incidence rate function as @xmath15 , where @xmath16 is the population size , the classical sirs model considered in @xcite takes the form @xmath17 where @xmath18 , @xmath19 , and @xmath20 represent the birth / death rate , the recovery rate and the loss of immunity rate , respectively . when @xmath21 , system ( [ paper3_eq1 ] ) becomes an sir model . assuming that the total population size is constant , that is , @xmath22 , the above system can be reduced to a 2-dimensional model : @xmath23 moreover , it is assumed in @xcite that the function @xmath15 , denoting the incidence rate , satisfies the following three conditions : @xmath24 & \frac{\partial f(s,\,i,\,n)}{\partial i}>0 , \quad \frac{\partial f(s,\,i,\,n)}{\partial s}>0 , \quad \forall \ ; s,\,i > 0 \label{paper3_eq3b}\\[0.5ex ] & \frac{\partial^2 f(s,\,i,\,n)}{\partial i^2}\leq 0 , \quad \forall \ ; s,\,i > 0 \label{paper3_eq3c}.\end{aligned}\ ] ] [ paper3_eq3 ] the first two conditions ( [ paper3_eq3a ] ) and ( [ paper3_eq3b ] ) are necessary to ensure that the model is biologically meaningful . the third condition ( [ paper3_eq3c ] ) implies that the incidence rate @xmath15 , is concave with respect to the number of infectives . it is also assumed that @xmath25 evaluated at the uninfected equilibrium is proportional to the basic reproduction number @xmath1 @xcite , and thus should be a positive finite number @xcite . korobeinikov and maini first considered @xmath26 , or @xmath27 , and showed that forward bifurcation occurs in model ( [ paper3_eq2 ] ) with a concave incidence function . they further proved that the uninfected equilibrium @xmath28 and the infected equilibrium @xmath29 are globally asymptotically stable , when @xmath30 and @xmath4 , respectively . in the sections to follow , for an incidence rate function @xmath31 , satisfying ( [ paper3_eq3a ] ) and ( [ paper3_eq3b ] ) , we define @xmath31 as concave , if it satisfies ( [ paper3_eq3c ] ) ; as convex , if @xmath32 , @xmath33 ; and as convex - concave , if there exist @xmath34 , such that @xmath35 , @xmath36 , and @xmath37 , @xmath38 , @xmath39 , for @xmath40 , @xmath41 , @xmath42 . several models closely related to ( [ paper3_eq2 ] ) have been previously studied . for example , by adding a saturating treatment term to model ( [ paper3_eq2 ] ) with a concave incidence rate , @xcite showed that this model may yield backward bifurcation and hopf bifurcation . with an even more sophisticated nonlinear incidence rate function : @xmath13 , where @xmath43 , @xcite proved that a reduced 2-dimensional sirs model could exhibit backward bifurcation , hopf bifurcation , and even bogdanov - takens bifurcation and homoclinic bifurcation . although the choice of @xmath43 was not motivated by a specific physical process , this important result demonstrates that a nonlinear incidence rate can induce backward bifurcation , and further generate complex dynamics in a simple disease model . one of the focal points of our study will be a convex incidence function which arose in a 4-dimensional hiv antioxidant therapy model @xcite . in this model , the infectivity of infected cells was proposed to be an increasing function of the density of reactive oxygen species , which themselves increase as the infection progresses . in @xcite , meaningful parameter values were carefully chosen by data fitting to both experimental and clinical results . in this parameter regime , the model was observed to capture the phenomenon of viral blips , that is , long periods of undetectable viral load punctuated by brief episodes of high viral load . viral blips have been observed clinically in hiv patients under highly active antiretroviral therapy @xcite , and have received much attention in the research literature , both by experimentalists @xcite and mathematicians @xcite . nonetheless , the mechanisms underlying this phenomenon are still not thoroughly understood @xcite . we recently re - examined the model developed in @xcite , with the aim of providing new insight into the mechanism of hiv viral blips @xcite . focusing on the dynamics of the slow manifold of this model , we reduced the dimension of the 4-dimensional model by using quasi - steady state assumptions . after a further generalization and parameter rescaling process , a 2-dimensional in - host hiv model @xcite was obtained , given by @xmath44 where @xmath45 and @xmath46 denote the concentrations of the uninfected and infected cells respectively . the constant influx rate and the death rate of @xmath46 have been scaled to @xmath47 . the death rate of @xmath45 is @xmath48 . the 2-dimensional infection model above ( [ paper3_eq4 ] ) , reduced from the 4-dimensional hiv model @xcite , preserves the viral blips observed in the hiv model . importantly , system ( [ paper3_eq4 ] ) is equivalent to the sir model ( [ paper3_eq2 ] ) , except that the incidence function is convex , as we will show in section [ hopfconvex ] . this equivalence can be demonstrated if we set @xmath49 , @xmath50 , and @xmath51 with @xmath52 and @xmath53 . in this case , system ( [ paper3_eq2 ] ) is rescaled to @xmath54 which takes the same form as system ( [ paper3_eq4 ] ) . therefore , although system ( [ paper3_eq2 ] ) arises in epidemiology and system ( [ paper3_eq4 ] ) was derived as an in - host model , they are mathematically equivalent in this sense . we will refer to both systems ( [ paper3_eq2 ] ) and ( [ paper3_eq4 ] ) as infection models . in previous work @xcite , we analyze the recurrent behavior which emerges in system ( [ paper3_eq4 ] ) in some detail . recurrence is a particular form of oscillatory behavior characterized by long periods of time close to the uninfected equilibrium , punctuated by brief episodes of high infection @xcite . thus hiv viral blips are an example of recurrent behavior , but recurrence is a more general feature of many diseases @xcite . we have demonstrated that the increasing and saturating infectivity function of system ( [ paper3_eq4 ] ) is critical to the emergence of recurrent behaviour . this form of an infectivity function corresponds to a convex incidence rate function in the associated 2-dimensional infection model ( [ paper3_eq4 ] ) , and can likewise induce recurrence in this model . convex incidence has been previously suggested to model ` cooperation effects ' in epidemiology @xcite , or cooperative phenomena in reactions between enzyme and substrate , as proposed by @xcite . the rest of this paper is organized as follows . in section 2 , we study two 2-dimensional infection models , both closely related to system ( [ paper3_eq2 ] ) . we show that system ( [ paper3_eq2 ] ) with either ( a ) a concave incidence rate and saturating treatment term or ( b ) a convex incidence rate as shown in system ( [ paper3_eq4 ] ) , can exhibit backward bifurcation ; we then identify the necessary terms in the system equations which cause this phenomenon . in section 3 , we demonstrate that in both models , backward bifurcation increases the likelihood of a hopf bifurcation on the upper branch of the infected equilibrium . studying system ( [ paper3_eq4 ] ) in greater detail , we illustrate how the location of the hopf bifurcations and their directions ( supercritical or subcritical ) , determine the possible dynamical behaviors , concluding that backward bifurcation facilitates hopf bifurcation(s ) , which then underly the rich behaviours observed in these models . in section 4 , we explore backward bifurcation further , presenting an autoimmune disease model which exhibits negative backward bifurcation , that is , a bifurcation for which the turning point when @xmath5 is located in a region where one or more state variables is negative . although this bifurcation introduces two branches of the infected equilibrium , we demonstrate that , in the biologically feasible area , only forward bifurcation exists in this model . we then present a modification to this autoimmune model , motivated by the recent discovery of a new cell type , which generates a negative backward bifurcation and hopf bifurcation , and allows recurrent behavior to emerge . a conclusion is drawn in section 5 . in this section , we study backward bifurcation in two 2-dimensional infection models . in particular , we explore the essential terms and parameter relations which are needed to generate backward bifurcation . furthermore , we examine the convex incidence rate , and reveal its underlying role in determining the emergence of backward bifurcation . first , we consider the sir model with concave incidence , described by the following equations @xcite : @xmath55 where @xmath6 , @xmath7 and @xmath56 denote the number of susceptible , infective , and recovered individuals , respectively ; @xmath57 is the constant recruitment rate of susceptibles ; @xmath58 , @xmath59 , and @xmath60 represent the rates of natural death , recovery , and the disease - induced mortality , respectively . note that the function @xmath61 is an incidence rate of the form @xmath62 @xcite , when @xmath63 . here , @xmath9 is the infection rate , and @xmath64 measures the inhibition effect . since the variable @xmath56 is not involved in the first two equations , system ( [ paper3_eq5 ] ) can be reduced to a 2-dimensional model as @xmath65 in @xcite , an additional assumption regarding limited medical treatment resources is introduced to the above model , leading to a model with a saturating treatment term , given by @xmath66 where the real , positive parameter @xmath20 represents the maximal medical resources per unit time , and the real , positive parameter @xmath67 is the half - saturation constant . for simplicity , let the functions on the right - hand side of the equations in ( [ paper3_eq7 ] ) be @xmath68 and @xmath69 , respectively . then , the equilibrium solutions of system ( [ paper3_eq7 ] ) are obtained by solving the following algebraic equations : @xmath70 and @xmath71 , from which the disease - free equilibrium can be easily obtained as @xmath72 . for the infected equilibrium @xmath73 , @xmath74 is solved from @xmath75 as @xmath76 . then , substituting @xmath77 into @xmath78 yields a quadratic equation of the form @xmath79 which in turn gives two roots : @xmath80 , where , @xmath81 , @xmath82(d+\gamma+\epsilon ) + ( d k+\beta)\alpha-\beta\lambda$ ] , @xmath83d - \beta\lambda\omega$ ] for system ( [ paper3_eq7 ] ) . since all parameters take positive values , we have @xmath84 . to get the two positive roots essential for backward bifurcation , it is required that @xmath85 and @xmath86 . noticing that @xmath87 , we can see that the infection force , @xmath9 , the constant influx of the susceptibles , @xmath57 , and the effect of medical treatment @xmath88 are indispensible terms for backward bifurcation . the number of positive infected equilibrium solutions changes from two to one when the value of @xmath89 passes from negative to positive , which gives a critical point at @xmath90 , that is , @xmath91d = \beta\lambda\omega$ ] , which is equivalent to @xmath92 . on the other hand , we may infer the emergence of backward bifurcation without solving the equilibrium conditions . if we do not consider the medical treatment term @xmath88 and remove it from system ( [ paper3_eq7 ] ) , that leads to system ( [ paper3_eq6 ] ) , which is a typical example of an sir model studied by ( [ paper3_eq2 ] ) . by setting the incidence function as @xmath93 , we have @xmath94 ; @xmath95 and @xmath96 for all @xmath97 ; and @xmath98 for all @xmath97 . therefore , the incidence function @xmath99 , satisfies the conditions given in ( [ paper3_eq3 ] ) . in particular , the function is concave , and can only have one intersection point with the line @xmath100 in the @xmath7-@xmath6 plane , as shown in figure [ paper3_fig1](a ) . thus , the uniqueness of the positive infected equilibrium implies that backward bifurcation can not occur in this case . moreover , according to the result in @xcite , the uninfected and infected equilibria are globally asymptotically stable for @xmath101 and @xmath4 , respectively . no complex dynamical behavior happens in system ( [ paper3_eq6 ] ) . in contrast , when we introduce the loss of the infectives due to medical treatment , the dynamics of system ( [ paper3_eq7 ] ) differ greatly from system ( [ paper3_eq6 ] ) . in particular , backward bifurcation emerges and complex dynamical behaviors may occur . to clarify this effect , we denote the function induced by @xmath26 from ( [ paper3_eq7 ] ) as @xmath102 . note that @xmath103 is not an incidence rate . but , if we fix @xmath104 , there exist @xmath105 , such that @xmath106>0 $ ] , @xmath107 ; and @xmath108 , @xmath109 , @xmath110 , for @xmath40 , @xmath111 , @xmath112 . thus , @xmath113 actually has a convex - concave ` @xmath6 ' shape , and may have two positive intersection points with the ray line , @xmath114 , in the first quadrant ; see figure [ paper3_fig1](b ) . these intersections contribute the two positive equilibrium solutions that are a necessary feature of backward bifurcation . paper3_fig1a.pdf ( 55,100)(a ) ( 55,-6)@xmath7 ( 10,62)@xmath115 ( 50,40)@xmath114 paper3_fig1b.pdf ( 55,100)(b ) ( 55,-6)@xmath7 ( 10,87)@xmath116 ( 50,40)@xmath114 in summary we may conclude that the necessary terms which should be contained in system ( [ paper3_eq7 ] ) in order to have backward bifurcation are the constant influx @xmath57 , the infection force @xmath9 , and the saturating medical treatment @xmath117 . now we consider the 2-dimensional infection model ( [ paper3_eq4 ] ) which exhibits viral blips , studied in @xcite . the motivation for this model was a series of clinical discoveries indicating that viral infection can increase the density of a harmful chemical substance @xcite , thereby amplifying an associated biochemical reaction @xcite , and thus accelerating the infection rate @xcite . this cooperative phenomenon in viral infection is expressed by an increasing , saturating infectivity function : @xmath118 . according to the principle of mass action , the incidence function is then denoted as @xmath119 , which is a convex function with respect to the infectives density @xmath46 . to analyze the occurrence of possible backward bifurcation , we first examine the two equilibrium solutions from the following equations : @xmath120 where all parameters @xmath121 , @xmath122 , @xmath89 and @xmath48 are positive constants . it is easy to find the uninfected equilibrium @xmath123 , whose characteristic polynomial has two roots : @xmath124 , and @xmath125 , which gives @xmath126 . consequently , @xmath127 is stable ( unstable ) for @xmath128 . to find the infected equilibrium solution , setting @xmath129 yields @xmath130 , which is then substituted into @xmath131 to give the following quadratic equation : @xmath132 in order to have two real , positive roots , two conditions must be satisfied , that is , @xmath133 and @xmath134 , or in compact form , @xmath135 . the condition @xmath134 is equivalent to @xmath136 , which is a necessary condition for backward bifurcation . moreover , the positive influx constant , having been scaled to @xmath47 , is a necessary term for the positive equilibrium of @xmath46 . therefore , the positive influx rate term and the increasing and saturating infectivity function are necessary for backward bifurcation . in the rest of the subsection , we further examine the incidence function , @xmath137 without solving the equilibrium solutions . the incidence function @xmath138 obviously satisfies the condition ( [ paper3_eq3a ] ) , as well as the condition ( [ paper3_eq3b ] ) since @xmath139y > 0 $ ] and @xmath140x>0 $ ] for all @xmath141 . however , the second partial derivative of @xmath142 with respect to @xmath46 , @xmath143 for all @xmath141 , showing that @xmath142 is a convex function with respect to the variable @xmath46 . consequently , @xmath142 can only have one intersection with @xmath144 , implying that only one equilibrium solution would exist if we only consider the second equation in ( [ paper3_eq9 ] ) , as shown figure [ paper3_fig2 ] ( a ) . however , when considering both conditions given in ( [ paper3_eq9 ] ) for equilibrium solutions , we will have two intersection points between @xmath138 and @xmath145 . according to the first equation in ( [ paper3_eq9 ] ) , that is @xmath131 , we can use @xmath46 to express @xmath45 in the equilibrium state as @xmath146^{-1}$ ] . substituting @xmath147 into @xmath142 in ( [ paper3_eq11 ] ) , we obtain @xmath148[(a+b)y^2+(bc+d)y+cd]^{-1 } , \label{paper3_eq12}\ ] ] and @xmath149[(a+b)y^2+(bc+d)y+cd]^{-2}>0 $ ] for all @xmath141 . however , the sign of @xmath150 [ ( a+b)y^2+(bc+d)y+cd]^{-3}$ ] , could alter at the inflection point from positive to negative as @xmath46 increases . therefore , with appropriate parameter values , @xmath151 can have a convex - concave ` @xmath6 ' shape , yielding two intersection points with the ray line , @xmath152 , in the first quadrant of the @xmath45-@xmath46 plane , as shown in figure [ paper3_fig2 ] ( b ) . the above discussion , as illustrated in figure [ paper3_fig2 ] , implies that system ( [ paper3_eq4 ] ) can have two positive equilibrium solutions when @xmath5 , and thus backward bifurcation may occur . paper3_fig2a.pdf ( 50,-5)@xmath46 ( 50,100)(a ) ( 70,55)@xmath153 in ( [ paper3_eq11 ] ) ( 70,20)@xmath144 ( 10,55 ) and @xmath151 for the parameter values @xmath154 , @xmath155 , @xmath156 , and @xmath157 . the incidence functions are denoted by the solid lines , while the ray lines , determined by @xmath144 , are denoted by dotted lines : ( a ) the incidence function @xmath158 , showing one intersection point with @xmath145 with an inset , with a fixed value @xmath159 ; and ( b ) the incidence function @xmath151 , showing two intersection points with an inset.,title="fig : " ] ( 27,53)@xmath46 ( 7,20)(1,0)23 ( 30,20)(0,-1)15 paper3_fig2b.pdf ( 50,-5)@xmath46 ( 50,100)(b ) ( 15,65)@xmath151 in ( [ paper3_eq12 ] ) ( 70,95)@xmath144 ( 60,10 ) and @xmath151 for the parameter values @xmath154 , @xmath155 , @xmath156 , and @xmath157 . the incidence functions are denoted by the solid lines , while the ray lines , determined by @xmath144 , are denoted by dotted lines : ( a ) the incidence function @xmath158 , showing one intersection point with @xmath145 with an inset , with a fixed value @xmath159 ; and ( b ) the incidence function @xmath151 , showing two intersection points with an inset.,title="fig : " ] ( 78,8)@xmath46 ( 7,20)(1,0)18 ( 25,20)(0,-1)15 summarizing the discussions and results given in this section indicates that a disease model with a convex - concave incidence function may lead to backward bifurcation , which in turn implies : ( a ) the system has at least two equilibrium solutions , and the two equilibrium solutions intersect at a transcritical bifurcation point ; and ( b ) at least one of the equilibrium solutions is determined by a nonlinear equation . [ paper3_remark1 ] in the previous section , we studied backward bifurcation and established the necessary conditions for the occurrence of backward bifurcation in two models . in this section , we turn to hopf bifurcation , since it typically underlies the change of stability in the upper branch of the infected equilibrium , the key condition in determining whether a model can exhibit oscillation or even recurrence . again , we will present detailed studies for the two models . in this subsection , we study two cases of an infection model with concave incidence : system ( [ paper3_eq6 ] ) and ( [ paper3_eq7 ] ) . first , we discuss the equilibrium solutions and their stability by using the jacobian matrix , denoted by @xmath160 , and examining the corresponding characteristic polynomial , @xmath161 bifurcation analysis is conducted by choosing @xmath57 as the bifurcation parameter . first , we consider the case without saturating medical treatment , system ( [ paper3_eq6 ] ) . this system satisfies the three conditions in ( [ paper3_eq3 ] ) , and consequently , its uninfected equilibrium @xmath162 is globally asymptotically stable if @xmath163 , while the infected equilibrium @xmath164 emerges and is globally asymptotically stable if @xmath165 . therefore , for this case the system has only one transcritical bifurcation point at @xmath3 and no complex dynamics can occur . next , with the saturating treatment term , system ( [ paper3_eq7 ] ) violates the conditions established for model ( [ paper3_eq3 ] ) , but leads to the possibility of complex dynamical behaviors . in fact , evaluating the jacobian matrix @xmath166 at the uninfected equilibrium , @xmath167 , yields the characteristic polynomial in the form of ( [ paper3_eq13 ] ) , denoted by @xmath168 , with @xmath169 @xmath170 , and @xmath171 . this indicates that + @xmath172 when @xmath173 , and thus hopf bifurcation can not occur from @xmath127 . on the other hand , a static bifurcation can occur when @xmath174 , that is , @xmath175 , where the subscript ` @xmath6 ' refers to _ static bifurcation_. therefore , @xmath127 is stable ( unstable ) for @xmath176 , or @xmath177 , with @xmath178 @xcite . we will show that complex dynamical behaviors can emerge in system ( [ paper3_eq7 ] ) from the infected equilibrium @xmath179 , where @xmath180 is determined from the equation @xmath181 in ( [ paper3_eq8 ] ) . in the @xmath57-@xmath7 plane , the bifurcation diagram as shown in figure [ paper3_fig3 ] ( 1)-(4 ) , indicates a turning point on the curve with appropriate parameter values , determined by both the quadratic equation ( [ paper3_eq8 ] ) and the relation @xmath182 , which is equivalent to @xmath183 . solving @xmath183 yields the turning point of @xmath7 , denoted by @xmath184 ( ` @xmath185 ' means _ turning _ ) , taking the form @xmath186,\ ] ] where @xmath187 is obtained from @xmath188 , see ( [ paper3_eq8 ] ) . thus , when @xmath189 , the turning point of the quadratic curve appears above ( below ) the @xmath7-axis , meaning that backward bifurcation occurs for @xmath190 . evaluating the jacobian matrix at the infected equilibrium @xmath191 , and further denoting it as @xmath192 , we obtain the characteristic polynomial in the form of ( [ paper3_eq13 ] ) , with @xmath193 $ ] and @xmath194 $ ] , where @xmath195 and @xmath196 , with @xmath197 and @xmath198 , and @xmath199 and @xmath200 only contain positive terms ( their expressions are omitted here for brevity ) . therefore , we can rewrite @xmath201 $ ] . determining whether a hopf bifurcation can occur from @xmath202 is equivalent to finding whether @xmath203 remains positive when @xmath204 . ignoring the positive factors in the following subtraction yields @xmath205 where @xmath206 $ ] . thus , when @xmath207 , @xmath208 and @xmath209 have opposite signs , implying that when @xmath204 , @xmath203 could be positive only if @xmath209 is negative . therefore , the necessary condition for system ( [ paper3_eq7 ] ) to have a hopf bifurcation from the infected equilibrium @xmath191 is that @xmath209 is negative . in the remaining part of this subsection , we demonstrate various dynamics which may happen in system ( [ paper3_eq7 ] ) with different parameter values of @xmath64 , as shown in table [ paper3_table1 ] . taking other parameter values as @xmath210 , @xmath211 , @xmath212 , @xmath213 , @xmath214 , and @xmath215 , and solving the two equations @xmath204 and @xmath181 in ( [ paper3_eq8 ] ) gives the hopf bifurcation point candidates , @xmath216 , for which @xmath217 . since the formula for the transcritical bifurcation point @xmath218 has no relation with @xmath64 , @xmath219 is a fixed value pair in table [ paper3_table1 ] . bifurcation diagrams and associated numerical simulations are shown in figure [ paper3_fig3 ] corresponding to the five cases given in table [ paper3_table1 ] . the blue lines and red curves represent the uninfected equilibrium @xmath220 and infected equilibrium @xmath221 , respectively . the stable and unstable equilibrium solutions are shown by solid and dashed lines / curves , respectively . backward bifurcation occurs in cases @xmath47 , @xmath222 , and @xmath223 ( see table [ paper3_table1 ] ) , which are illustrated by the corresponding bifurcation diagrams in figures [ paper3_fig3](1 ) , ( 2 ) , and ( 3 ) , respectively . for cases @xmath47 and @xmath222 , only one hopf bifurcation occurs on the upper branch of the infected equilibrium @xmath221 , and this bifurcation point exists at the critical point @xmath224 for case @xmath47 and @xmath225 , for case @xmath222 . for case @xmath47 with @xmath226 , the simulated time history converges to @xmath127 with initial condition ic@xmath227 $ ] , shown in figure [ paper3_fig3](1a ) , but converges to @xmath191 with initial condition ic@xmath228 $ ] , shown in figure [ paper3_fig3](1b ) . this clearly indicates the bistable behavior when @xmath224 , and an overlapping stable region for both @xmath127 and @xmath191 exists ( see figure [ paper3_fig3](1 ) ) . the recurrent behavior for case 2 is simulated at @xmath229 with ic@xmath230 $ ] , shown in figure [ paper3_fig3](2a ) . for case 2 , @xmath225 , and an overlapping unstable parameter region for both @xmath127 and @xmath191 occurs between @xmath218 and @xmath231 ( see figure [ paper3_fig3](2 ) ) . for case @xmath223 , two hopf bifurcations occur on the left side of @xmath218 , and a stable part in the upper branch of @xmath191 exists when @xmath57 passes through the critical value @xmath232 . in this case , although backward bifurcation still exists and the turning point is also located above the @xmath57-axis , giving two branches of biologically feasible @xmath191 , only regular oscillating behavior is observed . the simulated time history is conducted at @xmath233 , with initial condition ic@xmath234 $ ] , shown in figure [ paper3_fig3](3a ) . for case @xmath235 , only forward bifurcation occurs in the biologically feasible region , and the turning point for backward bifurcation moves down to the fourth quadrant , that is , negative backward bifurcation occurs in this case . the whole upper branch of @xmath191 in the first quadrant is stable , therefore , no oscillations ( or recurrence ) can happen . finally , further increases to the value of @xmath64 change the shape of the red curves , as shown in figure [ paper3_fig3](5 ) , which again indicates that no biologically meaningful backward bifurcation or oscillations can occur . note that in figure [ paper3_fig3](5 ) a hopf bifurcation point exists on the lower branch of the equilibrium solution , which is biologically unfeasible since it is entirely below the horizontal axis . in conclusion , interesting dynamical behaviors can emerge in system ( [ paper3_eq7 ] ) if backward bifurcation occurs . .[paper3_table1 ] dynamics of system ( [ paper3_eq7 ] ) for different values of @xmath64 , with @xmath210 , @xmath211 , @xmath212 , @xmath213 , @xmath214 , @xmath215 , and a fixed transcritical bifurcation point @xmath219 . [ cols="^,<,<,<,<,<,<",options="header " , ] to illustrate the analytical results given in tables [ paper3_table2 ] and [ paper3_table4 ] , we provide the bifurcation diagrams in figures [ paper3_fig4 ] ( 1)-(8 ) . these figures depict the uninfected equilibrium @xmath220 and the infected equilibrium @xmath221 in blue and red , respectively . the solid and dashed lines differentiate stable and unstable states of the equilibrium solutions . the bifurcation points on the equilibrium solutions are highlighted by solid black dots . moreover , ` transcritical ' , ` turning ' , ` hopf@xmath236 ' , and ` hopf@xmath237 ' , are used to denote _ transcritical bifurcation _ , _ turning point _ , _ subcritical hopf bifurcation _ , and _ supercritical hopf bifurcation _ , respectively . simulated time histories are used to validate the analytical results , and to show different dynamical behaviors in each case listed in tables [ paper3_table2 ] and [ paper3_table4 ] . subcritical hopf bifurcation occurs in cases 1 - 3 , shown in figures [ paper3_fig4 ] ( 1)-(3 ) . @xmath238 is used in figure [ paper3_fig4 ] ( 1 ) for case 1 . choosing @xmath239 , we have @xmath240 $ ] and @xmath241 $ ] . the simulated solution converges to @xmath242 or @xmath243 , with initial condition taken as ic@xmath244 $ ] or ic@xmath245 $ ] , shown in figures [ paper3_fig4 ] ( 1d ) and ( 1c ) , respectively . figures [ paper3_fig4 ] ( 1a ) and ( 1b ) , on the other hand , show the unstable limit cycle bifurcating from the subcritical hopf bifurcation with ic@xmath245 $ ] . figure [ paper3_fig4 ] ( 2 ) corresponds to case 2 with @xmath246 . choosing @xmath247 $ ] yields recurrence , independent of the initial conditions , see , for example , the result given in figure [ paper3_fig4 ] ( 2b ) with ic@xmath248 $ ] . however , for @xmath249 , the simulated time history converges to @xmath250 , with an initial condition close to @xmath250 , such as ic@xmath251 $ ] as shown in figure [ paper3_fig4 ] ( 2a ) ; or shows recurrence with an initial condition far away from @xmath250 , such as ic@xmath252 $ ] , as shown in figure [ paper3_fig4 ] ( 2c ) . figure [ paper3_fig4 ] ( 3 ) plots the result for case 3 with @xmath253 , and shows a broader region between the transcritical and hopf bifurcation points , associated with a larger recurrent region . recurrence occurs independent of the initial conditions for @xmath254 $ ] , giving @xmath255 $ ] and @xmath256 $ ] , as shown in figures [ paper3_fig4 ] ( 3a ) and ( 3b ) , with ic@xmath257 $ ] and ic@xmath258 $ ] , respectively . but if we choose @xmath259 , we have @xmath260 $ ] and @xmath261 $ ] . the time history converges to @xmath250 with ic@xmath262 $ ] , or shows recurrence with ic@xmath263 $ ] , as shown in figure [ paper3_fig4 ] ( 3c ) and ( 3d ) , respectively . supercritical hopf bifurcations occur in cases 4 - 7 , as shown in figures [ paper3_fig4 ] ( 4)-(7 ) . figure [ paper3_fig4 ] ( 4 ) depicts the result for case @xmath235 with @xmath264 . only one supercritical hopf bifurcation happens in this case , and gives a large recurrent parameter region between the transcritical and hopf bifurcation points . although the simulated recurrent behavior does not depend on initial conditions , the recurrent pattern will fade out with the growth of the value of @xmath122 from the transcritical point to the hopf bifurcation point , see figures [ paper3_fig4 ] ( 4a ) and ( 4b ) with the same ic@xmath265 $ ] , but different values of @xmath122 : @xmath266 and @xmath267 , respectively . figure [ paper3_fig4 ] ( 5 ) shows the result for case 5 with @xmath268 . a transcritical bifurcation happens between two supercritical hopf bifurcations . the recurrent region still starts from the transcritical point and independent of the initial conditions , but is narrower than that shown in figure [ paper3_fig4 ] ( 4 ) . the simulated recurrent behavior for this case is conducted at ic@xmath269 $ ] and @xmath266 . figure [ paper3_fig4 ] ( 6 ) corresponds to case 6 with @xmath270 , and two supercritical hopf bifurcations occur on the right side of the transcritical bifurcation point , which makes the recurrent region even narrower and the recurrent pattern less obvious , as shown in the simulated time history with ic@xmath271 $ ] and @xmath266 . negative backward bifurcations occur in cases 7 and 8 , as shown in figure [ paper3_fig4 ] ( 7 ) and ( 8) . although two hopf bifurcations are still present in case 7 , see figure [ paper3_fig4 ] ( 7 ) , only a regular oscillating pattern exists . for case 8 , no hopf bifurcation happens in the biologically feasible part of @xmath250 , and therefore no more interesting dynamics occur . in general , backward bifurcation , which occurs above the horizontal axis , is much more likely to induce hopf bifurcation . a hopf bifurcation can only occur along the upper branch of @xmath221 , since @xmath220 only changes its stability at a transcritical bifurcation point , and any point on the lower branch of @xmath221 is a saddle node ( yu et al . , submitted for publication ) . moreover , hopf bifurcation can lead to a change in the stability of the upper branch of the infected equilibrium @xmath221 . thus the system further develops bistable , recurrent , or regular oscillating behavior , corresponding to cases @xmath272 in tables [ paper3_table2 ] and [ paper3_table4 ] , and in figures [ paper3_fig4 ] ( 1)-(7 ) . in particular , bistability happens when both equilibria @xmath220 and @xmath221 share a stable parameter region , see case @xmath47 in table [ paper3_table2 ] and figure [ paper3_fig4 ] ( 1 ) . as for recurrent behavior , we observe that recurrence is more likely to happen if the following two conditions are satisfied for the upper branch of @xmath221 : ( 1 ) the equilibrium remains unstable as the bifurcation parameter increases and crosses the trancritical point , where @xmath220 and @xmath221 intersect , such that the two equilibria share an unstable parameter range ; and ( 2 ) at least one hopf bifurcation occurs from @xmath221 . as shown in cases @xmath222-@xmath273 in table [ paper3_table2 ] , and the corresponding figures [ paper3_fig4 ] ( 2)-(5 ) , the common recurrent parameter region for both subcritical and supercritical hopf bifurcations starts beside the transcritical point , and is located entirely in the unstable parameter region of @xmath220 and @xmath221 . the simulated recurrent pattern becomes more pronounced if the value of the bifurcation parameter is close to the transcritical point , but approaches an oscillatory pattern as the parameter diverges from the transcritical point , as shown in figure [ paper3_fig4 ] ( 4a ) and ( 4b ) . in this common recurrent parameter region , recurrence occurs independent of initial conditions ; see figures [ paper3_fig4 ] ( 3a ) and ( 3b ) . in addition to the common recurrent region , for subcritical bifurcation , seen in table [ paper3_table2 ] for cases ( 2 ) and ( 3 ) and figures [ paper3_fig4 ] ( 2 ) and ( 3 ) , recurrence may also appear on the stable side of the subcritical hopf bifurcation point with an initial condition close to @xmath221 . moreover , the subcritical hopf bifurcation and the transcritical point should be close to each other for a clear recurrent pattern . when this is not the case , the periodic solutions show a more regular oscillating pattern , as compared in figures [ paper3_fig4 ] ( 2c ) and ( 3d ) . although two hopf bifurcation points occur in table [ paper3_table2 ] for case @xmath274 , see figure [ paper3_fig4 ] ( 5 ) , the transcritical point is located inside the unstable range of the upper branch of @xmath221 , between the two hopf bifurcation points . a recurrent pattern still characterizes the dynamical behavior in this case . however , if the unstable range of @xmath221 , between the two hopf bifurcation points , is located entirely in the unstable range of @xmath220 , and moves further away from the transcritical point , the recurrent motion gradually becomes a regular oscillation , as shown in figures [ paper3_fig4 ] ( 6 ) and ( 7 ) . summarizing the results and discussions presented in the previous two sections , we have the following observations . 1 . due to the fact that @xmath220 only changes its stability at the transcritical bifurcation point , and the fact that any point on the lower branch of @xmath221 is a saddle node , hopf bifurcation can only occur from the upper branch of @xmath221 . a hopf bifurcation may result in convergent , recurrent , bistable , or regular oscillating behaviors . backward bifurcation gives rise to two branches in the infected equilibrium @xmath221 . hopf bifurcation is more likely to happen when the turning point of the backward bifurcation is located on the positive part of the equilibrium solution in the bifurcation diagram , as shown in figures [ paper3_fig4 ] ( 2)-(6 ) . this means that we have two biologically feasible infected equilibria , which is essential to observe bistability , as shown in figure [ paper3_fig4 ] ( 1 ) . however , if the turning point on the infected equilibrium @xmath221 , or the backward bifurcation moves down to the negative part of a state variable in the bifurcation diagram , that is , negative backward bifurcation occurs , then hopf bifurcation is very unlikely to happen . although figure [ paper3_fig4 ] ( 7 ) shows an exceptional case , the parameter range for such a hopf bifurcation is very narrow . 4 . the bifurcation diagram for system ( [ paper3_eq4 ] ) with @xmath275 , shown in figure [ paper3_fig4 ] ( 8) , is a typical model with negative backward bifurcation . such negative backward bifurcation may occur in higher - dimensional systems . however , by considering more state variables , which make the system more complicated , hopf bifurcation can happen in the upper branch of the negative backward bifurcation . we will discuss this possibility in more detail in the next section by examining an autoimmune disease model . the results obtained in this section suggest the following summary . if a disease model contains a backward bifurcation on an equilibrium solution , then as the system parameters are varied , there may exist none , one or two hopf bifurcations from the equilibrium solution , which may be supercritical or subcritical . if further this equilibrium has a transcritical bifurcation point at which it exchanges its stability with another equilibrium , then recurrence can occur between the transcritical and hopf bifurcation points and near the transcritical point , where both equilibrium solutions are unstable , and bistability happens when hopf bifurcation makes a shared stable parameter region for both equilibria . [ paper3_remark2 ] in the previous section , we examined three cases of negative backward bifurcation : table [ paper3_table1 ] case 4 for system ( [ paper3_eq7 ] ) and table [ paper3_table2 ] case ( 7 ) and ( 8) for system ( [ paper3_eq4 ] ) . the analytical and numerical results showed that solutions typically converge to the infected equilibrium in these cases , and the parameter range for hopf bifurcation is very limited . as a result , negative backward bifurcation tends to give no interesting behavior . in this section , however , we shall explore an established autoimmune model @xcite in which negative backward bifurcation occurs . we demonstrate that after modification , the autoimmune model can also exhibit recurrence . the autoimmune model @xcite takes the form @xmath276 { \displaystyle}\frac{{\mathrm d } r_n}{{\mathrm d } t}= ( \pi_1 e + \beta)a - \mu_n r_n \\[1.0ex ] { \displaystyle}\frac{{\mathrm d } e}{{\mathrm d } t } = \lambda_e a - \mu_e e \\[1.0ex ] { \displaystyle}\frac{{\mathrm d } g}{{\mathrm d } t } = \gamma e - \tilde{v } g - \mu_g g , \end{array } \label{paper3_eq16}\ ] ] where mature papcs ( @xmath121 ) undergo maturation by intaking self - antigen ( @xmath277 ) , at rate @xmath278 , and are suppressed by specific regulatory t cells , t@xmath279 cells ( @xmath280 ) , at rate @xmath281 ; @xmath282 represents additional non - specific background suppression . the t@xmath279 cells are activated by mature papcs at a rate proportional to the number of auto - reactive effector t cells ( @xmath283 ) at rate @xmath284 , and by other sources at rate @xmath9 . active auto - reactive effector t cells ( @xmath283 ) come from the activation process initiated by mature papcs , at rate @xmath285 , then attack healthy body tissue and release free self - antigen ( @xmath277 ) at rate @xmath59 , which is ready for mature papcs to engulf ; the antigen engulfing rate is @xmath286 . the death rates of the populations @xmath121 , @xmath280 , @xmath283 , and @xmath277 are denoted by @xmath287 , @xmath288 , @xmath289 , and @xmath290 , respectively . following the steps described by zhang et al . ( submitted for publication ) , system ( [ paper3_eq16 ] ) , can be reduced via quasi - steady state analysis to a 2-dimensional system : @xmath291a - \sigma_1 r_n a , \\[1.5ex ] { \displaystyle}\frac{{\mathrm d } r_n}{{\mathrm d } t } = ( \frac{\pi_1 \lambda_e}{\mu_e}a + \beta ) a - \mu_n r_n . \end{array } \label{paper3_eq17}\ ] ] for simplicity , we set @xmath292 and @xmath293 . for the stability and bifurcation analysis , we choose @xmath285 as the bifurcation parameter . system ( [ paper3_eq17 ] ) has a disease - free equilibrium @xmath294 , which is stable if @xmath295 or @xmath296 ; and unstable if @xmath297 or @xmath298 . thus a static bifurcation occurs on @xmath220 when @xmath299 or @xmath300 . the disease equilibrium is given by @xmath301 , in which @xmath302 , and @xmath303 is given by the roots of the following equation , @xmath304 equation ( [ paper3_eq18 ] ) has two roots with negative signs if @xmath297 , with opposite signs if @xmath295 , and only one zero root if @xmath299 . this means that a negative backward bifurcation is possible in system ( [ paper3_eq17 ] ) with proper parameter values . we further examine the characteristic equation at @xmath221 , which shares the same form as equation ( [ paper3_eq13 ] ) , with @xmath305 and @xmath306 . solving @xmath307 and @xmath308 , gives the static bifurcation point of @xmath221 at @xmath309 or @xmath310 , which is a transcritical bifurcation point between @xmath220 and @xmath221 . moreover , hopf bifurcation can happen if and only if @xmath307 and @xmath311 , which can be satisfied only if @xmath312 . this implies that the positive branch of @xmath221 is stable for any positive values of @xmath288 . thus , this model can not exhibit recurrence , bistability , or even regular oscillation . the same conclusion was obtained in zhang et al . , ( submitted for publication ) for the original 4-dimensional model ( [ paper3_eq16 ] ) . however , a recent experimental discovery @xcite has revealed a new class of terminally differentiated t@xmath313 cells . as described in detail in zhang et al . , ( submitted for publication ) , introducing this cell population , denoted @xmath314 , into the model yields the full system @xmath315 \frac{{\mathrm d } r_n}{{\mathrm d } t}= ( \pi_1 e+\beta)a - \mu_n r_n - \xi r_n \\[0.5ex ] \frac{{\mathrm d } r_d}{{\mathrm d } t } = c \xi r_n - \mu_d r_d \\[0.5ex ] \frac{{\mathrm d } e}{{\mathrm d } t } = \lambda_e a - \mu_e e \\[0.5ex ] \frac{{\mathrm d } g}{{\mathrm d } t } = \gamma e - \tilde{v}g - \mu_g g \end{array } \ ] ] and quasi - steady state analysis then yields a reduced 3-dimensional model in the form @xmath316a - \sigma_1(r_n+dr_d)a , \\[0.3ex ] \frac{{\mathrm d } r_n}{{\mathrm d } t}= ( \frac{\pi_1 \lambda_e}{\mu_e}a+\beta)a - \mu_n r_n - \xi r_n , \\[0.3ex ] \frac{{\mathrm d } r_d}{{\mathrm d } t } = c \xi r_n - \mu_d r_d . \end{array } \label{paper3_eq19}\ ] ] again , here @xmath285 is chosen as the bifurcation parameter for stability and bifurcation analysis . it is easy to show that system ( [ paper3_eq19 ] ) still has a disease - free equilibrium @xmath220 as @xmath317 , and a disease equilibrium @xmath221 as @xmath318 , where @xmath319 , @xmath320 , and @xmath303 is determined from the following quadratic equation : @xmath321 , \ ] ] which gives two negative roots if @xmath322 , and two roots with opposite signs when @xmath323 . the critical point is determined by @xmath324 , which is actually the intersection point of @xmath220 and @xmath221 . the two equilibrium solutions exchange their stability at @xmath325 , leading to a transcritical bifurcation at @xmath326 . note that the negative backward bifurcation still happens in system [ paper3_eq19 ] . moreover , a hopf bifurcation occurs from the upper branch of @xmath221 , giving rise to oscillation and recurrence . realistic parameter values have been obtained in zhang et al . , ( submitted for publication ) , and are given as follows : @xmath327 \beta = 200,\ \ \mu_n = 0.1,\ \ \mu_e = 0.2,\ \ \gamma = 2000,\ \ \mu_g = 5,\ \ \mu_d = 0.2,\ \ c=8,\ \ d = 2,\ \ \xi=0.025 . \end{array}\ ] ] for the above parameter values , the hopf critical point is obtained at @xmath328 @xmath329 , while the turning point is at @xmath330 , and the transcritical bifurcation point is at @xmath331 . these three bifurcation points and the stability of equilibrium solutions are shown in the bifurcation diagram given in figure [ paper3_fig5](a ) , and the simulated recurrent time history is plotted in figure [ paper3_fig5](b ) for @xmath332 . paper3_fig5a.pdf ( -4,60)@xmath121 ( 75,45)@xmath285 ( 55,85)hopf ( 60,5)hopf ( 35,55)transcritical ( 7,46)turning ( 50,110)(@xmath333 ) paper3_fig5b.pdf ( 0,60)@xmath121 ( 0,35)@xmath280 ( 0,10)@xmath314 ( 50,-8)@xmath334 ( 50,82)(@xmath335 ) in summary , when negative backward bifurcation occurs , that is , the turning point is located in the negative state variable space , less complex dynamical behavior will be present . hopf bifurcation in a biologically feasible area does not happen in the reduced 2-dimensional system ( [ paper3_eq17 ] ) , nor in the original system ( [ paper3_eq16 ] ) ( zhang et al . , submitted for publication ) . however , if we increase the dimension of the system , hopf bifurcation and complex dynamical phenomena can emerge , as shown in our results for system ( [ paper3_eq19 ] ) . in this paper , we first review previous work on a reduced 2-dimensional infection model with a concave incidence rate @xcite . the authors proved that the disease equilibrium will emerge and be globally stable when the basic reproduction number @xmath1 is greater than @xmath47 . this means that no complex dynamical phenomenon can occur in such models . however , by adding an extra saturating treatment term to this simple 2-dimensional infection model , the resulting system ( [ paper3_eq6 ] ) considered in @xcite can exhibit backward bifurcation , which increases the parameter range for hopf bifurcation , which in turn leads to recurrent , bistable and regular oscillating behaviors . instead of adding an extra term , a 2-dimensional infection model with a convex incidence function can likewise show rich dynamics due to the occurrence of backward bifurcation , giving rise to two types of hopf bifurcation . biologically , a convex incidence rate implies that existing infection makes the host more vulnerable to further infection , showing a cooperative effect in disease progression . from the view point of mathematics , the convex incidence function enables backward bifurcation to occur on the positive branch of the disease equilibrium solution , which further generates hopf bifurcation . the location and direction of hopf bifurcation(s ) , determined by parameter values , can further give rise to bistable , recurrent , and regular oscillating behaviors . cooperative effects also occur during the progression of autoimmune disease . however , for an autoimmune model with negative backward bifurcation , in which the turning point is located on the negative state variable space , the biologically feasible parameter range in which hopf bifurcation may occur is limited . by introducing an additional state variable to the autoimmune model , recurrent phenomenon are once again observed . g. c. brown and r. hasibuan . conidial discharge and transmission efficiency of neozygites floridana , an entomopathogenic fungus infecting two - spotted spider mites under laboratory conditions . , 65(1):10 16 , 1995 . c. castillo - chavez , k. cooke , w. huang , and s. levin . on the role of long incubation periods in the dynamics of acquired immunodeficiency syndrome ( aids ) . part 2 : multiple group models . in carlos castillo - chavez , editor , _ mathematical and statistical approaches to aids epidemiology _ , volume 83 of _ lecture notes in biomathematics _ , pages 200217 . springer berlin heidelberg , 1989 . a. c. collier , r. w. coombs , d. a. schoenfeld , r. l. bassett , j. timpone , a. baruch , m. jones , k. facey , c. whitacre , v. j. mcauliffe , h. m. friedman , t. c. merigan , r. c. reichman , c. hooper , and l. corey . treatment of human immunodeficiency virus infection with saquinavir , zidovudine , and zalcitabine . , 334(16):10111018 , 1996 . g. dornadula , h. zhang , b. vanuitert , j. stern , l. livornese jr . , m. j. ingerman , j. witek , r. j. kedanis , j. natkin , j. desimone , and r. j. pomerantz . residual hiv-1 rna in blood plasma of patients taking suppressive highly active antiretroviral therapy . , 282(17):16271632 , 1999 . c. fraser , n. m. ferguson , f. de wolf , and r. m. anderson . the role of antigenic stimulation and cytotoxic t cell activity in regulating the long - term immunopathogenesis of hiv : mechanisms and clinical implications . , 268(1481):20852095 , 2001 . i. c. h. fung , m. gambhir , a. van sighem , f. de wolf , and g. p. garnett . the clinical interpretation of viral blips in hiv patients receiving antiviral treatment : are we ready to infer poor adherence ? , 60(1):511 , 2012 . n. j. garretta , v. apeaa , a. noria , i. ushiro - lumbb , a. r. oliverb , g. bailya , and d. a. clarkb . comparison of the rate and size of hiv-1 viral load blips with roche cobas taqman hiv-1 versions 1.0 and 2.0 and implications for patient management . , 53(4):354355 , 2012 . j. t. grennan , m. r. loutfy , d. su , p. r. harrigan , c. cooper , m. klein , n. machouf , j. s. g. montaner , s. rourke , c. tsoukas , b. hogg , j. raboud , and the canoc collaboration . magnitude of virologic blips is associated with a higher risk for virologic rebound in hiv - infected individuals : a recurrent events analysis . , 205(8):12301238 , 2012 . s. palmer , f. maldarelli , a. wiegand , b. bernstein , g. j. hanna , s. c. brun , d. j. kempf , j. w. mellors , j. m. coffin , and m. s. king . low - level viremia persists for at least 7 years in patients on suppressive antiretroviral therapy . , 105(10):38793884 , 2008 . s. palmer , a. p. wiegand , f. maldarelli , h. bazmi , j. m. mican , m. polis , r. l. dewar , a. planta , s. liu , j. a. metcalf , j. w. mellors , and j. m. coffin . new real - time reverse transcriptase - initiated pcr assay with single - copy sensitivity for human immunodeficiency virus type 1 rna in plasma . , 41(10):45314536 , 2003 .

in this article , we present a greedy algorithm based on a tensor product decomposition , whose aim is to compute the global minimum of a strongly convex energy functional . we prove the convergence of our method provided that the gradient of the energy is lipschitz on bounded sets . the main interest of this method is that it can be used for high - dimensional nonlinear convex problems . we illustrate this method on a prototypical example for uncertainty propagation on the obstacle problem . universit paris est , cole des ponts - paristech , cermics , + 6 & 8 avenue blaise pascal + 77455 marne - la - valle cedex 2 + france + \{cances , ehrlachv , lelievre}@cermics.enpc.fr + keywords : greedy algorithm ; high dimension ; obstacle problem ; uncertainty quantification .

1004.0095

the main motivation for this work comes from two important and challenging problems in contemporary scientific computing : * the uncertainty quantification for some nonlinear models in mechanics , and more precisely , for contact problems ; * the computation of some high - dimensional functions in molecular dynamics . concerning the first domain of application which is the main focus of this work , there is now a wide literature on the subject , ranging from specific questions related to the modeling of the noise sources ( in particular of their correlation ) , to dedicated methods for the evaluation of events with very small probabilities ( reliability ) . the focus of this paper is rather on the development of methods to compute efficiently a _ reduced model _ which rapidly gives the output of interest as a function of the random variable which enters the input parameters , in the context of contact problems in continuum mechanics . such a model can then be used to evaluate the distribution of the outputs ( for a given distribution of the input parameters ) , or to reduce the variance in a monte carlo computation for example . many methods have been proposed in the literature to attack this problem@xcite : stochastic collocation methods , galerkin methods , perturbation methods , etc . to be more specific , let us assume that the noise on the parameters of the model can be modeled by a possibly large number of random variables @xmath0 , so that the quantity of interest ( say the displacement field ) @xmath1 is a function of @xmath2 variables , where @xmath3 is the dimension of the physical space . the question is then how to approximate a function on such a high - dimensional space . the natural idea at the basis of many methods is to look for the solution to this problem as a linear combination of tensor products : @xmath4 where @xmath5 and @xmath6 are bases of vector spaces of dimension @xmath7 and @xmath8 respectively which are fixed a priori , and where @xmath9 are real numbers to be computed . this method leads to the resolution of a problem in a vector space of dimension @xmath10 which may be very large . this difficulty becomes all the more pregnant if @xmath11 is really large , so that the solution should be typically approximated as a sum : @xmath12 in this case , @xmath13 will be too large for a classical discretization method . the method we are studying is a way to circumvent this difficulty . the second application we have in mind is the computation of the solution to a high - dimensional poisson equation arising in molecular dynamics , called the committor function@xcite . mathematically , this function gives the probability for a stochastic process to reach a given region ( say @xmath14 ) before another one ( say @xmath15 ) . using feynman - kac formula , it can be shown that this function satisfies a poisson equation in a weighted sobolev space , with dirichlet boundary conditions ( namely @xmath16 on @xmath17 and @xmath18 on @xmath19 ) . typically , the stochastic process lives in a high - dimensional space ( @xmath3 is large ) , so that computing this function is a challenge . in both cases , the difficulty comes from the high - dimensionality of the function to approximate . the principle of the method we are interested in is : ( i ) to rewrite the original problem as a minimization problem : @xmath20 where @xmath21 is a functional defined on some hilbert space @xmath22 and ( ii ) to expand the solution in tensor products of lower - dimensional functions @xmath23 in practice , for each @xmath24 , the functions @xmath25 and @xmath26 are computed as linear combinations of the functions of the bases @xmath27 and @xmath28 so that @xmath29 and @xmath30 where for each @xmath31 , @xmath32 and @xmath33 . in the end , computing the approximation ( [ eq : approx ] ) leads to a problem of dimension @xmath34 which , provided that @xmath35 remains small enough , will hopefully be lower than the dimension of the problem obtained with the classical approach @xmath10 when the size of the bases @xmath7 and @xmath8 are large . the reduction of dimension is even more significant when we are in the case of equation ( [ devlpt ] ) . indeed , the approximation ( [ eq : approx ] ) can be adapted in this case in the following form : @xmath36 in this case , the overall dimension of the problem will be @xmath37 instead of @xmath38 in the classical approach . such a representation of a function as a sum of tensor product of other functions to avoid the curse of dimensionality have already been introduced in the literature . one approach consists in using the so - called sparse tensor product representation@xcite . if the solution @xmath39 we wish to approximate is sufficiently regular , one does not need to use fine discretizations in each direction . this idea can be used for example in galerkin - like discretizations . however , this method loses its efficiency in the case when the solution @xmath39 is not regular enough or when the mesh considered is complicated . we adopt another approach in this article . the principle of our method is to determine _ sequentially _ the pairs of functions @xmath40 which intervene in the approximation ( [ eq : approx ] ) through the following minimization problem : @xmath41 where @xmath42 and @xmath43 in ( [ eq : min_pb ] ) denote respectively hilbert spaces of functions depending only on the variable @xmath44 or only on the variable @xmath45 . to rewrite the two problems mentioned above as minimization problems on hilbert spaces , we penalize the constraints , namely the presence of the obstacle for the contact problem , or the dirichlet boundary conditions for the high - dimensional poisson problem . the method described above has been introduced by chinesta@xcite for solving high - dimensional fokker - planck equations , by nouy@xcite in the context of uncertainty quantification in mechanics , and is very much related to so - called greedy algorithms@xcite used in nonlinear approximation theory . the main contributions of this work are the following : * the convergence of the greedy algorithm ( [ eq : approx])-([eq : min_pb ] ) to the unique solution of ( [ eq : minimisation ] ) is proved , under the key assumptions that @xmath21 is strongly convex and that the gradient of @xmath46 is lispchitz on bounded sets ; * an exponential rate of convergence is obtained in the finite dimensional case ; * an adequate procedure to solve the minimization subproblem is proposed and tested on an academic test case . this paper can be seen as an extension of previous works on greedy algorithms @xcite which concentrate on the linear case , namely when @xmath47 , where @xmath48 is the norm of the hilbert space @xmath22 , and @xmath49 a continuous linear form on @xmath22 . we would like to stress that even if all the results and proofs are provided in the context of tensor products of two functions , our results can be easily generalized to the case of tensor products of more than two functions such as ( [ devlpt ] ) _ except for the results in section 5_. we have chosen not to present the results in this general setting for the sake of clarity . the paper is organized as follows . in section 2 , we introduce the general setting for the problem we consider and state the main result of this paper , namely the convergence of the greedy algorithm . section 2 also presents more precisely the two specific examples of application we have in mind . section 3 is devoted to the proof of the convergence . in section 4 , an exponential rate of convergence is proved , in the finite dimensional setting ( i.e. when @xmath42 and @xmath43 are finite dimensional spaces ) . section 5 shows that , under specific additional assumptions which are typically satisfied in the context of uncertainty quantification , the convergence results also hold if @xmath50 in is only a local minimum . finally , section 6 is devoted to a discussion of the numerical implementation , as well as to the presentation of test cases on a toy model . in this paper , we are interested in the convergence of a greedy algorithm for the minimization of high - dimensional nonlinear convex problems . we first introduce the general theoretical setting in which we prove the convergence , then describe two prototypical examples to which our analysis can be applied . throughout this article , @xmath11 and @xmath3 denote some positive integers , and @xmath51 and @xmath52 some open sets of @xmath53 and @xmath54 respectively . let @xmath42 and @xmath43 be hilbert spaces of real - valued functions respectively defined over @xmath51 and @xmath52 ( typically @xmath55 or sobolev spaces ) . let @xmath56 and @xmath57 be the norms of @xmath42 and @xmath43 . we introduce the following tensor product for all @xmath58 , @xmath59 which defines a real - valued function on @xmath60 . we also denote by @xmath61 . let @xmath22 be a hilbert space of real - valued functions defined on @xmath62 . the scalar product of @xmath22 is denoted by @xmath63 and the associated norm by @xmath64 . let @xmath46 be a differentiable real - valued functional defined on @xmath22 . for all @xmath65 , we denote by @xmath66 the gradient of @xmath46 at @xmath67 . we make the following assumptions : @xmath68 @xmath69 is a dense subset of @xmath22 for @xmath64 ; @xmath70 for all sequences of @xmath71 bounded in @xmath22 , there exists a subsequence which weakly converges in @xmath22 towards an element of @xmath71 ; @xmath72 the functional @xmath46 is strongly convex for @xmath64 , i.e. there exists a constant @xmath73 for which @xmath74 the functional @xmath46 is also said to be @xmath75-convex ; @xmath76 the gradient of @xmath46 is lipschitz on bounded sets : for each bounded subset @xmath77 of @xmath22 , there exists a nonnegative constant @xmath78 such that @xmath79 the unique global minimizer of @xmath46 on @xmath22 is denoted by @xmath39 . its existence and uniqueness are ensured by the @xmath75-convexity of the functional @xmath46 : @xmath80 we are going to study the following algorithm : the sequence @xmath81 is defined recursively by @xmath82 throughout this article , we will denote for all @xmath83 , @xmath84 our main result is the following theorem , whose proof is given in section 3 . under the assumptions @xmath68 , @xmath70 , @xmath72 and @xmath76 , the iterations of the algorithm are well - defined , in the sense that ( [ algorithm ] ) has at least one minimizer @xmath85 . moreover , the sequence @xmath86 strongly converges in @xmath22 towards @xmath39 . for each @xmath83 , the minimizer of ( [ algorithm ] ) is not unique in general . in particular , notice that the function @xmath87 is not convex . theorem 2.1 could be generalized to the case of tensor products of more than two hilbert spaces . indeed , let @xmath88 with @xmath89 . let @xmath90 be @xmath91 positive integers . let @xmath92 be @xmath91 open subsets of @xmath93 respectively . we consider @xmath91 hilbert spaces , @xmath94 of real - valued functions defined respectively on @xmath92 . let @xmath22 be a hilbert space of real - valued functions defined on @xmath95 . let @xmath46 be a real - valued differentiable functional defined on @xmath22 . we denote by @xmath96 . our algorithm can then easily be adapted provided that assumptions @xmath68 , @xmath70 , @xmath72 and @xmath76 are satisfied : @xmath97 are defined recursively by @xmath98 our convergence result also holds in this case . but for the sake of simplicity , we will limit our analysis to the case of only two hilbert spaces . let @xmath99 be the scalar product defined on @xmath69 as : for all @xmath100 , @xmath101 where @xmath102 and @xmath103 denote the scalar products of @xmath42 and @xmath43 respectively . let @xmath104 be the cross - norm associated to the scalar product @xmath99 . the tensor space of @xmath42 and @xmath43 , denoted as @xmath105 is then defined as the closure of @xmath69 for the product norm @xmath104 , @xmath106 let us point out that the hilbert space @xmath22 is not necessarily equal to @xmath105 , the tensor space of @xmath42 and @xmath43 associated to the tensor product ( [ tensor ] ) . indeed , an example where our analysis can be applied and where @xmath107 is given in section 2.2.2 ( see remark 2.5 . ) . however , the following inclusion relationship holds : @xmath108 . if @xmath42 and @xmath43 are discretized in finite - dimensional spaces of dimension @xmath7 and @xmath8 , our algorithm consists in solving several problems in dimension @xmath109 instead of solving one problem of dimension @xmath110 . thus , we can circumvent the curse of high - dimensionality . to prove that the general theoretical setting we described in section 2.1 is satisfied on the prototypical problems we present in this section , we need the following lemma , which is well - known in distribution theory@xcite . let @xmath111 be a distribution such that for any functions @xmath112 , @xmath113 then @xmath114 in @xmath115 . moreover , for any two sequences of distributions @xmath116 and @xmath117 such that @xmath118 in @xmath119 and @xmath120 in @xmath121 , @xmath122 in @xmath115 . an example of application of our algorithm is the study of uncertainty propagation on obstacle problems . we assume that uncertainty can be modeled by a set of @xmath11 random variables @xmath123 , @xmath124 , ... , @xmath125 , and that the random vector @xmath126 takes its values in @xmath51 . we consider also that the physical problem is defined over the domain @xmath52 , which is supposed to be a bounded subset of @xmath54 . if @xmath127 is a hilbert space of functions defined on @xmath52 , we denote by @xmath128 < + \infty \right\},\ ] ] where @xmath129 denotes the expectation with respect to the probability law of @xmath130 , and @xmath131 denotes the norm of @xmath127 . we endow @xmath132 with the scalar product defined by @xmath133 $ ] . a formulation of the obstacle problem with uncertainty is the following@xcite . let @xmath134 and @xmath135 . a membrane is stretched over the domain @xmath52 and is deflected by some random force having pointwise density @xmath136 for @xmath137 . at the boundary @xmath138 , the membrane is fixed and in the interior of @xmath52 the deflection is assumed to be bounded from below by the function @xmath139 ( a random obstacle ) . then the deflection @xmath140 is solution of the following obstacle problem with uncertainty ( see figure 1 ) : @xmath141 an equivalent formulation of this problem is the following . let us denote @xmath142 solving the obstacle problem ( [ form1 ] ) consists in solving the minimization problem @xmath143 where @xmath144 $ ] . one of the main difficulties of this kind of problems is their very high nonlinearity . many methods have been proposed to approximate the solution of these problems in the case without uncertainty@xcite . among them , penalization methods@xcite are among the most widely used . they consist in approximating the solution of a given obstacle problem by a sequence of solutions of penalized problems defined on the entire hilbert space . let @xmath145 be a parameter in @xmath146 . such a penalized problem associated with problem ( [ obstacle ] ) may be defined as @xmath147 where @xmath148_+^2 dx\right]$ ] . here and below , we denote by @xmath149_+$ ] the positive part of the real number @xmath150 , i.e. @xmath151_+ = 0 $ ] if @xmath152 and @xmath151_+ = a$ ] if @xmath153 . when @xmath145 goes to infinity , the solution @xmath154 of problem ( [ penalized ] ) strongly converges to the solution @xmath155 of problem ( [ obstacle ] ) . the goal of the algorithm we described in the previous section is to calculate the solution @xmath156 of this regularized problem for a given value of the parameter @xmath145 . let us check that the general theoretical setting we described in section 2.1 can be applied in this case . let us consider @xmath157 , @xmath158 , @xmath159 and @xmath160 for @xmath65 . we have @xmath161 . we endow @xmath162 with the scalar product defined by @xmath163 . in this case , we have @xmath164 and as a consequence assumption @xmath68 is obviously satisfied . besides , assumption @xmath70 is satisfied as well . if @xmath165 is such that @xmath166 is bounded , it is possible to extract a subsequence which weakly converges in @xmath22 towards an element @xmath167 . besides , there exists a non - negative constant @xmath168 such that for all @xmath169 , @xmath170 \\ & = & \mathbb{e}\left[|r_n(t)|^2\right ] \int_{\mathcal{x } } |\nabla_x s_n(x)|^2 dx \\ & = & \|r_n\|_{v_t}^2 \|s_n\|_{v_x}^2 \\ & \leq & c .\\\end{aligned}\ ] ] we can then choose @xmath171 such that @xmath172 and @xmath173 . the sequences @xmath174 and @xmath175 are then bounded in @xmath176 and @xmath162 respectively and we can extract subsequences which weakly converge in @xmath176 and @xmath162 towards @xmath177 and @xmath178 respectively . as the weak convergences in @xmath176 and @xmath162 imply the convergences in the distributional sense , the sequence @xmath172 necessarily converges towards @xmath179 in @xmath180 by lemma 2.1 . as the weak convergence in @xmath22 also implies the convergence in the sense of the distributions , we obtain , by uniqueness of the limit , @xmath181 . hence assumption @xmath70 is satisfied . the functional @xmath46 is differentiable and @xmath16-convex . indeed , for all @xmath65 , @xmath182_+^2dx\right]\right),\ ] ] is the sum of a @xmath16-convex function ( @xmath183 ) and of a convex function ( @xmath184_+^2dx\right]$ ] ) . the functional @xmath46 therefore obeys property ( [ alpha ] ) with @xmath185 . hence , assumption @xmath72 is satisfied . let us finally check that the gradient of @xmath46 is lipschitz . for all @xmath186 , @xmath187\right| \\ & & + \rho\left|\mathbb{e}\left [ \int_{\mathcal{x } } ( [ g(t , x)-v(t , x)]_+ - [ g(t , x ) -w(t , x)]_+)y(t , x)dx \right]\right| \\ & \leq & \|v - w\|_v\|y\|_v \\ & & + \rho \mathbb{e}\left [ \int_{\mathcal{x } } \left|[g(t , x)-v(t , x)]_+ - [ g(t , x ) -w(t , x)]_+\right| \left|y(t , x)\right|dx \right].\\\end{aligned}\ ] ] for @xmath188 , it can easily be seen that @xmath189_+-[b]_+|\leq |a - b|$ ] . this implies @xmath190 \\ & \leq & \|v - w\|_v\|y\|_v\\ & & + \rho \left(\mathbb{e}\left [ \int_{\mathcal{x } } |v(t , x)- w(t , x)]|^2 dx\right]\right)^{1/2}\left(\mathbb{e}\left [ \int_{\mathcal{x } } |y(t , x)]|^2 dx\right]\right)^{1/2 } .\\\end{aligned}\ ] ] the poincar inegality in @xmath191 implies that there exists a nonnegative constant @xmath192 such that for all @xmath193 , @xmath194|^2 dx\right]\right|^{1/2 } \leq d\|h\|_v.\ ] ] this yields @xmath195 hence , @xmath196 the functional @xmath46 then obeys property ( [ lipschitz ] ) with a constant @xmath197 independent of the bounded set considered . thus , our obstacle problem ( [ penalized ] ) falls into the general theoretical setting introduced in section 2.1 . there exist several variants of the obstacle problem which could be tackled with our algorithm . we refer to ref . @xcite or ref . @xcite for such examples . our algorithm may also be used to calculate the solution of other problems than obstacle problems . other examples are high - dimensional nonlinear poisson equations . a specific application where such high dimensional poisson equations arise is the calculation of the so - called committor function in molecular dynamics@xcite , which is an important quantity to compute reaction rates or to derive some effective dynamics for example . let @xmath198 . the committor is the solution to the following problem : @xmath199 where @xmath91 is typically large , @xmath17 and @xmath19 are disjoint smooth open sets of @xmath200 , @xmath201 is a given potential function such that @xmath202 and @xmath203 for @xmath204 , @xmath205 can be interpreted as the probability that the stochastic process @xmath206 solution to @xmath207 reaches @xmath208 before @xmath209 . here , @xmath210 denotes a @xmath91-dimensional brownian motion . let @xmath211 such that @xmath212 . in this example , we consider the case when @xmath213 is bounded . let @xmath51 and @xmath52 be open convex bounded subsets of @xmath53 and @xmath54 respectively such that @xmath214 and such that @xmath215 where @xmath216 denotes the lebesgue measure . we also assume that @xmath217 . in this case , the initial problem can be rewritten as a minimization problem set on @xmath218 instead of @xmath219 . indeed , as @xmath220 and @xmath221 is bounded , there exists constants @xmath222 such that for all @xmath223 , @xmath224 . and thus , we have @xmath225 if and only if @xmath226 , @xmath227 and @xmath228 . the penalized version of the committor problem then reads @xmath229 where @xmath230 for some @xmath231 . let us check that the general theoretical setting described in section 2.1 is relevant for this problem . in this case , we consider @xmath232 , @xmath233 and @xmath234 . the inner products that are defined over these hilbert spaces are the following . for all @xmath235 , @xmath236 , @xmath237 , @xmath238 @xmath239 @xmath240 let us point out that in this case , @xmath107 . indeed , for all @xmath241 , the @xmath22-norm of the tensor product @xmath242 reads @xmath243 which is not a cross - norm , equivalent to the norm induced by @xmath244 and @xmath245 over @xmath105 , which is @xmath246 indeed , let us consider @xmath247 , @xmath248 and @xmath249 for @xmath250 and @xmath251 . the sequence @xmath252 is bounded , but the sequence @xmath253 is not . assumption @xmath68 holds true , since @xmath254 is such that @xmath255 and @xmath256 is dense in @xmath257 . hence , @xmath258 is also dense in @xmath22 . let us prove that assumption @xmath70 also holds true . if @xmath165 is such that @xmath166 is bounded , we can extract a subsequence of @xmath259 which weakly converges in @xmath22 towards an element @xmath167 . besides , there exists a nonnegative constant @xmath168 such that for all @xmath169 , @xmath260 we can then choose @xmath171 such that @xmath172 and such that @xmath261 . the sequences @xmath174 and @xmath175 are then bounded in @xmath262 and @xmath263 and we can extract subsequences which weakly converge in @xmath262 and @xmath263 respectively towards @xmath264 and @xmath265 . as the weak convergences in @xmath262 and @xmath263 imply the convergences in the distributional sense , @xmath172 necessarily converges towards @xmath179 in the distributional sense by lemma 2.1 . as the weak convergence in @xmath22 also implies the convergence in the sense of the distributions , by uniqueness of the limit , @xmath266 . let us suppose @xmath267 . in that case , we have @xmath268 and @xmath269 . besides , we have @xmath270 hence @xmath271 as a consequence @xmath272 is finite and @xmath273 . hence @xmath274 . if @xmath275 , then obviously @xmath276 . hence , assumption @xmath70 holds true . the functional @xmath46 is differentiable and strongly convex . to prove this , it is sufficient to prove that there exists a constant @xmath277 such that for all @xmath278 , @xmath279 . indeed , there exists @xmath280 such that for all @xmath281 , @xmath282 . thus , there exists a constant @xmath283 such that , for all @xmath278 , @xmath284 to prove that the functional @xmath46 is strongly convex , it is sufficient to have the following inequality : there exists a constant @xmath285 such that for all @xmath286 , @xmath287 as @xmath51 and @xmath52 are bounded open convex subsets of @xmath53 and @xmath54 respectively , @xmath221 is then a bounded open convex subset of @xmath200 such that @xmath288 and inequality ( [ poinc ] ) is a well - known poincare - like inequality . hence , assumption @xmath72 is satisfied . let us check that the gradient of @xmath46 is lipschitz . for all @xmath289 , @xmath290 hence @xmath291 the functional @xmath46 therefore obeys property ( [ lipschitz ] ) with a constant @xmath292 independent of the bounded subset considered . thus , the committor problem falls into the general theoretical setting introduced in section 2.1 . we begin by proving that the iterations of the algorithm are well - defined . for this , we will need the following lemma . let @xmath293 be a function in @xmath22 . then there exists a pair @xmath241 such that @xmath294 if and only if @xmath295 . let @xmath167 and let us suppose that @xmath296 for all @xmath297 . for a given pair @xmath298 , for all @xmath299 , @xmath300 as a consequence , we have the following by letting @xmath301 go to @xmath18 : @xmath302 . this holds for all @xmath297 . hence , for all @xmath303 , we also have @xmath304 , and the density of @xmath69 in @xmath22 , which is ensured by assumption @xmath68 , yields @xmath305 . conversely , let us assume that @xmath306 . then , as @xmath46 is @xmath75-convex , @xmath293 is necessarily the global minimizer of @xmath46 and , in particular , we have for all @xmath307 , @xmath308 this concludes the proof . using this lemma , the following result can be proved : for all @xmath309 , there exists a solution @xmath310 to the minimization problem ( [ algorithm ] ) . moreover , @xmath311 if and only if @xmath312 , where @xmath313 is defined by ( [ undef ] ) . firstly , let us prove the existence of a minimizer for problem ( [ algorithm ] ) . let @xmath314 . for all @xmath315 , @xmath316 . so @xmath317 exists in @xmath318 . we then consider a minimizing sequence @xmath319 such that @xmath320 using ( [ alpha ] ) and the fact that @xmath321 , we have @xmath322 then the sequence @xmath323 is bounded in @xmath22 because @xmath324 is convergent and consequently bounded . as assumption @xmath70 is satisfied , we can then extract a subsequence ( which we still denote @xmath323 ) which weakly converges in @xmath22 towards an element of @xmath71 . in other words , there exist @xmath325 and @xmath326 such that @xmath323 weakly converges in @xmath22 towards @xmath327 . furthermore , as the functional @xmath46 is convex and continuous on @xmath22 , @xmath328 hence @xmath329 so that @xmath330 is a minimizer of problem ( [ algorithm ] ) . let us prove now that @xmath331 if and only if @xmath332 . if @xmath333 , we have @xmath334 for all @xmath241 such that @xmath335 as @xmath46 is strictly convex . so a minimizer @xmath179 of problem ( [ algorithm ] ) must necessarily satisfy @xmath336 . conversely , if @xmath312 , we have @xmath337 and from lemma 3.1 , there exists a pair @xmath241 such that @xmath338 . hence , @xmath339 and @xmath179 can not be equal to @xmath18 . for each @xmath83 , a minimizer @xmath85 of problem ( [ algorithm ] ) obeys the following euler equation : @xmath340 this result is obtained by considering the first - order conditions of the minimization problem ( [ algorithm ] ) . this will be useful in the proof of convergence . in this subsection , we present the different steps of the proof . the series @xmath341 and the sequence @xmath342 are convergent . let us set @xmath343 using ( [ algorithm ] ) , @xmath344 for all @xmath345 , and in particular , by taking @xmath346 , @xmath347 is a non - increasing sequence . moreover , it is bounded from below . indeed , for all @xmath309 , we have @xmath348 . thus , it is convergent . this implies that the sequence defined as @xmath349 is nonnegative , converges to @xmath18 , and satisfies @xmath350 . besides , the @xmath75-convexity of @xmath46 yields the following inequality : @xmath351 using the euler equations ( [ el ] ) , @xmath352 and thus , @xmath353 . hence the result . the sequence @xmath354 is bounded in @xmath22 . by @xmath75-convexity of the functional @xmath46 , we have @xmath355 thus @xmath356 therefore , the sequence @xmath357 is bounded in @xmath22 . the following estimate is essential for the proof of convergence . there exists a constant @xmath358 such that , for all @xmath83 and all @xmath345 , @xmath359 let @xmath360 be such that for all @xmath83 , @xmath361 . its existence is ensured by lemma 3.3 . let @xmath362 be such that for all @xmath83 , @xmath363 . let @xmath364 be the closed ball of @xmath22 centered at @xmath18 of radius @xmath365 . let @xmath49 be the lipschitz constant associated with @xmath77 in ( [ lipschitz ] ) . for all @xmath345 , we have @xmath366 then , by the convexity of @xmath46 , we have the following inequality @xmath367 which leads to @xmath368 let @xmath345 such that @xmath369 . we then have , by using ( [ lipschitz ] ) and ( [ ineg ] ) , @xmath370 the last line has been obtained by taking into account the fact that @xmath371 because of the euler equation ( [ el ] ) . thus , for all @xmath58 such that @xmath372 , @xmath373 as a consequence , @xmath374 let @xmath297 such that @xmath375 and @xmath376 such that @xmath377 . then , we have @xmath378 and , by setting @xmath379 , we obtain the following inequality for all @xmath241 such that @xmath380 , @xmath381 of course , this inequality also holds true for all @xmath345 such that @xmath382 . therefore , ( [ estimate ] ) holds with @xmath383 . we now state an elementary result which will be useful in the sequel . let @xmath384 be a sommable sequence of @xmath146 . then , there exists a subsequence of @xmath385 which converges to @xmath18 . if such a subsequence could not be extracted , it would imply @xmath386 thus , the series @xmath387 would diverge . hence the contradiction . we are now in position to complete the proof of theorem 2.1 . by lemma 3.2 , the sequence @xmath388 is convergent . let us denot its limit by @xmath389 . we want to prove that @xmath390 . firstly , for all @xmath309 , @xmath391 , since @xmath39 is the global minimizer of the functional @xmath46 . by letting @xmath35 go to infinity , we obtain @xmath392 it remains to prove that @xmath393 . let us first prove that @xmath394 weakly converges to @xmath18 in @xmath22 . let @xmath360 such that for all @xmath83 , @xmath361 . its existence is ensured by lemma 3.3 . let @xmath395 be the closed ball of @xmath22 centered at @xmath18 of radius @xmath396 . let @xmath49 be the lipschitz constant associated with @xmath77 in ( [ lipschitz ] ) . using ( [ lipschitz ] ) and the fact that @xmath397 , we have @xmath398 and as @xmath357 is bounded in @xmath22 by lemma 3.3 , we deduce that @xmath399 is also bounded in @xmath22 . we can then extract a subsequence of @xmath399 which weakly converges in @xmath22 towards @xmath167 . by using proposition 3.3 and by letting @xmath35 go to infinity in ( [ estimate ] ) , we deduce that @xmath400 for all @xmath345 . then , as @xmath69 is dense in @xmath22 with assumption @xmath68 , necessarily @xmath275 . thus the sequence @xmath399 weakly converges to @xmath18 in @xmath22 . as @xmath46 is convex , we have the following inequality for all @xmath83 , @xmath401 let us prove that we can extract a subsequence of @xmath402 which converges to @xmath18 . let @xmath309 . by using proposition 3.3 , @xmath403 as the sequence @xmath404 converges by lemma 3.2 , we have @xmath405 . furthermore , we can also extract a subsequence from @xmath406 which converges to @xmath18 ( see lemma 3.4 ) . we can then extract a subsequence from @xmath407 which converges to @xmath18 . by letting @xmath35 go to infinity in ( [ fin ] ) with this subsequence , we obtain that @xmath393 . we have thus proved that @xmath408 . besides , as the functional @xmath46 is @xmath75-convex , ( [ alpha ] ) yields the following inequality , @xmath409 which necessarily implies that @xmath410 converges to @xmath18 when @xmath35 goes to infinity , which proves that @xmath357 strongly converges towards @xmath39 in @xmath22 . in the case when @xmath42 and @xmath43 are finite - dimensional , we are able to prove that the algorithm converges exponentially fast . we assume that @xmath42 and @xmath43 are finite - dimensional and that assumptions @xmath68 , @xmath70 , @xmath72 and @xmath76 are fulfilled . then there exist two constants @xmath411 and @xmath412 such that for all @xmath83 , @xmath413 and @xmath414[1 ] let us denote by @xmath415 and @xmath416 . then we can consider that @xmath417 , @xmath418 and @xmath419 ( which is implied by @xmath68 ) . as the spaces are finite - dimensional , all the norms are equivalent , and we can consider without loss of generality that @xmath244 , @xmath245 and @xmath64 are equal to the frobenius norms of @xmath420 , @xmath421 and @xmath422 defined by : @xmath423 \|s\|_m^2 & = & s^t s,\\ [ 5pt ] \|u\|_{lm}^2 & = & \mbox{tr}(u^t u).\\ \end{array}\ ] ] notice that for all @xmath424 , @xmath425 let @xmath27 and @xmath28 be orthonormal bases of @xmath42 and @xmath43 respectively . then , @xmath426 forms an orthonormal basis of @xmath22 . our goal is to prove that there exists a constant @xmath412 such that for all @xmath83 , @xmath427 let @xmath83 . let us notice that @xmath428 as for all @xmath83 , @xmath429 , it is then sufficient with ( [ ratebeg ] ) to prove that there exists @xmath430 such that @xmath431 to have ( [ finalin ] ) with @xmath432 . let us notice that ( [ alpha ] ) and ( [ el ] ) yield @xmath433 besides , let @xmath434 such that for all @xmath83 , @xmath361 . its existence is ensured by lemma 3.3 . let @xmath435 be the closed ball of @xmath22 centered at @xmath18 of radius @xmath436 . let @xmath49 be the lipschitz constant of the gradient of @xmath46 associated to @xmath77 in ( [ lipschitz ] ) . using ( [ lipschitz ] ) and the fact that @xmath397 , we also have , @xmath437 with ( [ init ] ) and ( [ un2 ] ) , it is sufficient to prove that there exists a constant @xmath438 such that for all @xmath83 , @xmath439 in order to have ( [ suite ] ) and hence ( [ finalin ] ) . indeed , if ( [ kappa ] ) holds , we then have , using ( [ init ] ) , ( [ kappa ] ) and ( [ un2 ] ) , @xmath440 as the @xmath75-convexity of @xmath46 and the fact that @xmath397 yields @xmath441 inequalities ( [ un1 ] ) and ( [ un2 ] ) then imply that @xmath442 and then ( [ suite ] ) holds with @xmath443 . let us prove inequality ( [ kappa ] ) . from proposition 3.3 , estimate ( [ estimate ] ) holds true . as @xmath426 forms an orthonormal basis of @xmath22 , we obtain , using ( [ estimate ] ) , @xmath444 we then have the following estimate : @xmath445 the @xmath75-convexity of @xmath46 and estimate ( [ normeprime ] ) lead to @xmath446 besides , by using the fact that @xmath447 , we obtain @xmath448 which is ( [ kappa ] ) with @xmath449 for @xmath17 large enough . hence the result . this result can be generalized to the case of tensor products of more than two hilbert spaces . indeed , with the notation of remark 2.2 , and if we denote @xmath450 , estimate ( [ normeprime ] ) becomes @xmath451 and the proof still holds . we are able to extend the results of theorem 2.1 and theorem 4.1 in the case when @xmath50 in ( [ algorithm ] ) is only defined as a * local minimum which ensures the decrease of the energy , more precisely , when @xmath85 is defined recursively as : @xmath452 such that @xmath453 where @xmath313 is defined as in ( [ undef ] ) . * to extend these results , we will need an additional assumption ( which is naturally fulfilled in the finite dimensional case ) , see remark 5.2 below : @xmath454 there exist @xmath455 such that @xmath456 let us suppose that the assumptions @xmath68 , @xmath70 , @xmath72 , @xmath76 and @xmath454 hold true . then , the iterations of the algorithm described above are well - defined in the sense that ( [ localalgo ] ) has at least one local minimizer @xmath85 which satisfies ( [ decrease ] ) . moreover , the sequence @xmath457 strongly converges in @xmath22 towards @xmath39 . the proof is similar to the proof of theorem 2.1 given in section 3 except for proposition 3.3 which gives estimate ( [ estimate ] ) : @xmath458 this estimate is no longer true , but we have a similar result which will be enough to complete the proof . indeed , let us prove that there exists a constant @xmath459 such that @xmath460 let @xmath360 such that for all @xmath83 , @xmath361 . its existence is ensured by lemma 3.3 . let @xmath461 be the closed ball of @xmath22 centered at @xmath18 and of radius @xmath462 . let @xmath49 be the lipschitz constant associated to @xmath77 in ( [ lipschitz ] ) . let @xmath345 and @xmath314 . as @xmath50 is a local minimum of @xmath463 , there exists a constant @xmath464 such that for all @xmath465 , we have @xmath466 moreover , by convexity of the functional @xmath46 , we have the following inequality @xmath467 we deduce from ( [ locmin ] ) , ( [ convex ] ) and property ( [ lipschitz ] ) that , for all @xmath301 small enough so that @xmath468 , @xmath469 as @xmath50 is a local minimum of the functional @xmath470 , @xmath50 still obeys the euler equation ( [ el ] ) and thus @xmath471 . finally , we have @xmath472 dividing this expression by @xmath473 and letting @xmath301 go to zero , we obtain @xmath474 which leads to @xmath475 all this holds without the additional assumption ( [ svd ] ) for all @xmath241 . to derive estimate ( [ estimate2 ] ) , we use the additional assumption we made on @xmath64 : @xmath476 we can then choose @xmath477 and @xmath478 such that @xmath172 and @xmath479 and such that @xmath480 and @xmath481 . thus , @xmath482 and in the end , we obtain estimate ( [ estimate2 ] ) with @xmath483 . with this result , it is then possible to conclude as in the proof of theorem 2.1 . problem ( [ penalized ] ) falls into the scope of theorem 5.1 . on the other hand , this is not the case for problem ( [ poisson ] ) , for which property ( [ svd ] ) is not true ( see remark 2.5 ) . we were not able to prove a similar result in the case when @xmath64 does not satisfy property ( [ svd ] ) . here are two typical examples for which assumption @xmath454 holds : * in the case when @xmath164 , property ( [ svd ] ) holds with @xmath484 . this holds in uncertainty propagation problems where @xmath485 with @xmath127 an hilbert space of real - valued functions defined on @xmath52 . denoting by @xmath158 and @xmath486 , then @xmath487 . * in other cases , to find an approximation of the global minimum of the energy @xmath46 , the hilbert spaces @xmath42 and @xmath43 are usually discretized in finite - dimensional spaces . the problem can then be rewritten as a problem over @xmath417 , @xmath418 with @xmath488 , and then @xmath22 is naturally defined as the hilbert space @xmath419 . then , assumptions @xmath68 , @xmath70 , @xmath72 and @xmath76 are automatically satisfied on the discretized spaces . as all the norms are equivalent in finite dimension , the norms on @xmath420 , @xmath421 and @xmath422 induced by the norms defined over the original hilbert spaces @xmath42 , @xmath43 and @xmath22 are equivalent to the frobenius norms , defined by ( [ normkl ] ) . these norms satisfy property ( [ svd ] ) since for all @xmath424 , @xmath489 . hence , the norms induced by the norms defined on the original hilbert spaces automatically satisfy property ( [ svd ] ) even if the property is not satisfied in the continuous spaces . as in section 4 , we can prove that the algorithm defined by ( [ localalgo ] ) and ( [ decrease ] ) converges exponentially fast in finite dimension . let us consider the algorithm defined by ( [ localalgo ] ) and ( [ decrease ] ) . let @xmath488 . let @xmath490 , @xmath418 and @xmath419 . then there exist two constants @xmath411 and @xmath412 such that for all @xmath83 , @xmath413 and @xmath414 as the spaces are finite - dimensional , assumptions @xmath68 , @xmath70 , @xmath72 , @xmath76 and @xmath454 are automatically fulfilled ( see remark 5.2 ) and estimate ( [ estimate2 ] ) holds true . the proof is similar to the proof of theorem 4.1 . indeed , ( [ ratebeg ] ) , ( [ init ] ) , ( [ un2 ] ) and ( [ un1 ] ) still hold . then it is sufficient to prove an inequality similar to ( [ kappa ] ) to prove theorem 5.2 . however , as ( [ estimate2 ] ) holds instead of ( [ estimate ] ) , inequality ( [ normeprime ] ) is replaced by : @xmath491 and consequently , an inequality similar to ( [ kappa ] ) must be obtained in another way . let @xmath360 such that for all @xmath83 , @xmath361 . its existence is ensured by lemma 3.3 . let @xmath395 be the closed ball of @xmath22 centered at @xmath18 of radius @xmath396 . let @xmath49 be the lipschitz constant associated with @xmath77 in ( [ lipschitz ] ) . on the one hand , using the convexity of @xmath46 , ( [ lipschitz ] ) and the fact that @xmath397 , we have @xmath492 on the other hand , ( [ un1 ] ) , the convexity of @xmath46 , ( [ el ] ) and ( [ normeprime2 ] ) yield @xmath493 then , using ( [ sup ] ) , we have ( [ kappa ] ) with @xmath494 . we can conclude as in the proof of theorem 4.1 . the results given in this section may not stand when we consider more than two hilbert spaces . indeed , the scheme of the proof of theorem 5.1 can not be easily adapted and we do not necessarily have an estimate similar to ( [ estimate2 ] ) . in this section , we describe how we implemented the algorithm introduced in section 2 for the resolution of problem ( [ penalized ] ) in a very simple setting , namely a one - dimensional membrane problem with uncertainty . we present the numerical results we obtained . additional investigations to demonstrate the applicability and the efficiency of the procedure on high - dimensionnal problems are still required . we however refer to nouy@xcite for illustrations of the interest of the method for problems in high dimensions . let us recall problem ( [ penalized ] ) . let @xmath135 and @xmath134 . let us assume that the random variable @xmath130 has a probability density @xmath495 for @xmath496 . in other words , @xmath497 where @xmath498 is a measurable subset of @xmath51 . for a given value of the penalization parameter @xmath499 , we wish to calculate an approximation of the minimizer of @xmath500 where @xmath501_+^2 dx \right].\ ] ] in other words , @xmath502_+^2 p(t ) \,dx\,dt.\\\end{aligned}\ ] ] in this case , our algorithm can be rewritten in the following form . set @xmath503 and @xmath504 and define recursively @xmath505 as @xmath506 with @xmath507_+^2p(t)\,dx\,dt,\\\end{aligned}\ ] ] where @xmath508 indeed , @xmath509_+^2p(t)\,dx\,dt + \mathcal{e}_n(r\otimes s),\ ] ] where @xmath313 is defined as in ( [ undef ] ) . in fact , from theorem 5.1 , it is sufficient for @xmath50 to be a _ local minimum of @xmath510 such that : @xmath511_+^2 p(t)\,dx\,dt,\ ] ] which ensures ( [ decrease ] ) . _ we write the algorithm in the discrete case , and , for clarity , we restrict ourselves to the case of two open intervals @xmath51 and @xmath52 of @xmath318 . more precisely , @xmath512\underline{t } , \overline{t}[$ ] and @xmath513\underline{x } , \overline{x}[$ ] , with @xmath514 , such that @xmath515 and @xmath516 . let @xmath488 , which will denote respectively the number of degrees of freedom in the discretized spaces of @xmath42 and @xmath43 . let us introduce a subdivision @xmath517 such that @xmath518 and a subdivision @xmath519 such that @xmath520 . let @xmath521 and @xmath522 be functions such that @xmath523 and @xmath524 and let us consider @xmath525 and @xmath526 . for example , @xmath527 or @xmath528 finite elements satisfy these properties for all @xmath198 . our goal is to find an approximation of the function @xmath39 under the following form @xmath529 where @xmath530 . let @xmath531 be the symmetric positive definite matrix which corresponds to the discretization of the one - dimensional operator @xmath532 in @xmath43 , in other words , for all @xmath533 , @xmath534 let also @xmath535 and @xmath536 be the symmetric positive definite matrices defined as @xmath537 and @xmath538 with discretization ( [ eq : discrete ] ) , the term @xmath539 is then equal to @xmath540 similarly , if we denote by @xmath541 the matrix defined as , for all @xmath542 and @xmath543 , @xmath544 the term @xmath545 is approximated by @xmath546 the approximation of the term @xmath547_+^2 p(t ) \,dx\,dt$ ] is more subtle . indeed , let us approximate the function @xmath548 in the discretized space @xmath549 by @xmath550 given relationships ( [ eq : phii ] ) and ( [ eq : psij ] ) , a natural way to define @xmath551 is the following @xmath552 it then holds @xmath553_+^2 p(t ) \,dx\,dt \approx \frac{\rho}{2 } \int_{\mathcal{x}\times \mathcal{t } } \left [ \sum_{i=1}^l \sum_{j=1}^m ( g^{ij } - u^{ij } ) \phi_i(t)\psi_j(x ) \right]_+^2p(t)\,dx\,dt.\ ] ] let @xmath293 be the function defined as @xmath554_+$ ] . from assumptions ( [ eq : phii ] ) and ( [ eq : psij ] ) , it holds @xmath555_+ , \ ; \forall 1\leq i \leq l,\ ; 1\leq j \leq m.\ ] ] thus , we perform a supplementary approximation , that is to approximate the function @xmath293 itself as @xmath556_+ \phi_i(t)\psi_j(x).\\\end{aligned}\ ] ] finally , it holds @xmath557_+^2 p(t ) \,dx\,dt & \approx & \frac{\rho}{2 } \int_{\mathcal{x}\times \mathcal{t } } \left(\sum_{i=1}^l \sum_{j=1}^m \left[g^{ij } - u^{ij } \right]_+\phi_i(t)\psi_j(x)\right)^2 p(t)\,dx\,dt,\\ & = & \mbox{tr}(\phi [ g - u]_+ \psi [ g - u]_+^t).\\\end{aligned}\ ] ] we also made the following mass lumping approximation , that is to consider that @xmath558 and @xmath559 . then , the discretized problem ( [ penalized ] ) can be rewritten as @xmath560_+:[g - v]_+,\\ \end{array}\ ] ] where for @xmath561 , @xmath562 this problem is equivalent to : @xmath563_+.\ ] ] for each function @xmath564 and @xmath565 , we denote by @xmath566 and @xmath567 , their coordinates in the bases @xmath568 and @xmath28 , which are given by @xmath569 and @xmath570 our algorithm can then be rewritten as : choose a threshold @xmath571 and set @xmath572 , @xmath573 . at iteration @xmath574 : 1 . find @xmath575 and @xmath576 two vectors respectively in @xmath420 and @xmath421 such that : @xmath577 with @xmath578_+ : [ g_{n-1 } - rs^t]_+ . \\\end{aligned}\ ] ] 2 . set @xmath579 and @xmath580 . 3 . if @xmath581_+\| \geq \varepsilon$ ] , proceed to iteration @xmath582 . otherwise , stop . the remaining question is : how can we compute @xmath583 at step 1 ? this critical step is described in the following section . let us first describe a method which has been proposed by nouy@xcite and chinesta@xcite , that is the fixed - point procedure and which we use in our final numerical implementation ( see section 6.2.2 ) . we present this algorithm in a particular case . let us consider @xmath417 , @xmath418 and @xmath419 endowed with the frobenius norms defined by ( [ normkl ] ) . we fix a given matrix @xmath584 . let us define the energy functional as @xmath585 for @xmath586 . in this particuler case , applying the greedy algorithm described above consists in computing the singular value decomposition of the matrix @xmath587 . in this particular case , the greedy algorithm can be rewritten in the following form . choose a threshold @xmath588 and set @xmath589 . at iteration @xmath574 , 1 . find two vectors @xmath590 and @xmath591 respectively in @xmath420 and @xmath421 such that @xmath592 2 . set @xmath593 . 3 . if @xmath594 , proceed to iteration @xmath582 . otherwise , stop . the euler equation associated to this problem can be rewritten as @xmath595 \|r_n\|_{v_t}^2 s_n & = & ( m_{n-1})^t r_n.\\ \end{array } \right .\ ] ] the method which is generally used@xcite to solve these euler equation is a fixed - point algorithm , which simply reads ( for a fixed @xmath35 ) : at iteration @xmath596 , compute two vectors @xmath597 such that @xmath598 \|r_n^{(q+1)}\|_{v_t}^2 s_n^{(q+1 ) } & = & ( m_{n-1})^t r_n^{(q+1)}.\\ \end{array } \right .\ ] ] one can check@xcite that this procedure is similar to the power method to compute the largest eigenvalue ( and associated eigenvector ) of the matrix @xmath599 . one could think of transposing this fixed - point procedure to the case of the obstacle problem we consider in this article . in our case , the euler equation @xmath600_+^t r_n , \\[5pt ] ( ds_n : s_n ) r_n & = & f_{n-1 } s_n + \rho [ g_{n-1 } - r_n s_n^t]_+ s_n .\\ \end{array } \right .\ ] ] could be solved _ a priori _ with a fixed point algorithm , which , at iteration @xmath91 , might be written as @xmath601_+^t r_n^{(q ) } , \\[5pt ] ( ds_n^{(q+1)}:s_n^{(q+1 ) } ) r_n^{(q+1 ) } & = & f_{n-1 } s_n^{(q+1 ) } + \rho\left[g_{n-1 } - r_n^{(q ) } \left(s_n^{(q+1)}\right)^t\right]_+ s_n^{(q+1)}. \\ \end{array } \right .\ ] ] unfortunately , we have not been able to make this fully - explicit fixed point algorithm converge for large values of the parameter @xmath145 . we therefore have decided to resort to a minimization procedure . the approach we adopt then is the following . we choose an initial pair @xmath602 and then perform a quasi - newton algorithm to find a local minimum of the function @xmath603_+ : \left[g_{n-1 } - rs^t\right]_+.\ ] ] the main difficulty is to find a proper initial pair @xmath604 such that @xmath605_+ : \left[g_{n-1 } - r_n^{(0)}s_n^{(0)t}\right]_+\ ] ] @xmath606_+ : [ g_{n-1}]_+,\ ] ] to ensure that the energy decreases ( see ( [ decrease ] ) ) . let us describe our approach in the continuous setting with the notation used in section 4 . it consists in finding a pair @xmath607 such that @xmath608 we notice that for @xmath345 , and @xmath609 , we have @xmath610 for @xmath611 small enough . the idea is then to find a pair @xmath241 such that @xmath612 , so that there exists @xmath613 small enough for which @xmath614 . then , @xmath615 is a good initial guess . let us first consider the pair @xmath616 such that @xmath617 in other words , we consider @xmath618 the first term of the singular value decomposition of @xmath619 in @xmath22 . the euler equations then imply @xmath620 and therefore , @xmath621 by taking @xmath622 , there exists then @xmath609 small enough such that @xmath623 in the discrete case associated to problem ( [ penalized ] ) , @xmath624 is obtained by taking the first term of the singular value decomposition of the matrix @xmath625_+$ ] . this can be done with a fixed point procedure similar to ( [ fixedpoint ] ) . once we have this initial guess @xmath626 , we perform a quasi - newton algorithm to minimize the energy . the computations are done with the software scilab@xcite and the quasi - newton procedure is performed via the _ optim procedure of scilab . _ let us point out that this procedure is intrusive in general . in this section , we present the results we obtained with this algorithm on the following membrane problem . we suppose @xmath627 . we consider a random variable @xmath130 following a uniform law of probability on the interval @xmath628 . we wish to study problem ( [ obstacle ] ) with the following values for @xmath629 and @xmath630 , @xmath631_+ + ( t-1)[\sin(3\pi x)]_-.\ ] ] the negative part of @xmath632 , i.e. @xmath151_- = 0 $ ] if @xmath153 , and @xmath151_- = -a$ ] if @xmath152 , is denoted by @xmath151_-$ ] . the above problem models a rope attached at @xmath633 and @xmath634 subjected to gravity and resting upon obstacles whose altitudes are given by @xmath635 . the quantity @xmath1 then represents the altitude of the rope at abscissa @xmath45 when @xmath636 . this problem is approximated by problem ( [ penalized ] ) with parameter @xmath637 . the problem is discretized with a regular mesh and @xmath638 finite elements in each direction . discretization parameters are chosen as @xmath639 . figure 2 represents the altitude of the obstacles given by @xmath635 for @xmath640 ^ 2 $ ] . the algorithm described in the previous sections is then applied with the following stopping criterion : @xmath641_+\|_v < 10^{-4}$ ] with @xmath642 for @xmath643 . figure 3 represents the evolution of @xmath644 and of @xmath645_+\|_v)$ ] . [ rate ] [ ] [ ] [ 0.6]number of iterations [ ] [ ] [ 0.7]@xmath646 [ ] [ ] [ 0.6]number of iterations [ ] [ ] [ 0.7]@xmath647_+\|_v)$ ] we can see that our algorithm captures very quickly the main modes of the solution and that both the energy and the @xmath22-norm of the residue @xmath641_+\|_v$ ] converges exponentially fast , as predicted by theorem 5.1 . figure 4 represents the results obtained for the solution @xmath1 . figure 5 and 6 represents @xmath1 and @xmath635 for some special values of @xmath130 . [ fig3 ] [ ] [ ] [ 0.4]@xmath648 [ ] [ ] [ 0.4]@xmath649 [ ] [ ] [ 0.4]@xmath648 [ ] [ ] [ 0.4]@xmath649 as we can observe , the solution does not exactly satisfies the constraint @xmath650 . this is due to the fact that we approximate a solution @xmath651 of the penalized problem ( [ penalized ] ) for @xmath637 . this is the main drawback of our method : we do not approximate directly the solution of the initial obstacle problem but the solution of a close regularized problem . indeed , if we try to further increase the parameter @xmath145 , we face the main drawback of penalization methods , that is the ill - conditioning of the resulting matrices . in this article , we presented a greedy algorithm based on variable decomposition aiming at computing the global minimum of a strongly convex energy functional . we proved that , provided that the gradient of the energy is lipschitz on bounded sets , and that the hilbert spaces considered satisfy assumptions @xmath68 and @xmath70 , then the approximation given by our algorithm strongly converges towards the desired result . one of the main advantage of the algorithm is that it can deal with highly nonlinear problems . we also proved that in finite dimension , this algorithm converges exponentially fast . we applied this algorithm in the context of uncertainty quantification on obstacle problems . in this frame , we considered regularizations of this kind of problems by penalization methods . indeed , the obstacle problem can be approximated by a global minimization problem defined on the entire hilbert space of some strongly convex energy functional where the constraints of the initial problem are replaced by penalization terms in the expression of the functional . our algorithm gives a good approximation of the solutions of the regularized problem . however , the problem of ill - conditioned matrices , which is inherent to penalization methods , limits the accuracy with which we can approach the solution of the initial obstacle problem . one way to circumvent this problem is to use augmented lagrangian methods ( see ref . @xcite ) instead of penalization methods . indeed , the former algorithms converge towards the true solution of the initial obstacle problems . the adaptation of our algorithm to such methods is work in progress . another extension of our work would be to consider other problems than obstacle problems . in ref . @xcite , a similar algorithm based on proper generalized decomposition is used to study uncertainty quantification upon a burger type equation . we believe that it could be possible to extend our proof of convergence in the case of such hyperbolic systems . we would like to thank the michelin company for financial support . a. ammar , b. mokdad , f. chinesta and r. keunings , a new family of solvers for some classes of multidimensional partial differential equations encountered in kinetic theory modeling of complex fluids , _ journal of non - newtonian fluid mechanics _ * 139 * ( 2002 ) 153176 . c. le bris , t. lelivre and y. maday , results and questions on a nonlinear approximation approach for solving high - dimensional partial differential equations , _ constructive approximation _ * 30 * ( 2009 ) 621651 . a. nouy , recent developments in spectral stochastic methods for the numerical solution of stochastic partial differential equations , _ archives of computational methods in engineering _ * 16 * ( 2009 ) 251285 .

mcduff and schlenk have recently determined exactly when a four - dimensional symplectic ellipsoid symplectically embeds into a symplectic ball . similarly , frenkel and mller have recently determined exactly when a symplectic ellipsoid symplectically embeds into a symplectic cube . symplectic embeddings of more complicated structures , however , remain mostly unexplored . we study when a symplectic ellipsoid symplectically embeds into a polydisc . we prove that there exists a constant depending only on ( here , is assumed greater than ) such that if is greater than , then the only obstruction to symplectically embedding into is the volume obstruction . we also conjecture exactly when an ellipsoid embeds into a scaling of for greater than or equal to , and conjecture about the set of such that the only obstruction to embedding into a scaling of is the classical volume . finally , we verify our conjecture for .

1409.2385

let @xmath13 and @xmath14 be symplectic manifolds . a _ symplectic embedding _ of @xmath13 into @xmath14 is a smooth embedding @xmath15 such that @xmath16 . it is interesting to ask when one symplectic manifold embeds into another . for example , define the ( open ) four - dimensional symplectic _ @xmath17 and define the ( open ) _ symplectic ball _ these inherit symplectic forms by restricting the standard form @xmath19 on @xmath20 in @xcite , mcduff and schlenk determined exactly when a four - dimensional symplectic ellipsoid @xmath0 embeds symplectically into a symplectic ball , and found that if @xmath21 is small , then the answer involves an infinite staircase " determined by the odd index fibonacci numbers , while if @xmath21 is large then all obstructions vanish except for the volume obstruction . to give another example , define the ( open ) four - dimensional _ polydisc _ @xmath22 where @xmath23 are real numbers and the symplectic form is again given by restricting the standard symplectic form on @xmath24 . frenkel and mller determined in @xcite exactly when a four - dimensional symplectic ellipsoid symplectically embeds into a _ cube _ @xmath25 and found that part of the expression involves the pell numbers . cristofaro - gardiner and kleinman @xcite studied embeddings of four - dimensional ellipsoids into scalings of @xmath26 and also found that part of the answer involves an infinite staircase determined by a recursive sequence . here we study symplectic embeddings of an open four - dimensional symplectic ellipsoid @xmath0 into an open four - dimensional symplectic polydisc @xmath1 . by scaling , we can encode this embedding question as the function @xmath27 where , @xmath28 and @xmath8 are real numbers that are both greater than or equal to @xmath29 . the function @xmath30 always has a lower bound , @xmath31 , the volume obstruction . our first theorem states that for fixed @xmath8 , if @xmath28 is sufficiently large then this lower bound is sharp , i.e. all embedding obstructions vanish aside from the volume obstruction : [ thm : rigidity ] if @xmath32 , then @xmath33 . this is an analogue of a result of buse - hind @xcite concerning symplectic embeddings of one symplectic ellipsoid into another . from the previously mentioned work of mcduff - schlenk , frenkel - mller , and cristofaro - gardiner - kleinman , one expects that if @xmath28 is small then the function @xmath30 should be more rich . our results suggest that this is indeed the case . for example , we completely determine the graph of @xmath34 ( see figure 1 ) : [ thm:13/2 ] for @xmath35 , @xmath36 and is equal to this lower bound for all @xmath28 except on the following intervals : \(i ) @xmath37 for all @xmath38 $ ] \(ii ) for @xmath39 , @xmath40 : @xmath41 , \\ \\ \dfrac{26 + 4k}{25 + 2k } & \quad a\in[13 + 2k,\beta_{k } ] , \end{array } \right.$ ] where @xmath42 , @xmath43 , @xmath44 , @xmath45 , @xmath46 , @xmath47 , @xmath48 , @xmath49 , @xmath50 , and @xmath51 . interestingly , the graph of @xmath52 has only finitely many nonsmooth points , in contrast to the infinite staircases in @xcite . this appears to be the case for many values of @xmath8 . for example , we conjecture what the function @xmath30 is for all @xmath53 see conjecture [ conj2 ] . our proofs rely on the following remarkable theorem of frenkel and mller @xcite . let @xmath54 be the sequence ( indexed starting at @xmath55 ) of all non - negative integer linear combinations of @xmath28 and @xmath8 , arranged with repetitions in non - decreasing order , and let @xmath56 be the sequence whose @xmath57 term is @xmath58 where @xmath59 . write @xmath60 if each term in the sequence @xmath54 is less than or equal to the corresponding term in @xmath61 . frenkel and mller show that embeddings of an ellipsoid into a polydisc are completely determined by the sequences @xmath62 and @xmath63 : ( frenkel - mller @xcite ) [ them:(frenkel - muller ) ] there is a symplectic embedding @xmath64 if and only if @xmath60 . to motivate the sequences @xmath62 and @xmath63 , note that @xmath63 is the sequence of _ ech capacities _ of the symplectic ellipsoid @xmath0 while @xmath62 is the sequence of ech capacities of the symplectic polydisc @xmath1 . the ech capacities are a sequence of nonnegative ( possibly infinite ) real numbers , defined for any symplectic four - manifold , that obstruct symplectic embeddings . we will not discuss ech capacities here ; see @xcite for a survey . theorem [ them:(frenkel - muller ) ] is equivalent to the statement that the ech capacities give sharp obstructions to embeddings of an ellipsoid into a polydisc . we wish to thank daniel cristofaro - gardiner for his helpful explanations , reference suggestions , encouragement and patience . we also thank the nsf , michael hutchings and uc berkeley for providing the opportunity to work on symplectic embedding problems this summer . we begin by describing the machinery that will be used to prove theorem [ thm : rigidity ] . let @xmath65 be a rational number . in @xcite , mcduff shows that there is a finite sequence of numbers @xmath66 called the _ ( normalized ) weight sequence for @xmath67 _ , such that @xmath68 embeds into a symplectic ellipsoid if and only if the disjoint union @xmath69 embeds into that ellipsoid . to describe the weight sequence , let @xmath70 where @xmath71 and @xmath72 . the @xmath73 are the multiplicities of the entries @xmath74 and come from the continued fraction expansion @xmath75 $ ] . the entries of [ equ : weightseq ] are defined as follows : @xmath76 . important results of the weight sequence include that @xmath77 and @xmath78 where for all @xmath79 , @xmath80 and @xmath81 . we will also make use of a helpful operation , @xmath82 , as in @xcite . suppose @xmath83 and @xmath84 are sequences indexed with @xmath85 , starting at 0 . then , @xmath86 . a useful application of @xmath82 is the following lemma : ( mcduff @xcite ) for all @xmath87 , we have @xmath88 . more generally , for all @xmath89 , we have @xmath90 . this lemma together with the weight sequence and scaling implies that @xmath91 similar to mcduff @xcite , this machinery allows us to reduce theorem [ thm : rigidity ] to a ball - packing problem . we begin by noting that the ech capacities for @xmath92 are @xmath93 . where the terms @xmath94 of this sequence are of the form @xmath95 and for each @xmath4 there are @xmath96 entries occurring at @xmath97 similarly , for the sequence @xmath98 , each term @xmath99 is of the form @xmath100 where @xmath101 by scaling and continuity , we can study @xmath102 with @xmath65 rational . so , we can prove that the volume obstruction is the only obstruction when @xmath103 by showing that @xmath104 for said @xmath28 values . by [ equ : n terms ] and [ equ : m terms ] , it is therefore sufficient to show that @xmath105 whenever @xmath106 are nonnegative integers such that @xmath107 we do so by considering the following cases : _ case 1 . _ @xmath108 . in this case , the cauchy - schwartz inequality along with [ equ : ai ] implies [ eq1 ] . _ @xmath109 . this case along with [ eq2 ] implies @xmath110 . so , we need @xmath111 . it follows that @xmath112 now let @xmath113 . we see that [ equ : m terms ] is equivalent to @xmath114 it is easy to see that @xmath115 for all such @xmath116 values . as such , we can apply @xmath113 to [ equ : initialbound ] to get @xmath117 hence the desired result . @xmath118 [ sharpbound ] we allow @xmath113 in the statement of theorem 1.4 . however , if we show @xmath115 for all @xmath119 , then we can use this @xmath4 in [ equ : initialbound ] to achieve a sharper bound for a. we begin by computing @xmath52 on the regions where it is linear . we first compute the values of @xmath4 at certain points . these will eventually be the points @xmath28 where @xmath52 is not differentiable . [ prop : criticalvalues ] we have : @xmath120 to prove the proposition , the main difficulty comes from the fact that that applying theorem [ them:(frenkel - muller ) ] in principle requires checking infinitely many ech capacities . our strategy for overcoming this difficulty is to study the growth rate of the terms in the sequences @xmath62 and @xmath63 . we will find that in every case needed to prove proposition [ prop : criticalvalues ] , one can bound these growth rates to conclude that only finitely many terms in the sequences need to be checked . this is then easily done by computer . the details are as follows : _ step 1 . _ for the sequence @xmath54 , let @xmath121 be the largest @xmath116 such that @xmath122 . similarly , for the sequence @xmath61 , let @xmath123 be the largest @xmath124 such that @xmath125 . to show that @xmath126 , by theorem [ them:(frenkel - muller ) ] , we just have to show that for all @xmath127 , we have @xmath128 . _ we can estimate @xmath121 by applying the following proposition : [ thm : ehrhart ] if @xmath129 and @xmath127 are all positive integers , then @xmath130 @xmath131 where @xmath132 . the number of terms in @xmath133 that are less than @xmath127 is the same as the number of lattice points @xmath134 in the triangle bounded by the positive @xmath135 and @xmath136 axes and the line @xmath137 . for integral @xmath127 , this number can be computed by applying the theory of ehrhart polynomials " . proposition [ thm : ehrhart ] follows by applying ( * ? ? ? . we will be most interested in this proposition in the case where @xmath138 . note that by the last two terms of the formula in proposition [ thm : ehrhart ] , @xmath139 is a periodic polynomial with period @xmath140 . we also need an argument to account for the fact that proposition [ thm : ehrhart ] is only for integer @xmath127 , whereas the argument in step @xmath29 involves real @xmath127 . to account for this we use an asymptotic argument . specifically , for @xmath141 , @xmath142 we bound the right hand side of from below by taking the floor function of @xmath127 . it is convenient for our argument to further bound this expression from below by @xmath143 where the @xmath144 are the coefficients of the right hand side of that do not involve @xmath127 or @xmath145 . this is the lower bound that we will use for @xmath146 . _ to get an upper bound @xmath123 for @xmath61 , recall that @xmath147 . for @xmath148 , we solve for @xmath149 in terms of @xmath150 and find : @xmath151 . considering @xmath152 , we can take the derivative of the left side of the inequality with respect to @xmath150 and then set the expression equal to 0 to maximize it . we do the same with @xmath149 to obtain : @xmath153 . by simplifying , we get that an upper bound for @xmath124 is : @xmath154 our strategy now is to check for each point in proposition [ prop : criticalvalues ] that we have @xmath155 asymptotically in @xmath127 for the corresponding @xmath156 . from there , we can check that for a sufficient number of terms , @xmath157 . _ since the rest of the proof amounts to computation , it is best summarized by the chart below . in the chart , @xmath158 and @xmath159 denote the coefficients of the quadratic terms in the upper and lower bounds from steps @xmath160 and @xmath161 , while @xmath162 and @xmath163 denote the corresponding coefficients of the linear terms . the @xmath127 column gives a sufficient number to check up to before the asymptotic bounds from the previous three steps are enough . note that if the @xmath164 coefficients in any row are equal , then linear coefficients are used to make an asymptotic argument ; this explains the appearance of the n / a"s in the table . it is simple to check by computer that the relevant @xmath63 and @xmath62 sequences in each row satisfy @xmath165 once one knows that the problem only has to be checked up to the @xmath127 in the @xmath127 column . code for this is contained in [ app : embedcode ] . the ech obstruction column gives an ech capacity that shows that one can not shrink @xmath166 further , i.e. the claimed embeddings are actually sharp . [ cols="^,^,^,^,^,^,^",options="header " , ] table 3.2 given @xmath167 $ ] for each integer @xmath168 $ ] , we have found that the upper and lower bounds of @xmath52 equal @xmath169 . thus , we have proven our claim for these intervals . to complete the proof of theorem [ thm:13/2 ] , we need to show that aside from the linear steps described in the previous section , the graph of @xmath52 is equal to the graph of the volume obstruction . to do this , we adapt some of the ideas from @xcite in a purely combinatorial way . this will be needed to complete the proof of theorem [ thm:13/2 ] . our combinatorial perspective on the techniques from @xcite borrows many ideas from @xcite . this section collects the main combinatorial machinery that will be used to complete the proof . the basic idea behind our proof will be to reduce to a ball packing problem , as in the proof of theorem [ thm : rigidity ] . the machinery we develop here will be useful for approaching this ball packing problem . we begin with two definitions : let @xmath170 where @xmath171 for @xmath172 and @xmath173 for all @xmath174 we say @xmath175 is the _ cremona transform_. we say @xmath176 is : \(i ) @xmath177 if @xmath178 for all @xmath79 , \(ii ) @xmath179 if @xmath180 implies @xmath181 and @xmath182 implies @xmath183 , \(iii ) @xmath184 if positive , ordered , and @xmath185 . it will be important to note that @xmath186 . we now define a product analogous to the intersection product in @xcite : @xmath187 we also define a vector @xmath188 that is motivated by the the standard anti - canonical divisor in the @xmath62-fold blow up of @xmath189 . @xmath190 the following is a combinatorial analogue of intersection positivity " that will be useful : [ 1 ] if @xmath191 is reduced , @xmath192 is positive , @xmath193 , and @xmath194 then @xmath195 . let @xmath196 be ordered @xmath192 . as @xmath197 we can assume without loss of generality that @xmath192 is ordered . if @xmath198 then @xmath199 for @xmath200 and @xmath201 as @xmath202 . we know that this expression is greater than or equal to @xmath203 as @xmath191 is reduced , and this is greater than or equal to @xmath55 . we now assume without loss of generality that @xmath204 . hence , @xmath205 for @xmath206 . let @xmath207 then @xmath208 as @xmath191 is reduced . this expression is equal to @xmath209 as @xmath210 . we now have the following chain of inequalities : @xmath211 @xmath212 @xmath213 @xmath214 @xmath215 in @xcite , cremona transformations preserve the intersection product . here we prove an analogous result . [ 2]@xmath216 @xmath217 @xmath218 @xmath219 @xmath220 the following three sets will also be useful : @xmath221 @xmath222 @xmath223 also observe : @xmath224 and @xmath225 . additionally , @xmath226 are invariant under permutations of @xmath227 . @xmath228 moreover , also define : [ c ] let @xmath2 be the set of @xmath191 such that @xmath229 and a. @xmath230 b. @xmath231 both li - li @xcite and mcduff - schlenk @xcite have found that compositions of cremona transformations and permutations can reduce certain classes . here we prove a combinatorial version of those lemmas . [ 3 ] if @xmath232 then by a sequence of cremona transforms and permutations of @xmath233 we can transform @xmath191 to @xmath234 where @xmath234 is reduced . we begin with some helpful results : [ .1 ] @xmath235 the fact that cr preserves a ) follows from the fact that cr preserves the intersection product . to complete the sublemma , note that if @xmath236 , then @xmath237 @xmath238 [ .2 ] if @xmath239 is some permutation , @xmath240 if @xmath236 , then @xmath241 [ .3 ] if @xmath242 , then @xmath243 if @xmath244 then @xmath245 for all @xmath246 . so , @xmath247 we also have @xmath248 as @xmath249 this implies that @xmath250 let @xmath251 denote the transformation @xmath175 followed by ordering the @xmath252 . fix @xmath253 let @xmath254 let @xmath255 it suffices to show @xmath256 for some @xmath257 assume not . then @xmath258 for all @xmath116 . by sublemmas [ .1 ] and [ .2 ] , @xmath259 . for @xmath260 @xmath261 thus , there exists @xmath116 such that @xmath262 . this contradicts sublemma [ .3 ] completing the proof that we may reduce @xmath191 we now prove a result analogous to ( * ? ? * proposition 1.2.12(i ) ) . [ 4 ] if @xmath263 by lemma [ 3 ] there exists @xmath264 , a composition of @xmath175 and permutations , such that @xmath265 reduced . for @xmath266 . so , @xmath267 let @xmath268 and @xmath269 . we note @xmath270 @xmath271 if @xmath272 then cauchy - schwarz shows @xmath273 otherwise , @xmath274 . then @xmath275 implies @xmath276 , so lemma [ 1 ] shows @xmath273 this completes the proof . by scaling , lemma [ 4 ] extends to @xmath191 that satisfy a ) and b ) of definition [ c ] with @xmath277 . we now use the combinatorial machinery from the previous section , together with a reduction to the ball packing problem , to prove the key lemma needed to complete the proof of theorem [ thm:13/2 ] , see part ( iii ) of lemma [ lem : keylemma ] below . to reduce to a ball packing problem , note that proposition 1.4 in frenkel - mller @xcite states that for rational @xmath28 , @xmath278 if and only if @xmath279 where @xmath280 denotes disjoint union . since , as explained in @xcite , one can compute the ech capacities of the disjoint union in terms of the @xmath82 operation , we know that the embedding in exists if and only if @xmath281 for the rest of the proof of theorem [ thm:13/2 ] , we are looking at intervals for @xmath28 on which the graph of @xmath4 is equal to the volume obstruction ; we therefore want to show that holds with @xmath282 ( of course for our proof one can specify @xmath283 , but we state things here in slightly greater generality ) . by an argument analogous to the argument used in the proof of theorem [ thm : rigidity ] it is sufficient to show @xmath284 implies @xmath285 for all @xmath286 non - negative integers let @xmath287 and @xmath288 for @xmath289 and let @xmath290 and @xmath291 for @xmath289 . hence , it is enough to show @xmath292 implies @xmath293 let @xmath294 . the previous condition is equivalent to @xmath295 . by lemma [ 4 ] it is sufficient to check the case @xmath296@xmath297 let @xmath298 be the set of @xmath299 satisfying and with @xmath300 non - negative integers . define @xmath301 by @xmath302 . we now have a series of lemmas , culminating in the key lemma [ lem : keylemma ] . [ 21 ] for @xmath303 a. @xmath304 b. @xmath305 if and only if @xmath306 c. @xmath307 implies @xmath308 d. let @xmath309 . @xmath310 @xmath311 \(i ) follows from @xmath312 and cauchy - schwarz . to prove ( ii ) note @xmath313 @xmath314 @xmath315 . to prove ( iii ) note @xmath316 @xmath317 @xmath318 if @xmath319 . to prove ( iv ) note @xmath320 @xmath321 @xmath322 . let @xmath303 and suppose that @xmath323 is the maximal nonempty open interval such that @xmath319 for all @xmath324.then there exists unique @xmath325 such that @xmath326 where @xmath327 is the length of @xmath328 and @xmath329 is the number of nonzero terms in @xmath149 . furthermore , @xmath330 for all @xmath324 . we adapt the proof of lemma 2.1.3 in @xcite . for @xmath331 is piecewise linear and is linear on open intervals that do not contain an element @xmath332 with length @xmath333 . therefore , if @xmath334 for all @xmath335 is linear on i. this is impossible as @xmath336 is concave and @xmath323 is bounded . thus there exists @xmath325 with @xmath337 . if @xmath338 then @xmath339 which implies @xmath340 which is impossible for @xmath324 . the proof of uniqueness is the same as in ( * ? ? ? 2.1.3 ) . [ lem : p ] let @xmath299 be in @xmath298 with @xmath319 for some @xmath28 . let @xmath341 be a block of @xmath342 consecutive integers such that @xmath328 is constant for @xmath343 . then : a. one of the following holds : @xmath344 @xmath345 @xmath346 b. there is at most one block of length @xmath342 on which the @xmath347 are not all equal . c. if there is a block @xmath348 of length @xmath342 on which the @xmath347 are not all equal then @xmath349 . see the proof of ( * ? ? ? 2.1.7 ) . here , mcduff and schlenk are considering the case of an ellipsoid into a ball , but their proof generalizes without change to our situation . [ 10 ] let @xmath303 be such that @xmath350 for some @xmath28 with @xmath351 . let @xmath352 be a block but not the first block of @xmath353 [ the first two terms of w(a ) are not considered to be part of any block ] . if this block is not the last block , then @xmath354 if this block is the last block , then @xmath355 b. always , @xmath356 similar to the proof of lemma [ lem : p ] , see the proof of ( * ? ? ? 2.1.8 ) where mcduff and schlenk s proof generalizes without change to our situation . [ lem : keylemma][22 ] assume that @xmath303 and @xmath319 for some @xmath28 with @xmath357 . assume further that @xmath358 . let @xmath359 and let @xmath360 . then : a. @xmath361 b. @xmath362 c. let @xmath363 . then @xmath364 and @xmath365 \(i ) follows from @xmath366 . ( ii ) follows from the same argument as ( * ? ? ? 5.1.2 ) . from ( * sublemma . 5.1.1 ) @xmath367 so lemma [ 21 ] implies @xmath368 this also shows @xmath369 with the lemma [ lem : keylemma ] now shown , we can complete the proof of theorem [ thm:13/2 ] . we explain the computation on various intervals seperately . we now wish to prove that @xmath371 for @xmath372 $ ] . previously , we have proved @xmath373 if @xmath374 is not @xmath375 on the interval @xmath376 $ ] , then there exists @xmath303 such that @xmath377 for some @xmath372 $ ] . so , lemma [ 22 ] shows that there exists @xmath378 in @xmath376 $ ] with @xmath379 and @xmath326 . let @xmath380 . as @xmath381 we know @xmath382 . for @xmath383 , q\geq5 $ ] we know @xmath384 thus , lemma [ 22 ] shows @xmath385 hence , @xmath386 . we also note that for @xmath387 we have @xmath388 and @xmath389 . thus , lemma [ 22 ] shows @xmath390 . using our code ( see appendix a.2 ) we can reduce the possibilities for @xmath299 to 38 candidates . we can then use lemma [ 10 ] to reduce these 38 cases to 11 possible candidates which can easily verified to not be obstructive by simple calculations . we now will show @xmath371 for @xmath392 $ ] . previously , we have proved @xmath393 if @xmath374 is not @xmath375 on the interval @xmath394 $ ] , then there exists @xmath303 such that @xmath377 for some @xmath392 $ ] . then lemma [ 22 ] shows that there exists @xmath395 $ ] with @xmath396 and @xmath326 . let @xmath397 with @xmath398 . for @xmath395 $ ] we know @xmath399 thus , lemma [ 22 ] shows @xmath400 . hence , @xmath401 . we can then verify these cases directly using our code ( see appendix a.2 ) to check these cases and we find no obstructions . we will now show @xmath371 for @xmath403 $ ] . previously , we have proved @xmath404 if @xmath374 is not @xmath375 on the interval @xmath405 $ ] , then there exists @xmath303 such that @xmath377 for some @xmath403 $ ] . then lemma [ 22 ] shows that there exists @xmath406 $ ] with @xmath396 and @xmath326 . let @xmath397 with @xmath398 . for @xmath406 $ ] we know @xmath407 thus , lemma [ 22 ] shows @xmath408 . hence , @xmath409 . we can then verify these cases directly using our code ( see appendix a.2 ) to check these cases and we find no obstructions . for @xmath411 , \sqrt{q+29}\geq\sqrt{q+{\left\lfloor a \right\rfloor}+2 } \text { and } \delta\geq21 - 15\sqrt{\frac{21}{13}}.$ ] hence , lemma [ 22 ] implies @xmath412 which implies @xmath413 . we can then verify these cases directly using our code ( see appendix a.2 ) to check these cases and we find no obstructions . we will apply remark [ sharpbound ] . as @xmath415 remark [ sharpbound ] implies we only need to verify @xmath416 for all @xmath417 for @xmath418 this completes the proof @xmath419 we now present some conjectures concerning exactly when an ellipsoid embeds into a polydisc . to consider an interesting refinement of theorem 1.1 , define @xmath420 . theorem 1.1 implies @xmath421 . for @xmath422 @xmath423 @xmath424 @xmath425 @xmath426 this implies @xmath423 experimental evidence seems to suggest that for @xmath427 this bound is sharp . [ conj1 ] for @xmath427 @xmath428 the methods used to compute the graph of @xmath374 should extend for the most part to any b. in light of those techniques , experimental evidence , and a conjecture regarding @xmath30 for b an integer by david frenkel and felix schlenk relayed to us by daniel cristofaro - gardiner , we offer a conjecture regarding the graph of @xmath30 for @xmath429 , see figure @xmath160 . [ conj2 ] for @xmath430 except on the following intervals @xmath431\ ] ] for @xmath432 @xmath433\ ] ] @xmath434\ ] ] where @xmath435 + for integers m if @xmath436 $ ] let @xmath437 @xmath438\ ] ] @xmath439\ ] ] where @xmath440 and @xmath441 . we note that conjecture [ conj2 ] implies conjecture [ conj1 ] for @xmath429 . furthermore , we prove that the conjecture is a lower bound for @xmath30 . for @xmath442 and @xmath443\ ] ] for @xmath432 @xmath444\ ] ] @xmath445\ ] ] where @xmath446 + for integers m if @xmath436 $ ] let @xmath437 @xmath447\ ] ] @xmath448\ ] ] where @xmath440 and @xmath449 we know that @xmath450 because symplectic embeddings are volume preserving . we also have @xmath451 additionally , for @xmath452 $ ] @xmath453 @xmath454 , k=0\ ] ] @xmath455 , k=1\ ] ] @xmath456 , k\geq2.\ ] ] we also have for @xmath457 @xmath458 @xmath459.\ ] ] furthermore , if @xmath460 $ ] for some @xmath461 $ ] @xmath462 @xmath463.\ ] ] we also have for @xmath464 $ ] @xmath465 @xmath466.\ ] ] this completes the proof . .... function m = embed(b , c , d , x ) l = length(nembed(x , b ) ) ; y = zeros(1,l ) ; w = nembed(x , b ) ; t = membed(l , c , d ) ; for i=1:l if w(i)<=(t(i)+10 ^ -10 ) y(i)=1 ; else m=0 ; break end end m = min(y ) ; function y = nembed(x , b ) ; y = zeros(1,x+2 ) ; for i=1:x+2 y(i)=floor((x+b-(i-1))/b ) ; end m = sum(y ) ; z = zeros(1,(x+1)^2 ) ; for i=1:x+1 for j=1:x+1 z(i+(x+1)*(j-1))=i-1+(j-1)*b ; end end l = sort(z ) ; y = zeros(1,m ) ; for i=1:m y(i)=l(i ) ; end function q = membed(n , c , d ) q = zeros(1,n ) ; for k=0:n-1 w = zeros(1,ceil(sqrt(k+1)-1 ) ) ; for i=0:(ceil(sqrt(k+1)-1 ) ) w(i+1)=c*i+d*(ceil((k+1)/(i+1))-1 ) ; end q(k+1)=min(w ) ; end .... + @xmath470\text{:=}\text{module}[\{\text{aa}=a , m , i=2,l , u , v\}},\\ \indent { m=\text{continuedfraction}[\text{aa}];}\\ \indent { l=\text{table}[1,\{j , m[[1]]\}];}\\ \indent { \{u , v\}=\{1,\text{aa}-\text{floor}[\text{aa}]\};}\\ \indent{\text{while}[i\leq \text{length}[m],l=\text{join}[l,\text{table}[v,\{j , m[[i]]\}]];}\\ \indent \indent{\{u , v\}=\{v , u - m[[i ] ] v\};}\\ \indent \indent{i\text{++}];}\\ \indent{\text{return}[l]]}\\ $ { \text{genw[a\ _ ] outputs } $ ( c\lambda,\lambda)$ \text { joined with the weight sequence of a.}}\\ { \text{genw}[\text{a$\_$},\text{c$\_$ } ] \text{:=}\text{join}[\{c*\text{sqrt}[a/(2*c)],\text{sqrt}[a/(2*c)]\},w[a]]}\\ { \text{p2}[\text{k$\_$}]\text{:=}\text{module}[\{\text{kk}=k,\text{pp},\text{t0},i\},\text{t0}=\text{table}[0,\{u,1,k\}];}\\ \indent{\text{t0p}=\text{replacepart}[\text{t0},1,1];}\\ \indent{\text{t11}=\text{table}[1,\{u,1,k\}];}\\ \indent{\text{t1m}=\text{replacepart}[\text{t11},0,-1];}\\ \indent{\text{pp}=\{\text{t0},\text{t0p},\text{t11},\text{t1m}\};}\\ \indent{\text{return}[\text{pp}]]}\\ { \text{difference}[\text{m$\_$}]\text{:=}\text{module}[\{v = m,\text{vn},\text{v1},l , l=\{\},d,\text{pp},i , j , n\},l=\text{length}[v];}\\ \indent{\text{if}[l==1,l=\text{p2}[v[[1]]]];}\\ \indent{\text{if}[l>1,\text{vn}=v[[-1]];}\\ \indent \indent{\text{v1}=\text{delete}[v,-1];}\\ \indent \indent{d=\text{difference}[\text{v1}];}\\ \indent \indent{\text{pp}=\text{p2}[\text{vn}];}\\ \indent \indent{i=1;}\\ \indent \indent${\text{while}[i\leq\text{length}[d],j=1;}\\ \indent \indent\indent{\text{while}[j\leq\text{length}[\text{pp}],}$\\ \indent \indent\indent\indent{n=\text{join } [ d[[i]],\text{pp}[[j]]];}\\ \indent\indent\indent\indent{l=\text{append}[l , n];}\\ \indent\indent\indent\indent{j\text{++}];}\\ \indent\indent\indent{i\text{++}]}\\ \indent{];}\\ \indent{\text{return}[l]]}\\ { \text{sol}[\text{a$\_$},\text{d$\_$},\text{c$\_$}]\text{:=}\text{module}[\{\text{aa}=a,\text{dd}=d,\text{cc}=c , a , m , f , d , i , v , l=\{\}\},}\\ \indent{a=\text{continuedfraction}[\text{aa}];}\\ \indent{m=\text{join}[\{1,1\},a];}\\ \indent{f=\text{floor}[\text{dd}/((1+\text{cc})*\text{sqrt}[\text{aa}/(2*\text{cc } ) ] ) \text{genw}[\text{aa},\text{cc}]];}\\ \indent{d=\text{difference}[m];}\\ \indent{i=1;}\\ \indent${\text{while}[i\leq \text{length}[d],v=\text{sort}[f+d[[i]],\text{greater}];}$\\ \indent\indent{\text{sv}=\text{sum}[v[[j]],\{j,1,\text{length}[v]\}];}\\ \indent\indent${\text{if}[\{\text{sv},v.v\}==\{3*\text{dd}-1,\text{dd}{}^{\wedge}2 + 1\}\&\&v[[-1]]>0\&\&}\\ \indent\indent\indent{\text{genw}[\text{aa},\text{cc}].v/\text{dd}\geq ( ( 1+\text{cc})*\text{sqrt}[\text{aa}/(2*\text{cc})]),l=\text{append}[l , v]];}\\ \indent\indent{i\text{++}];}\\ \indent{\text{return}[\{\text{dd},\text{union}[l]\}]]}\\ { \text{solless}[\text{a$\_$},\text{d$\_$},\text{c$\_$}]\text{:= } } { \text{module}[\{\text{aa}=a,\text{dd}=d,\text{cc}=c , d=1,\text{ld},l=\{\}\},}\\ \indent{\text{while}[d\leq d,\text{ld}=\text{sol}[\text{aa},d,\text{cc}];}\\ \indent\indent{\text{if}[\text{length}[\text{ld}[[2]]]>0,l=\text{append}[l,\text{sol}[\text{aa},d,\text{cc}]]];}\\ \indent\indent{d\text{++}];}\\ \indent{\text{return}[l]]}\\ { \text{sol2}[\text{a$\_$},\text{d$\_$},\text{c$\_$}]\text{:=}\text{module}[\{\text{aa}=a,\text{dd}=d,\text{cc}=c , a , m , f , d , i , v , l=\{\}\},}\\ \indent{a=\text{continuedfraction}[\text{aa}];}\\ \indent{m=\text{join}[\{1,1\},a];}\\ \indent{f=\text{floor}[\text{dd}/((1+\text{cc})*\text{sqrt}[\text{aa}/(2*\text{cc } ) ] ) \text{genw}[\text{aa},\text{cc}]];}\\ \indent{d=\text{difference}[m];}\\ \indent{i=1;}\\ \indent{\text{while}[i\leq \text{length}[d],v=\text{sort}[f+d[[i]],\text{greater}];}\\ \indent\indent{\text{sv}=\text{sum}[v[[j]],\{j,1,\text{length}[v]\}];}\\ \indent\indent{\text{if}[\{\text{sv},v.v\}==\{3*\text{dd}-1,\text{dd}{}^{\wedge}2 + 1\}\&\&}\\ \indent\indent\indent{\text{genw}[\text{aa},\text{cc}].v/\text{dd}\geq ( ( 1+\text{cc})*\text{sqrt}[\text{aa}/(2*\text{cc})]+10{}^{\wedge}-10),}\\ \indent\indent{l=\text{append}[l , v]];}\\ \indent{i\text{++}];}\\ \indent{\text{return}[\{\text{dd},\text{union}[l]\}]]}\\ { \text{solless2}[\text{a$\_$},\text{d$\_$},\text{c$\_$}]\text{:= } } { \text{module}[\{\text{aa}=a,\text{dd}=d,\text{cc}=c , d=1,\text{ld},l=\{\}\},}\\ \indent{\text{while}[d\leq d,\text{ld}=\text{sol2}[\text{aa},d,\text{cc}];}\\ \indent\indent{\text{if}[\text{length}[\text{ld}[[2]]]>0,l=\text{append}[l,\text{sol2}[\text{aa},d,\text{cc}]]];}\\ \indent\indent{d\text{++}];}\\ \indent{\text{return}[l]]}\\ { \text{solutions}[\text{a$\_$},\text{b$\_$}]\text{:=}\text{solutions}[a , b,\text{min}[a,\text{floor}[\text{sqrt}[b]]]]}\\ { \text{solutions}[\text{a$\_$},\text{b$\_$},\text{c$\_$}]\text{:= } } { \text{module}[\{a = a , b = b , c = c , i , m , k , j , v , l=\{\}\},\text{if}[a{}^{\wedge}2<b , l=\{\}];}\\ \indent{\text{if}[a{}^{\wedge}2==b,}\\ \indent\text{if}[a > c , l=\{\},l=\{\{a\}\}]];\\ \indent{\text{if}[a{}^{\wedge}2>b , i=1;}\\ \indent\indent{m=\text{min}[\text{floor}[\text{sqrt}[b]],c];}\\ \indent\indent{\text{while}[i\leq m , k=\text{solutions}[a - i , b - i{}^{\wedge}2,i];}\\ \indent\indent\indent{j=1;}\\ \indent\indent\indent{\text{while}[j\leq \text{length}[k],v=\text{prepend}[k[[j]],i];}\\ \indent\indent\indent\indent{l=\text{append}[l , v];}\\ \indent\indent\indent{j\text{++}];}\\ \indent\indent{i\text{++}]];}\\ \indent{\text{return}[\text{union}[l]]]}\\ { t \text{:=}\text{table}[0,\{u,1,18\}]}\\ { \text{t1 } \text{:=}\text{replacepart}[t,1,1]}\\ { \text{t2 } \text{:= } \text{replacepart}[t,1,2]}\\ { \text{t3 } \text{:=}\text{replacepart}[t,1,3]}\\ { \text{t4}\text{:= } \text{replacepart}[t,-1,18]}\\ { p=\{t,\text{t3},\text{t4},\text{t1},\text{t1}+\text{t3},\text{t1}+\text{t4},\text{t2},\text{t2}+\text{t3},\text{t2}+\text{t4},\text{t1}+\text{t2},}\\ { \text{t1}+\text{t2}+\text{t3},\text{t1}+\text{t2}+\text{t4}\}}\\ { \text{a1}[\text{d$\_$}]\text{:=}\text{floor}[d*13/15]}\\ { \text{a2}[\text{d$\_$ } ] \text{:= } \text{floor}[d*2/15]}\\ { \text{a3}[\text{d$\_$ } ] \text{:= } \text{round}[d*2/15/\text{sqrt}[16.1/13]]}\\ { q[\text{d$\_$}]\text{:=}\text{table}[\text{a3}[d],\{u,1,18\}]}\\ { \text{q1}[\text{d$\_$ } ] \text{:=}\text{replacepart}[q[d],\text{a1}[d],1]}\\ { \text{q2}[\text{d$\_$ } ] \text{:= } \text{replacepart}[\text{q1}[d],\text{a2}[d],2]}\\ { \text{sum}[\text{l$\_$}]\text{:=}\text{sum}[l[[j]],\{j,1,\text{length}[l]\}]}\\ { \text{squaresum}[\text{l$\_$ } ] \text{:=}\text{sum}[(l[[j]]){}^{\wedge}2,\{j,1,\text{length}[l]\}]}\\ { \text{solutions2}[\text{d$\_$ } ] \text{:=}\text{module}[\{\text{dd}=d , i , k\},}\\ \indent{l=\{\};}\\ \indent{\text{for}[i = 1,i<13,i\text{++},}\\ \indent\indent{k=\text{solutionsalt}[3*\text{dd}-1-\text{sum}[\text{q2}[\text{dd}]+p[[i]]],}\\ \indent\indent\indent{\text{dd}{}^{\wedge}2 + 1-\text{squaresum}[\text{q2}[\text{dd}]+p[[i]]],\text{q2}[\text{dd}]+p[[i]]];}\\ \indent\indent{l=\text{join}[l , k];}\\ \indent\indent{]}\\ \indent{]}\\ { \text{solverify}[\text{d$\_$ } ] \text{:=}\text{module}[\{k=\text{length}[\text{sol}[d]],t=\text{sol}[d],p=\{\}\},}\\ \indent{\text{if}[k\text{==}0,p = p,\text{if}[\text{length}[t[[1]]]>1,}\\ \indent\indent { \text{for}[i=1,i\leq k , i\text{++},}\\ \indent\indent\indent{\text{if}[t[[i]][[2]]\geq t[[i]][[3 ] ] } \\ \indent\indent\indent\indent{\&\&\text{if}[\text{length}[t[[i]]]\geq 19,t[[i]][[18]]\geq t[[i]][[19]],\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t[[i]]]\text{$>$=}20,}\\ \indent\indent\indent\indent{\text{length}[t[[i]]]\geq 22 \&\&t[[i]][[19]]\text{$<$=}((t[[i]][[20]])+1),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t[[i]]]\text{$>$=}21,t[[i]][[20]]\text{==}((t[[i]][[21]])),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t[[i]]]\text{$>$=}22,t[[i]][[20]]\text{==}((t[[i]][[22]])),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t[[i]]]\text{$>$=}23,t[[i]][[22]]\text{$<$=}((t[[i]][[23]])+1),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{test4}[t[[i]]],}\\ \indent\indent\indent\indent{p=\text{append}[p , t[[i]]]]],}\\ \indent\indent\indent{\text{if}[t[[2]]\geq t[[3 ] ] } \\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t]\geq 19 , t[[18]]\geq t[[19 ] ] , \text{true}]}\\ \indent\indent\indent\indent{\&\&\text{if}[\text{length}[t]\geq 20 , \text{length}[t]\geq 22 \&\&t[[19]]\leq ( t[[20]]+1),\text{true}]}\\ \indent\indent\indent\indent{\&\&\text{if}[\text{length}[t]\text{$>$=}21,t[[20]]\text{==}((t[[21]])),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t]\text{$>$=}22,t[[20]]\text{==}((t[[22]])),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{if}[\text{length}[t]\text{$>$=}23,t[[22]]\text{$<$=}((t[[23]])+1),\text{true}]}\\ \indent\indent\indent\indent{\&\ & \text{test4}[t],}\\ \indent\indent\indent\indent{p=\text{append}[p , t]]]];}\\ \indent\indent\indent\indent{\text{return}[p]]}\\ { \text{allsol}[\text{d$\_$ } ] \text{:=}\text{module}[\{\text{dd}=d , p=\{\}\},}\\ \indent{\text{for}[j=2,j\leq \text{dd},j\text{++},}\\ \indent\indent{\text{print}[\text{join}[\{j\},\text{solverify}[j]]]]]}\\ { \text{sol}[\text{d$\_$}]\text{:=}\text{module}[\{e = d\},}\\ \indent{\text{solutions2}[e];}\\ \indent{l]}\\ { \text{alpha}[\text{bb$\_$},\text{kk$\_$}]\text{:=}\text{module}[\{b=\text{bb},k=\text{kk}\},}\\ \indent{\text{if}[k==0,2*b-1/2,}\\ \indent\indent{\left.\left.\text{if}\left[k==1 , \frac{2 b ( 1 + 4 b)}{-1 + 4 b},\frac{(-1 + 4 b+2 k)^2}{8 b}\right]\right]\right]}\\ { \text{beta}[\text{bb$\_$},\text{kk$\_$}]\text{:=}\text{module}[\{b=\text{bb},k=\text{kk}\},}\\ \indent{\left.\text{if}\left[k==0,\frac{2 b ( 1 + 4 b)}{-1 + 4 b},\frac{32 \left(b^3 + 2 b^2 k+b k^2\right)}{(-1 + 4 b+2 k)^2}\right]\right]}\\ { \text{solutionsalt}[\text{a$\_$},\text{b$\_$},\text{v$\_$}]\text{:=}}\\ \indent{\text{if}[a\geq 0\&\&b\text{$>$=}0,\text{solutionsalt}[a , b,\text{min}[a,\text{floor}[\text{sqrt}[\text{max}[0,b]]]],v],\{\}]}\\ { \text{solutionsalt}[\text{a$\_$},\text{b$\_$},\text{c$\_$},\text{v$\_$}]\text{:=}}\\ \indent{\text{module}[\{a = a , b=\text{max}[0,b],c = c , i , m , k , j , v,\text{vv}=v , l=\{\}\},}\\ \indent\indent{\text{if}[a{}^{\wedge}2<b , l=\{\}];}\\ \indent\indent{\text{if}[a{}^{\wedge}2==b,\text{if}[a > c \| a==0,l=\{\},l=\{\text{join}[\text{vv},\{a\}]\}]];}\\ \indent\indent{\text{if}[a{}^{\wedge}2>b , i=1;}\\ \indent\indent\indent{m=\text{min}[\text{floor}[\text{sqrt}[b]],c];}\\ \indent\indent\indent{\text{while}[i\leq m , k=\text{solutions}[a - i , b - i{}^{\wedge}2,i];}\\ \indent\indent\indent\indent{j=1;}\\ \indent\indent\indent\indent{\text{while}[j\leq \text{length}[k],v=\text{prepend}[k[[j]],i];}\\ \indent\indent\indent\indent\indent{\quad v=\text{join}[\text{vv},v];}\\ \indent\indent\indent\indent{l=\text{append}[l , v];}\\ \indent\indent\indent\indent{j\text{++}];}\\ \indent\indent\indent{i\text{++}]];}\\ \indent{\text{return}[l]]}\\ { \text{test4}[\text{v$\_$}]\text{:=}\text{module}[\{\text{vv}=v , k=0,t = v[[18]]\},}\\ \indent{\text{while}[\text{length}[\text{vv}]\geq ( 19+k ) \&\ & ( \text{vv}[[19+k]]+1)\geq \text{vv}[[19]],}\\ \indent\indent{t=(t-\text{vv}[[19+k]]);k\text{++}];}\\ \indent{t<\text{sqrt}[k+2]]}\\ { \text{verify2}\text{:=}\text{for}[\text{qq}=1,\text{qq}\leq 11,\text{qq}\text{++},}\\ \indent{\text{for}[\text{pp}=1,\text{pp}<2*\text{qq},\text{pp}\text{++},}\\ \indent\indent{\text{if}[\text{gcd}[\text{pp},\text{qq}]==1 \&\ & 15028/841\text{$<$=}17+\text{pp}/\text{qq}\leq 961/52,}\\ \indent\indent\indent{\text{print}[\{17+\text{pp}/\text{qq},\text{solless}[17+\text{pp}/\text{qq},}\\ \indent\indent\indent\indent{15/2*\text{sqrt}[(17+\text{pp}/\text{qq})/13]/}\\ \indent\indent\indent\indent\indent{(17+(\text{pp}-1)/\text{qq}+1 - 15*\text{sqrt}[(17+\text{pp}/\text{qq})/13])*}\\ \indent\indent\indent\indent(\text{sqrt}[\text{qq}+20]-1 ) , { 13/2]\}]]]]}\\ { \text{verify3}\text{:=}\text{for}[\text{qq}=1,\text{qq}\leq 6,\text{qq}\text{++},}\\ \indent{\text{for}[\text{pp}=1,\text{pp}<2*\text{qq},\text{pp}\text{++},}\\ \indent\indent{\text{if}[\text{gcd}[\text{pp},\text{qq}]==1 \&\ & 18772/961\text{$<$=}19+\text{pp}/\text{qq}\leq 1089/52,}\\ \indent\indent\indent{\text{print}[\{19+\text{pp}/\text{qq},\text{solless}[19+\text{pp}/\text{qq},}\\ \indent\indent\indent\indent{15/2*\text{sqrt}[(19+\text{pp}/\text{qq})/13]/}\\ \indent\indent\indent\indent\indent{(19+(\text{pp}-1)/\text{qq}+1 - 15*\text{sqrt}[(19+\text{pp}/\text{qq})/13])*}\\ \indent\indent\indent\indent(\text{sqrt}[\text{qq}+22]-1 ) , { 13/2]\}]]]]}\\ { \text{verify4}\text{:=}\text{for } [ \text{qq}=1,\text{qq}<8,\text{qq}\text{++},}\\ \indent{\text{for}[\text{pp}=1,\text{pp}\text{$<$=}6*\text{qq},\text{pp}\text{++},}\\ \indent\indent{\text{if}[\text{gcd}[\text{pp},\text{qq}]==1,}\\ \indent\indent\indent{\text{print}[\{21+\text{pp}/\text{qq},\text{solless}[21+\text{pp}/\text{qq},}\\ \indent\indent\indent\indent{15/2*\text{sqrt}[(21+\text{pp}/\text{qq})/13]/}\\ \indent\indent\indent\indent\indent{(22+(\text{pp}-1)/\text{qq}-15*\text{sqrt}[(21+\text{pp}/\text{qq})/13])}\\ \indent\indent\indent\indent{(\text{sqrt}[\text{qq}+29]-1),13/2]\ } ] ] ] ] } $ ]

this paper combines data from the three preceding papers in order to analyze the multi - waveband variability and spectral energy distribution of the seyfert 1 galaxy ngc 4151 during the december 1993 monitoring campaign . the source , which was near its peak historical brightness , showed strong , correlated variability at x - ray , ultraviolet , and optical wavelengths . the strongest variations were seen in medium energy (.5 kev ) x - rays , with a normalized variability amplitude ( nva ) of 24% . weaker ( nva = 6% ) variations ( uncorrelated with those at lower energies ) were seen at soft-ray energies of kev . no significant variability was seen in softer ( 0.11 kev ) x - ray bands . in the ultraviolet / optical regime , the nva decreased from 9% to 1% as the wavelength increased from 1275 to 6900 . these data do not probe extreme ultraviolet ( 1200 to 0.1 kev ) or hard x - ray ( 250 kev ) variability . the phase differences between variations in different bands were consistent with zero lag , with upper limits of.15 day between 1275 and the other ultraviolet bands ,.3 day between 1275 and 1.5 kev , and day between 1275 and 5125 . these tight limits represent more than an order of magnitude improvement over those determined in previous multi - waveband agn monitoring campaigns . the ultraviolet fluctuation power spectra showed no evidence for periodicity , but were instead well - fitted with a very steep , red power - law ( ) . if photons emitted at a primary " waveband are absorbed by nearby material and reprocessed " to produce emission at a secondary waveband , causality arguments require that variations in the secondary band follow those in the primary band . the tight interband correlation and limits on the ultraviolet and medium energy x - ray lags indicate that the reprocessing region is smaller than.15 lt - day in size . after correcting for strong ( factor of ) line of sight absorption , the medium energy x - ray luminosity variations appear adequate to drive the ultraviolet / optical variations . however , the medium energy x - ray nva is 24 times that in the ultraviolet , and the single - epoch , absorption - corrected x - ray/-ray luminosity is only about 1/3 that of the ultraviolet / optical / infrared , suggesting that at most/3 of the total low - energy flux could be reprocessed high - energy emission . the strong wavelength dependence of the ultraviolet nvas is consistent with an origin in an accretion disk , with the variable emission coming from the hotter inner regions and non - variable emission from the cooler outer regions . these data , when combined with the results of disk fits , indicate a boundary between these regions near a radius of order lt - day . no interband lag would be expected as reprocessing ( and thus propagation between regions ) need not occur , and the orbital time scale of day is consistent with the observed variability time scale . however , such a model does not immediately explain the good correlation between ultraviolet and x - ray variations . * multiwavelength observations of * + * short time - scale variability in ngc 4151 . * + * iv . analysis of multiwavelength continuum variability * + 0.4 cm r.a . edelson , t . alexander , d.m . crenshaw , s . kaspi , m.a . malkan , b.m . peterson , r.s . warwick , j . clavel , a.v . filippenko , k . horne , k.t . korista , g.a . kriss , j.h . krolik , d . maoz , k . nandra , p.t . obrien , s.v . penton , t . yaqoob , p . albrecht , d . alloin , t.r . ayres , t.j . balonek , p . barr , a.j . barth , r . bertram,g.e . bromage , m . carini , t.e . carone,f .- z . cheng , k.k . chuvaev, m . dietrich , d . dultzin - hacyan , c.m . gaskell , i.s . glass , m.r . goad , s . hemar , l.c . ho , j.p . huchra , j . hutchings , w.n . johnson , d . kazanas , w . kollatschny , a.p . koratkar , o . kovo , a . laor , g.m . macalpine , p . magdziarz , p.g . martin , t . matheson , b . mccollum , h.r . miller , s.l . morris , v.l . oknyanskij , j . penfold , e . prez , g.c . perola , g . pike,r.w . pogge , r.l . ptak , b .- c . qian , m.c . recondo - gonzlez , g.a . reichert , j.m . rodrguez - espinoza , p.m . rodrguez - pascual , e.l . rokaki , j . roland , a.c . sadun , i . salamanca , m . santos - lle , j.c . shields,j.m . shull,d.a . smith , s.m . smith , m.a.j . snijders,g.m . stirpe , r.e . stoner , w .- h . sun , m .- h . ulrich , e . van groningen , r.m . wagner,s . wagner , i . wanders , w.f . welsh , r.j . weymann , b.j . wilkes , h . wu , j . wurster , s .- j . xue , a.a . zdziarski , w . zheng , and z .- l . zou 0.4 cm = 1.em = -1.em department of physics and astronomy , 203 van allen hall , university of iowa , iowa city , ia 52242 . electronic mail : [email protected] school of physics and astronomy and the wise observatory , the raymond and beverly sackler faculty of exact sciences , tel - aviv university , tel - aviv 69978 , israel . astronomy program , computer sciences corporation , nasa goddard space flight center , code 681 , greenbelt , md 20771 . department of astronomy , university of california , math - science building , los angeles , ca 90024 . department of astronomy , the ohio state university , 174 west 18th avenue , columbus , oh 43210 . department of astronomy , university of leicester , university road , leicester le17rh , united kingdom iso project , european space agency , apartado 50727 , 28080 madrid , spain . department of astronomy , university of california , berkeley , ca 94720 . school of physics and astronomy , university of st . andrews , north haugh , st . andrews ky169ss , scotland , united kingdom . department of physics and astronomy , university of kentucky , lexington , ky 40506 . department of physics and astronomy , the johns hopkins university , baltimore , md 21218 . laboratory for high energy astrophysics , nasa goddard space flight center , greenbelt , md 20771 . center for astrophysics and space astronomy , university of colorado , campus box 389 , boulder , co 80309 . universitts - sternwarte gttingen , geismarlandstrasse 11 , d-37083 gttingen , germany . centre detudes de saclay , service dastrophysique , orme des merisiers , 91191 gif - sur - yvette cedex , france . department of physics and astronomy , colgate university , hamilton , ny 13346 . mailing address : lowell observatory , 1400 west mars hill road , flagstaff , az 86001 . computer sciences corporation , nasa goddard space flight center , code 684.9 , greenbelt , md 20771 . space sciences laboratory , university of california , berkeley , ca 94720 , and eureka scientific , inc . current address : 28740 w. fox river dr . , cary , il 60013 . center for astrophysics , university of science and technology , hefei , anhui , people s republic of china . crimean astrophysical observatory , p / o nauchny , 334413 crimea , ukraine . deceased , 1994 november 15 . landessternwarte , knigstuhl , d-69117 heidelberg , germany . universidad nacional autonoma de mexico , instituto de astronomia , apartado postal 70 - 264 , 04510 mexico d.f . , mexico . department of physics and astronomy , university of nebraska , lincoln , ne 68588 . south african astronomical observatory , p.o . box 9 , observatory 7935 , south africa . space telescope science institute , 3700 san martin drive , baltimore , md 21218 . harvard - smithsonian center for astrophysics , 60 garden street , cambridge , ma 02138 . dominion astrophysical observatory , 5071 west saanich road , victoria , b.c . v8x 4m6 , canada . naval research laboratory , code 4151 , 4555 overlook sw , washington , dc 20375 - 5320 . physics department , technion - israel institute of technology , haifa 32000 , israel . department of astronomy , university of michigan , dennison building , ann arbor , mi 48109 . astronomical observatory , jagiellonian university , orla 171 , 30 - 244 cracow , poland . canadian institute for theoretical astrophysics , university of toronto , toronto , on m5s 1a1 , canada . department of physics and astronomy , georgia state university , atlanta , ga 30303 . sternberg astronomical institute , university of moscow , universitetskij prosp . 13 , moscow 119899 , russia . department of physics and astronomy , university of calgary , 2500 university drive nw , calgary , ab t2n 1n4 , canada , and department of mathematics , physics , and engineering , mount royal college , calgary t3e 6k6 , canada . istituto astronomico delluniversit , via lancisi 29 , i-00161 rome , italy . mailing address : 816 s. lagrange rd . , lagrange , il 60525 . department of physics and astronomy , bowling green state university , bowling green , oh 43403 . shanghai observatory , chinese academy of sciences , people s republic of china . facultad de ciencias , dept . fsicas , universidad de oviedo , c/ calvo sotelo , s / n . oviedo , asturias , spain . nasa goddard space flight center , code 631 , greenbelt , md 20771 . instituto de astrofsica de canarias , e-38200 la laguna , tenerife , spain . esa iue observatory , p.o . box 50727 , 28080 madrid , spain . royal observatory edinburgh , university of edinburgh , blackford hill , edinburgh eh93hj , united kingdom . institut dastrophysique , 98 bis boulevard arago , f-75014 paris , france . department of physics and astronomy and bradley observatory , agnes scott college , decatur , ga 30030 . royal greenwich observatory , madingley road , cambridge cb3 0ez , united kingdom . laeff , apdo . 50727 , e-28080 madrid , spain . steward observatory , university of arizona , tucson , az 85726 . hubble fellow . jila ; university of colorado and national institute of standards and technology , campus box 440 , boulder , co 80309 . iram , 300 rue de la piscine , 38046 saint martin dheres , france . mailing address : rue elysee reclubus 1 bis . , 38100 grenoble , france . osservatorio astronomico di bologna , via zamboni 33 , i-40126 , bologna , italy . institute of astronomy , national central university , chung - li , taiwan 32054 , republic of china . european southern observatory , karl schwarzschild strasse 2 , 85748 garching , germany . astronomiska observatoriet , box 515 , s-751 20 uppsala , sweden . department of physics , keele university , keele st5 5bg , staffordshire , united kingdom . observatories of the carnegie institution of washington , 813 santa barbara street , pasadena , ca 91101 . beijing astronomical observatory , chinese academy of sciences , beijing 100080 , people s republic of china . n. copernicus astronomical center , bartycka 18 , 00 - 716 warsaw , poland . = 0em = 1.5em

astro-ph9605082

two of the most constraining observed properties of active galactic nuclei ( agn ) are their large luminosities over a broad range of energies ( @xmath1-ray through infrared ) and their rapid variability ( implying a small source size unless the emission is highly beamed ) . the inferred large energy densities have led to a standard model of the ultimate energy source being the release of gravitational potential energy of matter from an accretion disk surrounding a supermassive black hole ( e.g. , rees 1984 ) . although this general model has broad support , the specific physical processes that produce the complex , broadband spectral energy distributions ( seds ) observed from agn have not been clearly identified . it is believed that a mix of processes is important . the ultraviolet and optical emission may be primary radiation from an accretion disk ( shields 1978 ; malkan & sargent 1982 ; malkan 1983 ) . in low - luminosity objects starlight will contribute as well . thermal dust emission is an important ingredient of the infrared band ( barvainis 1987 ; sanders et al . the high - energy ( x - ray and @xmath1-ray)emission is not well understood . there are a variety of models for their origin ranging from electromagnetic cascades in an @xmath7 pair plasma ( zdziarski et al . 1990 ) to thermal comptonization models ( haardt & maraschi 1993 ; haardt , maraschi , & ghisellini 1994 ) . furthermore , gas near the central source may reprocess at least some of the primary radiation via compton scattering , absorption , and fluorescent processes ( guilbert & rees 1988 ; lightman & white 1988 ; george & fabian 1991 ; matt , fabian , & ross 1993 ) . determination of the mix of physical processes that produce these large , broadband luminosities is a major unresolved issue in agn research , and multi - waveband variability studies are potentially highly constraining . causality arguments imply that if emission in a secondary " band is produced when photons from a primary " band are reprocessed in material near the central engine , then variations in the secondary band could not be seen to lead those in the primary band . furthermore , if the emission in any given waveband is a combination of two independent components ( with presumably independent variability behavior ) , then measurement of broadband spectral variability might allow them to be separated . finally , if a characteristic variability time scale could be measured , it could be compared with those indicative of different physical processes ( e.g. , with the expected viscous , orbital , light - travel time scales ) . in spite of the potential power of this approach , it has not until recently been exploited because of the very large amount of telescope time required . in several experiments designed to measure the size of the broad - line region in the seyfert 1 galaxy ngc 5548 , variations at @xmath01400 were seen to track those at @xmath02800 and @xmath05000 to within @xmath212 day ( clavel et al . 1991 ; peterson et al . 1991 ; korista et al . this was taken to imply an ultraviolet - optical propagation time that is too short to be associated with any dynamics mediated by viscosity , such as variations in the mass inflow rate , in a standard @xmath8-disk ( krolik et al . 1991 ; but see 4.2 . similar problems were noted in ultraviolet and optical monitoring of ngc 4151 by ulrich et al . krolik et al . ( 1991 ) suggested that variation in the different wavebands in ngc 5548 were coordinated by a photon signal . nandra et al . ( 1992 ) suggested that this signal might be x - ray heating ( reprocessing ) . several authors constructed specific models of x - ray illuminated accretion disks to account for the ngc 5548 data ( collin - souffrin 1991 ; rokaki & magnan 1992 ; molendi , maraschi , & stella 1992 ; rokaki , collin - souffrin , & magnan 1993 ) as well as for ngc 4151 ( perola & piro 1994 ) . a strong test of the idea that the ultraviolet is produced by reprocessing x - ray photons could be made by measuring the time relationship between fluctuations in the ultraviolet and the x - rays , but previous attempts ( e.g. , clavel et al . 1992 ) lacked adequate temporal resolution . in order to attempt this test , an international consortium of agn observers undertook a campaign to intensively monitor a single seyfert 1 galaxy , ngc 4151 , at ultraviolet , x - ray , @xmath1-ray , and optical wavelengths for @xmath010 days in december 1993 . these data are described in detail in the three preceding papers ( paper i crenshaw et al . 1996 ; paper ii kaspi et al . 1996 ; and paper iii warwick et al . 1996 ) ; they are summarized in the following section . in this paper , the multi - wavelength data are analyzed in combination . the measurement of the multi - waveband variability , temporal correlations , phase lags , and the broadband optical through@xmath1-ray sed are analyzed in 3 and the scientific implications are briefly discussed in 4 . the aim of these observations was to monitor intensively ngc 4151 across the accessible optical - through-@xmath1-ray region during the period from mjd 22.5 to 32.2 . ( the modified julian date , mjd , is defined as mjd = jd 2,449,300 . all dates refer to the center points of observations unless otherwise noted . ) while scheduling difficulties , satellite malfunctions , and other minor problems did cause some gaps and perturbations in this plan , this campaign still produced the most intensive coordinated observations to date of an agn across these high - energy wavebands . along with daily observations before and after this period ( not analyzed in this paper ) , the _ international ultraviolet explorer _ ( _ iue _ ) observed ngc 4151 nearly continuously from mjd 22.6 to 31.9 . a pair of swp and lwp spectra , spanning the range 12002000 and 20003000 , respectively , was obtained every @xmath070 min , excluding a daily @xmath02 hr gap at about modulo mjd 0.40.5 . _ rosat _ observed the source twice per day , with nearly even sampling from mjd 22.5 to 28.0 . a satellite malfunction caused _ rosat _ to go into safe mode for most of the second half of the campaign , although it did make one final observation at mjd 32.2 . _ asca _ was also plagued by satellite problems ( this time in the first half of the campaign ) but four 10 ksec observations were successfully obtained at mjd 26.0 , 27.5 , 29.1 , and 31.6 . compton gamma - ray observatory ( _ cgro _ ) observations were made with the osse instrument during the period mjd 22.734.6 . ground - based optical spectra were obtained for a 2 month period that included these dates . figure 1 gives the light curves in nominal 100 kev , 1.5 kev , 1275 , 1820 , 2688 and 5125 wavebands . these bands are explicitly defined below , and further details of the sampling characteristics are given in table 1 . a total of 18 optical spectra were obtained during the intensive monitoring period ; 10 at perkins observatory ( osu ) and eight at wise observatory . the wise data covered 42106990 with @xmath05 resolution through a 10@xmath9 slit , while the osu data covered 44805660 at @xmath09 resolution through a 5@xmath9 slit . these data were intercalibrated to remove the effects of ( presumably non - variable ) extended emission and instrumental offsets from the light curves . continuum fluxes in the 4600@xmath1040 and 5125@xmath1025 bands and h@xmath11 line fluxes were measured from both sets of spectra , while 6200@xmath1030 and 6925@xmath1025 continuum fluxes and h@xmath8 line fluxes were measured only from the wise data . the wise spectrograph projects to @xmath12 on the sky , and approximately 25% of the light at a wavelength of 5125 is due to starlight from the host galaxy ( see 3.5 . ) . the uncertainties in measurements of the optical fluxes are at about the 1% level ( see paper ii for details ) . the 5125 data from both telescopes were combined in the correlation analyses and light curves . during the intensive monitoring period , _ iue _ obtained a total of 176 spectra with the swp camera and 168 with the lwp camera . observations were spaced as closely as possible , leading to fairly even sampling of @xmath018 spectra per day in each camera , with a @xmath02 hour period each day , during which no data were obtained because of earth occultation and high particle background . one - dimensional spectra were extracted using the tomsips package ( ayres 1993 ) . continuum fluxes were measured by summing over relatively line - free , @xmath030 wide bands centered at 1275 , 1330 , 1440 , 1820 , 1950 , 2300 , and 2688 . uncertainties , taken to be the standard error in the band , were typically 12% , consistent with the observed epoch - to - epoch dispersion in the measured fluxes . line fluxes ( c iv , he ii , and c iii ] ) were measured by fitting multiple gaussians . ( see paper i and penton et al . 1996 for details of the _ iue _ data reduction . ) the continuum light curves measured at 1275 , 1820 , 2688 , and 5125 were used in the light curves and time - series analyses , all of the ultraviolet / optical continuum data were used only in the variability amplitude and zero lag correlation analyses , and the emission line data were not used in this paper . the _ rosat _ pspc made a total of 13 observations of ngc 4151 between mjd 22.5 and mjd 28.0 . although the center points of the first 12 observations were almost evenly spaced ( every 0.5 day ) , the integration times varied from 0.8 ksec to 6.4 ksec . the two soft ( 0.10.4 kev and 0.50.9 kev ) bands , which showed no significant variability during the observations , are apparently dominated by emission from an extended component ( e.g. , elvis et al . 1983 ; morse et al . . however , the hardest _ rosat _ band ( 12 kev ) showed variations significantly in excess of the measured errors . the _ rosat _ spectra were well - fitted by a non - variable thermal bremsstrahlung ( @xmath13 kev ) component that contributes most of the flux below 1.4 kev and a heavily absorbed ( @xmath14 @xmath15 ) power - law with fixed slope ( @xmath16 ) and variable normalization ( @xmath17 ph @xmath15 s@xmath18 kev@xmath18 ) that dominates at medium x - ray energies . however , the proportional counter data have low resolution and thus a range of other models can not be ruled out . the four _ asca _ spectra have higher resolution , allowing more detailed spectral analysis . these data were compatible with a model that includes warm and cold absorbers , thermal bremsstrahlung , a power - law and a 6.4 kev iron line . again , this is not a unique solution . the iron line shows significant broadening to the redward wing , suggesting a gravitational redshift that would constrain the material to lie very near the black hole ( yaqoob et al . the 12 kev _ asca _ data were used in the light curves along with the _ rosat _ points , with a 10% uncertainty added to account for possible calibration differences . the harder ( 210 kev ) _ asca _ data were not used for variability analysis because only four epochs were obtained . ngc 4151 was continuously observed with the osse instrument on _ cgro_. the broadband 50150 kev count rate showed weak but significant variability . the raw data were calibrated by convolving the detector response function with a model assuming a power - law with an exponential cutoff . see paper iii for further details on the x - ray and @xmath1-ray data . ngc 4151 was near its historical peak brightness during this campaign . the average ( absorption - corrected ) flux in the 2 - 10 kev band was @xmath19 . this is bright compared to the compilation of results from _ exosat _ and _ ginga _ reported in yaqoob et al . ( 1993 ) , which ranged from 0.8 to @xmath20 . the mean 1275 flux of @xmath21 erg @xmath15 s@xmath18 @xmath18 is higher than the range of @xmath22 erg cm@xmath18 s@xmath18 @xmath18 seen in 19781990 ( edelson , krolik , & pike 1990 ; courvosier & palatini 1992 ) , although the ultraviolet flux was a bit higher during the march 1995 astro-2 campaign ( kriss et al . during the period mjd 22.532.3 , the light curve of ngc 4151 was densely sampled by _ iue _ , _ rosat _ , _ cgro _ , and ground - based telescopes . the initial analysis of these data involved comparing fractional variability amplitudes during this period as a function of observing frequency . the normalized variability amplitude ( nva , or @xmath24 ) was computed as follows . for each band , the mean ( @xmath25 ) and standard deviation ( @xmath26 ) of the flux points and the mean error level ( @xmath27 ) were measured . because the nva is intended to be free of instrumental effects , it was determined by subtracting in quadrature the measured mean error from the standard deviation , and then dividing by the mean flux . that is , @xmath28 note that this is essentially the same procedure used to derive @xmath24 in peterson et al . ( 1991 ) and @xmath29 in edelson ( 1992 ) . these quantities are given in table 2 , along with the observing band , number of observations ( @xmath30 ) and the difference in days between the first and last observation ( @xmath31 ) during this intensive period . figure 2 is a plot of nva as a function of observing waveband . the nva shows a strong dependence on photon energy , increasing from @xmath32% in the optical to @xmath04% in the lwp and @xmath33% in the swp . the variability in the swp is clearly stronger than at longer wavelengths , but because of the lower variability levels and stronger starlight corrections , it is impossible to say if the optical and lwp have significantly different non - stellar nvas . " this strong wavelength dependence has been seen previously in ngc 4151 and other agn ( e.g. , edelson , krolik & pike 1991 ) , and is suggestive of the superposition of a non - variable soft component that dominates in the optical / infrared and a variable hard component that dominates in the ultraviolet . the nva also shows interesting behavior at high energies . the soft _ rosat _ bands show no evidence for any significant variations , with nva formally undefined as the observed variation levels are smaller than the instrumental errors . the medium energy 1.5 kev x - ray band shows the strongest variability seen in any band , with @xmath34% . this large difference is apparently due to the superposition of two components : a soft , extended ( and therefore non - variable ) component seen in spectra and hri images ( elvis et al . 1983 ) , and a harder , strongly variable component . however , this strongly variable component must cut off at some higher energies , because the _ cgro _ data show weaker variability , with @xmath35% in the 50150 kev band . unfortunately , there are two important energy ranges in which the variability properties of ngc 4151 during this campaign were not measured : the hard x - ray 250 kev gap between _ rosat _ and _ cgro _ ( the 210 kev _ asca _ data can not be used to measure a meaningful nva with only four points ) and the extreme ultraviolet 1200 ( 10 ev ) to 100 ev gap in the mostly unobservable region between _ iue _ and _ rosat_. the former waveband could contain the bulk of the luminosity of the putative primary emission component , and the latter , the bulk of the disk luminosity . these gaps do limit the power of these data ( see 4 ) , although we note that this campaign has produced the most densely sampled grid in time and energy band obtained of any agn to date . table 2 also gives the monochromatic variable luminosity , defined as @xmath36 . ( @xmath37 is the observed monochromatic luminosity , as defined in 3.5 . ) this parameter behaves somewhat differently than the nva . in particular , the medium energy x - rays , which have the strongest nva ( @xmath38% ) actually show the lowest value of @xmath39 ( @xmath40 erg s@xmath18 ) . note however that the x - rays appear to be heavily absorbed ( @xmath09095% , see 3.5 . ) , so the intrinsic value of this quantity is actually much larger . these are the most densely sampled ultraviolet observations of any agn obtained to date , allowing examination of the short time scale variability properties . these data were used to measure the fluctuation power density spectra ( pds ) in four ultraviolet bands ( 1275 , 1330 , 1820 , and 2688 ) , on time scales of @xmath00.25 day . a regular grid of spacing 0.1 day was created , and the value at each point was taken to be the average of the nearest 12 points , weighted as @xmath41 . this resampling , which was necessary to mitigate problems introduced by the periodic @xmath00.1 day earth occultation gaps , destroys all information on shorter time scales . however , as seen below , the pds are dominated by noise on these shorter time scales , so this is not a problem . only the continuous data were used , and the mean was not subtracted . these resampled data were used to compute the pds in the four ultraviolet bands , the results of which are shown in figure 3 . the pds of agn show no discrete features that would indicate periodicity . instead , the fluctuation power is spread out over a wide range of temporal frequencies . it is common practice to paramterize the pds of agn by the function @xmath42 , where @xmath43 is the pds at temporal frequency @xmath44 , and @xmath45 is the power - law slope . these pds have @xmath46 , making them the reddest " agn pds yet measured . by comparison , the ultraviolet pds of ngc 5548 had a slope @xmath47 on time scales of weeks to months ( krolik et al . this indicates that the variability power is falling off rapidly at short time scales , and the bulk of the variability power is on time scales of days or longer . the x - ray pds ngc 4151 has @xmath47 time scales of hours to days and @xmath48 on longer time scales ( papadakis & mchardy 1995 ) . the fact that the ( rather noisy ) x - ray and ultraviolet pds appear to have different slopes over the same range of time scales may indicate that different processes power the variability at the two wavebands , although systematic differences in the data and reduction techniques for the two data sets make direct comparison difficult . the four pds in figure 3 look similar to within the noise . there do however appear to be some qualitative differences in the character of variations in different bands that can be discerned by direct examination of the light curves . in particular , the ultraviolet data give the impression that the most rapid variations occur at the shortest wavelengths , and are somewhat smeared out " to longer wavelengths . ( the @xmath00.2 day spikes " in the 2688 light curve may be due to instrumental effects . ) a similar effect has been seen at longer time scales in _ hst _ monitoring of ngc 5548 ( korista et al . such an effect would not be apparent in the above pds analysis . however , it is unclear how much of this is intrinsic and how much is due to instrumental differences , since the variability - to - noise " ratio ( that is , @xmath49 ) , decreases toward longer wavelengths in the ultraviolet , so that this experiment was less sensitive to rapid variations at longer ultraviolet wavelengths . because there was no measurable interband phase lag ( see 3.4 ) , a slightly different view may be obtained by measuring the zero lag correlations between variations in different wavebands . to investigate this , fluxes in a number of comparison bands ( 100 kev , 1.5 kev , 1330 , 1820 , 2688 , and 5125 ) were correlated with the 1275 fluxes . for each of the comparison bands , each flux point was paired with the average of the two 1275 flux points measured immediately before and after that point , and 1275 uncertainties were taken to be the average of the uncertainties measured for the two data points . the results are plotted in figure 4 . the number of points ( @xmath30 ) , linear correlation coefficient ( @xmath50 ) , spearman rank correlation coefficient ( @xmath51 ) , @xmath52-statistic , and y - intercept are tabulated in table 3 . the 100 kev fluxes ( @xmath53 ) show no correlation with 1275 ( or with any other band for that matter ) . this could suggest that the processes producing the @xmath1-rays are unrelated to those at work at lower energies , especially given the relatively low nva of the @xmath1-rays . the data are strongly correlated on these short time scales in all other cases , although the optical and x - ray correlations are a bit marginal , with @xmath54 for 18 points and @xmath55 for 17 points , respectively . note that perola et al . ( 1986 ) found a similar ultraviolet / x - ray zero lag correlation in previous data over longer ( @xmath01 year ) time scales , although it was claimed that corrrelation broke down at the high flux levels seen in this experiment . following the regression line to zero flux at 1275 yields a positive excess at all wavebands except 1.5 kev x - rays . this means that if one would model the flux at each waveband as a combination of a non - variable component and a variable component produced by reprocessing emission from the other waveband , the variable component is always at shorter wavelengths , and the non - variable component becomes a progressively larger fraction of the total flux as the wavelength increases . the most detailed analysis undertaken with these data was to measure temporal cross - correlation functions between wavebands . if emission in one band is reprocessed to another without feedback , measurement of an inter - band lag would provide an important confirmation , and indicate which were the primary and secondary bands . this intensive monitoring of ngc 4151 is clearly better suited for this test than any previous campaign . unfortunately , even in these data , the 1.5 kev x - rays , which show the strongest variations , are clearly undersampled , as they were observed at a temporal frequency only one - tenth that in the ultraviolet . the differences between the sampling , variability levels , and signal - to - noise ratios in the different wavebands present problems for measuring the interband lag . below , a detailed analysis is presented of two independent techniques used to measure the interband correlations and lags : the interpolation and discrete correlation functions . cross - correlation functions were computed using the interpolation cross - correlation function ( iccf ) method of gaskell & sparke ( 1986 ) , as subsequently modified by white & peterson ( 1994 ) . as described by gaskell & peterson ( 1987 ) , the cross - correlation function was determined by first cross - correlating the real observations from one time series with values interpolated from the second series . the calculation was then performed a second time , using the real values from the second series and interpolated values from the first series . the final cross - correlation function was then taken to be the mean of these two calculated functions . a sampling grid spacing of 0.05 day was used because that is approximately the mean interval between the ultraviolet observations . because the x - ray and optical continuum light curves are much less well - sampled than the ultraviolet light curves , the x - ray and optical cross - correlations with the 1275 ultraviolet light curve were performed by interpolating only in the ultraviolet light curve . that is , the computed cross - correlation functions are based on the real x - ray and optical points and interpolated ( or regularized ) ultraviolet points , and no interpolation of the x - ray or optical data was performed . the cross - correlation functions for the various continuum bands were similarly calculated by the @xmath56-transformed discrete correlation function ( zdcf ) method ( alexander 1996 ) , which is related to the discrete correlation function ( dcf ) of edelson & krolik ( 1988 ) . this method is rather more general than the iccf as ( a ) it does not require any assumptions about the continuum behavior between the actual observations , and ( b ) it is possible to reject data pairs at zero lag ( e.g. , two points in different wavebands , measured from one spectrum ) , and thus avoid false zero lag correlations that might arise from correlated flux errors . the zdcf differs in a number of ways from the dcf , a notable feature being that the data are binned by equal population rather than into time bins of equal width @xmath57 . one advantage of this technique is that it allows direct estimation of the uncertainties on the lag without the use of more assumption - dependent monte carlo simulations . the maximum likelihood error estimate on the position of the true cross - correlation function peak ( @xmath58 ) is a 68% fiducial confidence interval , meaning that it contains the peaks of 68% of the likelihood - weighted population of all possible cross - correlation functions ( alexander 1996 ) . phrased more loosely , the confidence interval is where 68% of the cross - correlation functions that are consistent with the zdcf points are likely to reach their peaks . the accuracy of this error estimate is limited by the assumption that the distributions of the true cross - correlation points around the zdcf points are independently gaussian , which is only approximately true , but it does not require any further _ a priori _ assumptions about the nature of the variations or transfer function between bands . the various columns in table 4 give the results obtained by cross - correlating the specified light curve with the 1275 light curve . the parameter @xmath59 is the maximum value of the cross - correlation function , which occurs at a lag @xmath60 . table 4 also gives the centroid of the cross - correlation function @xmath61 , which is based on all points with @xmath62 . the computed cross - correlation functions are shown in figure 5 ( iccf ) and figure 6 ( zdcf ) . the measured range of @xmath63 of the various bands with 1275 ( 0.20 to + 0.26 day for the 1.5 kev x - rays , 0.14 to + 0.01 day for 1820 , 0.05 to + 0.22 day for 2688 , 0.78 to 0.53 day for 5125 ) are all consistent with zero measurable lag . thus , we have assigned conservative upper limits on the lags of 1275 with 1.5 kev of @xmath64 day , with other ultraviolet bands of @xmath65 day , and with 5125 of @xmath66 day . the physical significance of these limits is discussed in 4 . figure 7 gives a broadband @xmath1-ray optical snapshot sed , constructed by combining nearly - simultaneous observations consisting of the wise spectrum taken on mjd 25.6 , _ spectra lwp 26907 and swp 49441 , taken near mjd 26.0 , the _ rosat _ observation centered at mjd 26.0 , and _ asca _ observation number 1 , centered near mjd 26.0 . because of the weak variability and low signal to noise in the @xmath1-ray band , the _ cgro _ spectrum was integrated over the entire campaign to produce the high - energy sed . additional ( non - simultaneous ) infrared data from edelson , malkan & rieke ( 1987 ) were used in the spectral fits ( 4.2 . ) but are not shown in figure 7 . the data plotted are monochromatic luminosities , @xmath37 , as a function of frequency , @xmath67 . the quantity @xmath37 is given by @xmath68 where @xmath69 is the flux density and @xmath70 mpc is assumed to be the distance to ngc 4151 ( tully 1989 ) . the sed of any seyfert 1 galaxy is of course a combination of a number of components , and in particular , we note that starlight from the underlying sab galaxy of ngc 4151 contributes a significant amount of the optical / infrared flux . peterson et al . ( 1995 ) estimated that the underlying galaxy produced a flux of 19 mjy at 5125 in the @xmath71 wise aperture , in good agreement with the independent result reached in paper ii of this series . this corresponds to approximately 25% of the 5125 flux at the light levels seen in this campaign , so the 5125 nva of the nuclear component ( excluding starlight ) is approximately 1.33 times the observed value quoted in table 2 . for a normal sab galaxy , approximately 42% of the light at 6925 , 31% of the light at 6200 , 21% of the light at 4600 , and 4% of the light at 2688 would be due to starlight . finally , we note one other important feature in the sed : the soft x - rays show significant absorption due to line - of - sight gas intrinsic to ngc 4151 ( evident below 4 kev ) , and in our own galaxy ( evident below 0.3 kev ) . spectral fits indicate that approximately 9095% of the 12 kev luminosity appears to have been absorbed by gas along the line of sight ( see paper iii for details ) , although this result is model - dependent . this would imply that the intrinsic 12 kev luminosity is 1020 times that observed . possible sources of the intrinsic absorption include partial covering by cold gas ( e.g. , holt et al . 1980 ; yaqoob , warwick & pounds 1989 ; yaqoob et al . 1993 ) or absorption in a warm , ionized medium ( weaver et al . 1994a , b ; warwick , done & smith 1996 ) . similar behavior has been seen in the sed of ngc 3783 obtained during the world astronomy day " campaign ( alloin et al . 1995 ) . a smaller fraction ( @xmath035% ) of the 210 kev flux is apparently absorbed , but unfortunately , the data are inadequate to characterize the variability in this band . ngc 4151 was monitored intensively with _ iue _ , _ rosat _ , _ asca _ , _ cgro _ and ground - based optical telescopes for @xmath010 days in december 1993 . these observations provided the most intensively sampled multi - wavelength light curve of an agn to date . the major new observational results are as follows : 1 . ngc 4151 showed strong variability from the optical through @xmath1-ray bands during this period . the optical / ultraviolet / x - ray light curves are similar but not identical , with no detectable lags between variations in the different bands . the upper limits on the lags are , between 1275 and 1.5 kev , @xmath20.3 day ; between 1275 and 5125 , @xmath21 day ; and between 1275 and the other ultraviolet bands , @xmath20.15 day . the 100 kev variations are not clearly related to those in any other band . the strongest variability was seen in the medium energy x - rays , with the 12 kev band showing an nva of 24% . the variations were systematically weaker at lower energies , with nvas of 9% , 5% , 4% and 0.5% at 1275 , 1820 , 2688 and 5125 , respectively . however , the 100 kev light curve showed an nva of only 6% , and the soft ( @xmath721 kev ) x - rays showed no detectable variability . the observed luminosity ( as opposed to fractional ) variability was seen to be lowest in the 1.5 kev band , apparently due to strong x - ray absorption along the line of sight . the pds of all four ultraviolet bands are similar , showing no evidence for periodicity , but is instead being well modeled as a power - law with @xmath73 . the broadband sed shows a strong deficit in the soft / mid - x - ray , which is well - fitted as a factor of @xmath015 absorption by gas on the line - of - sight . finally , the character of the multi - wavelength variability in the seyfert 1 ngc 4151 appears to differ markedly from the only other agn monitored with similar intensity . the bl lac object pks 2155304 was observed continually for @xmath03 days at ultraviolet ( urry et al . 1993 ) , x - ray ( brinkmann et al . 1994 ) , and optical ( courvoisier et al . 1995 ) wavebands . multi - wavelength analysis by edelson et al . ( 1995 ) found that the x - ray , ultraviolet and optical variations were almost identical in amplitude and shape , but that the x - ray variations led those in the ultraviolet and optical by @xmath023 hr . for these observations of ngc 4151 , and other seyfert 1s observed over longer time scales ( e.g. , clavel et al . 1991 ) , the light curves do show significant differences between wavebands , with nva a strong function of energy , but no lag has been measured between variations at different bands . this is an observed ( and therefore model - independent ) example of an intrinsic difference between seyfert 1s and bl lacs . these results have important implications for models that attempt to explain the ultraviolet emission from agn . there are currently two broad classes of such models . the first hypothesizes that the bulk of the ultraviolet luminosity is produced internally by viscosity in the inner regions of an accretion disk surrounding a central black hole , and the second , that the observed ultraviolet emission is produced in gas illuminated and heated by the source that we observe at high energies . of course , the true picture could be a combination of these processes , a hybrid in which both intrinsic emission from an accretion disk and reprocessing of x - ray emission are important ( and , indeed , may feed back upon each other ) , or conversely , it may be that neither of these models is relevant . although the specific processes responsible for agn emission have not been clearly identified , there is broad support for the general model of a black hole and accretion disk ( see 1 ) . in this model , the emission from normal ( non - blazar , radio - quiet ) seyfert 1s like ngc 4151 is relatively isotropic , not significantly doppler - boosted or beamed towards earth . in this case , it is possible to use variability to place relatively model - independent limits on the central black hole mass ( and therefore the size scale ) . the most general is the eddington limit , which requires only that the source be gravitationally bound and possess a high degree of spherical symmetry . the minimum central mass given by this limit is @xmath74 where @xmath75 and @xmath76 are the eddington luminosity and mass , @xmath77 is the thompson cross - section , and @xmath78 is the proton mass . as the integrated luminosity of ngc 4151 is @xmath79 erg s@xmath18 , the eddington mass is @xmath80 . for a source surrounding a black hole of mass @xmath81 , and schwarzschild radius @xmath82 , the minimum variability time scale ( @xmath83 ) can be estimated from the size ( @xmath84 ) of the smallest stable orbit , which is @xmath85 for a schwarzschild black hole . for @xmath86 , this implies @xmath87 cm or @xmath88 sec , which is not a significant constraint on data which were sampled every hour . a larger but more model - dependent constraint assumes keplerian orbits and uses the correlation between the widths of emission lines and the distances estimated from their lags to estimate the central mass . in the form in which this estimate is generally presented , the inferred mass is the true mass if the clouds travel on circular orbits . if the clouds are gravitationally bound , but non - gravitational forces ( e.g. radiation pressure or hydrodynamics ) affect cloud motions , the real mass is greater than this estimate ; if the motions are unbound , the real mass is smaller than this estimate . clavel et al . ( 1987 , 1990 ) used this method to derive a central mass of @xmath89 for ngc 4151 . this corresponds to a smallest stable orbit of @xmath90 lt - min . again , this relatively weak constraint is not significant for these data . if the ultraviolet / optical continuum is optically thick thermal emission , the variability amplitudes can be used to constrain the temperature distribution . the simplest non - trivial general case is a flat , azimuthally symmetric disk with a local blackbody temperature that drops with radius as some power - law : @xmath91 . when @xmath8 is 3/4 , this is a fair approximation of a standard accretion disk , except at the smallest radii . by contrast , in a disk that radiates predominantly by reprocessed energy ( see 4.3 . ) , @xmath8 can be smaller , depending on the geometry of both the disk and the source of the primary radiation . if the local emission is described by a blackbody , the contribution of an annulus to the total disk flux density at a given frequency @xmath67 is @xmath92 where @xmath93 , and the starting and ending points of the integral are defined by the temperatures at the inner and outer radii of the ring . the increasing amplitude of variations with observing frequency is then naturally attributed to changing emission from the hotter regions ( that is , the inner disk radii ) . there is a test of the simplest case that produces simultaneous multi - wavelength variations , in which the emission from the outer disk ( @xmath94 ) is constant , and all of the variability is produced by a complete , simultaneous modulation of the emission from the inner disk ( @xmath95 ) . the only free parameter is the radius that separates the variable and constant parts of the disk , and this can be determined from the nva at a single wavelength . for example , using the 1275 variability amplitude of 8.6% gives @xmath96 for @xmath97 or @xmath98 for @xmath99 . this corresponds to a boundary at the disk radius where the temperature is 215,000 or 78,000 k , respectively . the last two columns of table 2 show how the percentage flux would drop if all of the emission inside this radius ( the putative variable component ) went to zero , including the effects of starlight . there is good agreement between the simple thermal model and the observed wavelength dependence of the variability amplitude for the standard accretion disk case , @xmath100 . this is not a strong function of how the inner boundary is chosen ; for the @xmath99 case , truncating the integration to between @xmath101 and @xmath98 changes the result by only 20% . the success of this simple exercise motivated the fitting of the ultraviolet / optical sed with a standard model of a geometrically thin , optically thick accretion disk ( e.g. , following the formalism of sun & malkan 1989 ) . aside from the starlight , no additional long - wavelength component was included in the models , so no attempt was made to fit fluxes longward of 2 @xmath102 m . if the disk is assumed to be viewed face - on , and the black hole is spinning rapidly ( kerr metric ) , the best - fit model parameters are @xmath103 and @xmath104 yr@xmath18 . if on the other hand the black hole is assumed stationary ( schwarzschild case ) , the best - fit parameters are @xmath105 and @xmath106 yr@xmath18 . in either case , the accretion rate corresponds to 0.6% of the eddington rate . the black hole mass inferred from the schwarzschild fit agrees with the keplarian value obtained by clavel et al . ( 1987 , 1990 ) . these fits give higher weight to the higher signal - to - noise optical continuum than to the ultraviolet continuum where the disk light dominates , which in turn requires a hotter disk and consequently a lower black hole mass ( @xmath107 ) . all the disk fits would give significantly larger black hole masses if the disk had a non - zero inclination . integrating the multi - waveband disk emission out to a boundary radius of @xmath108 lt - day shows that approximately 30% , 20% and 5% of the total disk flux at 1275 , 2688 , and 5125 is produced in the inner disk . ( these numbers refer to the kerr model , but are not very model - dependent ; malkan 1991 . ) after correcting for the effects of galactic starlight in the @xmath109 aperture , the 5125 emission from the inner disk falls to @xmath04% . for @xmath110 , this radius corresponds to @xmath111 these fractions are approximately the same as the largest peak - to - peak variations observed in these bands during the intensive 10 day multi - waveband monitoring campaign . thus , one consistent explanation for the decline in variability with increasing wavelength is that the variations occur entirely inside the inner disk . at a distance @xmath112 lt - day from a @xmath113 black hole , the orbital time scale is @xmath01 day , whereas the dominant fluctuations clearly occur on longer time scales ( i.e. , the 1275 peak - to - peak variation was only 42% over the entire 10 day intensive campaign ) . thus , they could be associated with either orbital mechanics or thermal fluctuations in the inner disk . it must also be noted that this simple disk model does not produce any x - ray emission , although this is also likely to originate within the inner regions . these data also can be used to constrain models in which the ultraviolet radiation is produced in gas that is heated by the same x - ray continuum that we observe directly at higher energies . in this model , time variations in the ultraviolet and high energy fluxes should be closely coupled : the ultraviolet becomes stronger shortly after the high energy flux rises , and the high energy flux may itself respond to changes in the ultraviolet flux if the ultraviolet photons provide seeds for compton upscattering into the higher energy band . the delays in both cases are essentially due to the light travel time between the two source regions . consequently , the ( reprocessed ) ultraviolet emission should vary simultaneously with the ( primary ) x - rays on time scales longer than the round - trip light - travel time between the emitting regions ( e.g. , clavel et al . the strong correlation between the ultraviolet and x - ray variations therefore supports the reprocessing hypothesis . the lack of any lag within the ultraviolet could be explained in two ways : either the sub - regions responsible for variations in different wavelengths are likewise very close to each other , or , as with the accretion disk model , only the hottest part of the region is varying . in the context of this model , the lack of any detectable lag implies that the x - ray and ultraviolet emitting regions are separated by @xmath20.15 lt - day , ( @xmath114 lt - day , because the light must travel in both directions ) , so the bulk of the reprocessing must occur in the central regions . perola & piro ( 1994 ) applied a more detailed reprocessing model to earlier x - ray / ultraviolet observations and predicted that high time resolution monitoring would measure lags of order 0.030.1 day ( rather close to but still formally consistent with the measured limits of @xmath20.150.3 day ) . the reprocessing model is supported by the broad profile of the iron k@xmath8 line , which suggests relativistic effects associated with an origin very close to a central black hole ( yaqoob et al . however , the _ asca _ observations ( and previous medium x - ray observations ) found no evidence for any significant hard tail " in the x - ray spectrum ( paper iii ; maisack & yaqoob 1991 ) . while the presence of the iron line implies reprocessing by some material , the lack of a `` reflection hump '' suggests that the material is not optically thick to compton scattering , as would be expected in the putative reprocessing disk . an associated problem is the overall energy budget . the total x - ray/@xmath1-ray flux must be adequate to produce the observed ultraviolet / optical / infrared flux , and the variable high energy flux must also equal or exceed that at lower energies . the observed , integrated 0.11 @xmath102 m flux is @xmath115 erg @xmath15 s@xmath18 . as strong ly@xmath8 and c iv emission indicates that the ultraviolet bump extends to wavelengths substantially shorter than 1000 , the intrinsic , integrated flux is probably larger by a factor of @xmath03 , corresponding to a total ultraviolet / optical / infrared flux of order @xmath116 erg @xmath15 s@xmath18 . the observed , integrated 12 kev and 210 kev fluxes are @xmath117 erg @xmath15 s@xmath18 and @xmath118 erg @xmath15 s@xmath18 , respectively after correction for line - of - sight absorption ( which is particularly important at 12 kev ) , the intrinsic , integrated fluxes rise to @xmath119 erg @xmath15 s@xmath18 and @xmath120 erg @xmath15 s@xmath18 , for a total of @xmath121 erg @xmath15 s@xmath18 . this is not adequate to power the lower energies , but inclusion of the gro data and interpolating between asca and gro yields a ( rather uncertain ) integrated 1200 kev flux of order @xmath122 erg @xmath15 s@xmath18 . this would be of order the amount necessary to power the lower energies . however , the fact that the @xmath1-ray variations differ from those in all other wavebands suggests that the bulk of the 10200 kev emission ( which dominates the x - ray/@xmath1-ray flux ) arises in a component that does not fully participate in the reprocessing . in this case , the x - ray/@xmath1-ray luminosity would still be a factor of order @xmath03 too small , so at most @xmath01/3 of the observed infrared / optical / ultraviolet flux could be due to reprocessing . however , this can not be independently verified because the _ asca _ sampling above 2 kev is inadequate to characterize the variability , and indeed the entire 10 - 50 kev spectrum is interpolated , not observed . although the absolute x - ray luminosity changes are observed to be small compared to those in the ultraviolet / optical , spectral fits indicate that the x - rays are highly ( factor of @xmath01520 ) absorbed , and the intrinsic , absorption - corrected variations have more than enough power to drive the ultraviolet / optical variations . however , the nva , which measures the fractional ( as opposed to absolute ) variations is 24% in the x - rays , while it is only 9% , 5% , 4% , and 1% at 1275 , 1820 , 2688 , and 5125 , respectively . this indicates that at most 35% , 20% , 15% , and 4% of the emission at 1275 , 1820 , 2688 , and 5125 could be due to reprocessing , with the rest coming from a component with different ( slower ) variability . as with the previous analysis , this would indicate that at most @xmath01/3 of the ultraviolet / optical / infrared can be reprocessed emission from higher energies . 1 . ngc 4151 showed significant variability over time scales of days , so the emitting region must be smaller than of order a few light days across , assuming the emission is not beamed . the limits on the interband lags indicate that , if there is reprocessing of flux between bands , none of the emission regions could be larger than @xmath00.15 lt - day . the lower limit to the central black hole mass derived from the eddington limit is small ( @xmath123 ) , corresponding to a minimum variability time of only 10 sec , which is not a significant constraint . accretion disk fits to the sed yield a central black hole with a mass of @xmath124 , accreting well below the eddington limit . the fact that the low nvas decrease from medium energy x - ray to ultraviolet to optical wavebands is consistent with the accretion disk model if the bulk of the variable ultraviolet / optical emission originates in a region @xmath00.07 lt - day ( @xmath125 ) from the center . this model gives is no immediate explanation of the link between ultraviolet and x - ray variations . the reprocessing model predicts a strong correlation between ultraviolet and x - ray variability , which is observed . because the nvas become systematically smaller at longer wavelengths , and because the absorption corrected x - ray/@xmath1-ray luminosity is only much smaller than in the ultraviolet / optical / infrared , and at most @xmath01/3 of the lower energy emission could be produced by reprocessing . this suggests that perhaps the ultraviolet arises in a disk powered partially by illumination by an x - ray source and partially by internal viscosity and accretion . determining the exact mix of these emission components will require probing the ultraviolet / x - ray bands at the shortest accessible time scales . it is of particular importance to extend the coverage to the gap at x - ray energies harder than 2 kev . this is the goal of the upcoming coordinated _ xte / iue_/ground - based observations of ngc 7469 , scheduled for the middle of 1996 , and it is hoped that they will shed further light on this important question . the work of bmp was supported by nasa grants nagw-3315 and nag5 - 2477 . the work of dmc and the _ iue _ data reduction were supported by nasa adp grant s-30917-f . the work of avf and lch was supported by nsf grant ast-8957063 . this research has also been supported in part by nasa grants nag5 - 2439 , nag5 - 1813 , and nagw-3129 . a96 alexander , t. 1996 , preprint a95 alloin , d. et al . 1995 , a&a , 293 , 293 a93 ayres , t. 1993 , pasp , 105 , 538 b87 barvainis , r. 1987 , apj , 350 , 537 b94 brinkmann , w. et al . 1994 , a&a , 288 , 433 c87 clavel , j. et al . 1987 , apj , 321 , 251 c90 clavel , j. et al . 1990 , mnras , 246 , 668 c91 clavel , j. et al . 1991 , apj , 366 , 64 c92 clavel , j. et al . 1992 , apj , 393 , 113 c91 collin - souffrin , s. 1991 , a&a , 249 , 344 g91 george , i. m. , & fabian , a. c. 1991 , mnras , 249 , 352 c95 courvoisier , t. et al . 1995 , apj , 438 , 108 c96 crenshaw , d.m . 1996 , submitted to apj ( paper i ) e87 edelson , r.a . , malkan , m , & rieke , g. 1987 , apj , 321 , 233 ek88 edelson , r.a . , & krolik , j.h . 1988 , apj , 333 , 646 ekp91 edelson , r.a . , krolik , j.h . , & pike , g.f . 1990 , apj , 359 , 86 e92 edelson , r.a . 1992 , apj , 401 , 516 e95 edelson r. et al . 1995 , apj , 438 , 120 e83 elvis , m. , briel , v. & henry , j.p . , 1983 , apj , 268 , 105 gp87 gaskell , c.m . , & peterson , b.m . 1987 , apjs , 65 , 16 gs86 gaskell , c.m . , & sparke , l.s . 1986 , apj , 305 , 175 g88 guilbert , p. w. , & rees , m. j. 1988 , mnras , 233 , 475 hm93 haardt , f. , & maraschi , l. 1993 , apj , 413 , 507 h94 haardt , f. , maraschi , l. , & ghisellini , g. 1994 , apj , 432 , l95 h80 holt , s. et al . 1980 , apj , 241 , l13 k96 kaspi , s. et al . 1996 , submitted to apj ( paper ii ) ko95 korista , k.t . 1995 , apjs , 97 , 285 kr95 kriss , g. et al . 1995 , apjl , 454 , l7 k91 krolik , j.h . , horne , k. , kallman , t.r . , malkan , m.a . , edelson , r.a . , & kriss , g.a . 1991 , apj , 371 , 541 l92 lee , g. , kriss , g. a. , & davidsen , a. f. 1992 , in testing the agn paradigm , ed . s. s. holt , s. g. neff , & c. m. urry ( new york : aip ) , 159 l88 lightman , a. p. , & white , t. r. 1988 , apj , 335 , 57 my91 maisack , m. & yaqoob , t. 1991 , a&a , 249 , 25 m83 malkan , m. a. 1983 , apj , 268 , 582 m91 malkan 1991 , in structure and emission properties of accretion disks ; editions frontieres , eds . bertout , collin , lasota , and van , p. 165 m82 malkan , m. a. , & sargent , w. l. w. 1982 , apj , 254 , 22 m93 matt , g. , fabian , a. c. , & ross , r. r. 1993 , mnras , 262 , 179 m95 morse , j.a . 1995 , apj , 439 , 121 n92 nandra , k. , pounds , k.a . , stewart , g.c . , george , i.m . , hayashida , k. , makino , f. , & ohasi , t. 1991 , mnras , 248 , 760 p95 papadakis , i.e. & mchardy , i.m . , 1995 , mnras , 273 , 923 p96 penton , s. et al . 1996 , apj , in preparation p86 perola , g. et al . 1986 , apj , 306 , 508 p94 perola , g. & piro , l. 1994 , apj , 281 , 7 p95 peterson et al . 1995 , pasp , 107 , 579 p91 peterson , b.m . 1991 , apj , 368 , 119 p78 press , w. 1978 , comments on astrophysics , 7 , 103 r84 rees , m.j . 1984 , araa , 22 , 471 r93 rokaki , e. , collin - souffrin , s. , & magnan , c. 1993 , a&a , 272 , 8 r92 rokaki , e. , & magnan , c. 1992 , a&a , 261 , 41 s89 sanders , d. b. , et al . 1989 , apj , 347 , 29 s78 shields , g. 1978 , nature , 272 , 706 sm89 sun , w .- h . , & malkan , m.a . 1989 , apj , 346 , 68 t89 tully , b. 1989 , nearby galaxy catalogue u91 ulrich , m .- h . , et al . 1991 , apj , 382 , 483 u93 urry , c. m. et al . 1993 , apj , 411 , 614 wds95 warwick , r.s . , done , c. , & smith , d.a . 1995 , mnras 275 , 1003 w96 warwick , r.s . 1996 , submitted to apj ( paper iii ) w94a weaver , k.a . et al . 1994a , apj , 423 , 621 w94b weaver , k.a . et al . 1994b , apj , 436 , l27 wp94 white , r.j . , & peterson , b.m . 1994 , pasp , 106 , 879 y89 yaqoob , t. , warwick , r.s . & pounds , k. 1989 , mnras , 236 , 153 y93 yaqoob , t. et al . 1993 , mnras , 262 , 435 y95 yaqoob , t. , edelson , r.a . , weaver , k.a . , warwick , r.s . , mushotzky , r.f . , serlemitsos , p.j . , & holt , s.s . 1995 , apj , 453 , l81 a90 zdziarski , a. , ghisellini , g. , george , i. m. , svensson , r. , fabian , a. c. , & done , c. 1990 , apj , 363 , l1 * figure 1 * continuum light curves during the 10 day intensive monitoring period , for wavebands centered on 100 kev , 1.5 kev , 1275 , 1820 , 2688 , and 5125 . the light curves are on a common scale , but shifted in flux to present them in one figure , so a 10% flux change is shown . the x - ray data showed the strongest variability but the poorest sampling ; the interruption in the second half of the campaign was due to a spacecraft malfunction . the apparent short flares in the 2688 light curve are probably due to instrumental effects , and not intrinsic variability . * figure 2 * plot of nva as a function of observing frequency . note the strong correlation in the optical / ultraviolet band , where the nva rises from @xmath01% in the optical to @xmath09% in the ultraviolet . the open circles refer to the total variability ( @xmath126 ) , and the filled circles are the nva ( @xmath24 ) , which represent the variability after correction for instrumental uncertainties . the strongest nva is seen in medium energy ( 12 kev ) x - rays , with an nva of 24% , while there is no significant variability observed in the softer x - ray bands , and weaker variations ( nva = 6% ) at 100 kev . * figure 3 * pds of the resampled ultraviolet light curves in four bands . the units of the pds ( co - ordinate axis ) are erg@xmath127 @xmath128 @xmath129 , while the units of the ordinate are day@xmath18 . the pds , which all appear the same to within the errors , show no signs of periodic variability . instead , they are well - fitted by a power - law with slope @xmath46 . * figure 4 * plots of the correlation between continuum fluxes at bands centered on 1275 and 100 kev , 1.5 kev , 1330 , 1820 , 2668 and 5125 . the ultraviolet and @xmath1-ray data are not significantly correlated , but all of the other bands show a significant correlation . the solid lines are unbiased least squares fits to the data . the regression lines with the long - wavelength ultraviolet have a positive y - intercept , while the ultraviolet1.5 kev x - ray regression has a positive x - intercept . * figure 5 * interpolated cross - correlation functions between 1275 and 1275 ( top panel ; autocorrelation ) , 1.5 kev ( second panel ) , 1820 ( third panel ) , 2688 ( fourth panel ) , and 5125 ( bottom panel ) . in all cases , the peaks are consistent with zero lag . * figure 6 * discrete cross - correlation functions between 1275 and 1275 ( top panel ; autocorrelation ) , 1.5 kev ( second panel ) , 1820 ( third panel ) , 2688 ( fourth panel ) , and 5125 ( bottom panel ) . in all cases , the peaks are consistent with zero lag . * figure 7 * spectral energy distribution of ngc 4151 . plotted quantities are monochromatic luminosity ( @xmath130 ) as a function of frequency . optical , ultraviolet , x - ray , and @xmath1-ray data are taken from observations made near mjd 26 . the asterisks show the rms variability in the observing bands listed in table 2 , in terms of monochromatic luminosity . lcccccccc + + & & & & & & & & & & & & & & & & + 50 - 150 kev & 9 & 8.10 & 8.9 & 6.4 & 6.1 & 119 & & 1 - 2 kev & 17 & 9.72 & 26.5 & 11.4 & 23.9 & 6.7 & & 0.5 - 0.9 kev & 17 & 9.72 & 3.5 & 6.6 & & & & 0.1 - 0.4 kev & 17 & 9.72 & 3.0 & 5.0 & & & & 1275 & 176 & 9.20 & 8.6 & 0.9 & 8.6 & 235 & 8.6 & 8.6 1330 & 176 & 9.20 & 8.2 & 1.0 & 8.2 & 241 & 8.1 & 7.4 1440 & 176 & 9.20 & 6.7 & 1.5 & 6.6 & 179 & 7.2 & 6.5 1730 & 176 & 9.20 & 6.1 & 1.4 & 6.0 & 172 & 5.5 & 4.1 1820 & 176 & 9.20 & 5.0 & 1.3 & 4.8 & 124 & 5.1 & 3.6 1950 & 176 & 9.20 & 4.9 & 2.0 & 4.5 & 106 & 4.5 & 3.1 2300 & 168 & 9.15 & 5.3 & 2.0 & 4.9 & 108 & 3.6 & 1.9 2688 & 168 & 9.15 & 3.9 & 1.1 & 3.7 & 89 & 2.7 & 1.3 4600 & 17 & 7.94 & 1.3 & 0.9 & 0.9 & 23 & 1.0 & 0.3 5125 & 18 & 9.33 & 1.1 & 0.9 & 0.7 & 14 & 0.9 & 0.2 6200 & 8 & 8.91 & 1.5 & 0.6 & 1.4 & 33 & 0.6 & 0.2 6925 & 8 & 8.91 & 1.6 & 0.7 & 1.4 & 36 & 0.4 & 0.1 + + lrccrr + + & & & & & + 100kev & 9 & 0.33 & 0.37 & 1.04 & 1kev & 17 & 0.60 & 0.56 & 2.64 & @xmath135 1330 & 176 & 0.96 & 0.95 & 38.11 & @xmath136 1820 & 176 & 0.90 & 0.86 & 22.54 & @xmath137 2688 & 168 & 0.69 & 0.70 & 12.47 & @xmath138 5125 & 18 & 0.65 & 0.59 & 2.95 & @xmath139 + + lccccc + + & & & & & & & & & & + @xmath59 & iccf & 0.82 & 0.87 & 0.70 & 0.69 & zdcf & 0.71 & 0.87 & 0.65 & 0.57 @xmath60 ( days ) & iccf & @xmath140 & @xmath141 & @xmath142 & @xmath143 & zdcf & @xmath144 & @xmath143 & @xmath145 & @xmath143 @xmath61 ( days ) & iccf & @xmath146 & @xmath147 & @xmath148 & @xmath149 & zdcf & @xmath150 & @xmath151 & @xmath152 & @xmath153 @xmath63 ( days)&zdcf & @xmath154 & @xmath155 & @xmath156 & @xmath157 + +

we have carried out vla zeeman observations of hi absorption lines toward the hii region in the m17 giant molecular cloud complex . the resulting maps have 60 spatial resolution and 0.64 km s velocity separation . the hi absorption lines toward m17 show between 5 and 8 distinct velocity components which vary spatially in a complex manner across the source . we explore possible physical connections between these components and the m17 region based on calculations of hi column densities , line of sight magnetic field strengths , as well as comparisons with a wide array of previous optical , infrared , and radio observations . in particular , an hi component at the same velocity as the southwestern molecular cloud ( m17 sw ) km s seems to originate from the edge - on interface between the hii region and m17 sw , in un - shocked pdr gas . we have detected a steep enhancement in the 20 km s hi column density and line of sight magnetic field strengths ( ) toward this boundary . a lower limit for the peak 20 km s hi column density is/ /k while the peak is g . in addition , blended components at velocities of 11 km s appear to originate from shocked gas in the pdr between the hii region and an extension of m17 sw , which partially obscures the southern bar of the hii region . the peak/ and for this component are /k and g , respectively . comparison of the peak magnetic fields detected toward m17 with virial equilibrium calculations suggest that of m17 sw s total support comes from its static magnetic field and the other 2/3 from its turbulent kinetic energy which includes support from alfvn waves .

astro-ph9711014

in recent years it has become increasingly clear that magnetic fields play an important role in the process of star formation . ( see , for example , mouschovias & spitzer 1976 ; heiles et al . 1993 ; mckee et al . unfortunately , the experimental techniques for measuring the strength and direction of magnetic fields are few and these methods are observationally challenging . one such technique uses the zeeman effect in 21 cm hi absorption lines toward galactic hii regions . very large array ( vla ) observations of the zeeman effect in this line yield maps of the line - of - sight magnetic field strengths ( @xmath3 ) . these maps can then be compared to maps of the distribution and kinematics of ionized , atomic , and molecular gas to study the role of magnetic fields in star forming regions . moreover , estimates of the masses , column densities , and volume densities of interstellar material associated with the star - forming regions can be compared to field strengths measured via the zeeman effect . these comparisons are needed to determine the energetic importance of magnetic fields in star - forming regions . for studies of this type , the m17 hii region@xmath9molecular cloud complex is ideal because it has been extensively studied at many wavelengths . the m17 hii region is one of the strongest thermal radio sources in our galaxy . its distance has been estimated through photometric and kinematic means to be @xmath14 2.2 kpc ( chini , elssser , & neckel 1980 ; reifenstein et al . 1970 ; wilson et al . there are at least 100 stars in the m17 hii region with one o4 v ( kleinmann s star ) and three o5 v stars providing the bulk of the ionizing radiation ( felli , churchwell , & massi 1984 and references therein ; see fig . the hii region is made up of two distinct bar like structures which are @xmath15 pc long and @xmath16 to 1.5 pc wide . the southern bar has nearly twice the peak radio continuum flux as the northern bar , and is located just to the east of the m17 sw molecular cloud core . the near absence of optical radiation from the southern bar indicates that this region suffers much more optical extinction than the northern bar ( gull & balick 1974 , fig 1 ) . dickel ( 1968 ) and gately et al . ( 1979 ) estimate @xmath17 1 - 2 mag toward the northern bar . estimates of the visual extinction toward the southern bar range from @xmath18 near the radio continuum peak to @xmath19 further west toward the core of the adjacent molecular cloud ( m17 sw ) ( beetz et al . 1976 ; felli et al . 1984 ; gatley et al . 1979 ; thronson & lada 1983 ) . the low extinction toward the northern bar seems to indicate that the northern molecular cloud , first identified by lada et al . ( 1976 ) ( @xmath20 km s@xmath1 ) , is behind the hii region ( chrysostomou et al . 1992 ) . the most interesting characteristic of m17 is that the interface between the southern bar of the hii region and m17 sw is seen almost edge on ( fig . 1 ) . from models of the density gradient on the western edge of the hii region , icke , gatley , & israel ( 1980 ) estimated that this interface lies at an angle of @xmath21 with respect to the line of sight . although the transition from the hii region to the molecular cloud is quite sharp , the interface region is too wide in photodissociation region ( pdr ) tracers such as [ cii ] 158 @xmath8 m unless it is clumpy ( see eg . , stutzki et al . 1988 ) . the extended [ cii ] 158 @xmath8 m emission observed in m17 sw by stutzki et al . ( 1988 ) and boreiko , betz , & zmuidzinas ( 1990 ) requires 912 - 2000 uv photons to persist into the molecular cloud at least an order of magnitude further than the predicted uv absorption length scale . that is , atoms are being photoionized and molecules photodissociated further into the molecular cloud than one would expect for a homogeneous medium . ( see tielens & hollenbach 1985 for homogeneous model calculations . ) a possible explanation for this paradox is that the interface region is clumpy . evidence that m17 sw is clumpy has been observed with high angular resolution at many = 13.0 cm wavelengths . for example , felli et al . ( 1984 ) , found clumpiness in their @xmath22 resolution radio continuum observations of the m17 hii region . they suggest this clumpiness originates from high density neutral clumps surrounded by dense ionized envelopes . these clumps may be the remnants of clumps from the parent molecular cloud which have been overtaken by the hii region ionization front . massi , churchwell , & felli ( 1988 ) observed seven clumps in nh@xmath23 emission with 6 resolution toward the northwestern portion of the hii region@xmath9m17 sw interface . the average density and size of these clumps are @xmath24 @xmath25 and 0.05 pc . stutzki & gsten ( 1990 ) modeled their @xmath26 resolution c@xmath27o ( @xmath28 ) observation ( with a larger field of view than the nh@xmath23 observation ) with @xmath29 clumps with average densities of @xmath30 @xmath31 @xmath25 and diameters of @xmath14 0.1 pc . hobson ( 1992 ) also saw evidence for clumpiness in hco@xmath32 and hcn molecules at @xmath33 resolution , and wang et al . ( 1993 ) found clumps consistent with the c@xmath27o clumps in @xmath34 to @xmath35 resolution observations of cs and c@xmath36s molecules . the identification of clumps in m17 sw is significant for several reasons . for example , calculations of density are more complex due to its dependence on the optical depth of the observed species , and the assumed clump filling factor . as discussed previously , the lengthscale of the pdr also depends on the clumpiness of the region . therefore , the regions and morphology of photodissociated atomic gas are difficult to predict . also the random motion of such clumps can influence observed linewidths and the sizescale of turbulence within the molecular cloud . these considerations play an important role in our analysis of the hi gas and the line of sight magnetic fields calculated from it . in this paper we report the characteristics of the 21 cm continuum ( 3.1 ) , hi optical depths and column densities ( 3.2 ) , and the line of sight magnetic field strengths ( 3.3 ) derived from our vla hi zeeman effect observation . we discuss possible origins for the most prominent hi absorption components in 4.1 . in addition , we present comparisons of our 20 km s@xmath1 @xmath3 map with previous linear polarimetry observations in 4.2 , and we also compare expected magnetic field strengths from virial equilibrium arguments to our observed line of sight magnetic field strengths in 4.3 . our findings are summarized in 5 . our hi absorption data were obtained with the dnc configuration of the vla . key parameters from this observation can be found in table 1 . we observed both senses of circular polarization simultaneously . since the zeeman effect is very sensitive to small variations in the bandpass , we switched the sense of circular polarization passing through each telescope s if system every 10 minutes with a front - end transfer switch . in addition , we observed each of the calibration sources at frequencies shifted by @xmath37 1.2 mhz from the hi rest frequency to avoid contamination from galactic hi emission at velocities near those of m17 . the calibration , map making , cleaning , and calculation of optical depths were all carried out using the aips ( astronomical image processing system ) package of the nrao . the right ( rcp ) and left ( lcp ) circular polarization data were separately calibrated and mapped using natural weighting . these two data cubes were then combined to create stokes i ( rcp + lcp ) and stokes v ( rcp @xmath9 lcp ) cubes . the i and v cubes were cleaned using the aips task sdcln down to 70 mjy beam@xmath1 . bandpass correction was applied only to the i data since bandpass effects subtract out to first order in the v data . subsequent antenna leakage correction and magnetic field derivations were carried out using the miriad ( multichannel image reconstruction image analysis and display ) processing package from bima . figure 1 shows our 21 cm continuum contours with an optical image from gull & balick ( 1974 ) and simplified @xmath38co contours from thronson & lada ( 1983 ) superimposed . the total continuum flux at 21 cm is @xmath39 jy calculated inside the 1 jy beam@xmath1 ( 250 k ) contour level ( lowest contour in fig . 1 ) . this value is almost twice that obtained by lockhart & goss ( 1978 ) with a 2 beam at the owens valley interferometer . in addition , we estimate the total flux of the southern bar is 384 jy while that of the northern bar is 240 jy ( calculated for points inside the 5 jy beam@xmath1 contour level ) . our total flux inside the 5 jy beam@xmath1 contour ( 624 jy ) is similar to the 644 jy reported by lada et al . ( 1976 ) with the haystack 120 ft telescope and 620 jy observed by lbert & goss ( 1978 ) with the fleurs synthesis telescope at 21 cm . m17 hi optical depths were calculated assuming that the spin temperature @xmath10 is much less than the background continuum temperature @xmath40 of the hii region . in addition , optical depth calculations were restricted to positions where the continuum power is greater than 1 jy beam@xmath1 ( 250 k ) and the signal - to - noise ratio is higher than three . the complex nature of the absorption lines in m17 can be seen from the optical depth spectra displayed in figure 2 . the strengths of some components vary spatially over the source but remain distinct , while other components seem limited to particular regions . this complexity makes gaussian fitting very difficult . our profiles agree qualitatively with the gaussian fits reported by lockhart & goss ( 1978 ) made with a @xmath41 beam and 0.85 km s@xmath1 velocity resolution . they fit their m17 hi optical depth profiles with eight gaussians having center velocities ranging from 4.2 to 27.5 km s@xmath1 . in the following analysis we concentrate on the hi velocity component near 20 km s@xmath1 and the blended components between 11@xmath917 km s@xmath1 due to their high optical depths and their coincidence in velocity with other species observed in m17 sw ( see 4.1 for details ) . figure 3 displays the morphology of the hi optical depths toward the m17 hii region as a function of velocity ( every two adjacent channels were averaged ) . the steep increase of the hi optical depth toward the hii region@xmath9molecular cloud interface is clearly apparent . this effect is particularly noticeable near 20 km s@xmath1 . although the hi optical depths become saturated at the western boundary of the hii region ( particularly in the northwest ) , they reach values at least as high as 5 . optical depth maps in the 11@xmath917 km s@xmath1 range have their highest values further to the east and in the northern part of the source . the column density of hi toward m17 was found using the relationship @xmath42 where @xmath43 is the optical depth per unit frequency . values for @xmath4/@xmath10 summed across the entire hi velocity range , the 20 km s@xmath1 component ( from 17 to 24.5 km s@xmath1 ) and the 11@xmath917 km s@xmath1 blended component are displayed in figs . 4a , 4b , and 4c , respectively . at positions along the northwest and southwest portions of the source , these maps represent lower limits to @xmath4/@xmath10 owing to saturation effects . note the increase in the hi column density toward the hii region@xmath9molecular cloud interface on the southwestern side of the maps . there is also spatial agreement between the hi column density concentration prominent in the total @xmath4/@xmath10 map ( fig . 4a ) near @xmath44 , @xmath45 and the northern condensation seen = 9.5 cm = 9.0 cm = 16.0 cm in molecular gas ( see wang et al . 1993 and fig . the total @xmath46 at this position is 8.3 @xmath47 @xmath6/k . aside from line saturation , the main source of uncertainty in calculating hi column densities is in the value assumed for @xmath10 . upper limits to @xmath10 come from two sources . for one , the hi line widths in some cases are quite narrow . toward the southern end of the molecular ridge defined by cs emission ( wang et al . 1993 ) they are as little as 2.8 km s@xmath1 for the 20 km s@xmath1 component ( figs . 8 , 9 ) . this width implies a maximum temperature of 180 k at this position . also , along the periphery of the continuum source , hi absorption lines exist where the continuum brightness is as little as 250 k ( 1 jy beam@xmath1 ) . evidently , @xmath10 is less than about 200 k in many directions toward m17 . at the position illustrated in fig . 9 , the cs line has a width of about 2 km s@xmath1 . if hi and cs are mixed and at a common temperature , then @xmath48100 k. in fact , modeling of iras 12 @xmath8 m infrared data byhobson & ward - thompson ( 1994 ) require a hot dust component of @xmath14108 k to explain the overall 12 @xmath8 m emission from m17 . this hot dust can be linked to the pdr by the fact that warm emission at 12 @xmath8 m is likely to arise from a population of polycyclic aromatic hydrocarbons ( pahs ) ( dsert , boulanger & puget 1990 ) , and pah emission from pdr regions has been predicted by many authors ( e.g. hollenbach 1993 ) . another indication of temperature in the atomic gas comes from the observations of [ ci ] 370 @xmath8 m by genzel et al . they estimate a temperature of 50 k for [ ci ] 370 @xmath8 m gas near 20 km s@xmath1 . molecular temperature determinations may also be relevant if some of the hi is mixed with molecular gas ( 4.2 ) . bergin et al . ( 1994 ) estimate @xmath49 = 50 k from observations of low-@xmath50 co transitions , while harris et al . ( 1987 ) show that high @xmath50 co lines ( @xmath51 ) arise from pdr gas at about 250 k. meixner et al . ( 1992 ) found that the spatial extent and intensities of a wide array of far - infrared and submillimeter cooling lines could be adequately modeled by a three component gas consisting of very dense clumps @xmath52 @xmath25 , intermediate density interclump gas @xmath53 @xmath25 , and a tenuous extended halo @xmath54 @xmath25 . they obtained the best agreement with observations by assuming an interclump gas kinetic temperature of @xmath55 k where it is also likely that dissociated hi gas can be found . almost certainly , @xmath10 in the hi gas is quite variable , with values in the range 50 - 200 k most common . as described above , it is likely that no one spin temperature is applicable to all of the hi gas components seen in m17 . with a spin temperature of 50 k , a lower limit to the peak @xmath56 column density toward m17 is @xmath57 @xmath6 . spin temperatures between 50 k and 200 k yield peak column densities of @xmath58 @xmath6 for the 20 km s@xmath1 component and @xmath59 @xmath6 for the 11@xmath917 km s@xmath1 blended components . we can use such estimates of the hi column density to estimate what fraction of the total hydrogen column density arises from atomic gas . for example , felli et al . ( 1984 ) used the optical depth of the silicon absorption feature at 10 @xmath8 m to estimate that @xmath60=30 toward the ultra compact hii region m17-uc1 ( @xmath61 ) . using @xmath62 ( bertoldi & mckee 1992 ) , we estimate that the total column density of hydrogen is @xmath63 @xmath6 . our _ total _ hi column density toward m17-uc1 is @xmath64 @xmath6 ( @xmath10=50 k ) . unless saturation effects have resulted in a severe underestimate of @xmath4 , much of the gas along the line of sight to m17@xmath9uc1 is in the form of h@xmath65 . the hi zeeman effect is only sensitive to the line of sight magnetic field ( @xmath3 ) in most astrophysical situations since the zeeman splitting is a tiny fraction of the line width . since v@xmath66i@xmath67 , values for @xmath3 were obtained by fitting the derivative of the i profile to the v profile at each pixel with a least squares fitting routine ( for details see roberts et al . 1993 , crutcher et al . due to the complexity of the hi spectra , the number of channels fit for each velocity component is small . to obtain more realistic error estimates , forty channels outside of the line region were included in each fit . the resulting @xmath3 maps were subsequently masked where the continuum power is less than 1 jy beam@xmath1 and when @xmath3/@xmath68(@xmath3 ) @xmath69 3 . the quantity @xmath3/@xmath68(@xmath3 ) will subsequently be denoted @xmath70 . observations of m17 have been made for many astrophysically relevant molecular and atomic transitions . from these data it is known that m17 sw has a lsr velocity of 19 to 21 km s@xmath1 which corresponds to the velocity of one of our strongest hi absorption components . figure 5 shows an hi optical depth profile and several molecular emission profiles from stutzki et al . ( 1988 ) for a position at the hii region@xmath9molecular cloud interface . note the similarity in the center velocities ( @xmath14 20 km s@xmath1 ) of each species . this velocity coincidence strongly suggests that the 20 km s@xmath1 hi absorption component is closely associated with the molecular cloud , most likely arising in the pdr at the hii region - molecular cloud interface . further evidence for this association comes from the steep rise in the 20 km s@xmath1 hi optical depth toward the interface region ( fig . therefore , values of @xmath3 derived from the hi zeeman effect in this component are of direct relevance to the energetics of the molecular cloud and its immediate surroundings ( see4.1.2 ) . a significant magnetic field exists in the 20 km s@xmath1 component over much of the central region of the southern bar . the velocity range of the fit for this component is 17.0 to 24.6 km s@xmath1 . a distinct hi component at @xmath14 23 km s@xmath1 is also visible toward the northwestern corner of the source . a component at this velocity was also detected at this velocity by lada & chaisson ( 1975 ) in h@xmath65co . however , detailed inspection of the hi data cube shows that the channel range used to fit the 20 km s@xmath1 component was such that no magnetic field was detected in regions where the 23 km s@xmath1 component is distinct . therefore , the @xmath3 detected in the velocity range 17 to 24.6 km s@xmath1 is only representative of the field in the 20 km s@xmath1 hi component . a map of @xmath3 for the 20 km s@xmath1 component is shown in fig . 6 , and sample fits at representative pixels are shown in figs . 7a , 7b , and 7c . negative values indicate that @xmath3 is toward the observer . values of @xmath3 for this component ( with sufficiently high @xmath70 to pass our cutoff of 3 ) are centered near the continuum peak of the southern bar . the values for @xmath3 start out at @xmath71100 @xmath8 g on the eastern side of the southern bar , pass through a shallow saddle point and then rapidly rise to @xmath71450 @xmath8 g on the western side of the map near the hii region@xmath9molecular cloud interface . the maximum @xmath70 obtained for this component is 8.5 . the apparently rapid change in direction of the @xmath3 along the narrow band in the northwest portion of the map may be real but is approaching the limit of our sensitivity and will require future observation to confirm . note that the region of highest measured @xmath3 coincides closely with the ridge of cs emission observed by wang et al . ( 1993 ; fig . 8) . also , the 20 km s@xmath1 hi line in this region is very narrow , and its width and center velocity correspond closely to those of the cs lines ( fig . this velocity correspondence , like those of fig . 5 , provide further evidence of a close association between the 20 km s@xmath1 hi gas and the molecular gas . in addition , the narrowness of the lines in this region seem to exclude the possibility of the ridge being post - shock gas . the nature of the hi gas at 20 km s@xmath1 and its effect on the morphology of @xmath3 will be discussed in 4.1.2 . = 10.0 cm = 11.0 cm = 18.0 cm = 13.7 cm = 9.0 cm strong , primarily positive magnetic fields were also detected in blended velocity components between 11 to 17 km s@xmath1 . a map of this magnetic field is shown in fig . 10 and the fit to a few representative pixels can be seen in figs . 11a , 11b , and 11c . there are four main regions of significant @xmath3 for this blended component which appear primarily in the northeastern , north central and western parts of the map . the center velocity of this component ( defined by the velocity at which the derivative changes sign ) is somewhat different in each of these regions . they range from 14.4 km s@xmath1 in the northeast ( fig . 11a ) , 13.7 km s@xmath1 in the central northern feature ( fig . 11b ) , 14.0 km s@xmath1 in the northwest ( fig . 11c ) and 12.4 km s@xmath1 in the southwest . one interesting feature of the @xmath3 for this component is that the direction of the field changes from positive ( away from the observer ) to negative and back to positive across the map from east to west ( this trend can be seen in the fits shown in figs . 11 a , b , c ) . @xmath3 in the northwestern feature reaches values as high as + 550 @xmath8 g with @xmath70 as high as 14 . evidence for molecular gas near this velocity has been observed in a number of species . for example greaves , white , & williams ( 1992 ) reported a velocity component in this range in their @xmath38co @xmath72 and @xmath73co @xmath72 spectra . their component shifts from 12 km s@xmath1 in the southeast to 16 km s@xmath1 further to the southwest of their field of view ( the eastern portion of their maps overlaps the the western side of our maps ) . the positions where their low velocity component is centered at 12 km s@xmath1 corresponds to the southwestern region of our magnetic field map where the hi line center is 12.4 km s@xmath1 . rainey et al . ( 1987 ) and stutzki et al . ( 1988 ) also identified emission in this velocity range in @xmath38co @xmath72 and @xmath38co @xmath74 , respectively . massi et al . ( 1988 ) observed that two of their seven nh@xmath23 clumps near m17-uc1 have velocities of 15.5 km s@xmath1 and stutzki et al . ( 1988 ) observed 31 c@xmath27o clumps ( out of 180 ) in the 11@xmath917 km s@xmath1 velocity range . in addition to molecular gas , there is also [ cii ] 158 @xmath8 m emission at a velocity of @xmath14 17.8 km s@xmath1 ( boreiko et al . [ cii ] 158 @xmath8 m is an important tracer of pdr s and should exist in the same zone of a pdr as hi gas . the precise spatial correlation between these two atomic species for m17 is unknown . however , the low resolution ( 3.4 ) map of [ cii ] 158 @xmath8 m obtained by matsuhara et al . ( 1989 ) does seem to follow the same morphology as our summed n@xmath75/t@xmath76 ( see fig . 4a ) . a profile from one pixel at the [ cii ] 158 @xmath8 m peak from boreiko et al . ( 1990 ; spatial resolution 43 ) and hi are shown in fig . the [ cii ] 158 @xmath8 m profile appears to have a component at a lower velocity than their lowest reported velocity of 17.8 km s@xmath1 . the authors noted that there may have been some contamination from emission in the reference beam at this position . the possible significance of a lower velocity [ cii ] 158 @xmath8 m component will be discussed again in 4.1.3 . line of sight magnetic field detections were also made for three other velocity components . a significant @xmath3 was measured in the velocity range @xmath92.4 to 4.0 km s@xmath1 with the line center at @xmath77 km s@xmath1 . this detection is interesting because the @xmath70 in the southeastern region is as high as 10 and there is a rapid change in @xmath3 from negative to positive values along the northwestern border with m17 sw . this abrupt change in field direction corresponds to the behavior of the @xmath3 measured in the 20 km s@xmath1 component in this same region . the line of sight fields in this component range from @xmath14 @xmath9200 to + 100 @xmath8 g . detections were also made in the velocity range from 4.1 to 10.5 km s@xmath1 with a center velocity of @xmath78 km s@xmath1 and the maximum = 12.0 cm = 18.0 cm @xmath79 . reliable measurements of @xmath3 in this hi component range from @xmath14 @xmath9125 to + 100@xmath8 g . the regions of significant @xmath3 measured for this component do not particularly correspond to the morphology of the 21 cm continuum emission of m17 . in fact the highest fields and concentration of hi in this velocity range are to the southeast . one interesting possibility is that there is a connection between hi measured in this velocity range and the 7 km s@xmath1 and the cold cloud observed by riegel & crutcher ( 1972 ) at 21 cm over a large region toward the galactic center including the direction of m17 . crutcher & lien ( 1984 ) estimate that this cold cloud ( @xmath1420 k ) must be located within 150 pc of the sun . a sparse @xmath3 was also detected in the range 24.7 to 33.0 km s@xmath1 with a center velocity of @xmath80 km s@xmath1 and maximum @xmath70=6 . the @xmath3 in this hi component ranges from @xmath14 @xmath9200 to + 100 @xmath8 g . a component in this velocity range was also observed by greaves et al . 1992 in @xmath38co @xmath72 and @xmath73co @xmath72 emission . the ratio of these two species intensity seems to indicate the presence of a distinct cloud at this velocity . meixner et al . ( 1992 ) suggest that the coincidence of part of their [ oi ] 63 @xmath8 m map with the radio continuum emission indicates that hot ionized gas from the hii region is carving a bowl into the molecular cloud to the southwest ( m17 sw ) . the symmetry axis of the bowl lies in the plane of the sky , and the overlapping [ oi ] 63 @xmath8 m emission arises from a pdr on the front and/or back sides of the bowl . the bottom of the bowl ( at the southwestern edge of the southern bar ) is the portion of the hii region - molecular cloud interface viewed edge@xmath9on . further evidence of a bowl@xmath9like morphology is the high visual extinction toward the southern bar which presumably arises from the front side of the bowl . gull & balick ( 1974 ) also suggest a bowl morphology based on the double peaks separated by @xmath14 20 km s@xmath1 in their h76@xmath81 radio recombination line data . ( also , see goudis & meaburn 1976 . ) these two recombination line velocity components are thought to arise from streaming motions of newly - ionized gas at the front and back sides of the bowl . in many ways , this m17 interface model resembles the interface region between the orion nebula and the molecular cloud behind it . for orion , the interface lies in the plane of the sky with the ionizing stars directly in front of it . for m17 , the geometry is turned by 90@xmath82 . the interface layer is perpendicular to the plane of the sky , with the ionizing stars offset to the northeast . the densest accumulation of ionized gas in m17 lies along the `` working surface '' near the ionization front . therefore , the brightest region in the m17 continuum map , the southern bar , does not coincide in the sky with the ionizing stars as it does for the orion nebula . the clumpiness of the m17 interface has been observed in many molecular species ( 1 ) . it seems to closely resemble the scenario discussed by bertoldi & draine ( 1996 ) , whereby dense clumps are slowly photodissociated by a nearby hii region . in this case the hii region is partially embedded in ( and partially obscured by ) a bowl of clumpy neutral gas . the hi gas associated with m17 can arise , in principle , anywhere in front of the source . our data suggest that the 20 km s@xmath1 component originates primarily in the edge@xmath9on portion of the bottom of the molecular bowl . in this model the 20 km s@xmath1 hi arises from molecular gas that been partially photodissociated but otherwise dynamically undisturbed ( unshocked ) . on the other hand , the hi data for the 11@xmath917 km s@xmath1 component supports the idea that it arises from photodissociated , shocked gas from the front side of the molecular bowl which is streaming toward us . the association of different hi absorption components with shocked and unshocked gas has also been suggested for s106 ( roberts et al . 1995 ) and w3 ( roberts et al . evidence supporting this picture for m17 is given in 4.1.2 and 4.1.3 . the velocity range of the 20 km s@xmath1 hi component ( 17 to 24.6 km s@xmath1 ) includes most of the molecular gas in the m17 region ( e.g rainey et al . 1987 and 3.3.1 ) . therefore , hi components in this range are likely to be closely associated with the bulk of the m17 molecular gas . such an association is particularly clear toward the molecular ridge ( see fig . 8) where the center velocity of the 20 km s@xmath1 hi component closely matches that of a variety of molecular lines , including high density tracers such as cs ( figs . 5 , 9 , and 12 ) . this hi component is also relatively narrow ( 3 - 5 km s@xmath1 ) , and its center velocity changes little with position along the interface region . therefore , it most likely originates in quiescent gas that is dynamically unaffected by the hii region . much of this quiescent gas must lie in the edge - on hii region@xmath9m17 sw interface region at the bottom of the bowl described by meixner et al . hence the concentration of 20 km s@xmath1 hi gas and high @xmath3 ( @xmath83 @xmath8 g ) near the interface region ( figs . 3 , 4b , 6 , and 8) . further indication of the association between hi and molecular gas comes from similarities in their spatial distribution near the molecular ridge . 13 ( upper panel ) shows the relative intensity of several species along the ne - sw strip scan defined by stutzki et al . the strip lies perpendicular to the interface region and cuts across it ( ionizing radiation propagates from left to right ) . as shown in the figure , the increase in the 20 km s@xmath1 @xmath46 and @xmath3 ( lower panel ) matches well the increase in @xmath38co line strengths to the right of the hii region . in fact , @xmath46 , @xmath38co@xmath74 , and @xmath38co@xmath84 peak _ outside _ of the hii region as expected for an interface region viewed edge on . note that the actual peak in @xmath46 for the 20 km s@xmath1 component is not defined by our data since we can not detect hi absorption beyond the continuum source . notice that the integrated [ cii ] 158 @xmath8 m emission , also shown in fig . 13 , peaks _ inside _ ( to the left of ) the rise in the 20 km s@xmath1 @xmath46 . in a simple planar pdr model , [ cii ] 158 @xmath8 m emission coincides with hi and with molecular gas up to @xmath85 mag . ( see review byhollenbach , 1990 . ) however , the higher velocity resolution [ cii ] 158 @xmath8 m profiles of boreiko et al . ( 1990 ) reveal the presence of several [ cii ] 158 @xmath8 m velocity components . an examination of these profiles suggests that the @xmath14 20 km s@xmath1 [ cii ] 158 @xmath8 m component alone peaks further into the molecular cloud than the integrated [ cii ] 158 @xmath8 m emission plotted in fig . also , note that the rise in the 11@xmath917 km s@xmath1 @xmath4/@xmath10 , plotted in fig . 13 ( lower panel ) , peaks to the left of the 20 km s@xmath1 @xmath46 in a similar manner to the integrated [ cii ] 158 @xmath8 m emission . in fact , the contribution of lower velocity [ cii ] 158 @xmath8 m emission to the integrated emission plotted in fig . 13 may explain why it seems to peak to the left of the 20 km s@xmath1 @xmath46 . apparently , hi and [ cii ] 158 @xmath8 m gas is more broadly distributed along the strip in fig . 13 than is the gas at 20 km s@xmath1 alone . this shows that absorbing hi gas lies on the near side of the bowl , not just at the bottom . the close association between hi gas near 20 km s@xmath1 and the molecular ridge region could arise in at least two ways . for one , the hi may be well mixed with and a minor constituent of the molecular gas . in such a case , the zeeman effect near the ridge ( fig . 8) is directly representative of magnetic fields in the molecular gas . a second possibility is that the absorbing hi near 20 km s@xmath1 is confined to a thin photodissociated shell about the molecular ridge or to thin shells surrounding a multitude of molecular clumps within the ridge . in this case , the hi zeeman effect is also representative of the magnetic field in the molecular gas if the hi is photodissociated but dynamically undisturbed h@xmath65 . in either case , the hi zeeman effect near the = 6.8 cm = 8.0 cm molecular ridge must arise in relatively dense gas if approximate equilibrium exists between the magnetic energy density and that associated with non - thermal motions in the gas . such an equilibrium was suggested by myers & goodman ( 1988a , 1988b ) on the basis of limited observational data . in such a case , the gas density is given by @xmath86 ^ 2 { \rm cm}^{-3},\ ] ] where @xmath87 is the proton density , @xmath88 is the average field strength ( @xmath8 g ) , and @xmath89 is the fwhm contribution to the line width from non - thermal motions ( km s@xmath1 ; see also 4.3 ) . as an example we apply this equation to the position of fig . 5 ( @xmath90 east of molecular ridge ) . at this position , @xmath91 km s@xmath1 , and @xmath88 200 @xmath8 g . if @xmath92 k ( as estimated by bergin et al . 1994 from @xmath38co observations ) , @xmath89 = 3.7 km s@xmath1 and @xmath93 @xmath25 . this estimate of the local density in which the hi zeeman effect arises is conservative since the actual field strength @xmath88 almost certainly exceeds @xmath3 and @xmath94 . this value for @xmath87 based on @xmath3 is also in excellent agreement with the value of @xmath95 @xmath25 calculated toward m17 sw by goldsmith , bergin , & lis ( 1997 ) using c@xmath27o column densities and average line widths . however , at the position cited here , @xmath46 in the 20 km s@xmath1 component is only 3 @xmath47 @xmath6k , so @xmath96 @xmath6 for @xmath9750 k. a typical scale length for hi absorption features in the plane of the sky is of order 2 pc ( 3@xmath98 at 2.2 kpc ) . if the same scale length applies along the line of sight , then @xmath99 @xmath100 250 @xmath25 . that is , if the hi zeeman effect arises in gas at a density of order @xmath87 , then the hi could be about only a 1% constituent of largely molecular gas . this possibility exists because the @xmath101 estimated by stutzki et al . ( 1988 ) and goldsmith et al . ( 1997 ) near this direction ( @xmath102 @xmath6 ) is about 100 times higher than @xmath4 . alternatively , hi might not be mixed with h@xmath65 but confined to thin sheets or very dense , thin atomic envelopes surrounding the multitude of molecular clumps which have been observed in the interface region ( see 1 ) . note that this would still require the hi gas to occupy about 1% of the line of sight in the absorbing region . these two cases are the same ones cited above to explain the close association of hi gas with molecular gas in the ridge . the morphology and velocity structure of gas between 11@xmath917 km s@xmath1 suggest that it also lies very close to the hii region on the near side of the bowl and that it has been shocked and accelerated away from the ionized gas . the 11@xmath917 km s@xmath1 gas is primarily distributed in an arc that partially encircles the northern and southern bars of the hii region ( figs . 3 , 4c ) . this arc of hi column density corresponds well with a similar arc structure apparent in @xmath38co(@xmath1032 ) emission mapped by rainey et al . ( 1987 ) in the velocity range 14@xmath920 km s@xmath1 . this morphology suggests a very close relationship between gas in the 11@xmath917 km s@xmath1 velocity range and the hii region . the gas is blue - shifted by 2 - 8 km s@xmath1 relative to the hii region center velocity of 19 km s@xmath1 ( joncas & roy 1986 ) . also , hi in this velocity range shows complex velocity structure with numerous components appearing at various positions across the source ( figure 2 , 11 a , b , c ) . this complicated velocity field and magnetic field reversal ( see 3.3.2 ) suggest that gas in the 11@xmath917 km s@xmath1 range lies in an interface zone that has been dynamically affected by interactions with the hii region , unlike the 20 km s@xmath1 component discussed in 4.1.2 . overall , the association between atomic and molecular gas in this velocity range is much less clear than it is at 20 km s@xmath1 near the western interface region . this could in part be due to fact that few high resolution molecular observations have mapped the northern regions of m17 . however , judging from @xmath38co @xmath1040 , @xmath38co@xmath1052 , and @xmath73co observations ( lada 1976 ; rainey et al . 1987 ) which all show components in this velocity range , the density of molecular material in this region is low compared to the western interface . toward the northern bar in particular , where the 11@xmath917 km s@xmath1 hi magnetic fields are high ( see 3.3.2 ) , there is not always a molecular counterpart in the 11@xmath917 km s@xmath1 velocity range . in fact , for the positions shown in figs . 11a and 11b there is no evidence of molecular gas . such a variable ratio of atomic to molecular gas might be expected if gas in this velocity range has been subjected to a widely variable degree of dissociation in the interface region . although molecular column densities are low in the 11@xmath917 km s@xmath1 range , hi field strengths are very high , reaching values of over + 500 @xmath8 g ( fig . therefore , the assumption of equilibrium between magnetic and non - thermal energies leads to conclusions similar to those for gas near 20 km s@xmath1 ( 4.1.2 ) . that is , @xmath106 @xmath25 , and @xmath99 along the line of sight is about two orders of magnitude less than @xmath87 . yet the scarcity of molecular gas at positions of high field strength means that hi at these positions can not exist as a minor constituent in a largely molecular region . if the hi alone is to exist at a density near @xmath107 @xmath25 , it must reside in one or more thin layers that occupy only about 1% of the path length through the hi absorbing region . that is , the absorbing hi gas is distributed in a very clumpy fashion along the line of sight . this conclusion is consistent with numerous other indications of a highly clumped medium in the m17 region ( 1 ) . most likely , the absorbing hi gas is shocked , photodissociated gas lying in thin interface regions about ablating neutral clumps . the clumps , in turn , lie within mostly ionized gas , like the clumps modeled by bertoldi & draine ( 1996 ) or the clumps identified by felli et al . ( 1984 ) in high resolution m17 radio continuum maps . in these clumps , the pressure of the ionized gas drives a shock front into the neutral medium . therefore , a rough equivalence between the thermal energy density of the ionized gas and the magnetic energy density in the adjacent compressed atomic region is expected . evidently , this is the case . if the m17 ionized gas has a temperature of 8000 k ( subrahmanyan & goss 1996 ) and a density of @xmath107 @xmath25 ( felli et al . 1984 ) , then its thermal energy density is equivalent to the magnetic energy density in the thin hi layers for b @xmath100 500 @xmath8 g . several reasons exist to expect hii region driven motions of a few km s@xmath1 near m17 . observationally , rainey et al . ( 1987 ) fit their @xmath73co data on molecular clumps near the southern bar to a geometric expansion model about kleinemann s star ( fig . 1 ) . they obtain an expansion velocity of @xmath100 11 km s@xmath1 about this star . radio and optical recombination line observations often reveal split profiles shifted by about @xmath108 km s@xmath1 from the hii region rest velocity ( gull & balick 1974 ; joncas & roy 1986 ) . theoretically , bertoldi & drain ( 1996 ) estimated the shock - induced acceleration expected in the orion nebula pdr , obtaining a value of 3 km s@xmath1 . this value is somewhat sensitive to the ratio of post to pre - shock density , taken by bertoldi & draine ( 1996 ) as 2 . if the ratio for m17 is 4 , as suggested by the ratio of the square of the @xmath38co @xmath109 ( 5 km s@xmath1 ) line width ( harris et al . 1987 ) to the square of the cs @xmath110 line width ( 2.5 km s@xmath1 , wang et al . 1993 ) , then the shock induced acceleration is increased to 6 km s@xmath1 . also , bertoldi ) and bertoldi & mckee ( 1990 ) discuss a rocket effect upon ablating clumps of neutral gas exposed to ionizing radiation . they estimate that rocket acceleration should amount to a clump velocity of 5@xmath910 km s@xmath1 under typical circumstances . these estimates of shock speed , agree quite well with the 2@xmath98 km s@xmath1 blueshift of the 11@xmath917 km s@xmath1 componets . a number of studies have been made of linear polarization toward m17 . these include polarization at optical and near ir wavelengths from grain absorption and polarization in the far ir from grain emission . ( see references cited by dotson 1996 . ) such studies reveal the position angles of the magnetic field in the plane of the sky ( @xmath111 ) , but they yield little if any information about field strengths . therefore , they compliment zeeman effect studies , assuming the linear and circular polarization arise in the same regions . in general , field directions inferred from grain absorption do not agree with those inferred from grain emission . very likely , this disprepancy results from the different regions sampled by each technique , with grain emission data more likely representative of the field directions inside m17 sw ( goodman et al . 1995 ) . the study of polarized 100 @xmath8 m dust emission by dotson ( 1996 ) provides the most comprehensive available data set for comparison with hi zeeman effect results . in figure 14 we present a map of the position angle of the transverse magnetic field @xmath111 from dotson ( 1996 ) overlaid on our @xmath1420 km s@xmath1 @xmath3 grayscale . the resolution of the dotson study ( 35@xmath112 ) is comparable to that of the hi data . also , the 20 km s@xmath1 hi component likely samples at least part of the molecular ridge ( 4.1.2 ) as does the 100 micron emission . a principal conclusion of the dotson study is that the m17 magnetic field shows a significant degree of spatial coherence across m17 . the hi zeeman data for the 20 km s@xmath1 component corroborates this result ( figs . 6 , 8) . moreover , the eight - fold _ increase _ in @xmath3 from east to west toward the molecular ridge ( fig . 8) is matched by a _ decrease _ in fractional linear polarization from about 4% to 1% . along this same line the field has a nearly constant position angle ( predominantly east - west ) . since @xmath3 is perpendicular to both to the transverse magnetic field and the 100 @xmath8 m linear polarization vectors , one might expect to see a correlation between regions of maximum @xmath3 and minimum 100 @xmath8 m polarization . one obvious interpretation of the ir and hi zeeman data is that the magnetic field lines curve into the line of sight as one approaches the molecular ridge from the east . in such a case , @xmath3 may be a close approximation to the total field strength in the region of the molecular ridge sampled by hi . since the hi must lie in front of the hii region , the field lines in this picture wrap around the hii region on the front side of the bowl , becoming more aligned with the line of sight at the bottom of the bowl as they pass through the molecular ridge . this geometry is consistent with the inferences made by dotson from the linear polarization data alone . the origin of the field curvature may lie in the process of gravitational contraction that formed the molecular ridge , gathering an initially uniform field along the line of sight into an hourglass shape with the ridge at its waist . an hourglass shaped field structure was also suggested for the w3 region by roberts et al . ( 1993 ) , although for this source , the axis of the hourglass is nearly in the plane of the sky rather than along the line of sight . alternately , the present geometry of the m17 field may represent the bending effects of the hii region as it expands into the molecular cloud deforming a field originally along the line of sight into a bowl - like shape . it should be noted that the occurrence of decreasing linear polarization toward the the 100 @xmath8 m flux and column density maxima could result from effects other than a true minimum in the transverse magnetic field component . for example , optical depth effects , dust properties , grain alignment efficiency , or some combination of these can cause a decrease in the linear polarization independent of the magnetic field . on the other hand , the magnetic field itself could simply be more tangled in higher density regions ( jones , klebe , & dickey 1992 ; myers & goodman 1991 ) . in addition , the 760 @xmath8 m polarimetry by valle & bastien ( 1996 ) does not show a minimum in the linear polarization in this region of the molecular ridge . however , the 760 @xmath8 m polarization arises from cooler dust grains which could be spatially distinct along the line of sight from those emitting at 100 @xmath8 m . future polarimetry of this region at an intermediate wavelength from iso and = 16.0 cm higher resolution hi zeeman results may help to clarify this issue . the magnetic field in a cloud can be separately described in terms of a static component @xmath113 and a time - dependent or wave component @xmath114 . the static component connects the cloud to the external medium and also determines the total magnetic flux through the cloud , while the wave component is associated with mhd alfvn waves in the cloud ( e.g. arons & max 1975 ) . although the static component of the field can only provide support to the cloud perpendicular to the field lines , the turbulent or wave component can provide 3-d support ( mckee & zweibel 1995 , also see references therein ) . if a cloud is threaded by @xmath115@xmath113 , it can be completely supported perpendicular to @xmath113 if its mass does not exceed the magnetic critical mass @xmath116 ( mouschovias & spitzer 1976 ) . this criterion leads directly to the observationally relevant relation @xmath117 where @xmath118 is the average proton column density of the cloud . if @xmath113 @xmath119 @xmath120 ( i.e. @xmath121 ) , then the cloud is supported by @xmath113 , it is magnetically subcritical , and further evolution of the cloud perpendicular to the field can only occur via ambipolar diffusion . if @xmath113 @xmath69 @xmath120 , ( i.e. @xmath122 ) , then @xmath113 can not fully support the cloud , the cloud is magnetically supercritical , and internal motions must supply additional support if the cloud is stable . in this case , further evolution of such a cloud will be controlled in part by processes that create and dissipate internal motions . as discussed above , the second type of support provided by magnetic fields arises from alfvn waves generated in a turbulent velocity field . the fluctuating or wave component of the magnetic field @xmath114 organizes supersonic ( but sub - alfvnic ) motions in the gas so that highly dissipative shocks do not occur . theoretical studies suggest that equipartition between the wave s kinetic and magnetic potential energy densities almost certainly exists in molecular clouds ( zweibel & mckee 1995 , and references therein ) . that is @xmath123 @xmath114@xmath124 where @xmath125 is the non - thermal component of the velocity dispersion , which is assumed to arise exclusively from alfvn waves . these assumptions lead to the equivalent of equation ( 2 ) for @xmath114 , that is , @xmath126 where @xmath89 is the non - thermal line width ( fwhm ) and @xmath127 is the proton density . since zeeman effect measurements are generally sensitive to large scale fields ( on the order of the beam size ) , they are expected to be better indicators of @xmath113 than of @xmath114 . therefore , the formal usefulness of comparing equation ( 4 ) for @xmath114 to zeeman observations ( @xmath3 ) is not clear . however , limited observational comparisons of zeeman measurements with the @xmath114 predicted from equation ( 4 ) show that there may also be approximate equipartition between the wave s kinetic and static magnetic energy densities in molecular clouds ; ie . @xmath128 , where @xmath129@xmath113@xmath130 , or equivalently @xmath113@xmath100@xmath114 ( myers & goodman 1988a , 1988b ; bertoldi & mckee 1992 ; mouschovias & psaltis 1995 ) . in any case it is worth noting that @xmath114 must be @xmath131 @xmath113 to prevent the formation of super - alfvnic shock waves , which are highly dissipative ( mckee & zweibel 1995 ) . we can assess the role of magnetic fields in the support of m17 sw from our hi zeeman measurements combined with other measurements of the clouds basic properties . goldsmith , bergin , & lis ( 1997 ) estimate that the mass of the m17 sw is @xmath132 ( including helium ) on the basis of their c@xmath27o maps . using the geometric mean of the extent of their m17 sw c@xmath27o maps ( 1.9 pc @xmath0 2.2 pc ) @xmath133 pc as the @xmath134 of the core region , and their integrated c@xmath27o column density , we estimate @xmath69@xmath118@xmath135 @xmath6 and @xmath69@xmath136@xmath137 @xmath25 ( assumes filling factor of 1 ) . although @xmath89 is often estimated from the @xmath138 of optically thin lines , we believe a more accurate estimate for the internal motions of the m17 sw core as a whole comes from the velocity dispersion of the center velocities of the clumps themselves . stutzki & gsten ( 1990 ) identified 180 , @xmath139 pc c@xmath27o clumps in m17 sw and provide center velocities for each . the dispersion of these center velocities is 2.7 km s@xmath1 , equivalent to @xmath89 = 6.3 km s@xmath1 . from these parameters and equations ( 3 ) and ( 4 ) , @xmath120@xmath140 @xmath8 g and @xmath114@xmath100 620 @xmath8 g . these estimates indicate that ( 1 ) unless the static field is four or more times higher than our maximum values for @xmath3 ( 500 @xmath8 g ) , then m17 sw as a whole is magnetically supercritical and ( 2 ) the estimate for @xmath114 is in good agreement with our highest values for @xmath3 . therefore , the evolution of the m17 sw core is not entirely controlled by @xmath113 and ambipolar diffusion , and internal motions must play a role in its support and dynamics . note that these estimates are not necessarily valid for individual clumps within the core ( bertoldi & mckee 1992 ) . another perspective on cloud support can be gained from a simplified form of the virial theorem that simultaneously takes into account magnetic effects from @xmath113 and @xmath114 ( but not external pressure or rotation ) . this form is @xmath141 where @xmath142 is the gravitational energy term , @xmath143 is the magnetic energy associated with @xmath113 , and 3@xmath144 is the energy contribution from all internal motions including the magnetic wave energy produced by @xmath114 ( see , for example , mckee et al . 1993 for a more detailed description of these terms . ) . note that 2 t , in the non - magnetic virial equation , has been replaced by 3 t because equipartition between internal motions and bw adds an additional term t to the energetics of the cloud . mckee et al . ( 1993 ) explain why virtually all self - gravitating clouds must be close to dynamically critical since their internal pressures would otherwise be comparable to external pressures , contrary to observation . if m17 sw is dynamically critical , that is , on the verge of collapse , then the virial equation above can be rewritten in the form @xmath145^{1/2}.\ ] ] given the m17 sw cloud parameters estimated at the beginning of this section , @xmath144/@xmath146@xmath142@xmath146 @xmath100 0.3 , so @xmath113 @xmath147@xmath120 @xmath148 @xmath8 g , and @xmath149 . given the inevitable errors in estimating virial terms , these values are quite uncertain . however , they can be regarded as a possible equilibrium model for the energetics of the m17 sw core . in this model , internal motions in the core are sub - virial since @xmath144/@xmath142 @xmath69 1/2 , and the cloud is magnetically supercritical . that is , the cloud is supported in part by internal motions , with the remainder of the support coming from @xmath113 . moreover , the magnitude of @xmath113 in this model is comparable to @xmath114 ( 620 @xmath8 g ) , so that the magnetic support provided by the static and wave components of the field are comparable . this suggests that @xmath113@xmath100@xmath114 is a good assumption for m17 sw ( see above ) . although our estimate for @xmath113 using this model is quite close to the highest measured values of @xmath3 ( @xmath150 @xmath8 g ) , this agreement must be considered somewhat fortuitous since estimates of @xmath113 are very sensitive to uncertainties in cloud parameters and , of course , @xmath3 is just one component of the field . apart from equations ( 3 ) , ( 4 ) , and ( 6 ) , one other method has been used to estimate field strengths in self - gravitating clouds . letting 2@xmath144@xmath151@xmath142@xmath146 we obtain the following mass estimate : @xmath152,\ ] ] where @xmath81 parametrizes the extent to which a magnetic field is needed to support the cloud , ie . @xmath153 for clouds in which the internal motions are sub - virial . substituting this estimate for the cloud s mass into equation ( 4 ) we obtain @xmath154~ \mu{\rm g},\ ] ] where @xmath155 is in pc and @xmath89 is in km s@xmath1 . note that this formulation is similar to those found in myers & goodman ( 1988a , 1988b ) , mouschovias & psaltis ( 1995 ) ( do not include @xmath81 ) , and bertoldi & mckee ( 1992 ) ( using @xmath81 ) . comparing the mass computed from equation ( 7 ) , @xmath156 , to the c@xmath27o mass of m17 sw ( @xmath157 ) , we find @xmath158 and @xmath114@xmath159 @xmath8 g , again , remarkably similar to the peak measured value of @xmath3 . the fact that the field strength predicted by equation ( 8) is relatively well matched by @xmath3 may be another indication that @xmath114@xmath100@xmath113 for most clouds ( see above ) . although formulations similar to equation ( 8) have often been used to estimate magnetic fields in the literature ( see above ) , we consider equation ( 6 ) to yield a somewhat better field estimate , at least in so far as comparison with zeeman observations are concerned . this is because 1 . ) if we assume @xmath160 ( myers & goodman 1988a , 1988b ) , equation ( 8) is essentially the same as equation ( 4 ) , both of which estimate @xmath114 . there is evidence ( this work , and references cited above ) that @xmath114@xmath100@xmath113 , but this assertion will require more observational evidence to confirm . 2 . ) in addition , the assumption that @xmath81=1 , excludes entirely the need for support from @xmath113 , contrary to our observational zeeman evidence , ie . @xmath8 g . the inclusion of @xmath81 in the equation solves this delima , but creates a new one , since the estimate of @xmath81 only describes the degree of under - virialization in the cloud , and does not distinguish between support from @xmath113 and @xmath114 , while equation ( 8) is formally an estimate of @xmath114 alone . for these reasons equation ( 6 ) may be the best method of estimating the magnetic field based on cloud parameters for the purpose of comparison with zeeman measurements even though all the estimates found here ( ie . [ 4 ] , eq . [ 6 ] , and eq . [ 8 ] ) are similar . the hi absorption lines toward m17 show complicated profiles with 5 to 8 distinct velocity components . in general , the highest hi column densities are concentrated toward the hii region@xmath9m17 sw boundary . this effect is particularly noticeable in the velocity range of the 20 km s@xmath1 hi component . a lower limit to the maximum hi column density summed across the whole hi velocity range is n@xmath161= 1.0@xmath162 @xmath6 . atomic and molecular emission temperature measurements suggest that @xmath10 lies between 50 and 200 k. a magnetic field has been detected at the same velocity as m17 sw ( @xmath163 km s@xmath1 ) which peaks steeply toward the m17 hii region@xmath9m17 sw interface to values of @xmath3= @xmath9450 @xmath8 g . analysis of this hi component s line width , along with its velocity and spatial coincidence with molecular density tracers along the m17 sw molecular ridge suggest that it originates in unshocked pdr gas . comparison of the proton density implied by the observed @xmath3 and the 20 km s@xmath1 hi density show that the hi gas is only a @xmath14 1% constituent of the gas along the line of sight toward m17 . at the same time its distribution along the stutzki et al . ( 1988 ) strip scan across the hii region@xmath9m17 sw boundary agrees well with that of molecular gas . therefore , we suggest that the 20 kms hi gas lies in thin , dense shells around the numerous molecular clumps which have been observed in m17 sw , or it must be well mixed with fairly dense ( @xmath164 @xmath25 ) interclump molecular gas . in addition , the region of maximum @xmath3 for this component agrees spatially with a minimum in 100 @xmath8 m polarimetry which samples the transverse magnetic field . another significant @xmath3 was observed in the blended component at 11@xmath917 km s@xmath1 with values reaching @xmath165 550 @xmath8 g . the morphology of high @xmath4 for this component is quite different from that at 20 km s@xmath1 , with the greatest concentrations found further east and toward the northern parts of the source . from these components morphology , line width , and evidence that there is a significant amount of obscuring material on the front side of the m17 hii region , we suggest that the 11@xmath917 km s@xmath1 components originate in shocked gas that is streaming toward us along the line of sight . estimates of the shock speeds which can be produced by the m17 hii region agree well with these components blueshift ( 2@xmath98 km s@xmath1 ) with respect to the rest velocity of the region . like the 20 km s@xmath1 hi component , the 11@xmath917 km s@xmath1 hi gas can only comprise @xmath14 1% of the line of sight gas , and may be confined to photodissociated hi shells surrounding molecular clumps close to the hii regions front side . using various virial arguments we have estimated that m17 sw is sub - virial ( @xmath144/@xmath142@xmath69 1/2 ) and magnetically supercritical ( @xmath166 ) . in our model @xmath13 of m17 sw s total support is provided by its static magnetic field @xmath113 ( @xmath148 @xmath8 g ) , with the rest of the support arising from internal motions including support from the wave component of the field @xmath114 . estimates of the static ( @xmath113 ) and wave ( @xmath114 ) components of the magnetic field indicate that they are of approximately the same magnitude ( @xmath167 @xmath8 g ) and agree well with our highest values of @xmath3 ( @xmath168@xmath8 g ) . subsequent analysis of higher resolution vla hi zeeman effect data will give better sensitivity to @xmath3 if the field is tangled , and allow us to better differentiate separate velocity components . in addition , observations of the zeeman effect in oh may also help untangle the magnetic field data at this interface since it tends to trace somewhat higher density gas . c. brogan thanks nasa / epscor for fellowship support through the kentucky space grant consortium . t. troland acknowledges support from nsf grant ast 94 - 19220 , and r. crutcher acknowledges support from nsf grant ast 94 - 19227 . arons , j. , & max , c. e. 1975 , , 196 , l77 beetz , m. , elssser , h. , poulakos , c. , weinberger , r. 1976 , a&a , 50 , 41 bergin , e. a. , goldsmith , p. f. , snell , r. l. , & ungerechts , h. 1994 , , 431 , 674 bertoldi , f. 1989 , , 346 , 735 lrrrrcrrrrr frequency & 1420 mhz observing date & 1996 june 2 total observing time & 8 hr primary beam hpbw & 30synthesized beam hpbw & 61.0@xmath0 44.5phase and pointing center(b1950 ) & @xmath169 , @xmath1701400frequency channels per polarization & 256 total bandwidth & 781.25 khz ( 165.05 km s@xmath1 ) velocity coverage & -67.5 to + 97.5 km s@xmath1 frequency resolution & 3.052 khz ( 0.64 km s@xmath1 ) rms noise in line channels & 40 mjy beam@xmath1 rms noise in continuum & 80 mjy beam@xmath1 @xmath171 to @xmath172 & 1 mjy beam@xmath1 = 0.25 k angular to linear @xmath173 & 1= 0.6 pc position of 21 cm continuum peak & @xmath174 , @xmath1701306 [ tab1 ]

we describe the mathematical underpinnings of the biorthogonal von neumann method for quantum mechanical simulations ( pvb ) . in particular , we present a detailed discussion of the important issue of non - orthogonal projection onto subspaces of biorthogonal bases , and how this differs from orthogonal projection . we present various representations of the schrdinger equation in the reduced basis and discuss their relative merits . we conclude with illustrative examples and a discussion of the outlook and challenges ahead for the pvb representation .

1603.03962

the quantum dynamics of many physical systems involves multiparticle continuua . examples range from ionization and high harmonic generation in multielectron atoms to molecular processes such as photodissociation and chemical reactions of polyatomics . the large hilbert space of these systems makes full simulations extremely challenging . it is therefore of great interest to find a compact representation for quantum dynamics in the continuum where accuracy can be easily controlled . an attractive option is to represent states as phase space objects , so that only those areas of phase space that are actually occupied are needed for the calculation . the most commonly used phase space representation , the wigner representation , has proven to be expensive computationally , although there has been some recent progress @xcite . an alternative is the husimi representation , although this too is awkward for direct numerical calculations @xcite . a third option for a phase space representation is the von neumann lattice @xcite . this representation consists of a discrete lattice of phase space gaussians , one per each cell of area @xmath0 where @xmath1 is planck s constant and @xmath2 is the number of degrees of freedom . initial work using this representation found it to be poorly convergent @xcite . about ten years ago , poirier and coworkers @xcite found a way to converge the method and have applied it recently to several challenging applications @xcite . independently , a few years later , our group discovered a different way to converge the von neumann lattice , based on modifying the gaussians to be periodic and band - limited @xcite . we named the method pvn , _ periodic von neumann_. the pvn basis is guaranteed to have exactly the same accuracy as a discrete fourier representation @xcite , while having the flexibility of a gaussian basis set . since the pvn basis is non - orthogonal , a key aspect of the mathematical formulation involves the concept of biorthogonal bases . we define the biorthogonal von neumann ( bvn ) basis as the basis biorthogonal to pvn . since a crucial part of our method is to exchange the roles of the pvn and bvn bases we refer to the method more generally as periodic von neumann with biorthogonal exchange or pvb . in this work we further develop the mathematical underpinnings of the pvb method . in particular , we derive new relations for the reduced hilbert space associated with non - orthogonal bases when projecting down from the full hilbert space . furthermore , we analyze the possible forms of the schrdinger equation in the reduced basis . these developments become crucial when applying the methodology to more challenging systems , such as the helium atom in strong fields@xcite . the paper is organized as follows . in section [ sub : review - of - pvb ] we present a brief review of the pvb approach as applied to the full hilbert space . we begin by clarifying the underlying hilbert space spanned by the fourier grid method . we then define the von neumann lattice of gaussians . by projecting the von neumann basis onto the hilbert space spanned by the fourier grid we generate the periodic von neumann ( pvn ) basis ( section [ sub : the - von - neumann ] ) . since the pvn basis is non - orthogonal we can define the basis that is biorthogonal ( section [ sub : bi - orthogonal - bases ] ) . in section ( [ sub : pvn , pvb ] ) we show that the role of the pvn basis and the biorthogonal basis must be exchanged to obtain a sparse representation . we conclude section [ sub : review - of - pvb ] with a discussion of the possible forms of the schrdinger equation in the pvb representation . having established the formalism on the full hilbert space , we turn to the representation on a reduced space . section [ sub : the - reduced - basis ] introduces the reduced pvb method , allowing one to represent only those regions of phase space actually occupied by the current state . we then define the biorthogonal bases for the reduced subspace ( section [ sub : bi - orthogonal - bases - for ] ) and analyze the projection into the reduced subspace ( section [ sub : the - projector ] ) , highlighting the differences between orthogonal and non - orthogonal projections . in section [ sub : the - tdse - for - the - reduced_state ] we derive the various forms of the schrdinger equation in the reduced representations and discuss approximations and performance considerations . section [ sec : examples ] gives some illustrative examples and section [ sec : conclusions - and - outlook ] gives an outlook and some ideas for future work . the pvb representation , developed in@xcite , projects the von neumann lattice of gaussians ( section [ sub : the - von - neumann ] ) onto the subspace spanned by the fourier grid ( section [ sub : the - fourier - grid ] ) . as the projected gaussians form a non - orthogonal basis , we define the basis biorthogonal to the projected gaussians ( section [ sub : bi - orthogonal - bases ] ) . together these building blocks define the pvb representation ( section [ sub : pvn , pvb ] ) . we then present several alternative but equivalent formulations of the schrdinger equation in the pvb representation ( section [ sub : schroedinger - equation - in - pvb ] ) . the pseudospectral fourier method@xcite ( also known as the periodic sinc dvr ( discrete variable representation@xcite ) ) is the underlying hilbert space on which we construct the pvb method . a comprehensive exposition can be found in @xcite . functions with support on a finite segment @xmath3 $ ] may be assumed , without loss of generality , to be cyclic , and reside in a hilbert space spanned by @xmath4 limiting bandwidth to @xmath5 implies @xmath6 $ ] where @xmath7 this defines a rectangular area of phase space of area @xmath8 which is spanned by the _ spectral basis _ , @xmath9 , with @xmath10 . with an inner product defined as @xmath11 this constitutes the _ fourier grid ( fg ) _ hilbert space , @xmath12 . any @xmath13 can be expanded as @xmath14 , allowing us to represent @xmath15 as a column vector of the expansion coefficients , @xmath16 @xmath17 @xmath18 . we shall drop the basis designation when implied by context . define the _ fourier grid points , _ or _ sampling points _ , as the set of @xmath19 equidistant points @xmath20 with @xmath21 $ ] . define the _ pseudo - spectral basis _ of @xmath12 as the set of _ pseudo - spectral functions _ @xmath22 such that for @xmath23 , @xmath24 . @xmath15 may now be represented by its _ sampling vector _ , @xmath25 @xmath26 + @xmath27 . by expanding the spectral basis functions in the pseudo - spectral basis , one may derive the explicit form of the latter , @xmath28 , with @xmath29 ( * ? ? ? * supplementary material 1 ) . these are periodic @xmath30 functions , which are localized around @xmath31 with @xmath32 . the pseudo - spectral basis functions satisfy the normalization condition @xmath33 @xcite . thus , functions in @xmath12 may be represented in the pseudo - spectral basis by their sampling vector . unlike the spectral basis representation , this does not require integration to compute the coefficients . the existence of the pseudo - spectral basis , allowing reconstruction of a function from sampled values , is equivalent to the shannon - nyquist theorem for band - limited functions . the projector into the fg hilbert space is @xmath34 when @xmath35 , the projector @xmath36 minimizes the distance to the projected state , and consequently maximizes the overlap , i.e. @xmath37 ( see@xcite ) . the pseudo - spectral functions are the projection of the sampling functions into @xmath12 , i.e. @xmath38 . defining the _ collocation _ or _ sampling pseudo - projection _ , @xmath39 generally @xmath40 , unless @xmath13 . for functions that are almost within @xmath12 , one may opt to accept the easy - to - compute @xmath41 as an approximation of @xmath36 , which requires costly integration . note that @xmath12 and the associated pseudo - spectral functions are just one possible choice for the phase space underlying pvb . see @xcite for alternative possibilities . consider a lattice of gaussians in the @xmath42 plane . let @xmath43 indicate the center of gaussian no . @xmath44 , and let @xmath45 be the spacing between the lattice sites , with @xmath46 . we define @xmath47 if the lattice spans the infinite plane in the @xmath48-@xmath49 phase space it is known as the _ von neumann lattice _ @xcite . however , in any real calculation the lattice must be truncated to a finite domain @xmath50 . this leads to a lack of convergence , as discovered independently in the context of quantum mechanics@xcite and the time - frequency analog in signal processing @xcite . however , if one projects the von neumann lattice onto the fg basis one builds in periodic boundary conditions and obtains the same convergence as for the fg . define a cyclic formulation of the gaussians , @xmath51 the cyclic projected gaussians , known as the _ periodic von neumann _ or _ pvn basis _ , constitute a _ periodic gabor basis _ , where all basis functions are related to each other by shifts in @xmath48 and @xmath49 @xcite . for the conditioning of the overlap matrices ( eq . [ eq : overlap - matrices ] ) it is beneficial to choose the von neumann lattice points such that @xmath52 are a subset of the fg sample points and @xmath53 are a subset of the spectral basis frequencies . the representation of the pvn basis functions in the @xmath54 basis is given by @xmath55 the fg defines an area of @xmath56 in phase space . therefore one may intuitively assign a phase space area of @xmath57 to each of the @xmath19 gaussians of the pvn basis , and consider them phase space pixels . defining @xmath58 , each such pixel covers a phase space area of @xmath59 . in this section we consider the general theory of non - orthogonal bases and their biorthogonal bases . in the next section we will specialize to the periodic von neumann basis and its biorthogonal basis . in anticipation of that section , we use the notation @xmath60 and @xmath61 here . any set of @xmath19 linearly independent vectors @xmath62 in @xmath12 may serve as a non orthogonal basis of @xmath12 . let @xmath63 be represented in the @xmath64 orthogonal basis of @xmath12 by the invertible matrix @xmath60 . let @xmath65 be a similarly defined non - orthogonal basis of @xmath12 , represented in @xmath64 by @xmath61 . the bases @xmath63 and @xmath65 are considered _ biorthogonal bases _ ( a reciprocal relationship ) if @xmath66 the relation is reciprocal , i.e. @xmath67 if , and only if , @xmath63 is orthogonal , then so is @xmath65 and @xmath68 . according to eq . [ eq : bi - otrho prelim_g_first ] , the rows of @xmath69 represent the bra states @xmath70 , while the columns of @xmath61 represent the ket states . the representation in the @xmath61 basis of any state @xmath71 associated with the sampling vector @xmath72 is @xmath73 with the elements @xmath74 denoting the overlaps @xmath75 . the _ completeness relation _ for @xmath12 is @xmath76 by construction , the projector in eq . [ eq : gb completeness ] spans the full fg hilbert space , and therefore is equal to the projectors in eq . [ eq : p_into_the_fg ] . we define the _ overlap matrices _ of the @xmath60 and @xmath61 bases , respectively : @xmath77 from these relations it follows that @xmath78 norms of vectors in @xmath61 and @xmath60 are computed via @xmath79 and @xmath80 . note that the @xmath60 and @xmath61 bases can not be independently normalized . many wavefunctions of interest , whether bound states or traveling wavepackets , are fairly well - localized in phase space . therefore a representation whose coefficients are @xmath81 , is expected to be sparse , i.e. to have many near - zero elements . consider a fourier grid hilbert space @xmath12 . [ eq : gb completeness ] , any state @xmath71 can be represented as @xmath82 we shall use the term pvn to denote this representation . alternatively , any state in @xmath12 can be represented as @xmath83 this defines the _ biorthogonal von neumann representation , pvb_. note that while the @xmath84 states are highly localized in phase space , the @xmath85 states are highly delocalized . as a result , the pvb representation is sparse , while the pvn representation is not . the pvb and pvn bases both span @xmath12 and therefore these representations contain exactly the same information . from this point onward , we shall assume all states are in @xmath12 , and default to the pseudo - spectral representation unless otherwise noted . we will also drop the explicit vector notation for states . we now turn to the form of the time independent and time dependent schrdinger equations ( tise and tdse , respectively ) in the pvb basis . starting from the tise , @xmath86 , where @xmath87 is the energy , and using eqs . [ eq : bi - otrho prelim_g_first ] and [ eq : psi_b ] , gives @xmath88 similarly , the tdse @xmath89 takes the form : @xmath90 the term @xmath91 is a similarity transformation of @xmath92 and therefore all eigenvalues are real and all evolutions are unitary . one can make hermiticity in eq . [ eq : tise_in_b ] explicit by multiplying it from the left by @xmath93 and using eq . [ eq : gb completeness ] , which transforms it into a generalized eigenvalue problem , @xmath94 one may rewrite the tdse in four distinct ways , all strictly equivalent . we denote a hamiltonian taking a state in basis @xmath95 and returning a state in basis @xmath96 by @xmath97 , @xmath98 although the forms are mathematically equivalent , they require different computational efforts . @xmath99 is quick to compute as the integrations on both sides are with the highly - local gaussians . this is counter - balanced by the need to compute @xmath100 . this issue shall be revisited in section [ sub : the - tdse - for - the - reduced_state ] . as discussed above , many of the coefficients of @xmath74 are negligible , since the phase space - localized state does not overlap with many of the localized gaussians . we therefore define a reduced hilbert subspace , spanned by a subset of the @xmath85 vectors , whose coefficients are all above some pre - defined threshold ( section [ sub : definition_of_reduced ] ) . next we define the biorthogonal bases for this reduced space ( section [ sub : bi - orthogonal - bases - for ] ) and the projector into it ( section [ sub : the - projector ] ) . finally we mention a different subspace , spanned by a subset of @xmath101 vectors ( section [ sub : check - check ] ) , which will be useful in later discussions . consider a hilbert space @xmath12 , of dimension @xmath19 , spanned by the orthogonal pseudo - spectral basis @xmath64 and a set of biorthogonal bases @xmath65 , @xmath63 , represented in @xmath64 by the @xmath103 matrices @xmath61 and @xmath60 , with @xmath60 being the periodic von neumann gaussian lattice , as defined in eq . [ eq : g_jk ] . a state @xmath104 that is localized in phase space will be sparse in the @xmath61-representation , i.e. @xmath74 will have many near - zero elements . for the sake of notational convenience , let us assume the first @xmath105 coefficients in @xmath74 are significant , while the remaining @xmath106 are negligible . in such a case , we can save computational resources by reducing the vector @xmath74 of length @xmath19 to a vector @xmath107 of length @xmath108 . define the _ reduced subspace _ , @xmath109 as the hilbert space spanned by the first @xmath105 vectors of @xmath61 , i.e. all states whose @xmath61-basis representation is strictly zero in entries @xmath110 . @xmath111 note that generally , @xmath102 is quite different from @xmath112 as is evident when @xmath113 . we shall discuss @xmath114 further in [ sub : check - check ] . define _ the complementary subspace_,@xmath115 , by @xmath116 requiring that _ every vector in the complementary subspace @xmath117 is orthogonal to every vector in the reduced subspace , @xmath102_. as the @xmath61 basis is non - orthogonal , there is no partitioning of the @xmath61 functions that allows one to span both subspaces separately . however , the complementary space may be written as @xmath118 by construction , @xmath102 is rank @xmath105 , @xmath115 is rank @xmath106 , and @xmath119 due to the biorthogonality of their respective base vectors . therefore @xmath120 and we conclude that indeed @xmath121 . given any state @xmath71 , we may therefore decompose it as @xmath122 where @xmath123 is the _ reduced state _ and @xmath124 _ the complementary state_. projection may be viewed as subtracting @xmath124 from @xmath104 , a view that will be made explicit in section [ sub : the - projector ] . we now turn to the biorthogonal bases for the reduced subspace , @xmath125 we define these bases in a manner similar to that of the unreduced bases . let us denote the first @xmath105 columns of the @xmath61 matrix with the matrix @xmath126 , i.e. @xmath127 is the @xmath64 representation of the basis defining @xmath102 . @xmath128 will be defined as a basis which is biorthogonal to @xmath127 , and spans the same hilbert space as @xmath127 ; i.e. the pair @xmath128 and @xmath127 satisfy the completeness relation on the subspace . we stress that @xmath127 is the natural basis of @xmath102 , while @xmath128 is defined in terms of @xmath127 . one should note that while @xmath129 is a truncated version of @xmath61 , the @xmath130 matrix is not a simple truncation of @xmath60 , but rather that individual @xmath131 functions are modified by a projection into the subspace spanned by @xmath129 . this will be discussed in detail later in this section . the biorthogonality requirement is @xmath132 the requirement that @xmath127 and @xmath128 span the same subspace , is equivalent to the statement that @xmath128 vectors are expressible as linear combinations of @xmath127 vectors , and vice versa , i.e. @xmath133 @xmath134 ( i.e. @xmath135 , etc ) . the biorthogonality requirement is satisfied by the right pseudo - inverse of @xmath136 , @xmath137 and eq . [ eq : red_completeness_req_1 ] and [ eq : red_completeness_req_2 ] are satisfied by @xmath138 where @xmath139 and @xmath140 where @xmath141 therefore @xmath128 and @xmath127 are biorthogonal bases , both spanning @xmath102 . as biorthogonality is a reciprocal relation , eq . [ eq : g_tilde_def ] is echoed by @xmath142 . using eq . [ eq : tilde_s ] and [ eq : tilde_s_and_invs ] , we may relate @xmath128 and @xmath127 using @xmath143 : @xmath144 and @xmath145 . note that the biorthogonality requirement alone , eq . [ eq : red_bi_ortho_req ] , is insufficient to uniquely define @xmath128 . for example , both the gaussians , @xmath101 , and the vectors @xmath146 of @xmath128 are biorthogonal to the @xmath147 vectors spanning the reduced subspace , i.e. @xmath148 . but while @xmath149 and @xmath150 span the same subspace , @xmath151 and @xmath152 do not ( except when @xmath153 ) . at the extreme , when @xmath113 , it is clear that @xmath154 and @xmath155 span different ( trivial ) hilbert spaces . similarly to the definitions of @xmath127 and @xmath128 , we define the basis for the complementary hilbert space @xmath115 with @xmath156 of size @xmath157 as the last @xmath106 columns of @xmath60 , and @xmath158 . the discussion above , and particularly eq . [ eq : g_tilde_def ] , leads to the conclusion that the columns of @xmath128 are no longer exactly gaussians , but have been deformed as the result of the basis reduction . this is depicted in fig . [ fig : deformed_gaussians ] . note that gaussians that are far away from the reduction boundary are not as affected as the gaussians closer to the boundary . this is consistent with the intuitive notion that the localized gaussians should not be affected by changes in areas of phase space into which they do not significantly extend . this intuition is made precise in the following section . , which is centered around a non - edge location in the phase space , is almost unchanged . on the right , we see a heavily deformed gaussian , i.e. the function is delocalized throughout the edge region . this results from the projection of the @xmath130 basis functions into the subspace spanned by @xmath129 . the deformation is much more significant for gaussians whose centers are near the edge . [ eq : i_am_so_tired ] and the detailed discussion following eq . [ eq : tildep_subtractive_on_gk ] for an explanation the states are plotted as heat maps , where the value of each cell of the von neumann lattice is the absolute value of the overlap of the state plotted ( here the modified gaussians ) , with the gaussian centered at that cell of the lattice , @xmath159.[fig : deformed_gaussians ] ] schematically , the reduced @xmath143 matrix is constructed from the @xmath160 matrix by the following circuitous path , @xmath161{pdf_figures / s_matrix_path_1.pdf}\ ] ] one starts with the unmodified @xmath162 , inverts to get @xmath163 , truncates to get @xmath164 , and then inverts back to get @xmath165 . the chain of relationships suggest that although @xmath60 is simpler than @xmath61 , @xmath128 is more complicated than @xmath127 . specifically , in the multi - dimensional case the columns of the @xmath128 matrix do not decompose into products of their one - dimensional counterparts , while the columns of the @xmath127 matrix do decompose into @xmath166 components . the reduced @xmath167 matrix is a subset ( of rows and columns ) of the unreduced @xmath100 matrix , but the reduced @xmath168 matrix is not a subset of the full @xmath160 , due to the inversion in eq . [ eq : s - flow - diag ] . non - decomposability of the multi - dimensional @xmath143 and @xmath128 matrices has significant impact on the computational resources required to compute the reduced hamiltonian . the issue is discussed in depth in section [ sub : the - tdse - for - the - reduced_state ] . we summarize the identities involving the reduced @xmath143 matrix that are counterparts of the identities for the unreduced @xmath160 matrix , eq . [ eq : overlap - matrices ] : @xmath169 we can now expand a state using the reduced basis , echoing eq . [ eq : psi_expanded_in_b ] , @xmath170 providing us with a representation - free form of the projector from @xmath12 to @xmath102 . in dirac notation , @xmath171 with the second equation following by hermiticy of the projection . note similarity to eqs . [ eq : psi_expanded_in_g ] and [ eq : psi_expanded_in_b ] . recall that @xmath172 . the projector in the @xmath64 representation ( i.e. both input and output are represented on the fourier grid ) is given by @xmath173 where we omit the basis designation @xmath64 . one may directly show that @xmath174 is idempotent , i.e. @xmath175 , and hermitian . moreover , the projector @xmath174 transforms a general state to the state closest to it within the subspace @xcite . the idempotent property of @xmath174 requires that the bra matrix in eq . [ eq : p_in_fg ] , @xmath176 , be biorthogonal to the ket matrix , @xmath127 . therefore , the projector is non - orthogonal if and only if the bases are non - orthogonal , which is the case if and only if @xmath127 and @xmath128 are not identical . the projection operator may be rewritten in several representations . we shall denote by @xmath177 the projection operator from the full hilbert space @xmath12 into the reduced subspace @xmath102 , where the fourier grid representation of the input basis for @xmath12 is the matrix @xmath95 , and the fourier grid representation for @xmath102 is the matrix @xmath96 . @xmath177 may be square or rectangular , depending on the choice of output basis . the projection operator transforms as do all operators , using @xmath178 when @xmath96 is not square , @xmath179 should be taken to be the left pseudo - inverse of @xmath96 . to better understand the difference between this non - orthogonal projector and a standard orthogonal projector we shall examine two representations : @xmath180 where the projection operation will be viewed as an additive operation , and @xmath181 , where the projection operation will be viewed as a subtractive operation . given a state in the @xmath61 basis , @xmath182 , the application to @xmath104 of eq . [ eq : p_dirac ] gives @xmath183 note that all @xmath184 vectors contribute to the coefficients in the reduced subspace , as indicated by the @xmath185 term of eq . [ eq : p_tildeb_b_applied ] . the participation of all @xmath184 vectors in the coefficients is reflected in the @xmath180 representation of the projector , as depicted in fig . [ fig : p_bb ] . we derive the explicit form of @xmath180 from eq . [ eq : p_yx ] , using eq . [ eq : red_bi_ortho_req ] , giving @xmath186 examining the top panel of fig . [ fig : p_bb ] , the right block of @xmath180 depicts the overlap of all @xmath61 vectors of the unreduced basis , including @xmath187 , with the @xmath151 subset . this added contribution is unique to non - orthogonal projections . if this were an orthogonal projection , only the basis vectors defining the reduced basis , @xmath151 , which make up the identity matrix appearing in the left square , would play a part . recalling eq . [ eq : h - straight - sum ] , @xmath121 , we express the identity operator as a sum of projectors , @xmath188 where , by analogy to @xmath189 , we define @xmath190 . using @xmath191 , @xmath192 we shall apply both sides of eq . [ eq : tildep_subtractive ] to @xmath193 . on the l.h.s , [ eq : p_dirac ] gives @xmath194 applying the r.h.s of eq . [ eq : tildep_subtractive ] to @xmath195 and equating it with eq . [ eq : i_am_so_tired ] results in @xmath196 note that @xmath189 , applied to @xmath197 is zero , as the state is part of the basis spanning the complementary subspace . we now have a new insight into the deformation of the gaussians due to the basis reduction , as depicted in fig . [ fig : deformed_gaussians ] : it is the result of subtraction of @xmath198 vectors , weighted by the overlap of the gaussians inside the reduced subspace with the gaussians spanning the complementary space . this clarifies why gaussians at the boundary of the reduced subspace are modified more they have a larger overlap with gaussians outside the reduced subspace . applying @xmath189 to a superposition state @xmath199 and using eq . [ eq : i_am_so_tired ] gives @xmath200 gaussians spanning the complementary subset are dropped , and the coefficients of the remaining gaussians are unchanged but now serve as coefficients of the modified gaussians . if this were an orthogonal basis , these coefficients would have been unchanged without modifying the underlying basis . an equivalent view is provided by examining the projector representation @xmath181 . define the matrix @xmath201 allowing us to write @xmath202 combining eqs . [ eq : p_in_fg ] , [ eq : p_yx ] and [ eq : g_tilde_def ] , and using eqs . [ eq : b_red ] and [ eq : bi - otrho prelim_b_first ] , we obtain @xmath203 this uniquely simply form is depicted in the bottom panel of fig . [ fig : p_bb ] . @xmath180 and @xmath181 provide two complementary views of the non - orthogonal projection . with @xmath180 , projection is an additive process : the vectors of the reduced basis , @xmath204 are the same as their unreduced counterparts , @xmath205 , as is the case for orthogonal projections . however , the projection modifies their coefficients due to additive contributions by @xmath206 , which is unique to non - orthogonal projections . with @xmath181 , projection is a subtractive process : if @xmath199 , then the coefficients @xmath207 are unchanged by the projection and the rest are removed completely , as in an orthogonal projection . however , the gaussian basis vectors are modified by subtraction proportional to their overlap with the gaussians spanning the complementary subspace , changing @xmath101 to @xmath146 . this modification is again unique to non - orthogonal projections . ( top panel ) the @xmath180 projector , i.e. the operator projecting the state @xmath74 , into the reduced hilbert space @xmath102 and representing the result in the @xmath127 basis . notice the block structure . the left square is the identity matrix , since if @xmath208 it remains unchanged by the projection . the right block provides insight into the unique additive contributions characteristic of non - orthogonal projections : as @xmath209 and @xmath210 overlap , when projecting into @xmath102 , all @xmath211 vectors have some overlap into the reduced subspace , and therefore contribute to the projected vector . ( bottom panel ) the @xmath181 projector . the right block is empty , denoting the subtraction of all @xmath156 vectors , which span the complementary subspace @xmath115 , from the @xmath60 representation of the state . the left block is the identity matrix , denoting the coefficients in this block remain unchanged . however , since @xmath212 , the vector represented by these coefficient does change , due to the subtraction of the @xmath156 subspace from the @xmath213 vectors , resulting in the @xmath128 basis . ] ( top panel ) the @xmath180 projector , i.e. the operator projecting the state @xmath74 , into the reduced hilbert space @xmath102 and representing the result in the @xmath127 basis . notice the block structure . the left square is the identity matrix , since if @xmath208 it remains unchanged by the projection . the right block provides insight into the unique additive contributions characteristic of non - orthogonal projections : as @xmath209 and @xmath210 overlap , when projecting into @xmath102 , all @xmath211 vectors have some overlap into the reduced subspace , and therefore contribute to the projected vector . ( bottom panel ) the @xmath181 projector . the right block is empty , denoting the subtraction of all @xmath156 vectors , which span the complementary subspace @xmath115 , from the @xmath60 representation of the state . the left block is the identity matrix , denoting the coefficients in this block remain unchanged . however , since @xmath212 , the vector represented by these coefficient does change , due to the subtraction of the @xmath156 subspace from the @xmath213 vectors , resulting in the @xmath128 basis . ] transitions between the unreduced and reduced bases , represented on the middle and outer circles , respectively , with dimension @xmath19 and @xmath105 . dashed arrows indicate a projection ( surjective ; with information loss ) , continuous arrows are injective ( one - to - one but not necessarily on - to ) mappings . the nomenclature @xmath177 is used to indicate the projection from representation @xmath95 of @xmath12 into representation @xmath96 of @xmath102 . this diagram may be used , for example , to construct @xmath180 by going through the following series of transformations : @xmath214 , from the @xmath61 basis to @xmath60 using the transformation matrix @xmath100 , @xmath215 projecting with @xmath216 , @xmath217 transforming back to @xmath127 using @xmath143 , @xmath218 transforming to base @xmath61 using @xmath219 . multiplying the matrices appearing from right to left gives @xmath220 . ] -1.5cm-1.5 cm @xmath221 we have defined the reduced subspace @xmath102 in eq . [ eq : red_subspace_def ] as the hilbert space spanned by @xmath222 . one may define a different reduced subspace , @xmath223 , as in eq . [ eq : check_hilbert ] . as discussed in section [ sub : bi - orthogonal - bases - for ] , these subspaces are not identical ( unless @xmath153 ) , as is evident when @xmath113 and @xmath224 and @xmath225 . following the discussion in sections [ sub : definition_of_reduced ] , [ sub : bi - orthogonal - bases - for ] and [ sub : the - projector ] one may define the projector into this subspace and the subspace s biorthogonal bases . let us denote the first @xmath105 columns of the @xmath60 matrix with the matrix @xmath226 , i.e. @xmath213 is the @xmath64 representation of @xmath227 , the basis spanning @xmath114 . to satisfy the biorthogonality requirement , eq . [ eq : bi - otrho prelim_g_first ] , in the @xmath114 basis , one must define @xmath228 @xmath229 with @xmath230 . the projector into @xmath114 may then be defined as @xmath231 with @xmath232 . therefore , in the @xmath64 representation the projector is @xmath233 note both @xmath115 and @xmath114 are spanned by gaussians . specifically , @xmath115 is the reduced subspace spanned by the set of gaussians complementary to @xmath213 . it must be stressed that @xmath234 . as discussed in section [ sub : pvn , pvb ] , the representation of a state as a sum of of gaussians ( pvn ) is not sparse . therefore , given a well - localized state @xmath104 , if one wishes to use the @xmath114 subspace , one can expect that a very large number of functions will be needed . this is in stark contrast to @xmath235 , where well - localized states are well - approximated by a small number of functions . thus , given a fixed size of the representation , one will achieve much higher accuracies with @xmath236 than with @xmath237 . this will be explicitly demonstrated when we discuss the various representations of the reduced hamiltonian and their accuracy , in section [ sub : the - tdse - for - the - reduced_state ] . the schrdinger equation for the reduced state requires projecting the hamiltonian into the reduced subspace . in this section we consider three possibilities from both the theoretical and practical viewpoints : projecting into the @xmath102 subspace ( section [ sub : h_bb ] ) , into the @xmath114 subspace ( section [ sub : h_check_check ] ) and performing a pseudo - projection into a combination of the two ( section [ sub : h_hybrid ] ) . finally we benchmark the respective accuracies of the possible projections ( section [ sub : h_comparing ] ) . to arrive at the reduced hamiltonian , we apply the @xmath174 projection on both the input and output of the hamiltonian . as the state is represented in the @xmath127 basis , one can either convert the vector to @xmath64 , project into @xmath102 , apply @xmath92 , project the resulting state again into @xmath102 , and transform it to @xmath127 , or , equivalently , first apply the projection and then the base transformation . these two alternatives correspond to the following two expressions for @xmath238 : @xmath239 utilizing eq . [ eq : p_in_fg ] , [ eq : p_bb_1 ] and [ eq : g_tilde_def ] @xmath240 the tise and tdse then take the form @xmath241 @xmath242 _ eq . [ eq : h_tilde_b_tilde_b ] , [ eq : tise_red ] and [ eq : tdse_red ] will be the key working equations for the remainder of this article . _ the formulation using only @xmath127 matrices ( r.h.s . of [ eq : h_tilde_b_tilde_b ] ) is appealing , as @xmath243 is the naturally reduced matrix in @xmath174 . an equivalent , more heuristic approach , is to replace all matrices in any of the forms of eq . [ eq : h_gb_bb_bg_gg ] by their reduced counterparts , @xmath244 , @xmath245 , @xmath246 , etc . all four variations transform to the reduced hamiltonian in eq . [ eq : h_tilde_b_tilde_b ] . finally , we note that @xmath247 is similar to a hermitian matrix , in that the product of matrices to the left and right of @xmath92 in eq . [ eq : h_tilde_b_tilde_b ] produce the identity , @xmath248 . therefore eigenvalues are real and evolution is unitary . moreover , combining eq . [ eq : h_tilde_b_tilde_b ] and [ eq : tise_red ] the tise may be reformulated as a generalized eigenvalue equation with hermitian matrices . @xmath249 thus avoiding the matrix inversion required to compute @xmath143 . this form is solvable with an iterative eigensolver , such as arnoldi . unfortunately , in this form it is not possible to implement the eigensolver with just matrix - vector multiplications ( i.e. it is not possible to avoid matrix - matrix multiplications ) @xcite . this is in contrast with the pseudo - projection discussed in section [ sub : h_hybrid ] , which can be implemented with just matrix - vector multiplications . while accurate and mathematically rigorous , computing the @xmath247 form in eq . [ eq : h_tilde_b_tilde_b ] is time consuming . most significantly , calculation of elements of the reduced hamiltonian @xmath250 can be laborious , particularly in the multi - dimensional case , as the @xmath61 functions are non - localized . fortunately , there are symmetry considerations and numerical techniques which can accelerate this computation by several orders of magnitude . alternatively , using eq . [ eq : h_tilde_b_tilde_b ] , [ eq : b_red ] , [ eq : overlap - matrices ] and [ eq : overlap - matrices - red ] , one may rewrite @xmath247 as @xmath251 eq . [ eq : h1_on_the_fly ] may be quicker to compute then [ eq : h1_def ] : due to the locality of @xmath60 and the fact that elements of @xmath92 are usually functions of either position or momentum ( but rarely both ) , the vast majority of @xmath252 elements are vanishingly small . note that @xmath60 and @xmath100 decompose dimensionally . indeed , one would still have to calculate @xmath143 , which requires the inversion of @xmath253 , but this may be accelerated , if one is able to store the matrix . if we then use an iterative algorithm for solving the tise ( e.g. arnoldi ) and the tdse ( e.g. taylor propagator ) , one may use only matrix - vector type operations to further accelerate the process . similar considerations have been noted in @xcite . in the previous section we presented the projection of the tise and tdse into the @xmath102 subspace . there are , however , additional alternatives we should consider . first , we shall examine the possible choices for state representation : in addition to the @xmath127 representation which we have been using , there are have three additional a priori candidates : @xmath128 , @xmath254 and @xmath213 . then , we shall consider various projections for the hamiltonians . this analysis is of some importance , as alternatives to eq . [ eq : h_tilde_b_tilde_b ] are currently in use by the community . for example , @xmath255 , defined in eq . [ eq : h2 ] using a hybrid of @xmath174 and @xmath256 is used in @xcite . the first alternative state representation , @xmath128 , spans @xmath102 just as @xmath127 does and therefore provides the same accuracy . however , it has a disadvantage : the coefficients are determined by an overlap with @xmath127-s , which are highly non - local . as a result , the coefficients are not expected to drop - off sharply at the boundary of the reduced subspace , where the wavepacket is absent . as a result , we can not use their declining values to determine the selection of basis functions for the reduced subspace , as one may do with an amplitude cutoff threshold for @xmath107 . this issue is of primary concern , especially for the tdse , where the coefficients of the state at the boundary of the reduced phase space serves determine how to modify the subspace over time . we therefore conclude that the @xmath128 is unfit for our purposes . the two remaining options , @xmath254 and @xmath213 , are disqualified because the subspace to which such states belong , @xmath257 , is spanned by gaussians , and hence does not provide a sparse representation for localized states . in other words , many gaussians ( or many @xmath254 vectors ) are required to accurately represent a localized state ( see section [ sub : check - check ] ) . we therefore conclude that @xmath127 is the only viable option for state representation . next we proceed to examine the possible projections of the hamiltonian . until now we have been considering projection into @xmath102 . the other natural alternative is projection into @xmath114 . this may be performed either by acting with @xmath237 on @xmath92 on the left and the right , or by using the replacement rules @xmath258 , @xmath259 , @xmath260 on any of the hamiltonian forms in eq . [ eq : h_gb_bb_bg_gg ] . this leads to @xmath261 note that we would like to continue to represent states in the @xmath127 representation , in accordance with the above discussion of its compactness . then the tise and tdse take the form @xmath262 @xmath263 however , because the subspace into which @xmath264 is projected does not span the subspace defined by the @xmath107 , the accuracy may be significantly lower than that of @xmath247 ( see fig . [ fig : h1v2v3v4 ] and the discussion in section [ sub : h_comparing ] ) . nevertheless , @xmath264 greatly benefits from having a highly local hamiltonian transform , @xmath265 , making it usable when fast low - accuracy results are acceptable . with the strict application of @xmath174 leading to an accurate but slow @xmath247 , and the application of @xmath256 leading to an inaccurate alternative , in the following section we explore the possibility of a hybrid projection that combines some of the benefits of both . recall the discussion in section [ sub : bi - orthogonal - bases - for ] regarding the modification of the @xmath128 gaussians when near the edge of the reduced subspace . one may hypothesize that if we extend the reduced subspace beyond the minimal volume required to represent the state , so that gaussians near the boundary do not have significant overlap with the state , we may achieve a good approximation while still using the unmodified gaussians . in the nomenclature established in section [ sub : check - check ] , we wish to replace @xmath128 with @xmath213 , and possibly avoid the non - local @xmath127 matrix altogether . we note that @xmath266 , i.e. they are biorthogonal . however , as they do not span the same subspace the completeness relations do not hold . recalling eq . [ eq : h_tilde_b_tilde_b ] , @xmath267 and eq . [ eq : h4 ] , @xmath268 , we shall define the hybrid form @xmath269 with the corresponding tise and tdse : @xmath270 @xmath271 this form is advantageous over eq . [ eq : h_tilde_b_tilde_b ] in that no matrix inversion is necessary ( implying there is no generalized eigenvalue problem to solve for the tise ) , and the localized nature of @xmath213 can be used to simplify the calculation of multi - dimensional terms in @xmath272 . @xmath255 is similar to a hermitian matrix ( as @xmath266 ) and hence diagonalizable with real eigenvalues . [ eq : h2 ] may be reformulated using eq . [ eq : g_check ] , [ eq : b_red ] and [ eq : overlap - matrices ] as @xmath273 this form allows the tdse to be solved using only matrix - vector type operations . @xmath100 decomposes into a product of one - dimensional matrixes and @xmath252 may be calculated analytically by expanding the potential term as a taylor series around the center of the gaussian . such an approach may be sufficiently quick to eliminate the need to store the reduced hamiltonian , by making use of an iterative eigensolver as described in @xcite along with a taylor propagator for the tdse . a final possible hybrid is @xmath274 but we do not see any immediate usefulness in this form . accuracy of the ground state energy of the 1d harmonic oscillator as a function of the reduced basis size ( lower @xmath48-axis ) or reduced - basis boundary width ( upper @xmath48-axis ) computed using different reduced hamiltonians . as expected from the theoretical discussion , @xmath275 is orders of magnitude more accurate then @xmath276 , until the reduced subspace is increased to the point where the state does not significantly overlap any of the distorted gaussians , and the distinction between @xmath128 and @xmath213 becomes moot . @xmath277 and @xmath278 are significantly less accurate . ] we now compare the accuracy of the various reduced hamiltonians introduced in the previous sections . from fig . [ fig : h1v2v3v4 ] we can conclude that @xmath247 is indeed the most accurate . this is expected as it contains no approximations beyond the amplitude cutoff used to define the reduced subspace . @xmath255 provides two orders of magnitude lower accuracy for the same reduced basis size or alternatively , requires a boundary width of @xmath279 lattice cells to achieve equivalent accuracy . the latter translates to @xmath280 to @xmath281 larger reduced basis size for 1d problems , and is expected to grow exponentially with higher dimensions , making it an unappealing option for high - dimensional problems . the lower accuracy of @xmath282 may be understood when we compare equations [ eq : h2 ] to [ eq : h_tilde_b_tilde_b ] , i.e. @xmath272 to @xmath283 , and recall the deformation of the gaussians by the basis reduction , as depicted in fig . [ fig : deformed_gaussians ] . @xmath282 ignores this deformation , instead acting with the unmodified gaussians . if we enlarge the boundary of the reduced subspace sufficiently so that there is little overlap between the state represented and the deformed gaussians , then @xmath284 . therefore , @xmath282 may achieve accuracy equal to @xmath238 provided the reduced subspace is expanded . @xmath285 and @xmath264 provide almost identically poor performance , never achieving an accuracy of @xmath286 , except when utilizing the full grid , where the difference between the various reduced hamiltonians disappears . this is again expected , as by using the projector @xmath256 instead of @xmath174 on the input state to the hamiltonian , we are projecting into a subspace spanned by gaussians ( section [ sub : check - check ] ) , which is non - sparse . as analyzed in @xcite , in this case the full hilbert space is required to achieve highly accurate results . to conclude , for highly accurate results , @xmath247 is the form of choice . in some instances where high accuracy is not paramount , or one opts for a significant increase in the reduced subspace size , @xmath255 may allow for greater speed . we provide some illustrative examples of the use of the pvb representation for both the tise and tdse . the algorithms used to solve the time dependent and time independent schrdinger are described in detail in @xcite and an application of pvb to the challenging problem of double ionization of helium , is given in @xcite . the hamiltonian for the morse oscillator , in atomic units , is @xmath287 the pvb representation for the @xmath288 eigenstate is shown in fig . [ fig : ex_states_morse ] . the mean error of the first @xmath289 states as a function of the reduced basis size is depicted in fig . [ fig : ex_inv_comp_morse ] . the behavior of @xmath282 compared to @xmath238 for the morse oscillator is similar to their behavior for the harmonic oscillator ground state ( fig . [ fig : h1v2v3v4 ] ) . if the reduced basis is large enough , so that the basis functions near the subspace boundary have negligible overlap with the state , then @xmath290 is almost equal to @xmath291 and the results using @xmath282 do not differ significantly from the results using @xmath238 . in other words , @xmath292 is similar to @xmath293 . at the other extreme , if the hilbert space is pruned too drastically , parts of phase space with significant population are left out of the reduced subspace . in that case , the errors induced by the excessive pruning become more significant than the errors caused by approximating @xmath130 by @xmath213 and the distinction between @xmath238 and @xmath282 becomes negligible . we conclude that @xmath282 may be useful for low accuracy applications . eigenstate of the one dimensional morse oscillator , eq . . pixel colors represent the absolute value squared of the pvb expansion coefficients . only the basis functions corresponding to colored pixels are part of the reduced subspace.[fig : ex_states_morse ] ] eigenstates in a one dimensional morse oscillator , as a function of the ratio between the reduced and unreduced basis sizes . the system supports @xmath294 bound states ; the size of the reduced basis is @xmath295 . the error is shown for the two hamiltonians , @xmath238 and @xmath282 . eigenstate error is computed as the infidelity with respect to the eigenstate computed with the unreduced hilbert space . ] the observations of section [ sec : ex_morse ] are applicable to higher dimensional systems . we repeat the previous analysis for a coupled two - dimensional harmonic oscillator described by @xmath296 figure [ fig : ex_inv_comp_cho2d ] shows the error of the first @xmath297 eigenstates of this system as a function of the subspace size . the results are qualitatively similar to those of fig . [ fig : ex_inv_comp_morse ] . for a large reduced space , the results of @xmath238 and @xmath282 are similar , with the results of @xmath282 actually slightly more accurate . the slightly higher accuracy of @xmath282 may originate from the additional inversion of @xmath298 that is required for @xmath238 . for intermediate basis sizes , the error of @xmath282 is larger than for @xmath238 , until the error introduced by excessive pruning of the basis dominates at small basis sizes . eigenstates in a two dimensional coupled harmonic oscillator , eq . , as a function of the ratio of the reduced and unreduced basis sizes . the error is shown for the two hamiltonians , @xmath238 and @xmath282 . ] we now turn to the time - dependent case . when propagating a wavefunction in time , the reduced basis changes as the phase space occupied by the state evolves . to compare the accuracy of @xmath238 and @xmath282 , we consider a two dimensional double well potential with strong coupling between @xmath48 and @xmath299 , with @xmath300 and the initial wave packet is given by @xmath301 the wave packet was propagated from @xmath302 to @xmath303 . the potential surface and the initial and final wave packets are shown in fig . [ fig : ex_dwell_states ] . at the final time , the wave packet has spread across the barrier and oscillates in @xmath48 and @xmath299 . the wave packet was also investigated at @xmath304 ( not shown ) , when the packet is at the barrier but still retains a compact form . coordinate is proportional to the absolute value squared of its amplitude.[fig : ex_dwell_states ] ] figure [ fig : ex_inv_comp_dwelldyn ] compares the accuracy of the dynamics computed using @xmath238 and @xmath282 . here we observe a significant deterioriation in @xmath282 performance . specifically , @xmath282 fails to achieve @xmath305 accuracy , both at @xmath306 and @xmath307 , unless more than half of the unreduced hilbert space is used . this is in stark contrast to @xmath238 , which can achieve better than @xmath308 accuracy with less than @xmath280 of the hilbert space . moreover , the @xmath309 accuracy of @xmath282 for a very small basis size at @xmath306 is lost for @xmath307 . both aspects of this reduction in @xmath282 accuracy can be explained by recognizing the importance of the low - amplitude boundary dynamics as the wavepacket tunnels across the barrier . the gaussians at the boundary are the first to be removed in the pruning process , as depicted in fig . [ fig : deformed_gaussians ] . therefore , @xmath282 may not be used where the low - amplitude boundary is essential for getting acceptable accuracy . in contrast , @xmath238 can be used in such situations with no difficulty , as long as the amplitude cutoff is sufficiently low so as to retain the tunneling amplitude . and @xmath282 , at two times @xmath306 and @xmath307 . ] in this paper we presented the mathematical underpinnings of the pvb ( biorthogonal von neumann ) method for quantum mechanical simulations . the pvb method exploits the phase phase space localization of the von neumann basis to provide a sparse representation of quantum mechanical states that spans only the part of phase space where there is significant amplitude . this in turn can lead to significant computational savings in both cpu and memory . a detailed analysis was given of the subtle issues of projection onto subspaces of biorthogonal bases . two complementary ways of understanding this projection were provided . the first focuses on the basis functions : it was shown that under projection one of the biorthogonal bases remains unchanged while the other becomes significantly distorted . we showed that the distortions may be viewed as arising from subtraction of the @xmath156 gaussians that span the complementary subspace . this explains why gaussians near the boundary of the reduced phase space boundary are significantly distorted , while gaussians far from the boundary are essentially unperturbed . the second way to understand the effect of non - orthogonal projection focuses on the coefficients : in the @xmath156 basis all @xmath310 basis vectors contribute to the coefficients , since all basis vectors overlap one another due to the non - orthogonality of the basis . we then analyzed the various representations of the schrdinger equation in the reduced basis and approximations thereto . we concluded that for high - accuracy applications @xmath311 ( eq . [ eq : h1_def ] ) is the preferred form , although it comes which a relatively high computational cost . for medium to low accuracy applications , an approximate form , @xmath312 ( eq . [ eq : h2 ] ) may be used . several numerical examples were brought , showing the relative merits of @xmath238 and @xmath282 . a more challenging application of pvb , the double ionization of helium , is presented in @xcite . despite the significant methodological progress further development is possible . specifically , one may further reduce the representation by decomposing multi - dimensional objects into a sum - of - products , and truncating the sum when the correlation is sufficiently low . this strategy is used by the potfit algorithm to decompose the potential , but a similar approach could be used for the wavefunction and the reduced hamiltonian . this is a challenging problem , however , as the dynamics continuously modify the reduced basis , which generally is not easily decomposable . further areas of research include the correspondence between pvb and other phase space representations , including the discrete husimi and wigner representation . we also plan to explore the treatment of particle symmetries ( bosonic , fermionic ) in multi - particle implementations of the pvb method . beyond method development , there is a wide range of problems which are amenable to the pvb methodology , including high - harmonic generation , multi - electron ionization , photodissociation and chemical reactions . we intend to explore these applications in the near future . to conclude , pvb is an accurate , scalable and efficient method for quantum dynamics simulations , and it is our hope that it will find its place as part of the standard quantum numerics toolbox . the authors thank tucker carrington for useful correspondence . this work was supported by the israel science foundation ( 533/12 ) , the minerva foundation with funding from the federal german ministry for education and research and the koshland center for basic research . h. r. larsson acknowledges support by the deutscher akademischer austauschdienst , studienstiftung des deutschen volkes and the fonds der chemischen industrie . @ifundefined cabrera , r. ; bondar , d. i. ; jacobs , k. ; rabitz , h. a. _ phys . * 2015 * , _ 92 _ , 042122 esposti , m. d. ; graffi , s. in _ the mathematical aspects of quantum maps _ ; esposti , m. d. , graffi , s. , eds . ; springer berlin heidelberg : berlin , heidelberg , 2003 ; chapter mathematical aspects of quantum maps , pp 4990 neumann , j. _ mathematische annalen _ * 1931 * , _ 104 _ , 570578 gabor , d. _ journal of the institution of electrical engineers - part iii : radio and communication engineering _ * 1946 * , _ 93 _ , 429441 davis , m. j. ; heller , e. j. _ j. chem . phys . _ * 1979 * , _ 71 _ , 33833395 poirier , b. ; salam , a. _ j. chem . phys . _ * 2004 * , _ 121 _ , 17041724 lombardini , r. ; poirier , b. _ phys . e _ * 2006 * , _ 74 _ , 036705 halverson , t. ; poirier , b. _ chemical physics letters _ * 2015 * , _ 624 _ , 3742 halverson , t. ; poirier , b. _ j. phys . a _ * 2015 * , _ 119 _ , 1241712433 , pmid : 26418314 dimler , f. ; fechner , s. ; rodenberg , a. ; brixner , t. ; tannor , d. j. _ new journal of physics _ * 2009 * , _ 11 _ , 105052 shimshovitz , a. ; tannor , d. j. _ phys . lett . _ * 2012 * , _ 109 _ , 070402 takemoto , n. ; shimshovitz , a. ; tannor , d. j. _ j. chem . * 2012 * , _ 137 _ , 011102 whittaker , e. t. _ proceedings of the royal society of edinburgh _ * 1915 * , _ 35 _ , 181194 nyquist , h. _ american institute of electrical engineers , transactions of the _ * 1928 * , _ 47 _ , 617644 shannon , c. _ proceedings of the ire _ * 1949 * , _ 37 _ , 1021 kosloff , r. in _ numerical grid methods and their application to schrdinger s equation _ ; cerjan , c. , ed . ; springer netherlands : dordrecht , 1993 ; chapter the fourier method , pp 175194 assmat , e. ; machnes , s. ; tannor , d. j. _ arxiv:1502.05165 _ * 2015 * , marston , c. c. ; balint - kurti , g. g. _ j. chem . * 1989 * , _ 91 _ , 35713576 fattal , e. ; baer , r. ; kosloff , r. _ phys . rev . * 1996 * , _ 53 _ , 12171227 colbert , d. t. ; miller , w. h. _ j. chem . phys . _ * 1992 * , _ 96 _ , 19821991 tannor , d. j. _ introduction to quantum mechanics : a time - dependent perspective , by david j. tannor . isbn-10 1 - 891389 - 23 - 8 ; isbn-13 978 - 1 - 891389 - 23 - 8 . published by university science books , sausalito , ca usa , 2007 . _ ; university science books , 2007 ref . , supplementary material 1 note the different definition of the norm from that in ref . genossar , t. ; porat , m. _ ieee transactions on systems , man and cybernetics _ * 1992 * , _ 22 _ , 449460 tannor , d. j. ; takemoto , n. ; shimshovitz , a. _ advances in chemical physics 156 ( 2014 ) 134 _ ; 2014 shimshovitz , a. ; bacic , z. ; tannor , d. j. _ j. chem . phys . _ * 2014 * , _ 141 _ , 234106 brown , j. ; carrington , t. _ j. chem . phys . _ * 2015 * , _ 143 _ , 044104 fechner , s. ; dimler , f. ; brixner , t. ; gerber , g. ; tannor , d. j. _ opt . express _ * 2007 * , _ 15 _ , 1538715401 daubechies , i. _ ieee transactions on information theory _ * 1990 * , _ 36 _ , 9611005 the transformation from the fourier to the pvn basis can be accomplished using approximately @xmath313 operations using a fast gabor transform @xcite , which is closely related to the fast fourier transform . bai , z. ; demmel , j. ; dongarra , j. ; ruhe , a. ; van der vorst , h. in _ templates for the solution of algebraic eigenvalue problems _ ; bai , z. , demmel , j. , dongarra , j. , ruhe , a. , van der vorst , h. , eds . ; society for industrial and applied mathematics , 2000 machnes , s. ; assmat , e. ; tannor , d. j. _ to be published _ * 2016 * , wang , l. ; chen , c .- t . ; lin , w .- c . _ ieee transactions on image processing _ * 1994 * , _ 3 _ , 8792

we describe convex hulls of the simplest compact space curves , reducible quartics consisting of two circles . when the circles do not meet in complex projective space , their algebraic boundary contains an irrational ruled surface of degree eight whose ruling forms a genus one curve . we classify which curves arise , classify the face lattices of the convex hulls and determine which are spectrahedra . we also discuss an approach to these convex hulls using projective duality .

1612.09382

convex algebraic geometry studies convex hulls of semialgebraic sets @xcite . the convex hull of finitely many points , a zero - dimensional variety , is a polytope @xcite . polytopes have finitely many faces , which are themselves polytopes . the boundary of the convex hull of a higher - dimensional algebraic set typically has infinitely many faces which lie in algebraic families . ranestad and sturmfels @xcite described this boundary using projective duality and secant varieties . for a general space curve , the boundary consists of finitely many two - dimensional faces supported on tritangent planes and a scroll of line segments , called the edge surface . these segments are stationary bisecants , which join two points of the curve whose tangents meet . we study convex hulls of the simplest non - trivial compact space curves , those which are the union of two circles lying in distinct planes . zero - dimensional faces of such a convex hull are extreme points on the circles . one - dimensional faces are stationary bisecants . it may have two - dimensional faces coming from the planes of the circles . it may have finitely many nonexposed faces , either points of one circle whose tangent meets the other circle , or certain tangent stationary bisecants . [ fig : firsttaste ] shows some of this diversity . in the convex hull on the left , the discs of both circles are faces , and every face is exposed . in the oloid in the middle , the discs lie in the interior , @xmath0 of each circle is extreme , and no endpoint of an arc is exposed . in the convex hull on the right , there are two nonexposed stationary bisecants lying on its two - dimensional face , which is the convex hull of one circle and the point where the other circle is tangent to the plane of the first . these objects have been studied before . paul schatz discovered and patented the oloid in 1929 @xcite , this is the convex hull of two congruent circles in orthogonal planes , each passing through the center of the other . it has found industrial uses @xcite , and is a well - known toy . a curve in @xmath1 may roll along its edge surface . when rolling , the oloid develops its entire surface and has area equal to that of the sphere @xcite with equator one of the circles of the oloid . other special cases of the convex hull of two circles have been studied from these perspectives @xcite . this paper had its origins in subsection 4.1 of @xcite , which claimed that the edge surface for a general pair of circles is composed of cylinders . our results show that this is only the case when the two circles either meet in two points or are mutually tangent in all other cases , the edge surface has higher degree and it is an irrational surface of degree eight when the circles are disjoint in @xmath2 . this is related to problem 3 on convexity in @xcite , on the convex hull of three ellipsoids in @xmath1 . an algorithm was presented in @xcite ( see the video @xcite ) , using projective duality . we sketch this in section [ s : duality ] , and also apply duality to the convex hull of two circles . in section [ s : background ] we recall some aspects of convexity and convex algebraic geometry , and show that the convex hull of two circles is the projection of a spectrahedron . we study the edge surface and the edge curve of stationary bisecants of complex conics @xmath3 in section [ s : relaxation ] . we show that the edge curve is a reduced curve of bidegree @xmath4 in @xmath5 , and if @xmath6 and neither is tangent to the plane of the other , we show that the edge surface has degree eight . we also classify which curves of bidegree @xmath4 arise as edge curves to two conics . all possibilities occur , except a rational curve with a cusp singularity and a maximally reducible curve . in section [ s : circles ] we classify the possible arrangements of two circles lying in different planes in terms that are relevant for their convex hulls . we determine the face lattice and the real edge curve of each type and show that these convex hulls are spectrahedra only when the circles lie on a quadratic cone . we review some fundamental aspects of convexity and convex algebraic geometry , summarize our results about convex hulls of pairs of circles and their edge curves and show that any such convex hull is the projection of a spectrahedron . the convex hull of a subset @xmath7 is @xmath8 a set @xmath9 is _ convex _ if it equals its convex hull . a point @xmath10 is _ extreme _ if @xmath11 . a compact convex set is the convex hull of its extreme points . a convex subset @xmath12 of a convex set @xmath9 is a _ face _ if @xmath12 contains the endpoints of any line segment in @xmath9 whose interior meets @xmath12 . a _ supporting hyperplane _ @xmath13 is one that meets @xmath9 with @xmath9 lying in one of the half - spaces of @xmath14 defined by @xmath13 . a supporting hyperplane @xmath13 supports a face @xmath12 of @xmath9 if @xmath15 , and it _ exposes _ @xmath12 if @xmath16 . not all faces of a convex set are exposed . the boundary of the convex hull of two coplanar circles in fig . [ f : circlesinplane ] consists of one arc on each circle and two bitangent segments . an endpoint @xmath17 of an arc is not exposed . the only line supporting @xmath17 is the tangent to the circle at @xmath17 , and this line also supports the adjoining bitangent . ( 120,60 ) ( 0,0 ) ( 104,54)@xmath17 a fundamental problem from convex optimization is to describe the faces of a convex set , determining which are exposed , as well as their lattice of inclusions ( the _ face lattice _ ) . for more on convex geometry , see @xcite . convex algebraic geometry is the marriage of classical convexity with real algebraic geometry . a real algebraic variety @xmath18 is an algebraic variety defined over @xmath19 . if @xmath18 is irreducible and contains a smooth real point , then its real points are zariski - dense in @xmath18 , so it is often no loss to consider only the real points . conversely , many aspects of a real algebraic variety are best understood in terms of its complex points . studying the complex algebraic geometry aspects of a question from real algebraic geometry is its _ algebraic relaxation_. this relaxation enables the use of powerful techniques from complex algebraic geometry to address the original question . we also consider _ semialgebraic sets _ , which are defined by polynomial inequalities . by the tarski - seidenberg theorem on quantifier elimination @xcite , the class of semialgebraic sets is closed under projections and under images of polynomial maps . a closed semialgebraic set is _ basic _ if it is a finite intersection of sets of the form @xmath20 , for @xmath21 a polynomial . motivating questions about convex algebraic geometry were raised in @xcite . a fundamental convex semialgebraic set is the cone of positive semidefinite matrices ( the _ @xmath22 cone _ ) . these are symmetric matrices with nonnegative eigenvalues . the boundary of the @xmath22 cone is ( a connected component of ) the determinant hypersurface and every face is exposed . spectrahedron _ is an affine section @xmath23 of this cone . write @xmath24 to indicate that @xmath25 . parameterizing @xmath26 shows that a spectrahedron is defined by a _ linear matrix inequality _ , @xmath27 where @xmath28 are real symmetric matrices . images of spectrahedra under linear maps are _ spectrahedral shadows_. semidefinite programming provides efficient methods to optimize linear objective functions over spectrahedra and their shadows , and a fundamental question is to determine if a given convex semialgebraic set may be realized as a spectrahedron or as a spectrahedral shadow , and to give such a realization . scheiderer showed that the convex hull of a curve is a spectrahedral shadow @xcite , and recently showed that there are many convex semialgebraic sets which are not spectrahedra or their shadows @xcite . since the optimizer of a linear objective function lies in the boundary , convex algebraic geometry also seeks to understand the boundary of a convex semialgebraic set . this includes determining its faces and their inclusions , as well as the zariski closure of the boundary , called the _ algebraic boundary_. this was studied for rational curves @xcite and for curves in @xmath1 by ranestad and sturmfels @xcite . they showed that the algebraic boundary of a space curve @xmath29 consists of finitely many tritangent planes and a ruled _ edge surface _ composed of stationary bisecant lines . a _ stationary bisecant _ is a secant @xmath30 to @xmath29 ( @xmath31 ) such that the tangent lines @xmath32 and @xmath33 to @xmath29 at @xmath34 and @xmath35 meet . for a general irreducible space curve of degree @xmath36 and genus @xmath37 , the edge surface has degree @xmath38 @xcite . for example , suppose that @xmath29 is a general space quartic ( see ( * ? ? ? 5.5 ) or ( * ? ? ? * ex . 2.3 ) ) . this is the complete intersection of two real quadrics @xmath39 and @xmath40 , and has genus one by the adjunction formula ( * ? ? ? v.1.5.2 ) . its edge surface has degree @xmath41 and is the union of four cones . in the pencil of quadrics @xmath42 for @xmath43\in{\mathbb{p}}^1 $ ] that contain @xmath29 , four are singular and are given by the roots of @xmath44 . here , the quadratic forms @xmath45 are expressed as symmetric matrices . each singular quadric is a cone and each line on that cone is a stationary bisecant of @xmath29 . a general point of @xmath29 lies on four stationary bisecants , one for each cone . the union of two circles in different planes is also a space quartic , but it is not in general a complete intersection ( the complex points of a complete intersection are connected ) . we therefore expect a different answer than for general space quartics . we give a taste of that which is to come . [ th : one ] let @xmath46 and @xmath47 be circles in @xmath1 lying in different planes . their convex hull is a spectrahedron if and only if the scheme @xmath48 has length 2 . when the complex points of the circles are disjoint and neither is tangent to the plane of the other , the edge surface is irreducible and has degree eight . its rulings are parameterized by a smooth curve of genus one in @xmath5 . a general point of @xmath49 lies on two stationary bisecants . this is proven in lemma [ l : stbis ] , and in theorems [ th : edgecurve ] , [ th : degreeeight ] and [ th : issectrhedron ] . we study stationary bisecants and edge surfaces in the algebraic relaxation of our problem of two circles , replacing circles in @xmath1 by smooth conics in @xmath50 . a conic @xmath29 in @xmath51 spans a plane . let @xmath46 and @xmath47 be conics spanning different planes , @xmath52 and @xmath53 , respectively . a stationary bisecant is spanned by points @xmath54 and @xmath55 with @xmath56 whose tangent lines @xmath57 and @xmath58 meet . set @xmath59 . [ l : stbis ] a point @xmath54 lies on two stationary bisecants unless the tangent line @xmath57 meets @xmath47 . if the tangent line meets @xmath47 , then it is the unique stationary bisecant through @xmath17 unless @xmath60 or @xmath57 lies in the plane @xmath53 of @xmath47 . when @xmath61 , all lines in @xmath53 through @xmath17 are stationary bisecants . see fig . [ f : stationarybisecants ] for reference . ( 330,117)(-15,-10 ) ( -15,-10 ) ( 179,87.8)(-3,-1)45 ( 179,88.2)(-3,-1)45 ( 218,62.3)(-2,-1)28 ( 218,62.7)(-2,-1)28 ( 30.5,27.8)(3,1)50 ( 30.5,28.2)(3,1)50 ( 25,52)@xmath53 ( 72,98)@xmath46 ( 115,26)@xmath47 ( 123,101)@xmath17 ( 228,21)@xmath62 ( 247,13)@xmath63 ( 148,11)@xmath64 ( 222,39)@xmath65 ( 181,86)stationary bisecants through @xmath17 ( 179,88)(-3,-1)47 ( 179,85)(-1,-1)12 ( 219,62)@xmath66 ( 218,62.5)(-2,-1)29 ( 30.5,28)(3,1)52 ( -9,19)tangent stationary ( -9,7)bisecant consider the tangent line @xmath57 for @xmath54 . either @xmath67 in case @xmath68 , let @xmath62 be the point where @xmath57 meets @xmath53 . there are further cases . when @xmath69 , there are two tangents to @xmath47 that meet @xmath62 , and the lines through @xmath17 and each point of tangency ( @xmath70 in fig . [ f : stationarybisecants ] ) give two stationary bisecants through @xmath17 . if @xmath55 and @xmath56 , then the tangent line @xmath57 is the only stationary bisecant through @xmath17 . in case @xmath71 , the tangent line @xmath57 meets every tangent to @xmath47 , and every line in @xmath53 through @xmath17 ( except @xmath72 if @xmath60 ) meets @xmath47 and is therefore a stationary bisecant . if @xmath60 , then the tangent line @xmath72 is a limit of such lines . when @xmath46 is tangent to the plane @xmath53 at a point @xmath17 , the pencil of lines in @xmath53 through @xmath17 are _ degenerate stationary bisecants_. when @xmath73 , a general line in the pencil meets @xmath47 twice so that the map from @xmath47 to this pencil has degree two . lines that meet @xmath46 and @xmath47 in distinct points are given by points @xmath74 with @xmath56 . the _ edge curve @xmath75 _ is the zariski closure of the set of points @xmath76 such that @xmath77 is a stationary bisecant . as a smooth conic is isomorphic to @xmath78 , the edge curve is a curve in @xmath79 . subvarieties of products of projective spaces have a multidegree ( see @xcite ) . for a curve @xmath29 in @xmath79 this becomes its bidegree @xmath80 , where @xmath81 is the number of points in the intersection of @xmath29 with @xmath82 for @xmath62 general and @xmath83 the number of points in the intersection of @xmath29 with @xmath84 for @xmath17 general . as @xmath82 has bidegree @xmath85 and @xmath84 bidegree @xmath86 , the intersection pairing on curves in @xmath79 , expressed in terms of bidegree , is @xmath87 a curve of bidegree @xmath80 is defined in homogeneous coordinates @xmath88,[u : v])$ ] for @xmath79 by a bihomogeneous polynomial that has degree @xmath81 in @xmath89 and @xmath83 in @xmath90 . [ th : edgecurve ] the edge curve @xmath75 has bidegree @xmath4 . in the projection to @xmath46 , two points of @xmath75 map to a general point @xmath54 , by lemma [ l : stbis ] . thus the intersection number of @xmath75 with @xmath91 is 2 , and vice - versa for @xmath92 , for general @xmath55 . consequently , @xmath75 has bidegree @xmath4 . we compute the defining equation of @xmath75 to give a second proof . first , parameterize the conics . let @xmath93 for @xmath94 be two quadruples of homogeneous quadrics that each span @xmath95 . each quadruple gives a map @xmath96 whose image is a conic @xmath97 . the plane @xmath98 of @xmath97 is defined by the linear relation among @xmath99 , and we assume that @xmath100 . in coordinates , if @xmath101\in{\mathbb{p}}^1 $ ] , then @xmath102\ = \ [ f_{i,0}(s , t):f_{i,1}(s , t):f_{i,2}(s , t):f_{i,3}(s , t)]\,\ ] ] is the corresponding point of @xmath97 . its tangent line is spanned by @xmath103 and @xmath104 , where @xmath105 , as @xmath106 , for a homogeneous quadric @xmath21 . the points @xmath107 $ ] and @xmath108 $ ] span a stationary bisecant when their tangents meet . equivalently , when @xmath109 as the first two rows have bidegree @xmath86 and the second two have bidegree @xmath85 , this form @xmath110 has bidegree @xmath4 . [ ex : edge_curves ] let @xmath46 and @xmath47 be unlinked unit circles where @xmath46 is centered at the origin and lies in the @xmath111-plane and @xmath47 is centered at @xmath112 and lies in the @xmath113-plane . choosing homogeneous coordinates @xmath114 $ ] for @xmath51 where @xmath115 , these admit parametrizations @xmath116\ \longmapsto\ [ s^2+t^2 : s^2-t^2 : 2st:0 ] \quad\mbox{and}\quad [ u : v]\ \longmapsto\ [ u^2+v^2 : 2u^2 + 4v^2 : 0 : 2uv]\,.\ ] ] dividing the determinant by @xmath117 gives the equation for the edge curve @xmath75 , @xmath118 which is irreducible . we draw @xmath75 below in the window @xmath119 in @xmath120 . @xmath121{pictures / unlinked_wide_curve}}\ ] ] the zariski closure of the union of all stationary bisecants is the _ edge surface @xmath122_. by lemma [ l : stbis ] , a general point of one of the conics lies on two stationary bisecants . therefore , each conic is a curve of self - intersections of @xmath122 , and the multiplicity of @xmath122 at a general point of a conic is @xmath123 . [ th : degreeeight ] the edge surface @xmath122 has degree eight when @xmath6 and neither is tangent to the plane of the other . the line @xmath59 meets each conic in two points , and therefore @xmath122 in at least four points . any other point @xmath124 lies on a stationary bisecant @xmath125 between a point @xmath17 of @xmath46 and a point @xmath62 of @xmath47 . as @xmath126 , we have @xmath127 , and similarly @xmath128 . thus @xmath129 , but @xmath63 is not a stationary bisecant , a contradiction . each of the four points of @xmath130 has multiplicity two on @xmath122 by lemma [ l : stbis ] and the observation preceding the statement of the theorem . thus @xmath122 has degree eight . for a second proof , let @xmath125 be a general line that meets @xmath122 transversally . the points of @xmath131 lie on stationary bisecants that meet @xmath125 . we count these . let @xmath132 be the curve whose points are pairs @xmath76 such that the secant line spanned by @xmath17 and @xmath62 meets @xmath125 . the stationary bisecants that meet @xmath125 are the points of intersection of @xmath133 and @xmath75 . we compute the bidegree of @xmath133 . fix a point @xmath54 with @xmath134 . secant lines through @xmath17 rule the cone over @xmath47 with vertex @xmath17 . as this quadratic cone meets @xmath125 in two points , @xmath135 . the symmetric argument with a point of @xmath47 , shows that @xmath133 has bidegree @xmath4 . by , @xmath133 meets @xmath75 in @xmath136 points . this proves the theorem . the arguments in this proof using intersection theory are similar to arguments used in the contributions @xcite in this volume . each irreducible component @xmath29 of the edge curve @xmath75 gives an algebraic family of stationary bisecants and an irreducible component @xmath137 of the edge surface @xmath122 . if @xmath29 has bidegree @xmath80 , then the corresponding component @xmath137 of @xmath122 has degree at most @xmath138 . this is not an equality when the intersection @xmath139 has a basepoint or when the general point of @xmath137 contains two stationary bisecants . this occurs when one circle is tangent to the plane of the other , and there are one or more components of degenerate stationary bisecants . [ ex : edge_curves2 ] the real points of the edge curve of example [ ex : edge_curves ] had two connected components ( the picture showed a patch of @xmath120 ) . thus the real points of the edge surface has two components . stationary bisecants corresponding to the oval in the center of lie along the convex hull , which shown on the left below . the others bound a non - convex set that lies inside the convex hull . we display it in an expanded view on the right below . @xmath140{pictures / unlinked}\qquad\qquad \includegraphics[height=85pt]{pictures / unlinked_inside}\ ] ] the planes of the circles meet in the @xmath34-axis . for sufficiently small @xmath141 , the line defined by @xmath142 meets @xmath122 transversally . near each point of a circle lying on the @xmath34-axis it meets @xmath122 in two points , one for each of the two families of stationary bisecants passing through the nearby arc of the circle . these eight points are real . a curve of bidegree @xmath4 on @xmath79 has arithmetic genus one , by the adjunction formula ( * ? ? ? v.1.5.2 ) . if smooth , it is an irrational genus one curve . another way to see this is that the projection to a @xmath78 factor is two - to - one , except over the branch points , of which there are four , counted with multiplicity . indeed , writing its defining equation as a quadratic form in the variables @xmath143 for the second @xmath78 factor , its coefficients are quadratic forms in the variables @xmath89 of the first @xmath78 . the projection to the first has branch points where the discriminant vanishes , which is a quartic form . by elementary topology , a double cover of @xmath144 with four branch points has euler characteristic zero , again implying that it has genus one . [ l : any4points ] for every set @xmath145 of four points of @xmath46 , there is a conic @xmath47 such that the projection to @xmath46 of the edge curve is branched over @xmath145 . let @xmath17 be the point of intersection of two of the tangents to @xmath46 at points of @xmath145 and @xmath62 be the point of intersection of the other two tangents ( see fig.[f : jinvariant ] ) . since the tangent @xmath146 at any point @xmath147 meets @xmath47 ( in one of the points @xmath17 or @xmath62 ) , lemma [ l : stbis ] implies this is the unique stationary bisecant involving the point @xmath65 . thus the points of @xmath145 are branch points of the projection to @xmath46 of the edge curve . ( 265,100 ) ( 0,0 ) ( 121,65)@xmath46 ( 18,68)@xmath52 ( 61,25)@xmath17 ( 207,33)@xmath62 ( 117.5,47)@xmath145 ( 119.7,39.3)(0,1)6.7 ( 149,39.8333)(-3,1)26 ( 160,55.8333)(-6,-1)37 ( 97.5,53.1666)(6,-1)19 ( 166 , 1)@xmath47 [ r : families ] there are three families of conics @xmath47 giving an edge curve branched over @xmath145 . these correspond to the three partitions of @xmath145 into two parts of size two . each partition determines two points @xmath148 on the plane @xmath52 of @xmath46 where the tangent lines at the points in each part meet . the corresponding family is the collection of conics @xmath47 that meet @xmath52 transversally in @xmath17 and @xmath62 . if both @xmath46 and @xmath149 are real , and we choose an affine @xmath1 containing the points @xmath17 and @xmath62 , then we may choose @xmath47 to be a circle . the isomorphism class of a smooth genus one curve is determined by its @xmath150-invariant @xcite . this may be computed from the branch points @xmath145 of any degree two map to @xmath78 . explicitly , choose coordinates on @xmath78 so that the branch points @xmath145 are @xmath151 . then the @xmath150-invariant is @xmath152 we have the following corollary of lemma [ l : any4points ] . for every conic @xmath46 and every @xmath153 , there is a conic @xmath47 such that the edge curve has @xmath150-invariant @xmath154 . when @xmath46 and @xmath149 are real , @xmath47 may be a circle . we now classify the possible edge curves @xmath75 to a pair of conics @xmath46 and @xmath47 lying in distinct planes @xmath52 and @xmath53 . we show in the proof of lemma [ l : fivepossibilities ] that every component of @xmath75 is reduced . if @xmath155 is reducible , then @xmath156 . thus the bidegrees of the components of @xmath75 form a partition of @xmath4 . if @xmath75 is irreducible , then either it is smooth of genus one or singular of arithmetic genus one and hence rational . any curve of bidegree @xmath157 or @xmath158 is rational . table [ t : components ] gives the different possibilities , along with pictures of a real curve . @xmath159 ' '' '' + ' '' '' & & & + & @xmath160&@xmath160 & @xmath161 & @xmath161 ' '' '' + & & & @xmath162&@xmath162 + ' '' '' & & & & + [ th : edgecurves ] all types of @xmath4-curves of table [ t : components ] occur as the edge curve of a pair of conics @xmath163 lying in distinct planes except a curve with a cusp and a reducible curve @xmath159 with four components . for existence , see tables [ t : galleryi ] , [ t : galleryii ] and [ t : galleryiii ] , which display edge curves of two circles in all possible configurations . we rule out edge curves with a cusp and reducible edge curves of type @xmath159 . we first analyze the singularities of edge curves . [ l : fivepossibilities ] the edge curve @xmath75 is reduced . a point @xmath74 is a singular point of @xmath75 only if @xmath164 or @xmath61 or @xmath165 . there are five possibilities for @xmath148 and the tangents , up to interchanging the conics @xmath46 and @xmath47 . 1 . @xmath164 and the tangent to each conic at @xmath17 does not lie in the plane of the other . 2 . @xmath164 with @xmath166 , but @xmath167 . 3 . @xmath164 with both @xmath166 and @xmath168 . @xmath56 and @xmath166 , but @xmath167 . then @xmath169 is a stationary bisecant . @xmath56 and @xmath170 is @xmath171 , and is a stationary bisecant . let @xmath74 be a point on a curve @xmath75 of bidegree @xmath4 . if the fiber of @xmath75 in one of the projections from @xmath76 , say to @xmath47 , has exactly two points , then @xmath75 is smooth at @xmath76 . indeed , as @xmath75 is a @xmath4 curve , @xmath172 is either @xmath92 or one double or two simple points , and if two , then @xmath75 is smooth at each point . consequently , there are three possibilities for points of @xmath75 in the fibers of the projections to @xmath46 and @xmath47 containing a singular point @xmath76 . either 1 . @xmath76 is the only point of @xmath75 in both fibers , 2 . @xmath76 is the only point in one fiber and the other fiber is a component of @xmath75 , or 3 . both fibers are components of @xmath75 . now let @xmath75 be the edge curve , which is smooth at any point @xmath76 where there is another point in one of the two fibers of projections to @xmath97 . lemma [ l : stbis ] implies that there are two points in @xmath75 over a general point of either conic , so every component of @xmath75 is smooth at a generic point and therefore @xmath75 is reduced . by lemma [ l : stbis ] and the analysis above , a point @xmath173 is singular if and only if both tangents meet the other conic for otherwise there is a second point in one of the fibers . if @xmath166 , then every line in @xmath53 through @xmath17 is a stationary bisecant , so @xmath75 contains @xmath91 , which has bidegree @xmath86 . if @xmath165 , then @xmath75 contains @xmath92 , which has bidegree @xmath85 . if neither occurs but @xmath75 is singular at @xmath76 , then we are in case @xmath68 . when @xmath164 and we are not in case @xmath68 , then , up to interchanging the indices 1 and 2 , we are in either case @xmath71 or case @xmath174 . when @xmath56 , so that one circle is tangent to the plane of the other , then we are in either case @xmath175 or @xmath176 . we need only to rule out that the edge curve @xmath75 has type @xmath159 or has a cusp . by lemma [ l : fivepossibilities ] , @xmath75 has a component @xmath91 of bidegree @xmath86 exactly when @xmath46 is tangent to the plane @xmath53 at the point @xmath17 . since @xmath177 , there is at most one such point of tangency on @xmath46 , so @xmath75 has at most one component of bidegree @xmath86 , and the same is true for a component of bidegree @xmath85 . thus the type @xmath159 can not occur for an edge curve . we show that if @xmath173 is a singular point in case @xmath68 of lemma [ l : fivepossibilities ] , then @xmath75 has a node at @xmath76 , ruling out a cusp and completing the proof . suppose that @xmath164 and the tangents to each conic at @xmath17 do not lie in the plane of the other . choose coordinates @xmath178 for @xmath51 so that @xmath52 is the plane @xmath179 , @xmath53 is the plane @xmath180 , @xmath57 is the line @xmath181 , @xmath72 is @xmath182 and @xmath183 $ ] . then we may choose parametrizations near @xmath17 for @xmath46 and @xmath47 of the form @xmath184\\\rule{0pt}{12pt } c_2\ \colon\ u & \longmapsto & [ \,0\,:\ , \beta u^2\,:\,u + \alpha u^2\,:\ , 1+\gamma u+\delta u^2\,]\ , , \end{array}\ ] ] for some @xmath185 where @xmath186 . the edge curve is defined by @xmath187 indeed , the matrix has rows @xmath188 , where @xmath189 is the parameterization of @xmath97 . the determinant vanishes when the tangent to @xmath46 at @xmath190 meets the tangent to @xmath47 at @xmath191 . the lowest order terms in , @xmath192 , have distinct linear factors when @xmath186 . thus @xmath75 has a node when @xmath193 , which is @xmath194 . we classify the relative positions of two circles in @xmath1 and show that the combinatorial type of the face lattice of their convex hull depends only upon their relative position . this relative position is determined by the combinatorial type of the face lattice and the real geometry of the edge curve . we use this classification to determine when the convex hull of two circles is a spectrahedron . let @xmath163 be circles in @xmath1 lying in distinct planes @xmath52 and @xmath53 , respectively . the intersection @xmath195 in @xmath2 is either two real points , two complex conjugate points , or @xmath46 is tangent to @xmath53 at a single real point . let @xmath196 be the number of real points in this intersection , and the same for @xmath197 . order the circles so that @xmath198 , and call @xmath199 $ ] the _ intersection type _ of the pair of circles . the configuration of the circles is determined by the order of their points along the line @xmath200 . for example , @xmath46 and @xmath47 have _ order type @xmath201 along @xmath63 _ when they have intersection type @xmath202 $ ] and meet @xmath63 in distinct points which alternate . if @xmath203 , then we write _ @xmath145 _ for that shared point . for example , if @xmath46 meets @xmath63 in two real points with @xmath47 tangent to @xmath63 at one , then this pair has order type @xmath204 . the intersection type may be recovered from the order type . a further distinction is necessary for intersection type @xmath205 $ ] , when both circles meet @xmath63 in two complex conjugate points . in @xmath2 either @xmath6 or @xmath206 . write @xmath207 for the order type in the first case and @xmath208 in the second . by lemma [ l : fivepossibilities ] the edge curve is smooth in order type @xmath207 and singular in order type @xmath208 . there are fifteen possible order types of two circles in @xmath1 . see tables [ t : galleryi ] , [ t : galleryii ] and [ t : galleryiii ] for the order types of circles and their convex hulls . the possible intersection types are @xmath205 $ ] , @xmath209 $ ] , @xmath210 $ ] , @xmath211 $ ] , @xmath212 $ ] , and @xmath202 $ ] . for @xmath205 $ ] , we noted two order types , and intersection types @xmath209 $ ] and @xmath211 $ ] each admit one order type , namely @xmath213 and @xmath214 , respectively . for @xmath210 $ ] , each circle @xmath97 is tangent to @xmath63 at a point @xmath215 . either @xmath216 or @xmath217 , so there are two order types , @xmath218 and @xmath219 . for @xmath212 $ ] , the line @xmath63 is secant to circle @xmath46 and @xmath47 is tangent to @xmath63 at a point @xmath220 . either @xmath220 is in the exterior of @xmath46 or it lies on @xmath46 or it is interior to @xmath46 . these give three order types @xmath63 , @xmath221 , @xmath204 , and @xmath222 , respectively . finally , for @xmath202 $ ] there are three order types when all four point are distinct , @xmath223 , @xmath201 , and @xmath224 . when one point is shared , we have @xmath225 or @xmath226 . finally , both points may be shared , giving @xmath227 . & ' '' '' + @xmath205 $ ] @xmath207 & @xmath211 $ ] @xmath214 & @xmath202 $ ] @xmath224 + & & + & & ' '' '' + @xmath202 $ ] @xmath223 & @xmath202 $ ] @xmath201 & @xmath202 $ ] @xmath226 + & & + ' '' '' & & + @xmath202 $ ] @xmath225 & @xmath202 $ ] @xmath227 & @xmath205 $ ] @xmath208 + & & + ' '' '' & + @xmath212 $ ] @xmath222 & @xmath212 $ ] @xmath204 & @xmath212 $ ] @xmath221 + & & + & ' '' '' & + @xmath209 $ ] @xmath213 & @xmath210 $ ] @xmath218 & @xmath210 $ ] @xmath219 + & & + the order type of the circles determines the combinatorial type of the face lattice of their convex hull @xmath9 . describing the face lattice means identifying all ( families of ) faces of @xmath9 , their incidence relations , and which are exposed / not exposed . throughout , _ @xmath228 _ is the disc of the circle @xmath97 . we invite the reader to peruse our gallery in tables [ t : galleryi ] , [ t : galleryii ] and [ t : galleryiii ] while reading this classification . our main result is the following . [ th : facelattice ] the order type of @xmath163 determines the combinatorial type of the face lattice of @xmath9 , as summarized in table [ t : facelattices ] . there are eleven distinct combinatorial types of face lattice . the combinatorial type of the face lattice , together with the real algebraic geometry of its edge curve , determines the order type . we determine the face lattice for each order type . some general statements are given in preliminary results which precede our proof of theorem [ th : facelattice ] . the statements are asymmetric , with the symmetric statement obtained by interchanging 1 and 2 . we first study the section @xmath229 of @xmath9 , which contains @xmath230 . [ l : section ] @xmath231 . since @xmath232 , @xmath233 and therefore a point @xmath234 is a convex combination @xmath235 ( @xmath236 with @xmath237 ) of points @xmath238 and @xmath239 . if @xmath240 , then as @xmath241 , we must have that @xmath242 . [ c:2face ] if @xmath243 , then @xmath244 . otherwise , @xmath245 is the convex hull of @xmath230 and the one or two points of @xmath246 exterior to @xmath230 . a point @xmath54 is an extreme point of @xmath245 if and only if @xmath46 and @xmath246 lie on the same side of @xmath57 . an extreme point @xmath54 of @xmath245 is not exposed if and only if @xmath57 meets @xmath247 . extreme points of @xmath245 are extreme points of @xmath9 and non - exposed points of @xmath245 are non - exposed in @xmath9 . finally , @xmath245 is a face of @xmath9 if and only if @xmath248 . the first two statements follow from lemma [ l : section ] . the next two about extreme points @xmath17 of @xmath245 follow as @xmath57 is the only possible supporting line to @xmath245 at @xmath17 . the next , about extreme points of @xmath9 and its section @xmath245 , follows by lemma [ l : section ] , and the last is immediate as @xmath47 lies on one side of @xmath52 if and only if @xmath249 . by lemma [ l : stbis ] , a general point @xmath54 lies on two stationary bisecants . if @xmath10 is extreme , then these may support one - dimensional _ bisecant faces _ of @xmath9 . we determine the bisecant faces meeting most extreme points . any plane supporting an extreme point @xmath54 contains @xmath66 . if such a plane does not meet @xmath47 , then @xmath17 is exposed . [ l : bisecantfaces ] let @xmath10 be an extreme point of @xmath9 . if @xmath66 neither meets @xmath47 nor lies in @xmath53 , then @xmath17 is exposed . such a point @xmath17 lies on one bisecant face if @xmath248 and two if @xmath250 . when there are two , one is on each side of @xmath52 . let @xmath54 be an extreme point of @xmath9 such that @xmath66 neither meets @xmath47 nor lies in @xmath53 . by corollary [ c:2face ] , @xmath46 and @xmath246 lie on the same side of @xmath57 in @xmath52 . in the pencil @xmath251 of planes containing @xmath57 , those meeting @xmath9 form an interval @xmath252 containing @xmath52 and an interval @xmath253 of planes meeting @xmath47 . each endpoint of @xmath253 is a plane containing a stationary bisecant through @xmath17 . our assumptions on @xmath17 and @xmath57 imply that @xmath254 , so that @xmath17 is exposed . if @xmath248 , then @xmath52 is one endpoint of @xmath252 and the other is an endpoint of @xmath253 , otherwise the endpoints of @xmath252 are the endpoints of @xmath253 and @xmath52 is an interior point , which proves the lemma . [ r : program ] corollary [ c:2face ] to identifies the 2-faces , extreme points and some non - exposed points of @xmath9 . lemma [ l : bisecantfaces ] identifies most exposed points and bisecant edges . the rest of the face lattice is determined in the proof of theorem [ th : facelattice ] . we first understand the boundary of each section @xmath255 . [ f : kappas ] shows the possibilities when @xmath256 is not the disc @xmath228 . ( 84,67)(0,-15 ) ( 0,0).,title="fig : " ] ( 22,23)@xmath228 ( 37,52)@xmath257 ( 37,-2)@xmath220 ( 79,26)@xmath258 ( 35,-15)(a ) ( 82,72)(0,-10 ) ( 0,0).,title="fig : " ] ( 12,27.5)@xmath228 ( 8,49)@xmath257(8,8)@xmath220 ( 72 , 0)@xmath258 ( 72,57)@xmath259 ( 35,-10)(b ) ( 104,74)(-7,-10 ) ( 0,5).,title="fig : " ] ( -7,30)@xmath258 ( 93,30)@xmath259 ( 25,57)@xmath260 ( 60,57)@xmath261 ( 25 , 5)@xmath257 ( 60 , 5)@xmath220 ( 42,10)@xmath228 ( 41,-10)(c ) there , @xmath258 and @xmath259 are points of the other circle on the boundary of @xmath256 and points @xmath262 are non exposed points of @xmath97 as @xmath263 meets @xmath258 or @xmath259 . the line segment between @xmath258 and @xmath259 is where the disc of the second circle meets @xmath98 . we give separate arguments for each order type . * order type @xmath207 . * by corollary [ c:2face ] , both discs are faces of @xmath9 , and every point of the circles is extreme . by lemma [ l : bisecantfaces ] , all points of the circles are exposed and each lies on exactly one bisecant face . the same description holds for order type @xmath208 . as the edge curve for order type @xmath207 is smooth , while that for order type @xmath208 is singular , the edge curve distinguishes these two order types . * order type @xmath214 . * since @xmath264 and @xmath265 , @xmath230 is the only 2-face . the section @xmath266 is similar to fig . [ f : kappas ] ( b ) , so the extreme points on @xmath47 form an arc @xmath267 whose endpoints are not exposed , each lying on one bisecant edge . the interior points of @xmath267 are exposed by lemma [ l : bisecantfaces ] and each lies on two bisecant edges . similarly , every point of @xmath46 is exposd and lies on one bisecant edge . the same description holds for order type @xmath222 . its edge curve is singular , while order type @xmath214 has a smooth edge curve . * order type @xmath224 . * since @xmath268 , @xmath9 has no 2-faces . since @xmath47 meets the interior of @xmath230 , corollary [ c:2face ] implies that every point of @xmath46 is extreme and @xmath47 has two intervals of extreme points . the four endpoints are not exposed and each lies on one bisecant edge . by lemma [ l : bisecantfaces ] , every point of @xmath46 and of the interior of the arcs on @xmath47 is exposed and lies on two bisecant edges . * order type @xmath223 . * by corollary [ c:2face ] , @xmath9 has no 2-faces and each circle has one arc of extreme points , as the sections @xmath256 are similar to fig . [ f : kappas ] ( a ) . no endpoint of an arc is exposed and each lies on one bisecant edge , and each interior point of an arc is exposed and lies on two bisecant edges . the same description holds for order types @xmath201 and @xmath226 . the edge curve in type @xmath223 has two real components as seen in example [ ex : edge_curves2 ] , while for type @xmath201 there is one real component , and both are smooth . the edge curve for or type @xmath226 is singular . * order type @xmath225 . * again , @xmath9 has no 2-faces . all points of @xmath46 are extreme and @xmath47 has an arc of extreme points whose endpoints are not exposed and each lies on one bisecant edge . all interior points of that arc and of @xmath46except possibly the shared point @xmath17are exposed and lie on two bisecant edges . the tangents @xmath57 and @xmath72 span a plane exposing @xmath17 and @xmath17 lies on no bisecant edges . * order type @xmath227 . * there are no 2-faces and as in type @xmath225 every point of the circles is extreme , and the non - shared points are exposed and each lies on two bisecant edges . each shared point is exposed by the plane spanned by the two tangents at that point and neither shared point lies on a bisecant edge . * order type @xmath204 . * the only 2-face is @xmath230 . every point of @xmath46 is extreme and @xmath47 has an arc @xmath267 of extreme points with one endpoint , say @xmath257 , the shared point where @xmath47 is tanget to @xmath52 . neither endpoint is exposed and @xmath220 lies on one bisecant edge ( the bisecant @xmath269 meets the interior of @xmath230 ) . by lemma [ l : bisecantfaces ] , every point of @xmath46 except @xmath257 lies on one bisecant edge and every interior point of @xmath267 lies on two bisecant edges , and all of these are exposed . * order type @xmath221 . * the only 2-face is @xmath245 and its shape is as in fig [ f : kappas ] ( a ) with the vertex @xmath258 where @xmath270 is tangent to @xmath52 . there is an arc @xmath267 of extreme points of @xmath46 whose endpoints are not exposed with each lying on a bisecant edge @xmath271 . the section @xmath266 has the same shape and @xmath47 has an arc @xmath272 of extreme points with neither endpoint exposed . the point @xmath259 lies on one bisecant edge along @xmath273 and @xmath258 lies on two bisecant edges @xmath271 . neither of the edges @xmath271 is exposed as @xmath52 is the only supporting plane of @xmath9 containing either edge . finally , by lemma [ l : bisecantfaces ] , interior points of the arcs are exposed , with those from @xmath267 lying on one bisecant edge and those from @xmath272 lying on two . in the order types of the last row of table [ t : galleryiii ] , the circle @xmath47 is tangent to @xmath52 at a point @xmath258 and the tangent @xmath274 does not meet the interior of @xmath230 . in the pencil of planes containing @xmath274 , @xmath52 and @xmath53 are the endpoints of an interval of planes meeting @xmath275 and of an interval of planes that meet @xmath9 only in @xmath276 . thus , both sections @xmath245 and @xmath266 are 2-faces of @xmath9 and the face @xmath276 is exposed . * order type @xmath213 . * here @xmath277 and @xmath265 . the 2-face @xmath266 has the same shape as in order type @xmath221 . the description of the points and bisecant edges meeting @xmath47 is also the same . by lemma [ l : bisecantfaces ] and the preceeding observation , every point of @xmath46 is exposed , and all lie on a unique bisecant edge except @xmath258 , which lies on the two non exposed bisecant edges @xmath271 . * order type @xmath218 . * this is the most complicated . each circle is tangent to the plane of the other , sharing a tangent line , and the description is symmetric in the indices @xmath278 and @xmath123 . the 2-faces are the sections @xmath245 and @xmath266 , with the description for each is nearly the same as for @xmath245 in order type @xmath221 . the exception is the bisecant edge @xmath279 lying along the shared tangent . this is exposed , but neither endpoint is exposed . it is also isolated from the other bisecant edges , which form a continuous family . * order type @xmath219 . * the two circles are mutually tangent at a point @xmath17 . the 2-faces are @xmath230 and @xmath270 , every point of either circle is extreme , including @xmath17 , and each ( except for @xmath17 ) lies on one bisecant edge . table [ t : facelattices ] summarizes the face lattices by order type . in it , when @xmath280 , @xmath215 is the point were @xmath97 is tangent to the plane of the other circle . p.15 - 2 - 1.3333 p.28 - 2 - 1.3333 p.40 - 2 - 1.3333 p.16 - 2 - 1.3333 * order type * & * 0-faces * & * 1-faces * & * 2-faces * + @xmath207 & points on @xmath49 & one family parameterized by @xmath46 & @xmath230 , @xmath270 + & points on @xmath46 and points on an arc of @xmath47 & & + & points on @xmath46 and points on two arcs of @xmath47 & & + & points on an arc of @xmath46 + and an arc of @xmath47 & one family parameterized by a 2-fold branched cover of an arc & + @xmath201 & + @xmath226 & + & points on @xmath46 and + an arc of @xmath47 & two families parameterized + by @xmath281 & + & & four families with two parameterized by each arc @xmath281 & + @xmath208 & + & points on @xmath46 and + an arc of @xmath47 & one family parameterized + by @xmath281 & + & points on an arc of @xmath46 + and an arc of @xmath47 & one family parameterized by + the arc of @xmath46 & + @xmath222 & + & points on @xmath46 and + an arc of @xmath47 & one family parameterized by + the arc on @xmath47 & @xmath230 , @xmath282 + & points on an arc of @xmath46 + and an arc of @xmath47 & one family parameterized by either arc , and an isolated bisecant @xmath283 & @xmath284 , @xmath282 + & & one family parameterized by either circle except the common point & + by theorem [ th : one ] , the convex hull @xmath9 is a spectrahedral shadow . we use our classification to describe when @xmath9 is a spectrahedron . [ quadriclemma ] let @xmath285 be conics in distinct planes @xmath52 and @xmath53 . if @xmath286 , then @xmath46 and @xmath47 lie on a pencil of quadrics . since @xmath163 lie on the singular quadric @xmath287 , we need only find a second quadric containing them . choose coordinates @xmath288 $ ] for @xmath51 so that @xmath52 is defined by @xmath289 and @xmath53 by @xmath179 . then @xmath46 and @xmath47 are given by homogeneous quadratic polynomials @xmath290 and @xmath291 . since @xmath292 , the forms @xmath293 and @xmath294 define the same scheme , so they are proportional . scaling @xmath37 if necessary , we may assume that @xmath295 . . then @xmath297 and @xmath298 , which implies that the quadric defined by @xmath299 contains both @xmath46 and @xmath47 . [ th : issectrhedron ] let @xmath46 and @xmath47 be circles in @xmath1 lying in distinct planes . then their convex hull is a spectrahedron only if they have order type @xmath300 or @xmath208 or @xmath219 . we have that @xmath286 in @xmath51 if and only if the circles have order type @xmath300 or @xmath208 or @xmath219 . by lemma [ quadriclemma ] , @xmath46 and @xmath47 lie on a pencil @xmath301 of quadrics . following example 2.3 in @xcite , this pencil of quadrics contains singular quadrics given by the real roots of @xmath302 . such a singular quadric is given by the determinant of a @xmath303 matrix polynomial @xmath304 , and the block diagonal matrix with blocks @xmath305 , @xmath306 , @xmath29 , and @xmath307 represents conv@xmath308 as a spectrahedron . by corollary [ c:2face ] , @xmath9 has a nonexposed face when a tangent line to one circle meets the other circle in a different point . this occurs for all the remaining order types of the circles @xmath46 and @xmath47 , except type @xmath207 where @xmath6 in @xmath51 . in this case , the edge curve is irreducible with two connected real components , and the edge surface meets the interior of conv@xmath308 ( as there are internal stationary bisecants ) . thus conv@xmath308 is not a basic semialgebraic set and thus not a spectrahedron . we sketch an alternative approach to studying the convex hull @xmath9 of two circles that uses projective duality . this is inspired by the paper @xcite and accompanying video @xcite that explains a solution to the problem of determining the convex hull of three ellipsoids in @xmath1 . points @xmath309 of the dual projective space @xmath310 correspond to planes @xmath13 of the primal space @xmath51 . a line @xmath311 represents the pencil of planes containing a fixed line @xmath312 , and a plane @xmath313 represents the net of planes incident on a point @xmath314 . the dual @xmath315 of a conic @xmath316 is the set of planes that contain a line tangent to @xmath29 . the pencil of planes containing the tangent line @xmath318 to @xmath29 is a line lying on @xmath317 that meets @xmath309 as @xmath319 . thus @xmath317 is a cone in @xmath310 with vertex @xmath309 . let @xmath314 be any point that is not on @xmath13 . then the curve @xmath320 is the set of planes through @xmath321 that contain a tangent line @xmath318 to @xmath29 . as there are two such planes that contain a general line @xmath63 through @xmath321@xmath63 meets two tangents to @xmath29the curve @xmath320 is a conic in @xmath313 and @xmath317 is the cone over that conic with vertex @xmath309 . let @xmath163 be circles in @xmath322 lying in distinct planes @xmath323 and let @xmath324 be the convex hull of @xmath49 . let @xmath321 be any point in the interior of @xmath9 . we will consider the hyperplane @xmath325 to be the hyperplane at infinity and set @xmath326 . this is an affine space that contains every hyperplane supporting @xmath9 as well as all those disjoint from @xmath9 , as every hyperplane incident on @xmath321 meets the interior of @xmath9 . it also contains the point @xmath327 corresponding to the hyperplane at infinity in @xmath328 . for @xmath94 , let @xmath329 be the cone in @xmath330 dual to the conic @xmath97 . if @xmath331 , then the vertex @xmath332 of @xmath329 lies at infinity ( @xmath333 ) and @xmath329 is a cylinder . neither dual cone contains the point @xmath327 . let @xmath334 be the closure of the component of @xmath335 containing @xmath327 . [ p : convex - dual ] points @xmath309 in the interior of @xmath334 are exactly those whose corresponding hyperplane @xmath13 is disjoint from @xmath9 . points of the boundary of @xmath334 correspond to supporting hyperplanes of @xmath9 , and @xmath334 is convex and bounded . choose coordinates @xmath336 for @xmath1 so that @xmath337 is the origin . an affine hyperplane is defined by the vanishing of an affine form @xmath338 , whose coefficients @xmath339 $ ] give homogeneous coordinates for @xmath340 . in these coordinates , @xmath327 is the point @xmath341 $ ] , @xmath313 has equation @xmath342 , and the points of the affine @xmath330 have coordinates @xmath343 $ ] , so that @xmath327 is the origin in @xmath330 . let @xmath344 and consider the linear map @xmath345 , @xmath346 since @xmath347 is a plane containing the origin @xmath321 , @xmath348 is a closed interval @xmath349 $ ] with @xmath350 in its interior , so that @xmath351 . thus the points @xmath352\ \ \mbox{for}\ \ -\tfrac{1}{\delta}<t<-\tfrac{1}{\epsilon}\ ] ] of @xmath340 are exactly the hyperplanes in @xmath328 parallel to @xmath347 that are disjoint from @xmath9 as @xmath353 for @xmath354 . all other planes parallel to @xmath355 meet @xmath9 , with @xmath356 and @xmath357 the hyperplanes in this family that support @xmath9 . these supporting hyperplanes necessarily lie on @xmath358 . thus the interior of @xmath334 is exactly the set of all hyperplanes disjoint from @xmath9 and its boundary is exactly the set of hyperplanes supporting @xmath9 . as @xmath321 lies in the interior of @xmath9 , there is a closed ball centered at @xmath321 of radius @xmath359 contained in the interior of @xmath9 . then for any unit vector @xmath360 , the numbers @xmath361 defined by @xmath362 $ ] satisfy @xmath363 . thus the coordinates of points @xmath364 $ ] in @xmath334 satisfy @xmath365 , proving that @xmath334 is bounded . let @xmath366 $ ] and @xmath367 $ ] be points of @xmath334 . then @xmath368 . since @xmath369 is convex , for every @xmath370 $ ] , if @xmath371 , then @xmath372 and so @xmath373 . this proves that @xmath334 is convex . points in the boundary ( @xmath374 ) of @xmath334 are hyperplanes supporting @xmath9 , and faces of @xmath334 correspond to exposed faces of @xmath9 . for example , @xmath375 if and only if the plane @xmath98 of @xmath97 supports a two - dimensional face of @xmath9 . points of the curve in @xmath376 where the cones @xmath377 and @xmath378 meet correspond to stationary bisecants , and line segments in the ruling of @xmath329 lying in @xmath379 correspond to the exposed points of @xmath97 in @xmath9 . this may be seen in fig . [ fig : duals ] , the intersection of two cones on the left has cone points corresponding to the planes of the discs in the boundary of the convex set on the left in fig . [ fig : firsttaste ] . in the center is the dual of the oloid . the origin @xmath321 is in the interior of the discs of the circles , so both cones @xmath329 are elliptical cylinders . on the right is the intersection of a cone with a horizontal cylinder meeting its vertex . the cylinder is dual to the vertical circle in the rightmost convex set in fig . [ fig : firsttaste ] . the vertex is the 2-dimensional face , and the two branches of the intersection curve at the vertex of the cone have limit the two non - exposed stationary bisecants . in @xcite , the authors sketch an exact algorithm ( beautifully explained in @xcite ) to compute the convex hull of three ellipsoids @xmath39 , @xmath40 , and @xmath380 in @xmath1 . their approach inspired the previous discussion . if the origin @xmath321 lies in the interior of an ellipsoid @xmath39 , then its dual @xmath381 is also an ellipsoid . if @xmath321 lies on @xmath39 , then its dual is a paraboloid and @xmath327 lies in the convex component of its complement . if @xmath321 is exterior to @xmath39 , then its dual is a hyperboloid of two sheets , and one of the convex components of its complement contains @xmath327 . choosing an origin @xmath321 in the interior of the convex hull @xmath9 of @xmath382 as in proposition [ p : convex - dual ] , @xmath334 is a bounded convex set that is the closure of the region in the complement of the duals containing the origin @xmath327 . the video @xcite describes the algorithm to compute @xmath9 when the origin @xmath321 lies in the interior of all three ellipsoids . in that case , the dual @xmath334 of the convex hull of the three ellipsoids is the intersection of the three dual ellipsoids @xmath383 . computing @xmath334 requires the computation of the curves where two dual ellipsoids intersect , and points where three dual ellipsoids meet , and then decomposing the dual ellipsoids along these curves into patches . 1 . points common to all three dual ellipsoids . these give tritangent planes in @xmath374 . 2 . points on curves given by the pairwise intersection of dual ellipsoids . they are bitangent planes and give bitangent edges . these form 1-dimensional families of 1-faces in @xmath374 . 3 . points on a single dual ellipsoid . these are tangent planes to an ellipsoid at a point of @xmath9 , and give a two - dimensional family of exposed points of @xmath9 coming from the corresponding ellipsoid . nicola geismann , michael hemmer , and elmar schmer , _ the convex hull of ellipsoids _ , scg 01 proceedings of the seventeenth annual symposium on computational geometry , acm , new york , 2001 , pp . 321322 . branko grnbaum , _ convex polytopes _ , second ed . , graduate texts in mathematics , vol . 221 , springer - verlag , new york , 2003 , prepared and with a preface by volker kaibel , victor klee and gnter m. ziegler . kristian ranestad and bernd sturmfels , _ the convex hull of a variety _ , notions of positivity and the geometry of polynomials ( petter brndn , mikael passare , and mihai putinar , eds . ) , springer , 2011 , pp .

the reflectivity properties of graphene with nonzero mass - gap parameter are investigated in the framework of dirac model using the polarization tensor in ( 2 + 1)-dimensional space - time . for this purpose , a more simple explicit representation for the polarization tensor along the real frequency axis is found . the approximate analytic expressions for the polarization tensor and for the reflectivities of graphene are obtained in different frequency regions at any temperature . we show that the nonzero mass - gap parameter has a profound effect on the reflectivity of graphene . specifically , at zero temperature the reflectivity of gapped graphene goes to zero with vanishing frequency . at nonzero temperature the same reflectivities are equal to unity at zero frequency . we also find the resonance behavior of the reflectivities of gapped graphene at both zero and nonzero temperature at the border frequency determined by the width of the gap . at nonzero temperature the reflectifities of graphene drop to zero in the vicinity of some frequency smaller than the border frequency . our analytic results are accompanied with numerical computations performed over a wide frequency region . the developed formalism can be used in devising nanoscale optical detectors , optoelectronic switches and in other optical applications of graphene .

1604.06873

graphene is a two - dimensional sheet of carbon atoms arranged in a hexagonal lattice , which possesses unusual mechanical , electrical and optical properties . at energies below a few ev the electronic excitations ( quasiparticles ) in graphene are well described by the dirac model , i.e. , display the linear dispersion relation , where the speed of light @xmath0 is replaced with the fermi velocity @xmath1 @xcite . this makes graphene a unique laboratory for testing some effects of quantum electrodynamics and quantum field theory , such as the thermal casimir force @xcite , the klein paradox @xcite , and the creation of particles from vacuum in strong external fields @xcite . it is important also that technological applications of graphene are numerous and diverse . one could mention graphene coatings used in the optical detectors @xcite , solar cells @xcite , transparent electrodes @xcite , corrosium protection @xcite , optical biosensors @xcite , optoelectronic switches @xcite , etc . in order to make the best use of graphene in both fundamental physics and its applications , it is necessary to investigate the reflectivity properties of this novel material over the wide frequency region . as was shown experimentally , in the region of visible light , where the photon energy is much larger than both thermal and fermi energies , the transparency and conductivity of graphene are defined by the fine structure constant @xmath2 @xcite and by the quantity @xmath3 @xcite , respectively . these facts were understood in theoretical studies of the conductivity of graphene using the current - current correlation function and the kubo formula ( see ref . @xcite and review in ref . @xcite ) . the obtained results for the in - plane ( longitudinal ) conductivity have been applied to investigate the reflectivity properties of graphene . this was done in the framework of spatially local approximation , i.e. , with no regard for dependence on the wave vector . the reflection coefficients and reflectivities of graphene with zero mass - gap parameter ( quasiparticle mass ) for the transverse magnetic ( tm ) , i.e. , @xmath4 polarized , electromagnetic waves have been found at both low @xcite and high @xcite frequencies . the case of gapless graphene deposited on a substrate was considered in ref . @xcite . the most fundamental quantity allowing full investigation of the reflectivity properties of graphene over the entire frequency range is the polarization tensor in ( 2 + 1)-dimensional space - time . it was found in refs . @xcite at zero and nonzero temperature , respectively , and used to investigate the casimir effect in graphene systems in refs . @xcite ( several calculations of the casimir force were performed also using the density - density correlation functions of graphene and other methods @xcite ) . we stress , however , that for calculation of the casimir force using the lifshitz theory one should use the reflection coefficients defined at the discrete imaginary matsubara frequencies @xcite . it was shown @xcite that the polarization tensor of ref . @xcite is valid strictly at the matsubara frequencies and can not be immediately continued either to the real or to the entire imaginary frequency axis . the polarization tensor of graphene valid at all complex frequencies , including the real frequency axis , was derived in ref . @xcite . at the imaginary matsubara frequencies it coincides with that of ref . @xcite , but , in contrast to ref . @xcite , allows immediate analytic continuation to the whole complex plane . using this representation for the polarization tensor , the reflectivity properties of gapless graphene were investigated at low , high and intermediate frequencies for both polarizations of the electromagnetic field tm and te ( i.e. , transverse electric or @xmath5 polarization ) . in so doing , some new properties were found which escaped notice in previous literature . the polarization tensor of ref . @xcite , has been also used to find the reflectivities of material plates coated with gapless graphene @xcite . in this paper , the polarization tensor of ref . @xcite is used to investigate the reflectivity properties of graphene with nonzero mass - gap parameter @xmath6 . it is common knowledge that the quasiparticles of pristine ( ideal ) graphene are massless . however , they acquire some small mass @xmath6 under the influence of defects of the structure , electron - electron interaction , and in the presence of substrates @xcite . the exact value of @xmath6 for some specific graphene sample usually remains unknown , but , according to some estimations @xcite , may achieve 0.1ev ( we measure masses and frequencies in the units of energy ) . here , we derive convenient explicit expressions for the polarization tensor of graphene with nonzero mass - gap parameter at any temperature . we demonstrate that under some conditions the nonzero mass - gap parameter may have a dramatic effect on the reflectivity properties of graphene . specifically , we find the resonance behavior of graphene reflectivities at the border frequency of incident electromagnetic waves @xmath7 . this effect occurs at both zero and nonzero temperature . at nonzero temperature the reflectivities of gapped graphene drop to zero at some frequency smaller than the border one . we find the asymptotic expressions for the reflectivities of gapped graphene at both low and high frequencies . at zero temperature the tm and te reflectivities of graphene with @xmath8 take the zero value at zero frequency . this is different from the case of gapless graphene , where both reflectivities at zero temperature are equal to nonzero constants depending on @xmath9 and on the angle of incidence @xcite . at nonzero temperature the reflectivities of gapped graphene go to unity with vanishing frequency . we also perform numerical computations of the reflectivities of gapped graphene over the wide range of frequencies at both zero and nonzero temperature and find the regions , where nonzero value of @xmath6 makes a profound effect on the obtained results . the paper is organized as follows . in sec . ii we derive convenient expressions for the polarization tensor of graphene with nonzero mass - gap parameter and present the general formalism . section iii contains calculation of the reflectivities of gapped graphene at zero temperature . an investigation of the impact of temperature on the reflectivity of gapped graphene is presented in sec . iv . in sec . v the reader will find our conclusions and discussion . the description of graphene by means of the polarization tensor in ( 2 + 1)-dimensional space - time is in fact equivalent @xcite to the method using the density - density correlation functions , which is more often used in atomic and condensed matter physics . the formalism of the polarization tensor , however , offers some advantages over the more conventional methods , especially at nonzero temperature , because it has been much studied and elaborated in the framework of thermal quantum field theory . it is conventional @xcite to notate the polarization tensor as @xmath10 , where in ( 2 + 1)-dimensional case @xmath11 , and @xmath12 . in our configuration the polarization tensor @xmath13 , where @xmath14 is the frequency of electromagnetic wave incident on graphene and @xmath15 is the angle of incidence . we assume that the graphene sheet is situated in the plane @xmath16 and the @xmath17 azis is perpendicular to it . then for real photons on a mass - shell the magnitude of the wave vector component parallel to graphene ( i.e. , perpendicular to the @xmath17 axis ) is defined as @xmath18 the amplitude reflection coefficients for tm and te polarizations of the electromagnetic field incident on a graphene sheet are given by @xcite @xmath19 & & r_{\rm te}(\omega,\theta_i)= \frac{\pi(\omega,\theta_i)}{2i\hbar\frac{\omega^3}{c^3}\sin^2\theta_i\ , \cos\theta_i -\pi(\omega,\theta_i ) } , \label{eq2}\end{aligned}\ ] ] where the following notation is introduced : @xmath20\ , . \label{eq3}\ ] ] note that so important quantity as the conductivity of graphene can be immediately expressed via the polarization tensor . thus , the spatially nonlocal ( i.e. , depending on the wave vector ) longitudinal and transverse conductivities of graphene are given by @xcite @xmath21 it is convenient to represent the quantities @xmath22 and @xmath23 in the form @xmath24 where @xmath25 and @xmath26 are defined at zero temperature and @xmath27 , @xmath28 are the thermal corrections which go to zero in the limiting case of vanishing temperature . the explicit expressions for the quantities @xmath25 and @xmath26 at the imaginary frequencies in the case of arbitrary mass - gap parameter @xmath6 have been found in ref . @xcite . in our case of real frequencies they can be presented in the form @xmath29 where @xmath30^{1/2}$ ] , @xmath31 and the function @xmath32 is given by @xcite @xmath33 { \rm arctanh}\frac{\hbar\omega\eta}{2mc^2 } , & \hbar\omega<\frac{2mc^2}{\eta } , \\[1 mm ] 4mc-\frac{2\hbar\omega\eta}{c}\left[1 + \frac{4m^2c^4}{\hbar^2\omega^2\eta^2}\right ] \left({\rm arctanh}\frac{2mc^2}{\hbar\omega\eta } + i\frac{\pi}{2}\right ) , & \hbar\omega\geq\frac{2mc^2}{\eta}. \end{array } \right . \label{eq7}\ ] ] note that for a gapless graphene ( @xmath34 ) only the second line in eq . ( [ eq7 ] ) is applicable at all @xmath14 . in this case the function @xmath32 is pure imaginary and has the negative imaginary part @xmath35 substituting eqs . ( [ eq6 ] ) and ( [ eq8 ] ) in eq . ( [ eq4 ] ) with account of eq . ( [ eq1 ] ) , one finds the conductivities of graphene at zero temperature @xmath36 which are in accordance with the previous knowledge ( see , for instance , refs . @xcite ) . now we consider thermal corrections to the 00 component of the polarization tensor and to its trace entering the definition ( [ eq3 ] ) of the quantity @xmath23 . according to eqs . ( 33)(36 ) of ref . @xcite , along the real frequency axis these thermal corrections can be written in the form @xmath37 , \label{eq12}\ ] ] where @xmath38 . here , the integration variable @xmath39 has the dimension 1/cm , @xmath40^{1/2}$ ] , @xmath41 is the boltzmann constant , @xmath42 is the temperature , and the following notations are introduced : @xmath43 . \nonumber\end{aligned}\ ] ] the expression for @xmath44 depends on the fulfilment of the following conditions : @xmath45^{1/2 } , & -i \left[-{q_{\lambda}}^{\!\!\!2}+\left(2\tilde{v}_f \frac{\omega}{c}q\sin\theta_i\right)^2\right]^{1/2 } , & \end{array}\right . \label{eq14}\ ] ] where @xmath46 in the case of @xmath8 under consideration here it is more convenient to represent the temperature correction ( [ eq12 ] ) as an integral with respect to the new variable @xmath47 : @xmath48 , \label{eq16}\ ] ] where @xmath49 the quantity @xmath50 defined in eq . ( [ eq15 ] ) takes the form @xmath51 it is easily seen that for @xmath52 the quantity @xmath50 is always negative and satisfies the condition @xmath53^{1/2}\sin\theta_i\ , . \label{eq19}\ ] ] thus , the first line in eq . ( [ eq14 ] ) is applicable leading to @xmath54 } \eta^2(\theta_i)\left ( \frac{\omega}{c}+2u\right)^2 + \tilde{v}_f^2\sin^2\theta_i\left[4\left(\frac{mc}{\hbar}\right)^2- \frac{\omega^2}{c^2}\eta^2(\theta_i)\right]\right\}^{1/2}. \label{eq20}\ ] ] the case @xmath55 is more complicated and deserves separate consideration . according to eq . ( [ eq18 ] ) in this case @xmath56 can be both positive and negative . in order to explicitly rewrite eq . ( [ eq14 ] ) in terms of the variable @xmath57 , we introduce the quantity @xmath58\sin^2\theta_i \label{eq21 } \\ & & ~~=\frac{\omega^2}{c^2}\left\ { \vphantom{\left[4\left(\frac{mc}{\hbar}\right)^2- \frac{\omega^2}{c^2}\eta^2(\theta_i)\right ] } \eta^2(\theta_i)\left ( \frac{\omega}{c}-2u\right)^2 + \tilde{v}_f^2\sin^2\theta_i\left[4\left(\frac{mc}{\hbar}\right)^2- \frac{\omega^2}{c^2}\eta^2(\theta_i)\right]\right\}. \nonumber\end{aligned}\ ] ] the sign of the quantity @xmath59 under different values of parameters is found by solving the equation @xmath60 with respect to the variable @xmath57 : @xmath61 . \label{eq23}\ ] ] if the condition @xmath62 is satisfied , eq . ( [ eq22 ] ) does not possess real roots and the function @xmath59 is always positive . in this case the first line in eq . ( [ eq14 ] ) is applicable and @xmath63 where @xmath59 is defind in eq . ( [ eq21 ] ) . now we consider the opposite condition @xmath64 when eq . ( [ eq22 ] ) has two different real roots @xmath65 defined in eq . ( [ eq23 ] ) [ they coincide in the case of equality in eq . ( [ eq26 ] ) ] . it is easily seen that under the condition ( [ eq26 ] ) @xmath66 . thus , in the regions @xmath67 and @xmath68 the function @xmath59 is nonnegative and the quantity @xmath69 is given by the first line of eq . ( [ eq14 ] ) . in so doing @xmath70 in the latter interval whereas it can be both negative and positive in the former . as to the region @xmath71 , here the function @xmath59 is negative and , as a consequence , @xmath69 is given by the second line of eq . ( [ eq14 ] ) . combining all these facts together , for the quantity @xmath69 one obtains @xmath72 simple analysis using eqs . ( [ eq15 ] ) and ( [ eq23 ] ) shows that for @xmath73 within the interval @xmath74 we have @xmath75 , whereas within the interval @xmath76 it holds @xmath77 . equations ( [ eq16])([eq18 ] ) , ( [ eq20 ] ) , ( [ eq21 ] ) , ( [ eq25 ] ) and ( [ eq27 ] ) are convenient for both analytic and numerical calculations using the polarization tensor of gapped graphene . to apply them for calculations of the reflection coefficients and reflectivities , it is convenient also to present an explicit expression for the quantity @xmath78 . it is obtained from eqs . ( [ eq3 ] ) , ( [ eq16 ] ) and ( [ eq17 ] ) and has the form @xmath79 , \label{eq30}\ ] ] where the quantity @xmath80 is given by @xmath81 . \label{eq31}\ ] ] the denominator @xmath44 in eq . ( [ eq30 ] ) is presented in eqs . ( [ eq20 ] ) and ( [ eq27 ] ) . in the limiting case @xmath82 the above expressions take a more simple form used in ref . @xcite to investigate the reflectivity properties of gapless graphene . at @xmath83 the polarization tensor of graphene is given by eqs . ( [ eq6 ] ) and ( [ eq7 ] ) . we consider first the region of frequencies ( [ eq24 ] ) . here , the first line of eq . ( [ eq7 ] ) is applicable which can be rewritten in the form @xmath84 . \label{eq32}\ ] ] this is real function in contrast with the case of gapless graphene [ in the latter case the function @xmath32 is pure imaginary , see eq . ( [ eq8 ] ) ] . substituting eq . ( [ eq6 ] ) with the real function @xmath32 in eq . ( [ eq2 ] ) , one obtains the reflection coefficients and reflectivities of gapped graphene at sufficiently low frequencies @xmath85 where @xmath32 is defined by eq . ( [ eq32 ] ) . it is seen that at the normal incidence ( @xmath86 ) @xmath87 as it should be . now we consider the asymptotic regime of very small frequencies satisfying the condition @xmath88 expanding eq . ( [ eq32 ] ) in powers of small parameter ( [ eq35 ] ) , one obtains @xmath89 substituting this in eq . ( [ eq33 ] ) with account of eq . ( [ eq35 ] ) , we arrive at @xmath90 note that in eq . ( [ eq37 ] ) and below we have also used @xmath91 . as can be seen in eq . ( [ eq37 ] ) , @xmath92 this is different from the case of gapless graphene at zero temperature . in the latter case , substituting eq . ( [ eq8 ] ) in eq . ( [ eq6 ] ) , one finds @xmath93 substituting these equations in eq . ( [ eq2 ] ) , after simple calculation , we obtain @xmath94 note that both reflectivities in eq . ( [ eq40 ] ) do not depend on the frequency and , thus , take nonzero values at @xmath95 , as opposed to eq . ( [ eq38 ] ) for a gapped graphene . we are coming now to the asymptotic region @xmath96 where the quantity on the left - hand side of eq . ( [ eq35 ] ) approaches unity from the left . it is convenient to present the frequency in the form @xmath97 where the limiting case ( [ eq41 ] ) corresponds to @xmath98 . substituting eq . ( [ eq42 ] ) in eq . ( [ eq32 ] ) , one finds that at @xmath98 @xmath99 then from eq . ( [ eq33 ] ) we conclude that in the limiting case ( [ eq41 ] ) @xmath100 the next region to consider is given by eq . ( [ eq26 ] ) . here , the second line of eq . ( [ eq7 ] ) is applicable , where the function @xmath32 has both real and imaginary parts @xmath101 , \nonumber \\[1 mm ] & & { \rm im}\phi(\omega,\theta_i)=-\frac{\pi\hbar\omega\eta}{c}\left(1 + \frac{4m^2c^4}{\hbar^2\omega^2\eta^2}\right ) . \label{eq45}\end{aligned}\ ] ] using eqs . ( [ eq2 ] ) and ( [ eq6 ] ) , we find the reflectivities in the form @xmath102}{\alpha^2\cos^2(\theta_i){\rm re}^2\phi(\omega,\theta_i)+ \left[2\hbar\frac{\omega}{c}\eta^2-\alpha\cos\theta_i { \rm im}\phi(\omega,\theta_i)\right]^2 } , \nonumber \\[1 mm ] & & { \cal r}_{\rm te}(\omega,\theta_i)= \frac{\alpha^2[{\rm re}^2\phi(\omega,\theta_i)+ { \rm im}^2\phi(\omega,\theta_i)]}{\alpha^2{\rm re}^2\phi(\omega,\theta_i)+ \left[2\hbar\frac{\omega}{c}\cos\theta_i-\alpha { \rm im}\phi(\omega,\theta_i)\right]^2}. \label{eq46}\end{aligned}\ ] ] now we consider the asymptotic behavior of the reflectivities at high frequencies @xmath103 in this case eq . ( [ eq45 ] ) leads to @xmath104 & & { \rm im}\phi(\omega,\theta_i)\approx -\pi\hbar\frac{\omega}{c}\eta(\theta_i ) , \label{eq48}\end{aligned}\ ] ] i.e. , to the same result as for a gapless graphene [ see eq . ( [ eq8 ] ) ] . then , the reflectrivities are again given by eq . ( [ eq40 ] ) or , neglecting by the small terms , as compared to unity , by @xmath105 note that these reflectivities do not depend on the frequency and on the mass - gap parameter . the remaining asymptotic region is given by eq . ( [ eq41 ] ) when the quantity on the left - hand side of eq . ( [ eq47 ] ) approaches unity from the right . in this case from the second line of eq . ( [ eq45 ] ) one obtains @xmath106 the asymptotic behavior of the real part of @xmath32 in the first line of eq . ( [ eq45 ] ) can be found similar to eqs . ( [ eq42 ] ) and ( [ eq43 ] ) with the same results for the reflectivity properties as in eq . ( [ eq44 ] ) . we are coming to numerical computations of the reflectivity properties of gapped graphene at zero temperature over wide frequency range of incident electromagnetic waves . computations are performed by using eqs . ( [ eq2 ] ) , ( [ eq6 ] ) and ( [ eq7 ] ) . taking into account that due to the conditions @xmath107 , @xmath108 the angular dependences of reflectivities are similar to those for gapless graphene @xcite , computations are performed at the normal incidence . in fig . 1 the three lines from right to left show the computational results for the tm and te reflectivities ( [ eq34 ] ) at the normal incidence calculated as functions of frequency for the graphene sheets with mass - gap parameter @xmath6 equal to 0.1 , 0.05 , and 0.01ev , respectively . the double logarithmic scale is used allowing to cover the frequency range from 1@xmath109ev to 20ev . the linear dependence of the logarithm of reflectivities on the logarithm of frequency at low frequencies satisfying the condition ( [ eq35 ] ) corresponds to the asymptotic regime where eq . ( [ eq37 ] ) is applicable . at the frequency value @xmath110 which is different for different @xmath6 [ recall that at the normal incidence @xmath111 , the narrow resonances are observed in fig . 1 . under the condition ( [ eq41 ] ) these resonances satisfy eq . ( [ eq44 ] ) . at high frequencies satisfying the condition ( [ eq47 ] ) the asymptotic equations ( [ eq49 ] ) are applicable . note that at @xmath86 one can neglect by @xmath112 , as compared with 2 in the denominators . it would be interesting to investigate in more detail an immediate vicinity of the resonance frequency ( [ eq51 ] ) , where the above asymptotic expressions do not apply . as an example , in fig . 2 we again plot the reflectivity at the normal incidence ( [ eq34 ] ) as a function of frequency for the graphene sheet with @xmath113 . now the computational results are plotted over a narrow frequency region from 0.14 to 0.30ev using the natural scale in the frequency axis . as is seen in fig . 2 , even over a so narrow range , the reflectivity of graphene varies by four orders of magnitude , whereas the width of the resonance peak remains unresolved . in fact the half - width of this resonance can be found analytically from the asymptotic representation ( [ eq43 ] ) . substituting eq . ( [ eq43 ] ) in eq . ( [ eq33 ] ) taken at @xmath86 , we find @xmath114 taking into account eq . ( [ eq51 ] ) , one obtains that @xmath115 when @xmath116 , i.e. , for @xmath117 . then , using eq . ( [ eq42 ] ) , we conclude that the half - width of the resonance is equal to @xmath118 this half - width is too small and makes the resonance under consideration unobservable . however , at zero temperature it might be possible to observe the change in the reflectivity of graphene by a factor of ten with increasing frequency of the incident light . thus , the reflectivity increases from 0.00013 to 0.00134 when @xmath14 increases from 0.16 to 0.199ev . in a similar way , with further increase of frequency from 0.201 to 0.3ev the reflectivity of graphene changes from 0.00141 to 0.000141 . we consider now more detailly the region of frequencies from 0 to 0.4ev , where the reflectivity varies by more than an order of magnitude for all values of @xmath6 under discussion . note that at room temperature ( @xmath119k ) it holds @xmath120ev , i.e. , the thermal energy is of the same order as @xmath121 for the smallest value of @xmath6 . in fig . 3 we plot the reflectivity of gapped graphene at the normal incidence as a function of frequency by the three lines from right to left for the mass - gap parameter @xmath122 , 0.05 , and 0.01ev , respectively . in this figure , both the reflectivity and frequency are plotted in their natural scales . because of this , only the lower parts of the resonance peaks are shown . as is seen in fig . 3 , an impact of the mass - gap parameter of graphene on the reflectivity properties decreases with decreasing @xmath6 and increasing frequency . for small mass - gap parameters @xmath123ev in the region @xmath124 the reflectivity is approximately equal to that obtained for @xmath125 at any temperature @xmath126k @xcite . the role of nonzero temperature in the case of gapped graphene is considered below . in this section , we consider the reflection coefficients ( [ eq2 ] ) , where the quantities @xmath22 and @xmath23 are given by eq . ( [ eq5 ] ) , i.e. , are defined at any nonzero temperature . in so doing , the zero - temperature contributions are presented in eqs . ( [ eq6 ] ) and ( [ eq7 ] ) and the thermal corrections to them are given in eqs . ( [ eq16 ] ) and ( [ eq30 ] ) with all necessary notations in eqs . ( [ eq17])([eq21 ] ) , ( [ eq25 ] ) , ( [ eq27 ] ) and ( [ eq31 ] ) . we begin with thermal correction to the polarization tensor of graphene calculated at frequencies satisfying eq . ( [ eq24 ] ) . from eqs . ( [ eq16 ] ) , ( [ eq17 ] ) , ( [ eq20 ] ) , ( [ eq21 ] ) and ( [ eq25 ] ) one obtains @xmath127 } 1+\frac{1}{2\omega\eta } \right . \nonumber \\ & & \times\left[- \frac{(2cu+\omega)^2-\tilde{v}_f^2\omega^2 \sin^2\theta_i}{\sqrt{(2cu+\omega)^2+\tilde{v}_f^2\omega^2 \left(\frac{4m^2c^4}{\hbar^2\omega^2\eta^2}-1\right ) \sin^2\theta_i}}\right . \nonumber \\[-2 mm ] & & \label{eq54 } \\[-0.3 mm ] & & + \left.\left . \frac{(2cu-\omega)^2-\tilde{v}_f^2\omega^2 \sin^2\theta_i}{\sqrt{(2cu-\omega)^2+\tilde{v}_f^2\omega^2 \left(\frac{4m^2c^4}{\hbar^2\omega^2\eta^2}-1\right ) \sin^2\theta_i}}\right]\right\}. \nonumber\end{aligned}\ ] ] now we impose a more stringent than eq . ( [ eq24 ] ) requirement , i.e. , @xmath128.under this requirement the quantity @xmath129 can not become too small and one can expand both square roots in eq . ( [ eq54 ] ) up to the first order of respective small parameters with the result @xmath130\right\}. \nonumber\end{aligned}\ ] ] then , using the smallness of @xmath131 , we arrive at @xmath132 note that eq . ( [ eq56 ] ) is also valid under a less stringent condition ( [ eq24 ] ) if one assumes that @xmath133 . this is because the main contribution to the integral in eq . ( [ eq54 ] ) is given by @xmath134 and hence the quantity @xmath129 can not become too small . both application regions of eq . ( [ eq56 ] ) are used below to obtain approximate expressions for the polarization tensor and reflectivity of gapped graphene at nonzero temperature . it is convenient to introduce two dimensionless parameters and new integration variable according to @xmath135 then eq . ( [ eq56 ] ) takes the form @xmath136 we consider , first , eq . ( [ eq58 ] ) under its validity condition @xmath137 and assume that @xmath138 as well . taking into account that the main contribution to the integral in eq . ( [ eq58 ] ) is given by @xmath139 , one can neglect by the small second term in the round brackets . then , after the integration , we arrive at @xmath140 the latter thermal correction does not depend on @xmath6 and coincides with that obtained in eq . ( 89 ) of ref . @xcite for a gapless graphene . note , however , that for a gapped graphene at @xmath141k from eqs . ( [ eq6 ] ) and ( [ eq36 ] ) , i.e. , in the case of small frequencies , one obtains @xmath142 this is real quantity , as opposed to the case of @xmath125 , where @xmath143 at small frequencies is pure imaginary . as is seen using eq . ( [ eq24 ] ) and the conditions @xmath144 , @xmath145 . as a result , for the tm reflectivity of graphene from eqs . ( [ eq2 ] ) and ( [ eq59 ] ) we find @xmath146 similar approximate calculation with account of eqs . ( [ eq6 ] ) and ( [ eq36 ] ) results in @xmath147 then , substituting this equation in eq . ( [ eq2 ] ) , one obtains @xmath148 unlike the case of gapped graphene at zero temperature , eqs . ( [ eq61 ] ) and ( [ eq61b ] ) result in @xmath149 similar to the case of gapless graphene at nonzero @xmath42 @xcite . below we demonstrate , both analytically and numerically , that this is general property of the reflectivity of gapped graphene at @xmath150 . now we return to eq . ( [ eq58 ] ) derived under the condition @xmath128 and consider it under additional assumptions @xmath151 and @xmath152 . under these assumptions one can neglect by unity , as compared with @xmath153 , and by @xmath154 , as compared with @xmath155 and @xmath156 . as a result , eq . ( [ eq58 ] ) takes the form @xmath157 & & \label{eq63}\\[-0.5 mm ] & & ~ = -8\alpha\sin^2\theta_i\frac{k_bt}{c}\left\{e^{-\mu}+\mu^2\left [ \frac{e^{-\mu}}{\mu}+{\rm ei}(-e^{-\mu})\right]\right\ } , \nonumber\end{aligned}\ ] ] where @xmath158 is the exponential integral . taking into account the asymptotic expansion @xcite @xmath159 , \label{eq64}\ ] ] one arrives at @xmath160 combining this expression with eq . ( [ eq60 ] ) , we find @xmath161 , \label{eq66}\ ] ] i.e. , the thermal correction is exponentially small . substituting this equation [ and similar for @xmath162 in eq . ( [ eq2 ] ) , we again obtain eq . ( [ eq62 ] ) in the limiting case of vanishing frequency . as is seen from eq . ( [ eq66 ] ) , the thermal correction to the 00 component of the polarization tensor is negative , whereas the zero - temperature contribution to it is positive . taking into account that they can be of the same order of magnitude , we find the value of frequency @xmath163 where the quantity @xmath22 vanishes @xmath164 at room temperature ( @xmath165ev ) only the maximum value of the mass - gap parameter under consideration here ( @xmath166ev ) leads to @xmath167 and can be considered as close to the application region of the approximation used . for these values of parameters we find from eq . ( [ eq67 ] ) the value @xmath168ev , which is more or less in agreement with the second application condition of our approximation ( @xmath152 ) . below we compare the obtained approximate value of @xmath163 with the results of numerical computations and find rather good agreement . just now we only note that vanishing of @xmath169 leads in accordance with eq . ( [ eq2 ] ) to vanishing of the reflectivities of graphene @xmath170 at the frequency @xmath171 . this does not happen either for a gapless graphene at any temperature or for a gapped graphene at @xmath141k . in the next subsection we consider the effect of vanishing reflectivity of graphene in more details by means of numerical computations . here , we continue with the approximate analytic expressions for the polarization tensor and consider the frequency region of higher frequencies defined in eq . ( [ eq26 ] ) . we start with the imaginary part of the thermal correction ( [ eq16 ] ) which arises only for @xmath55 and is obtained from the first line of eq . ( [ eq17 ] ) and the second line of eq . ( [ eq27 ] ) : @xmath172 where @xmath173 and @xmath65 are defined in eq . ( [ eq23 ] ) . note that in the case of equality in eq . ( [ eq26 ] ) the imaginary part ( [ eq68 ] ) vanishes . it is convenient to introduce new dimensionless integration variable @xmath174 then eq . ( [ eq68 ] ) takes the form @xmath175 where @xmath176 the integral in eq . ( [ eq71 ] ) can be calculated approximately under different relationships between @xmath14 and @xmath42 . we first assume that @xmath133 [ and also use @xmath91 ] . in this case both exponents in the denominator of eq . ( [ eq71 ] ) can be replaced with unities and after the integration one obtains @xmath178 for @xmath125 this expression coincides with eq . ( 83 ) of ref . @xcite obtained for gapless graphene . now we consider the opposite case @xmath179 . in this case it is convenient to rewrite eq . ( [ eq71 ] ) in the form @xmath180 now we can neglect by the second exponent in the denominator , as compared to the first , put @xmath91 and calculate the integral with the result @xmath181 , \label{eq75}\ ] ] where @xmath182 is the modified bessel function of the first kind . if the frequency is not too large , i.e. , @xmath179 but @xmath183 , one can put @xmath184 and @xmath185 . substituting this in eq . ( [ eq75 ] ) one finds @xmath186 note that for @xmath125 this result agrees with the estimation in eq . ( 70 ) of ref . @xcite made for a gapless graphene . we now turn our attention to the real part of the polarization tensor under the condition ( [ eq26 ] ) . it is given by eq . ( [ eq16 ] ) , where all notations are contained in the first line of eq . ( [ eq17 ] ) for @xmath187 , in eq . ( [ eq20 ] ) and in the first and third lines of eq . ( [ eq27 ] ) . using the dimensionless quantities @xmath109 and @xmath188 and the integration variable @xmath189 introduced in eq . ( [ eq57 ] ) , the real part of the polarization tensor takes the form @xmath190 where the integrals @xmath191 are defined as @xmath192 & & + \left.\left . \frac{(2v-\nu)^2-\tilde{v}_f^2\nu^2\sin^2\theta_i}{\sqrt{(2v-\nu)^2- \tilde{v}_f^2\nu^2a\sin^2\theta_i}}\right]\right\ } , \nonumber \\[1.5 mm ] i_2&\equiv & \int_{v^{(-)}}^{v^{(+)}}\frac{dv}{e^v+1}\left[1-\frac{1}{2\nu\eta } \frac{(2v+\nu)^2-\tilde{v}_f^2\nu^2\sin^2\theta_i}{\sqrt{(2v+\nu)^2- \tilde{v}_f^2\nu^2a\sin^2\theta_i}}\right ] , \nonumber \\[1.5 mm ] i_3&\equiv & \int_{v^{(+)}}^{\infty}\frac{dv}{e^v+1}\left\{1-\frac{1}{2\nu\eta}\left [ \frac{(2v+\nu)^2-\tilde{v}_f^2\nu^2\sin^2\theta_i}{\sqrt{(2v+\nu)^2- \tilde{v}_f^2\nu^2a\sin^2\theta_i}}\right.\right . \nonumber \\[1 mm ] & & -\left.\left . \frac{(2v-\nu)^2-\tilde{v}_f^2\nu^2\sin^2\theta_i}{\sqrt{(2v-\nu)^2- \tilde{v}_f^2\nu^2a\sin^2\theta_i}}\right]\right\}. \label{eq78}\end{aligned}\ ] ] here , the integration limits are @xmath193 and @xmath65 are presented in eq . ( [ eq23 ] ) , @xmath194 is defined below eq . ( [ eq6 ] ) and @xmath195 is introduced in eq . ( [ eq69 ] ) . we consider first the region of relatively low frequencies @xmath133 ( @xmath137 ) . note that under the condition ( [ eq26 ] ) assumed now the inequality @xmath137 can be valid only if @xmath138 is valid as well . now we take into account that the main contributions to all integrals ( [ eq78 ] ) are given by @xmath139 and that in our case the inequalities @xmath196 lead to @xmath197 . therefore only the integral @xmath198 contains the values @xmath139 in the integration region and , thus , @xmath198 alone determines the value of @xmath199 in this case . under the integral @xmath198 one can expand in powers of a small parameter @xmath200 because for @xmath139 we have @xmath201 and the quantity @xmath202 can not vanish . as a result , one obtains @xmath203 \approx -8\alpha\ln 2\frac{k_bt}{c}\sin^2\theta_i . \nonumber\end{aligned}\ ] ] comparing this with eq . ( [ eq59 ] ) , we conclude that under the condition @xmath138 ( @xmath204 ) there is a smooth transition between the frequency regions satisfying the inequalities ( [ eq24 ] ) and ( [ eq26 ] ) . at this point we can present the complete polarization tensor of graphene under the conditions @xmath205 . in this case the zero - temperature contribution @xmath143 is given by eqs . ( [ eq6 ] ) and ( [ eq48 ] ) , whereas the imaginary and real parts of thermal correction are found in eqs . ( [ eq73 ] ) and ( [ eq79 ] ) . the results is @xmath206 it is seen that under our conditions the imaginary part of this expression is much smaller than the real one and can be neglected . taking into account the similar result for @xmath207 and using eq . ( [ eq2 ] ) , for the reflectivities of graphene we return back to eqs . ( [ eq61 ] ) and ( [ eq61b ] ) obtained above in the frequency region @xmath152 . we will return to this fact below when discussing the results of numerical computations . now we consider the opposite case @xmath179 , i.e. , @xmath208 . in the region of frequencies satisfying eq . ( [ eq26 ] ) this is possible for both @xmath209 and @xmath210 . then we have @xmath211 and , thus , the major contribution to @xmath27 is given by the integral @xmath212 , where the upper integration limit @xmath213 can be replaced with infinity . taking into account that for @xmath139 we have @xmath214 , it is easy to expand the function under the integral @xmath212 , as was done above , and arrive at @xmath215 after the integration in eq . ( [ eq80 ] ) we have @xmath216 , \nonumber\end{aligned}\ ] ] where @xmath217 is the polylogarithm function . if we additionally assume that @xmath204 , we can neglect by the first and second terms on the right - hand side of eq . ( [ eq81 ] ) , as compared to the third one . then , by putting @xmath218 , one arrive at @xmath219 taking into account that @xmath220 , where @xmath221 is the riemann zeta function , this is in agreement with eq . ( 67 ) of ref . @xcite obtained for @xmath125 . then , the complete polarization tensor and also the reflectivities of gapped graphene in this case are the same for a gapless one [ we remind that under our conditions the imaginary part of the thermal correction to the 00 component of the polarization tensor is exponentially small according to eq . ( [ eq76 ] ) , whereas the dominant contributions to the imaginary parts of @xmath143 and @xmath26 do not depend on @xmath6 ] . under another additional condition , @xmath223 , the main contribution in eq . ( [ eq81 ] ) is given by the first term on the right - hand side and the result is @xmath224 in this case both the real and imaginary parts of the thermal correction are exponentially small [ see eq . ( [ eq76 ] ) ] , and we again return to the same reflectivities , as were obtained for a gapless graphene in ref . @xcite . now we perform numerical computations of the reflectivity of graphene at the normal incidence ( @xmath86 ) defined via the reflection coefficients by the first equalities in eq . ( [ eq33 ] ) . computations are performed using eqs . ( [ eq5])([eq7 ] ) , ( [ eq16])([eq21 ] ) , ( [ eq25 ] ) and ( [ eq27 ] ) . as noted in sec . iiib , under the conditions @xmath225 , @xmath108 the angular dependences of reflectivities are similar to the case of gapless graphene . they are presented in the above analytic expressions and in figs . 3 and 5 of ref . @xcite . taking into account the possibility of comparison with the measurement data , we consider the temperature at a laboratory , @xmath119k , and cover the maximally wide range of frequencies . in fig . 4 we present the computational results for the reflectivities @xmath226 as functions of frequency by the two solid lines 1 and 2 for graphene sheets with the mass - gap parameter @xmath6 equal to 0.1 and 0.05ev , respectively . the double logarithmic scale is used . the dashed lines 1 and 2 reproduce from fig . 1 the respective results at zero temperature . as is seen in fig . 4 , the reflectivity of graphene depends heavily on the fact that temperature is not equal to zero . first of all note that the reflectivity at zero frequency is equal to unity rather than zero , as it was at zero temperature ( see figs . 1 and 3 ) . this fact was already established analytically in the asymptotic expressions ( [ eq61 ] ) , ( [ eq62 ] ) and ( [ eq66 ] ) valid at small frequencies . second , as was noted above , in the region of sufficiently small frequencies the thermal correction @xmath227 in eq . ( [ eq65 ] ) is real and negative . taking into account that at small frequencies @xmath143 is real and positive [ see eqs . ( [ eq6 ] ) and ( [ eq32 ] ) ] , it is possible that @xmath22 turns into zero at some frequency @xmath163 depending on @xmath6 and @xmath42 . as is noted above , the reflection coefficients and reflectivities also vanish at the frequency @xmath163 . from fig . 4 it is seen that @xmath228 at @xmath229 and 0.03033ev for graphene sheets with @xmath122 and @xmath230ev , respectively . an approximate analytic expression for @xmath163 is obtained in eq . ( [ eq67 ] ) . as discussed above , at room temperature it can be applied only for the largest value @xmath166ev and even in this case it is slightly outside the validity region of the approximation used . however , comparing the numerical ( 0.02159ev ) and analytic ( 0.0234ev ) results for @xmath163 , one can conclude that the approximate expression ( [ eq67 ] ) leads to only 8.4% error . in the region of higher frequencies satisfying eq . ( [ eq26 ] ) for the mass - gap parameters considered in fig . 4 the computational results are similar to those obtained at @xmath141k . here , we observe the resonance peaks at the border frequencies and the asymptotic regime at high frequencies , which does not depend on @xmath14 . in this region the dashed line 1 coincides with the solid line 1 ( @xmath166ev ) , and the dashed line 2 ( @xmath231ev ) only minor deviates from the solid line 2 . one can conclude that in the range of low frequencies satisfying the condition ( [ eq24 ] ) the reflectivity properties of gapped graphene at zero and room temperature are quite different . now we perform numerical computations in order to investigate these differences in more details . in fig . 5 , we present the reflectivity of graphene with @xmath122 and 0.05ev at @xmath119k at the normal incidence in an immediate vicinity of the frequency @xmath163 where it drops to zero ( the lines 1 and 2 , respectively ) . for this purpose the natural scale is used along the frequency axis . the dashed lines 1 and 2 show the respective results at @xmath141k . as is seen in fig . 5 , near the frequencies @xmath163 it should be possible to observe the change in the reflectivity of graphene by several orders of magnitude . from fig . 4 it can be seen that with decreasing @xmath6 the interval between the frequency @xmath163 , where the reflectivity vanishes , and the border frequency @xmath232 , where it is equal to unity , becomes more narrow . to illustrate this tendency , in fig . 6 we plot the reflectivity of graphene with @xmath233ev at @xmath119k as a function of frequency in the double logarithmic scale ( the solid line ) . from fig . 6 it is seen that zero reflectivity is achieved at the frequency @xmath163 which is only slightly less than 0.002ev . because of this , the minimum and maximum values of the reflectivity virtually coincide . for not too high frequencies satisfying the condition @xmath133 and with exception of some vicinity of the border frequency @xmath232 , the solid line in fig . 6 is in a good agreement with the asymptotic expressions ( [ eq61 ] ) and ( [ eq61b ] ) taken at @xmath86 . in the same figure the reflectivity of graphene with @xmath233ev at zero temperature is shown by the dashed line . in this case the reflectivity has only the maximum value at 0.002ev and goes to zero with vanishing frequency . note that all the above results related to the region of frequencies ( [ eq24 ] ) are valid only for graphene with nonzero mass - gap parameter . specifically , the reflectivity of gapless graphene does not vanish at any frequency @xcite . finally , we perform numerical computations of the frequency @xmath163 , where the reflectivity of graphene vanishes , as a function of the mass - gap parameter @xmath6 . the computational results in the region of @xmath6 from 0.001 to 0.1ev at @xmath119k are shown in fig . 7 by the solid line . as is seen in fig . 7 , with increasing @xmath6 , the frequency @xmath163 increases , achieves the maximum value at @xmath234ev and then gradually decreases . in the same figure , the border frequency @xmath235 is plotted by the dashed line . it is seen that the frequency interval between the dashed and solid lines becomes narrower with decreasing @xmath6 ( as already was concluded from fig . 4 basing on only two values of @xmath6 ) . the dashed line remains above the solid line at all @xmath6 and almost coincides with it at @xmath236ev . this is in accordance with fig . 6 , where for @xmath233ev the frequencies , where reflectivity of graphene takes the minimum and maximum values , can not be discerned . in the foregoing , we have investigated the reflectivity properties of gapped graphene at both zero and nonzero temperature in the framework of the dirac model . it was shown that the presence of nonzero mass - gap parameter has a profound effect on the reflectivity of graphene . to find this effect , we have employed the polarization tensor of graphene found in ref . @xcite , which was further adapted and simplified for the case of gapped graphene . specifically , it was shown that at zero temperature the reflectivities of gapped graphene go to zero when the frequency vanishes . this is not the case for a gapless graphene , where the reflectivities at zero temperature are nonzero and depend only on the fine structure constant and the angle of incidence . we have shown also that at nonzero temperature the reflectivities of gapped graphene go to unity with vanishing frequency , as it is for a gapless graphene . another distinctive property of gapped graphene is that its reflectivities possess the narrow resonances having the maximum values equal to unity at the border frequency @xmath237 $ ] . this is the case at both zero and nonzero temperature and can be observed experimentally as a jump in the reflectivities of graphene when passing across the border frequency . at frequencies larger than the border frequency there is only minor influence of the mass - gap parameter on the reflectivities of graphene . the next remarkable property of gapped graphene at nonzero temperature is that in the vicinity of some frequency @xmath163 smaller than the border frequency its reflectivities drop to zero . this local minimum also can be observed as a jump in the reflectivities of graphene when passing across the frequency @xmath163 . there is no such effect for a gapless graphene or for a gapped graphene at @xmath141k . it is caused by the fact that the thermal correction to the polarization tensor takes negative values . all the above effects have been derived and investigated analytically in different frequency regions . as a result , the approximate analytic expressions for the reflectivities of graphene and for the frequency @xmath163 have been found . both the cases of the normal incidence and an arbitrary incidence angle were considered . the analytic results have been illustrated by numerical computations over a wide range of frequencies from @xmath238ev to 20ev . from the comparison between approximate analytic and computational results , the exactness of the former was estimated . the developed theory of the reflectivity of gapped graphene at any temperature is based on the first principles of quantum electrodynamics . it can be used in many prospective applications of graphene , such as optical detectors , optoelectronic switches and other mentioned in sec . i. in future it would be interesting to generalize this theory for the case of nonzero chemical potential ( see ref . @xcite for the polarization tensor of graphene taking the chemical potential into account ) and to apply it to graphene - coated substrates . the authors are grateful to m. bordag for helpful discussions . 99 a. h. castro neto , f. guinea , n. m. r. peres , k. s. novoselov , and a. k. geim , rev . phys . * 81 * , 109 ( 2009 ) . m. i. katsnelson , _ graphene : carbon in two dimensions _ ( cambridge university press , cambridge , 2012 ) . g. gmez - santos , phys . b * 80 * , 245424 ( 2009 ) . m. i. katsnelson , k. s. novoselov , and a. k. geim , nat . phys . * 2 * , 620 ( 2006 ) . d. allor , t. d. cohen , and d. a. mcgady , phys . d * 78 * , 096009 ( 2008 ) . c. g. beneventano , p. giacconi , e. m. santangelo , and r. soldati , j. phys a : math . * 42 * , 275401 ( 2009 ) . g. l. klimchitskaya and v. m. mostepanenko , phys . d * 87 * , 215011 ( 2013 ) . x. jiang , y. cao , k. wang , j. wei , d. wu , and h. zhu , surf . coatings technology * 261 * , 327 ( 2015 ) . , edited by r. vajtai ( springer , berlin , 2013 ) . s. watcharotone , d. a. dikin , s. stankovich et al . , nano lett . * 7 * , 1888 ( 2007 ) . l. f. dume , l. he , z. wang et al . , carbon * 87 * , 395 ( 2015 ) . t. kuila , s. bose , p. khanra , a. k. mishra , n. h. kim , and j. h. lee , biosens . bioelectr . * 29 * , 4637 ( 2011 ) . v. bludov , m. i. vasilevskiy , and n. m. r. peres , europhys . lett . * 92 * , 68001 ( 2011 ) . r. r. nair , p. blake , a. n. grigorenko , k. s. novoselov , t. j. booth , t. stauber , n. m. r. peres , and a. k. geim , science * 320 * , 1308 ( 2008 ) . a. b. kuzmenko , e. van neumen , f. carbone , and d. van der marel , phys . * 100 * , 117401 ( 2008 ) . gusynin , s. g. sharapov , and j. p. carbotte , phys . . lett . * 96 * , 256802 ( 2006 ) . v. p. gusynin , s. g. sharapov , and j. p. carbotte , int . b * 21 * , 4611 ( 2007 ) . l. a. falkovsky and s. s. pershoguba , phys . b * 76 * , 153410 ( 2007 ) . t. stauber , n. m. r. peres , and a. k. geim , phys . b * 78 * , 085432 ( 2008 ) . f. h. l. koppens , d. e. chang , and f. j. garca de abajo , nano lett . * 11 * , 3370 ( 2011 ) . m. bordag , i. v. fialkovsky , d. m. gitman , and d. v. vassilevich , phys . b * 80 * , 245406 ( 2009 ) . i. v. fialkovsky , v. n. marachevsky , and d. v. vassilevich , phys . b * 84 * , 035446 ( 2011 ) . m. bordag , g. l. klimchitskaya , and v. m. mostepanenko , phys . b * 86 * , 165429 ( 2012 ) . m. chaichian , g. l. klimchitskaya , v. m. mostepanenko , and a. tureanu , phys . a * 86 * , 012515 ( 2012 ) . g. l. klimchitskaya and v. m. mostepanenko , phys . b * 87 * , 075439 ( 2013 ) . g. l. klimchitskaya , v. m. mostepanenko , and bo e. sernelius , phys . b * 89 * , 125407 ( 2014 ) . g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , phys . rev b * 89 * , 115419 ( 2014 ) . g. l. klimchitskaya and v. m. mostepanenko , phys . rev a * 89 * , 052512 ( 2014 ) . g. l. klimchitskaya and v. m. mostepanenko , phys . b * 91 * , 045412 ( 2015 ) . g. l. klimchitskaya and v. m. mostepanenko , phys . rev b * 91 * , 174501 ( 2015 ) . d. drosdoff and l. m. woods , phys . b * 82 * , 155459 ( 2010 ) . d. drosdoff and l. m. woods , phys . a * 84 * , 062501 ( 2011 ) . bo e. sernelius , europhys . lett . * 95 * , 57003 ( 2011 ) . bo e. sernelius , phys . rev . b * 85 * , 195427 ( 2012 ) . a. d. phan , l. m. woods , d. drosdoff , i. v. bondarev , and n. a. viet , appl . * 101 * , 113118 ( 2012 ) . m. bordag , g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , _ advances in the casimir effect _ ( oxford university press , oxford , 2015 ) . g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , rev . * 81 * , 1827 ( 2009 ) . m. bordag , g. l. klimchitskaya , v. m. mostepanenko , and v. m. petrov , phys . d * 91 * , 045037 ( 2015 ) ; phys . rev . d*93 * , 089907(e ) ( 2016 ) . g. l. klimchitskaya , c. c. korikov , and v. m. petrov , phys . b * 92 * , 125419 ( 2015 ) ; phys . rev . b * 93 * , 159906(e ) ( 2016 ) . s. a. jafari , j. phys . : condens . matter * 24 * , 205802 ( 2012 ) . k. pyatkovsky , j. phys . : condens . matter * 21 * , 025506 ( 2009 ) . v. p. gusynin , s. g. sharapov , and j. p. carbotte , new j. phys . * 11 * , 095013 ( 2009 ) . v. p. gusynin and s. g. sharapov , phys . b * 73 * , 245411 ( 2006 ) . a. i. akhiezer and v. b. berestetskii , _ quantum electrodynamics _ ( interscience , new york , 1965 ) . s. s. schweber , _ an introduction to relativistic quantum field theory _ ( dover , new york , 2005 ) . m. i. katsnelson , eur . j. b * 51 * , 157 ( 2006 ) . i. s. gradshtein and i. m. ryzhik , _ table of integrals , series and products _ ( academic press , new york , 1980 ) . m. bordag , i. fialkovsky , and d. vassilevich , phys . rev . b * 93 * , 075414 ( 2016 ) . ( color online ) the reflectivity of gapped graphene at @xmath141k at the normal incidence is shown as a function of frequency by the three lines from right to left for the mass - gap parameter @xmath6=0.1 , 0.05 , and 0.01ev , respectively . ] ( color online ) the reflectivity of graphene with the mass - gap parameter @xmath6=0.1ev at @xmath141k at the normal incidence is shown as a function of frequency in the vicinity of the resonance frequency @xmath235 . ] ( color online ) the reflectivity of gapped graphene at @xmath141k at the normal incidence is shown as a function of frequency using the natural scales in the vicinity of resonance frequencies . the three lines from right to left are for the mass - gap parameter @xmath6=0.1 , 0.05 , and 0.01ev , respectively . ] ( color online ) the reflectivity of gapped graphene at @xmath119k at the normal incidence is shown by the solid lines 1 and 2 as a function of frequency for the mass - gap parameters @xmath122 and 0.05ev , respectively . the dashed lines 1 and 2 reproduce the respective results at @xmath141k . ] ( color online ) the reflectivity of gapped graphene at @xmath119k at the normal incidence is shown in the vicinity of frequencies where it drops to zero by the solid lines 1 and 2 as a function of frequency for the mass - gap parameters @xmath122 and 0.05ev , respectively . the dashed lines 1 and 2 reproduce the respective results at @xmath141k . ] ( color online ) the frequency , where the reflectivity of graphene at @xmath119k at the normal incidence vanishes , is shown by the solid line as a function of the mass - gap parameter . the dashed line shows the border frequency @xmath232 . ]

given a sample from a multivariate distribution , the uniform random variates generated independently and rearranged in the order specified by the componentwise ranks of the original sample look like a sample from the copula of . this idea can be regarded as a variant on baker s [ j. multivariate anal . 99 ( 2008 ) 23122327 ] copula construction and leads to the definition of the empirical beta copula . the latter turns out to be a particular case of the empirical bernstein copula , the degrees of all bernstein polynomials being equal to the sample size . necessary and sufficient conditions are given for a bernstein polynomial to be a copula . these imply that the empirical beta copula is a genuine copula . furthermore , the empirical process based on the empirical bernstein copula is shown to be asymptotically the same as the ordinary empirical copula process under assumptions which are significantly weaker than those given in janssen , swanepoel and veraverbeke [ j. stat . plan . infer . 142 ( 2012 ) 11891197 ] . a monte carlo simulation study shows that the empirical beta copula outperforms the empirical copula and the empirical checkerboard copula in terms of both bias and variance . compared with the empirical bernstein copula with the smoothing rate suggested by janssen et al . , its finite - sample performance is still significantly better in several cases , especially in terms of bias . _ ams 2010 subject classifications _ : 62g20 , 62g30 , 62h12 + _ keywords _ : copula , empirical copula , bernstein polynomial , empirical bernstein copula , checkerboard copula

1607.04430

let @xmath1 , @xmath2 , be independent and identically distributed random vectors , and assume that the cumulative distribution function , @xmath0 , of @xmath3 is continuous . by sklar s theorem ( sklar @xcite ) , there exists a unique copula , @xmath4 , such that @xmath5 where @xmath6 is the @xmath7th marginal distribution function of @xmath0 . for @xmath2 and @xmath8 , let @xmath9 be the rank of @xmath10 among @xmath11 ; namely , @xmath12 the vector of ranks is denoted by @xmath13 . the basic nonparametric estimator for the copula @xmath4 is the empirical copula ( deheuvels @xcite ) , and we use the version given by @xmath14^d.\ ] ] slightly different definitions are employed for instance in fermanian et al . @xcite and tsukahara @xcite , but the supremum distance between these empirical copula variants is at most @xmath15 . in section [ sec : simul ] , we will also use the so - called empirical checkerboard copula ; see . let @xmath16 be independent random variables which are uniformly distributed on the unit interval and are independent of the sample @xmath17 . for each @xmath8 , let @xmath18 be the order statistics based on @xmath19 . define @xmath20 one of the authors conceived that @xmath21 could be interpreted as a sample from some version of the empirical copula . although this was not quite correct , we found that picking one vector randomly from these @xmath22 vectors is equivalent to sampling from a smoothed version of the empirical copula , which we call the _ empirical beta copula _ in view of the marginal distribution of uniform order statistics . this idea may be regarded as baker @xcite s copula construction based on uniform order statistics , where @xmath23-tuples of order statistics are determined by the rank vectors @xmath24 , and the probability of choosing among them is equal to @xmath25 . the empirical beta copula arises as a particular case of the empirical bernstein copula when the degrees of all bernstein polynomials are set equal to the sample size . the bernstein copula and the empirical bernstein copula are introduced in sancetta and satchell @xcite , and the asymptotic behavior of the latter is studied in janssen et al . we give necessary and sufficient conditions for the empirical bernstein copula to be a genuine copula , conditions which hold for the empirical beta copula . we show asymptotic results for the empirical bernstein copula process under weaker assumptions than those in janssen et al . @xcite . the advantages of the empirical beta copula are that it is a genuine copula , that it does not require the choice of a smoothing parameter , and that simulating random samples from it is straightforward . furthermore , the corresponding empirical process converges weakly to a gaussian process under standard smoothness conditions on the underlying copula . for small samples , the empirical beta copula outperforms the empirical copula both in terms of bias and variance . compared with the empirical bernstein copula with polynomial degrees as suggested in janssen et al . @xcite , the empirical beta copula is still more accurate for several copula models , especially in terms of the bias . the paper is organized as follows . in section [ sec : cop ] , we define the empirical beta copula and prove its relation to the empirical bernstein copula together with some auxiliary results for the bernstein transformation . in section [ sec : asym ] , we formulate and prove the asymptotic results for the empirical bernstein copula process . a monte carlo simulation study is presented in section [ sec : simul ] , and we conclude the paper with some remarks in section [ sec : concl ] . some additional proofs are given in the appendix . we define the empirical beta copula and we show that the sampling scheme introduced in the preceding section amounts to the first stage of sampling from it . the empirical beta copula turns out to be a particular case of the empirical bernstein copula . necessary and sufficient conditions are given for the empirical bernstein copula to be a genuine copula , conditions which imply that the empirical beta copula is a genuine copula . finally , we provide a non - asymptotic bound for the difference between the empirical copula and the empirical beta copula . we continue to use the notation given in the introduction . to express mathematically the idea stated above , we replace each indicator function in the definition of the empirical copula by the cumulative distribution function of @xmath26 in conditionally on @xmath27 . since @xmath28 is the @xmath29th order statistic of an independent random sample of size @xmath22 from the uniform distribution on @xmath30 $ ] , its distribution is @xmath31 . thus we define the _ empirical beta copula _ by @xmath32^d,\ ] ] where , for @xmath33 $ ] and @xmath34 , @xmath35 here and henceforth , @xmath36 generically denote the order statistics based on @xmath22 independent random variables @xmath37 , uniformly distributed on @xmath30 $ ] . picking one element randomly from among the points @xmath38 , @xmath2 , in amounts to sampling from the empirical beta copula @xmath39 conditionally on @xmath40 . in the absence of ties , all @xmath23 margins of @xmath39 are equal to the uniform distribution on @xmath30 $ ] , so that @xmath39 is a genuine copula . indeed , for @xmath41 and @xmath42 $ ] , @xmath43 = { \operatorname{e}}\biggl[\frac{1}{n}\sum_{r=1}^n { \mathbf{1}}\{u_r\leq u_j\}\biggr]=u_j.\end{aligned}\ ] ] before we show that the empirical beta copula is a particular case of the empirical bernstein copula , we need to state and prove some auxiliary results on bernstein polynomials in general . put @xmath44,\ m\in{\mathbb{n}},\ s\in\{0,1,\ldots , m\}.\ ] ] for @xmath45 and a real array @xmath46 , consider the following bernstein polynomial : @xmath47^d.\ ] ] our objective in this subsection is to derive a ( necessary and ) sufficient condition on the coefficients @xmath48 for @xmath49 to be a copula . to this end , we need the following two lemmas . [ prop : prelim ] let @xmath50 and @xmath51 . 1 . we have @xmath52 for all @xmath53 $ ] if and only if @xmath54 for all @xmath55 . we have @xmath56 for all @xmath53 $ ] if and only if @xmath57 for all @xmath55 . proof the ` if ' parts follow from direct computation . the ` only if ' parts are an immediate consequence of the fact that the bernstein polynomials @xmath58 form a basis of the linear space of all real polynomials with degree not larger than @xmath22 . [ prop : bernstein-0 ] @xmath59 for all @xmath60^d$ ] if and only if @xmath61 . proof the ` if ' part is trivial . the ` only if ' part can be proven by induction on @xmath23 , using lemma [ prop : prelim](i ) both for the case @xmath62 as for the induction step . we leave the details to the reader . for @xmath41 , define the difference operator @xmath63 mapping a given array @xmath48 to the new array @xmath64 given by @xmath65 difference operators @xmath63 with distinct indices @xmath7 can be composed in the obvious way . in particular , @xmath66 where @xmath67 for each @xmath41 . furthermore , consider the following three conditions on a real array @xmath48 : 1 . @xmath68 as soon as @xmath69 for some @xmath41 ; 2 . @xmath70 for each @xmath41 and each @xmath71 ; 3 . @xmath72 for all @xmath73 . [ prop : bernstein - polynomial - copula ] if the conditions ( c.1 ) , ( c.2 ) and ( c.3 ) hold , then the bernstein polynomial @xmath49 is a copula . moreover , ( c.1 ) and ( c.2 ) are necessary for @xmath49 to be a copula . before commencing the proof , recall that a function @xmath74^d{\to}[0,1]$ ] is a copula if and only if the following conditions hold : a. @xmath4 is _ grounded _ ; that is , @xmath75 whenever at least one of the @xmath76 equals @xmath77 . b. @xmath78 for all @xmath8 . c. @xmath4 is _ @xmath23-increasing _ ; that is , for all @xmath79^d$ ] with @xmath80 for all @xmath8 , we have @xmath81 where the sum is taken over all @xmath82 such that @xmath83 for all @xmath8 . see nelsen @xcite or more explicitly , theorem 1.1 of mai and scherer @xcite . proof of proposition [ prop : bernstein - polynomial - copula ] since @xmath84 , we have @xmath85 it then follows from lemma [ prop : bernstein-0 ] that ( c.1 ) is equivalent to the groundedness of @xmath49 . since @xmath86 , we have @xmath87 by lemma [ prop : prelim](ii ) , condition ( c.2 ) is equivalent to the marginal condition ( ii ) above for @xmath49 . using the identity and by induction , one can easily find that @xmath88 if ( c.3 ) is satisfied , then it is obvious from the above expression that we have @xmath89 for all @xmath90 . since @xmath49 is infinitely differentiable , it thus follows from proposition [ prop : d - increasing - partial - derivative ] that @xmath49 is @xmath23-increasing . therefore ( c.1 ) , ( c.2 ) and ( c.3 ) imply that @xmath49 is a copula . the necessity of ( c.1 ) and ( c.2 ) for @xmath49 to be a copula has already been proved . the condition ( c.3 ) is not necessary for @xmath49 to be a copula . take @xmath91 and @xmath92 , and let @xmath93 for @xmath94 , @xmath95 , and @xmath96 , @xmath97 , @xmath98 , @xmath99 , @xmath100 , @xmath101 . then @xmath102 satisfies ( c.1 ) and ( c.2 ) , but not ( c.3 ) since @xmath103 . by direct inspection , it is easy to see that @xmath104 for all @xmath60 ^ 2 $ ] because @xmath105 is linear in @xmath106 . for a function @xmath107^d{\to}{\mathbb{r}}$ ] , the bernstein polynomial of order @xmath108 of @xmath109 is defined by @xmath110 when @xmath4 is a copula , @xmath111 is called the _ bernstein copula _ of @xmath4 . for the empirical copula @xmath112 , we call @xmath113 the _ empirical bernstein copula_. for a copula @xmath4 , the real array @xmath48 defined by @xmath114 satisfies conditions ( c.1)(c.3 ) . by proposition [ prop : bernstein - polynomial - copula ] , the bernstein copula @xmath111 of @xmath4 is therefore a copula itself . on the other hand , the empirical bernstein copula @xmath113 is not necessarily a copula . [ prop : bernstein - genuine - copula ] let @xmath112 be the empirical copula of a sample @xmath17 of @xmath23-variate vectors without ties in any of the @xmath23 components . let @xmath115 . the empirical bernstein copula @xmath113 is a copula if and only if all the polynomial degrees @xmath116 are divisors of @xmath22 . proof the empirical copula @xmath112 in is a cumulative distribution function on @xmath30^d$ ] without mass on the lower boundary of the unit hypercube . the real array @xmath48 defined by @xmath117 therefore satisfies conditions ( c.1 ) and ( c.3 ) above . by proposition [ prop : bernstein - polynomial - copula ] , @xmath113 is a copula if and only if that array @xmath48 also satisfies ( c.2 ) . let @xmath118 denote the integer part of the real number @xmath119 . for @xmath120 and @xmath67 , we have @xmath121 condition ( c.2 ) is fulfilled if and only if the right - hand side is equal to @xmath122 for all @xmath67 . setting @xmath123 shows that this requires @xmath124 to be a divisor of @xmath22 , and the latter condition is easily seen to be sufficient as well . now we show that the empirical beta copula @xmath39 in is a particular case of the empirical bernstein copula . [ lem : betabernstein ] let @xmath112 be the empirical copula of a sample @xmath17 of @xmath23-variate vectors without ties in any of the @xmath23 components . then @xmath125 . proof the beta cumulative distribution function @xmath126 in can be written in terms of bernstein polynomials @xmath127 as @xmath128.\ ] ] as a consequence , for @xmath129^d$ ] , @xmath130\prod_{j=1}^d p_{n , s_j}(u_j).\end{aligned}\ ] ] invoking the definition of the empirical copula @xmath112 in finally yields @xmath131 as required . proposition [ prop : bernstein - genuine - copula ] and lemma [ lem : betabernstein ] confirm that the empirical beta copula is itself a copula . we had already reached this conclusion in subsection [ subsec : beta ] by a direct argument . sancetta and satchell @xcite first introduced the bernstein copula . in that paper , it is falsely claimed ( p. 537 ) that a function @xmath4 on @xmath132^d$ ] is a copula if and only if @xmath4 is nondecreasing in all its arguments and satisfies the frchet bounds @xmath133 [ for a counterexample , see nelsen ( * ? ? ? * exercise 2.11 ) . ] our results above correct both the statement and the proof of their theorem 1 . we provide a deterministic , non - asymptotic bound for the difference between the empirical copula @xmath112 and the empirical beta copula @xmath39 . [ prop : nonasymp - bd ] let @xmath112 and @xmath39 be the empirical copula and the empirical beta copula , respectively , of a sample @xmath17 of @xmath23-variate vectors without ties in any of the @xmath23 components . we have @xmath134^d}|{\mathbb{c}}_n({\boldsymbol{u}})-{\mathbb{c}}_n^\beta({\boldsymbol{u}})| \leq d[n^{-1/2}(\log n)^{1/2}+n^{-1/2}+n^{-1}].\ ] ] proof consider the ranks @xmath9 as in . using the identity @xmath135 we have , for @xmath129^d$ ] , combining and , @xmath136 let @xmath137 and fix @xmath129^d$ ] . let @xmath138 be the set of indices @xmath2 such that @xmath139 $ ] for some @xmath8 . the set @xmath138 contains at most @xmath140 elements because for each @xmath8 , the ranks @xmath141 constitute a permutation of @xmath142 . it follows that @xmath143 let @xmath144 and @xmath33 $ ] . let @xmath145 denote a binomial random variable with @xmath22 trials and success probability @xmath146 . if @xmath147 , then , in view of and by hoeffding s inequality , @xmath148 similarly , if @xmath149 , then @xmath150 the two displays taken together imply that for @xmath151 $ ] , we have @xmath152 combine and to see that @xmath153 the right - hand side does not depend on @xmath154 and we still have the liberty of choosing @xmath155 . choosing @xmath156 yields . by proposition [ prop : nonasymp - bd ] , the empirical beta copula is uniformly consistent whenever the empirical copula is . nevertheless , the bound in is too weak to conclude that the empirical processes @xmath157 and @xmath158 are asymptotically the same . this is the topic of the section [ sec : asym ] . we provide asymptotic theory for the empirical bernstein copula process @xmath159 under weaker assumptions than those in janssen et al . @xcite . here @xmath160 is a sequence of multi - indices such that @xmath161 as @xmath162 . for @xmath115 and @xmath129^d$ ] , let @xmath163 be the law of the random vector @xmath164 , where @xmath165 are independent random variables , the law of @xmath166 being binomial@xmath167 for each @xmath41 . for asymptotic analysis , it is convenient to write the empirical bernstein copula as a mixture : @xmath168^d}{\mathbb{c}}_n({\boldsymbol{w}})\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) , \qquad { \boldsymbol{u}}\in [ 0 , 1]^d.\ ] ] the empirical copula process and the empirical bernstein copula process are defined by @xmath169 in view of lemma [ lem : betabernstein ] , the special case @xmath170 yields the empirical beta copula process : @xmath171 noting that @xmath172^d}c({\boldsymbol{w}})\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}})$ ] , we have @xmath173^d}{\mathbb{g}}_n({\boldsymbol{w}})\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) + \sqrt{n}\bigl(b_{{\boldsymbol{m}}}(c)({\boldsymbol{u}})-c({\boldsymbol{u}})\bigr).\ ] ] we call the first and second term on the right - hand side of the stochastic term and the bias term , respectively . we deal with these terms in subsections [ subsec : stoch ] and [ subsec : bias ] . the combination of both analyses in subsection [ subsec : thm ] then yields our main result , theorem [ thm : empbernstein ] below , on the asymptotic distribution of the empirical bernstein copula process . let @xmath174^d)$ ] be the banach space of real - valued , bounded functions on @xmath132^d$ ] , equipped with the supremum norm @xmath175 . the arrow @xmath176 denotes weak convergence in the sense used in van der vaart and wellner @xcite . if @xmath177 are independent and identically distributed random vectors from a common distribution with copula @xmath4 , then , provided @xmath4 satisfies condition [ cond : s ] below , we have @xcite @xmath178 in @xmath179^d)$ ] , where @xmath180 and where @xmath181 is a centered , gaussian process with continuous trajectories and covariance function @xmath182^d.\ ] ] more generally , if @xmath177 is a strictly stationary time series whose stationary @xmath23-variate distribution has copula @xmath4 satisfying condition [ cond : s ] , then , provided certain mixing or weak dependence conditions hold , we still have weak convergence in @xmath179^d)$ ] of the empirical copula process , as shown in bcher and volgushev @xcite : @xmath183 with @xmath184 a tight , centered gaussian process on @xmath30^d$ ] whose covariance structure depends on the full distribution of the time series . in the analysis of the stochastic term in , the weak convergence in is all that we need to know of the empirical copula process . [ lem : stochastic - term ] if @xmath185 in @xmath174^d)$ ] as @xmath186 , and if the limiting process @xmath184 has continuous trajectories almost surely , then , whenever @xmath187 as @xmath186 , @xmath188^d}\left|\int_{[0,1]^d}{\mathbb{g}}_n({\boldsymbol{w}})\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) -{\mathbb{g}}_n({\boldsymbol{u}})\right|={o_{\scriptscriptstyle p}}(1).\ ] ] proof let @xmath189 denote the maximum norm on @xmath190 . for @xmath191 , we have @xmath192^d } \lvert { \mathbb{g}}_n({\boldsymbol{w}})-{\mathbb{g}}_n({\boldsymbol{u } } ) \rvert \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}})\\ \leq 2\ , \lvert{\mathbb{g}}_n\rvert_\infty \ , \mu_{{\boldsymbol{m}},{\boldsymbol{u}}}(\{{\boldsymbol{w}}\colon \lvert { \boldsymbol{w}}-{\boldsymbol{u}}\rvert_\infty>{\varepsilon}\ } ) + \sup _ { \lvert { \boldsymbol{x}}-{\boldsymbol{y}}\rvert_\infty\leq{\varepsilon } } \lvert { \mathbb{g}}_n({\boldsymbol{x}})-{\mathbb{g}}_n({\boldsymbol{y } } ) \rvert.\end{gathered}\ ] ] for a given @xmath193 , we can find @xmath194 sufficiently small such that @xmath195 moreover , the weak convergence @xmath185 implies that @xmath196 as @xmath186 . finally , chebyshev s inequality yields @xmath197 which converges to zero as @xmath198 , uniformly in @xmath60^d$ ] . since @xmath199 was arbitrary , the conclusion follows . now we turn to the bias term on the right - hand side of : @xmath200^d } \bigl ( c({\boldsymbol{w}})-c({\boldsymbol{u } } ) \bigr ) \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}}).\ ] ] one - dimensional bernstein polynomials are well studied , and results on the accuracy of the approximation of a function by its associated bernstein polynomial are presented in lorentz @xcite and devore and lorentz ( * ? ? ? * chapter 10 ) . we need to extend these to the multidimensional case . [ lem : bias - term1 ] for a @xmath23-variate copula @xmath4 and a multi - index @xmath115 we have , writing @xmath201 , @xmath202 proof any copula is lipschitz with respect to the @xmath203-norm with lipschitz constant equal to @xmath204 . we obtain @xmath205^d } \lvert c({\boldsymbol{w}})-c({\boldsymbol{u } } ) \rvert \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \leq\sum_{j=1}^d\int_{[0,1]^d } \lvert w_j - u_j \rvert \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}}).\ ] ] fix @xmath41 and let @xmath166 be a binomial@xmath167 random variable . by the cauchy schwarz inequality , @xmath206^d } \lvert w_j - u_j \rvert \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \leq { \operatorname{e } } [ \lvert s_j / m_j - u_j \rvert ] \le \sqrt { { \operatorname{var}}(s_j / m_j ) } \le \frac{1}{2\sqrt{m_j}},\ ] ] which yields the conclusion . according to lemma [ lem : bias - term1 ] , the uniform convergence rate of @xmath111 to @xmath4 is at least @xmath207 as @xmath198 . with a bit of additional smoothness , this can be strengthened to @xmath208 . the following condition originates from segers @xcite . [ cond : s ] for each @xmath120 , the copula @xmath4 has a continuous first - order partial derivative @xmath209 on the set @xmath210^d\colon 0<u_j<1\}$ ] . [ lem : bias - term2 ] if the copula @xmath4 satisfies condition [ cond : s ] , then @xmath211 proof by monotonicity and lipschitz continuity of @xmath4 , we have @xmath212 . fix @xmath213^d$ ] and write @xmath214 for @xmath53 $ ] . the function @xmath215 is continuous and , by condition [ cond : s ] , continuously differentiable on @xmath216 with derivative @xmath217 for each @xmath7 such that @xmath218 , the corresponding term in the sum is zero no matter how @xmath219 is defined . if @xmath220 , then @xmath221 whenever @xmath222 , so that the partial derivative @xmath223 is well defined at @xmath224 by condition [ cond : s ] . by the fundamental theorem of calculus , we get @xmath225 hence , by fubini s theorem , @xmath226^d } \bigl ( c ( { \boldsymbol{w } } ) - c ( { \boldsymbol{u } } ) \bigr ) \ , { { \mathrm{d } } } \mu_{{\boldsymbol{m } } , { \boldsymbol{u } } } ( { \boldsymbol{w } } ) \\ & = \sum_{j=1}^d\int_0 ^ 1 \left ( \int_{[0,1]^d}(w_j - u_j ) \ , \dot{c}_j({\boldsymbol{w}}(t))\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \right ) \ , { { \mathrm{d}}}t.\end{aligned}\ ] ] we need to bound its absolute value uniformly in @xmath60^d$ ] . fix @xmath8 and @xmath53 $ ] . choose @xmath227 . first , let @xmath60^d$ ] be such that @xmath228 $ ] . since @xmath229 , we find @xmath230^d}(w_j - u_j ) \ , \dot{c}_j({\boldsymbol{w}}(t))\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}})\right\rvert \leq \int_{[0,1]^d } \lvert w_j - u_j \rvert \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \leq \sqrt{\frac{u_j(1-u_j)}{m_j } } < \sqrt{\frac{\delta}{m_*}}.\ ] ] second , let @xmath60^d$ ] be such that @xmath231 $ ] . then @xmath232 is well defined . since @xmath233^d}(w_j - u_j)\,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u } } } ( { \boldsymbol{w}})=0 $ ] , we obtain @xmath234^d}(w_j - u_j ) \bigl ( \dot{c}_j({\boldsymbol{w}}(t))-\dot{c}_j({\boldsymbol{u } } ) \bigr ) \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \right ) { { \mathrm{d}}}t.\ ] ] choose @xmath235 . let @xmath189 denote the maximum norm on @xmath190 . split the integral over @xmath236^d$ ] into two parts , according to whether @xmath237 is larger than @xmath238 or not . we have @xmath239^d}(w_j - u_j ) \bigl ( \dot{c}_j({\boldsymbol{w}}(t))-\dot{c}_j({\boldsymbol{u } } ) \bigr ) \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}})\right\rvert \\ \leq \sup _ { \lvert { \boldsymbol{w}}-{\boldsymbol{u}}\rvert_\infty\leq{\varepsilon } } \lvert \dot{c}_j({\boldsymbol{w}}(t))-\dot{c}_j({\boldsymbol{u } } ) \rvert \int_{[0,1]^d } \lvert w_j - u_j \rvert \ , { { \mathrm{d } } } \mu_{{\boldsymbol{m } } , { \boldsymbol{u}}}({\boldsymbol{w } } ) \\ + \int_{[0,1]^d } { \mathbf{1}}\{\lvert{\boldsymbol{w}}-{\boldsymbol{u}}\rvert_\infty > { \varepsilon}\ } \ , \lvert w_j - u_j \rvert \,{{\mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}}).\end{gathered}\ ] ] here we have used the inequality @xmath240 again . we now treat both terms on the right - hand side separately . * first , fix @xmath193 . by condition [ cond : s ] , we can choose @xmath241 sufficiently small such that , for each @xmath8 , @xmath242^d\\ \delta\leq u_j\leq 1-\delta } } \sup_{\lvert { \boldsymbol{w}}-{\boldsymbol{u}}\rvert_\infty\leq{\varepsilon } } \lvert \dot{c}_j({\boldsymbol{w}})-\dot{c}_j({\boldsymbol{u } } ) \rvert \leq \eta.\ ] ] furthermore , we have @xmath243^d } \lvert w_j - u_j \rvert \ , { { \mathrm{d } } } \mu_{{\boldsymbol{m } } , { \boldsymbol{u}}}({\boldsymbol{w } } ) \le 1 / ( 2\sqrt{m_*})$ ] by . * second , by the cauchy schwarz inequality , @xmath244^d } { \mathbf{1}}\ { \lvert { \boldsymbol{w}}-{\boldsymbol{u}}\rvert_\infty > { \varepsilon}\ } \ , \lvert w_j - u_j \rvert \ , { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \leq \sqrt { \frac { \mu_{{\boldsymbol{m}},{\boldsymbol{u}}}(\{{\boldsymbol{w}}\colon \lvert { \boldsymbol{w}}- { \boldsymbol{u}}\rvert_\infty > { \varepsilon}\ } ) } { m _ * } } .\ ] ] by the markov inequality and , the numerator of the right - hand side converges to zero as @xmath198 , uniformly in @xmath60^d$ ] . by the above bounds , it follows that @xmath245 . since @xmath246 and @xmath193 were arbitrary , the result follows . condition [ cond : s ] on @xmath4 is significantly weaker than the assumption that @xmath4 has bounded third - order partial derivatives , which janssen et al . ( 2012 ) require to prove their asymptotic results for the empirical bernstein copula . the latter assumption is violated by many common parametric copula families . janssen et al . ( 2012 ) in fact proved a pointwise @xmath247 convergence rate of the bias term under the assumption that @xmath4 has bounded third - order partial derivatives , but to obtain such a rate , the assumption is still too strong . in fact , a lipschitz condition on the first - order partial derivatives already suffices . [ lem : bias - term3 ] let @xmath4 be a copula satisfying condition [ cond : s ] . let @xmath248^d$ ] and suppose that there exists @xmath249 such that , for each @xmath41 such that @xmath250 , we have @xmath251 and @xmath223 is lipschitz on @xmath252^d : \lvert { \boldsymbol{w}}- { \boldsymbol{u}}\rvert_\infty \le { \varepsilon}\}$ ] . then @xmath253 proof if @xmath254 for some @xmath41 , then , since the binomial@xmath255 distribution is concentrated at @xmath124 , we can omit the @xmath7th coordinate altogether and pass to the appropriate @xmath256-dimensional margins of @xmath111 and @xmath4 . in this way , we can eliminate all margins @xmath7 such that @xmath254 . it follows that , without loss of generality , we can assume that @xmath257 . as in the proof of proposition [ lem : bias - term2 ] , we have the representation @xmath258^d } ( w_j - u_j ) \ , \bigl ( \dot{c}_j({\boldsymbol{w}}(t ) ) - \dot{c}_j({\boldsymbol{u } } ) \bigr ) \ , { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \right ) { { \mathrm{d}}}t.\ ] ] we fix @xmath259 and @xmath41 and split the integral over @xmath82 into two parts , according to whether @xmath260 is larger than @xmath238 or not . this yields the following bound : @xmath261^d } ( w_j - u_j ) \ , \bigl ( \dot{c}_j({\boldsymbol{w}}(t ) ) - \dot{c}_j({\boldsymbol{u } } ) \bigr ) \ , { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \right\rvert \\ \le \int_{[0 , 1]^d } m _ * \ , \lvert w_j - u_j \rvert \ , \lvert \dot{c}_j({\boldsymbol{w}}(t ) ) - \dot{c}_j({\boldsymbol{u } } ) \rvert \ , { \mathbf{1}}\ { \lvert { \boldsymbol{w}}- { \boldsymbol{u}}\rvert_\infty \le { \varepsilon}\ } \ , { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \\ + \int_{[0 , 1]^d } m _ * \ , \lvert w_j - u_j \rvert \ , { \mathbf{1}}\ { \lvert { \boldsymbol{w}}- { \boldsymbol{u}}\rvert_\infty > { \varepsilon}\ } { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w}}).\end{gathered}\ ] ] for the second term on the right - hand side , we used the fact that @xmath240 . we now bound the two integrals separately . * recall that @xmath262 , a convex combination of @xmath154 and @xmath82 . the lipschitz assumption on @xmath223 implies that there exists a constant @xmath263 such that @xmath264 we find that the first term on the right - hand side of is bounded by @xmath265^d } m _ * \ , \lvert w_j - u_j \rvert \ , \lvert w_k - u_k \rvert \ , { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) \le \frac{1}{4 } tld.\ ] ] here we applied the cauchy schwarz inequality , by which @xmath266^d } m _ * \ , \lvert w_j - u_j \rvert \ , \lvert w_k - u_k \rvert \ , { { \mathrm{d}}}\mu_{{\boldsymbol{m}},{\boldsymbol{u}}}({\boldsymbol{w } } ) & = { \operatorname{e } } [ m^ * \lvert s_j / m_j - u_j \rvert \ , \lvert s_k / m_k - u_k \rvert ] \\ & \le m^ * \ , \left ( { \operatorname{e } } [ ( s_j / m_j - u_j)^2 ] \ , { \operatorname{e } } [ ( s_k / m_k - u_k)^2 ] \right)^{1/2 } \\ & \le m^ * \left ( m_j^{-1 } u_j(1-u_j ) \ , m_k^{-1 } u_k(1-u_k ) \right)^{1/2 } \\ & \le \left ( u_j(1-u_j ) \ , u_k(1-u_k ) \right)^{1/2 } \le \frac{1}{4}.\end{aligned}\ ] ] * for the second term on the right - hand side of , we can bound the indicator function in the integrand by @xmath267 the resulting integrals have already been bounded above in the previous term . this finishes the proof of proposition [ lem : bias - term3 ] . by lemma [ lem : betabernstein ] , the empirical beta copula @xmath39 is equal to the empirical bernstein copula @xmath113 with @xmath268 for all @xmath41 . propositions [ lem : stochastic - term ] , [ lem : bias - term2 ] and [ lem : bias - term3 ] then immediately imply the following theorem and corollary . recall the empirical copula process @xmath269 , the empirical bernstein copula process @xmath270 , and the empirical beta copula process @xmath271 . in theorem [ thm : empbernstein ] , we assume weak convergence of @xmath272 to some limit process @xmath184 . see the discussion before proposition [ lem : stochastic - term ] for a justification of this condition and for the form of the limit process . in particular , the condition is automatically satisfied in the iid case when @xmath4 satisfies condition [ cond : s ] , and then the limit process is given by @xmath273 in . [ thm : empbernstein ] suppose that @xmath4 satisfies condition [ cond : s ] and that @xmath185 in @xmath174^d)$ ] as @xmath186 , the limiting process @xmath184 having continuous trajectories almost surely . let @xmath274 be multi - indices such that @xmath161 as @xmath162 . a. if @xmath275 , then , for all @xmath248^d$ ] that satisfy the condition in proposition [ lem : bias - term3 ] , we have @xmath276 b. if @xmath277 , then , in @xmath179^d)$ ] , @xmath278 proof recall the decomposition of @xmath279 in . the stochastic term is equal to @xmath280 , uniformly in @xmath129^d$ ] , by proposition [ lem : stochastic - term ] . \(i ) since @xmath281 , the bias term converges to zero at the given point @xmath154 . \(ii ) since @xmath282 , the bias term converges to zero uniformly in @xmath129^d$ ] thanks to proposition [ lem : bias - term2 ] . setting @xmath283 in theorem [ thm : empbernstein](ii ) yields the following corollary . [ cor : empbeta ] suppose that @xmath4 satisfies condition [ cond : s ] and that @xmath185 in @xmath174^d)$ ] as @xmath186 , the limiting process @xmath184 having continuous trajectories almost surely . then , in @xmath179^d)$ ] , @xmath284 the conclusion of this section is that under reasonable conditions , the empirical bernstein and empirical beta copulas have the same large - sample distribution as the empirical copula . the real advantage of the use of the smoothed versions is visible mainly for small samples . the difference can perhaps be quantified via higher - order asymptotic theory . instead , we assess the finite - sample performance through monte carlo simulations . in this section , we compare the finite - sample performance of the empirical beta copula with those of various other estimators by a monte carlo experiment . the simulations were performed in @xcite using the package @xcite . we show the results for five copulas , three bivariate ones and two trivariate ones : * the bivariate farlie gumbel morgenstern ( fgm ) copula with negative dependence ( @xmath285 ) ; * the bivariate independence copula ; * the bivariate gaussian copula with positive dependence ( @xmath286 ) ; * the trivariate @xmath287 copula with degrees - of - freedom parameter equal to @xmath288 and pairwise correlation parameters equal to @xmath289 , @xmath290 and @xmath291 ; * the trivariate nested archimedean copula with frank generators and with kendall s tau equal to @xmath292 at the upper node and @xmath293 at the lower node . see , e.g. , joe @xcite and nelsen @xcite for the explicit functional forms and properties of these copulas . we have run simulations for many other copula models , but the results were qualitatively the same as the ones showed here . we compared the following five estimators : * the empirical copula @xmath112 in ; * the empirical checkerboard copula @xmath294 in below ; * the empirical beta copula @xmath39 in ; * the empirical bernstein copula @xmath295 with @xmath296 and @xmath297 , the smallest integer not smaller than @xmath298 ; * the empirical bernstein copula @xmath295 with @xmath296 and @xmath299 as recommended by janssen et al . @xcite , see below . for each possible combination of estimator and model and for sample sizes @xmath22 from @xmath300 up to @xmath301 , we computed the integrated ( over the unit cube ) squared bias , the integrated variance , and the integrated mean squared error by monte carlo simulation . see appendix [ app : estim ] for the definitions of the three performance measures and a description of our computation method . as the simplest version of smoothed empirical copula , we introduce the _ empirical checkerboard copula _ defined by @xmath302 ( li et al . @xcite for @xmath303 ; carley and taylor @xcite for general @xmath23 ) . we note that this is a genuine copula , just like the empirical beta copula . in each of the graphs in figure [ fig : emp - checker - beta ] , the horizontal axis indicates the sample size ; the vertical axis indicates the integrated squared bias , the integrated variance and the integrated mean squared error for the left , middle and right panels , respectively ( the same comment also applies to the other figures ) . from the figures , one sees that the empirical copula performs worst with respect to all three measures . the empirical beta copula and the empirical checkerboard copula are comparable in terms of bias , but the variance of the former is smaller than the one of the latter , resulting in a smaller mean squared error for the empirical beta copula in all cases considered . note that the empirical checkerboard copula can be thought of as smoothing at bandwidth @xmath304 , whereas the empirical beta copula involves smoothing at bandwidth @xmath305 . in this subsection , we consider the empirical bernstein copula with all the orders @xmath306 equal to an integer @xmath299 . our first choice for the order of bernstein polynomial is @xmath297 , where @xmath307 denotes the smallest integer not smaller than @xmath119 . with this choice , we see the consequence of proposition [ prop : bernstein - genuine - copula ] because @xmath299 is a divisor of @xmath22 only when @xmath22 is a multiple of 3 . our second choice for @xmath299 is based on janssen et al . @xcite , who recommend , in the bivariate case , the following choice for @xmath299 : @xmath308 ( or the integer part thereof ) , where @xmath309 and with @xmath223 and @xmath310 the first- and second - order partial derivatives of @xmath4 with respect to @xmath76 , for @xmath311 . the following remarks on this choice @xmath312 are in order . * this choice requires the knowledge of @xmath4 through the derivatives @xmath223 and @xmath310 . to estimate these derivatives in small samples is not easy . * for the independence copula , we have @xmath313 , so that it is not clear how to define @xmath312 . * in lemma 3(i ) in @xcite , an @xmath314 term is written but subsequently neglected . however , the order of magnitude of this term could be larger than the remainder terms that are analyzed later on . in the simulations , we take the true ( but in practice unknown ) value for @xmath315 for the bivariate fgm and gaussian copulas . from figure [ fig : beta - bern3-bernjsv ] , it is observed that the performance of the empirical bernstein copulas are strongly affected by the periodicity due to the divisibility relation between @xmath22 and @xmath299 , especially in terms of bias . compared to the empirical bernstein copulas , the empirical beta copula has a smaller bias and a larger variance in every case . there is no clear ordering of the three in terms of the mean squared error . it is interesting to note that while the mean squared error of the empirical beta copula is smaller in positively dependent cases , it is larger in negatively dependent cases . at this point , we have no explanation for this phenomenon . the empirical beta copula can be considered as a simple but effective way of correcting and smoothing the empirical copula . no smoothing parameter needs to be chosen . the empirical beta copula is a special case of the empirical bernstein copula , but in contrast to the latter , it is always a genuine copula . moreover , it is extremely simple to simulate samples from it . the asymptotic distribution of the empirical beta copula is the same as that of the empirical copula , but in small samples , it performs better both in terms of bias and variance . moreover , there seems little to be gained from using bernstein smoothers at other polynomial degrees @xmath299 , except in special cases such as the independence copula . we also note that the only properties of the kernels @xmath163 that we needed in the asymptotic analysis were moment properties and concentration inequalities of their margins around the points @xmath76 . this implies that the approach may perhaps be extended to products of other families of smoothing kernels than beta kernels . our smoothing procedure might also have a beneficial effect on the accuracy of resampling schemes for the empirical copula process ( bcher and dette @xcite ) . more specifically , testing procedures based on the empirical copula typically rely on the bootstrap for the computation of the critical values of the test statistic . for finite samples , the accuracy is often not very good : the true type i error of the test may differ greatly from the nominal one . then , the question is how to construct a bootstrap or multiplier resampling scheme for the empirical beta and bernstein copulas . this is a topic for future research , together with the question of higher - order asymptotics for the various nonparametric copula estimators . j. segers gratefully acknowledges funding by contract `` projet dactions de recherche concertes '' no . 12/17 - 045 of the `` communaut franaise de belgique '' , by iap research network grant p7/06 of the belgian government ( belgian science policy ) , and by the `` projet de recherche '' no . frfc pdr t.0146.14 of the `` fonds de la recherche scientifique fnrs '' ( belgium ) . h. tsukahara is supported by jsps kakenhi grant number 15h03337 . we derive a convenient expression for the derivative of the bernstein - type polynomial ( making no claim of originality ) . . then we have @xmath317\\[5pt ] & = \begin{cases } -n(1-t)^{n-1 } & \text{if $ r=0$}\\[5pt ] \displaystyle\frac{n!}{(r-1)!(n - r)!}t^{r-1}(1-t)^{n - r } -\frac{n!}{r!(n - r-1)!}t^r(1-t)^{n - r-1 } & \text{if $ r=1,\ldots , n-1$}\\[10pt ] nt^{n-1 } & \text{if $ r = n$ } \end{cases}\end{aligned}\ ] ] using the above expression and summation by parts , it is easy to see that for any sequence @xmath318 , @xmath319 , the following identity holds : @xmath320 when a function is differentiable , we can give a condition for @xmath23-increasing property in terms of its partial derivative . [ prop : d - increasing - partial - derivative ] suppose that a function @xmath107^d{\to}[0,1]$ ] is infinitely differentiable on @xmath321 . then @xmath109 is @xmath23-increasing if and only if @xmath322 for all @xmath90 . _ proof_. suppose that @xmath109 is @xmath23-increasing . then , since @xmath109 is infinitely differentiable on @xmath321 , it follows that for @xmath90 and @xmath323 , @xmath324 , @xmath325 where the sum is taken over all @xmath82 such that @xmath218 or @xmath326 for all @xmath324 . conversely , suppose that @xmath322 . by the standard calculus , for all @xmath79^d$ ] with @xmath80 for all @xmath324 , we have @xmath327 where the sum is taken over all @xmath82 such that @xmath218 or @xmath328 for all @xmath324 . @xmath329 given a copula estimator @xmath330 , we consider in section [ sec : simul ] the following three performance measures : @xmath331^d } \bigl\ { { \operatorname{e } } [ { \widehat}{c}_n ( { \boldsymbol{u } } ) - c ( { \boldsymbol{u } } ) ] \bigr\}^2 \ , { \mathrm{d}}{\boldsymbol{u } } ; \\ \text{integrated variance : } \quad & \int_{[0 , 1]^d } { \operatorname{e}}\bigl [ \bigl\ { { \widehat}{c}_n ( { \boldsymbol{u } } ) - { \operatorname{e } } [ { \widehat}{c}_n ( { \boldsymbol{u } } ) ] \bigr\}^2 \bigr ] \ , { \mathrm{d}}{\boldsymbol{u } } ; \\ \text{integrated mean squared error : } \quad & \int_{[0 , 1]^d } { \operatorname{e}}\bigl [ \bigl\ { { \widehat}{c}_n ( { \boldsymbol{u } } ) - c ( { \boldsymbol{u } } ) \bigr\}^2 \bigr ] \ , { \mathrm{d}}{\boldsymbol{u}}.\end{aligned}\ ] ] to compute these , we apply the following trick . let @xmath332 and @xmath333 be independent replications of the estimator , and let the random vector @xmath334 be independent of the two estimators and uniformly distributed on @xmath30^d$ ] . then each of the three performance measures above can be written as a single expectation with respect to @xmath335 : @xmath336 ; \\ \text{integrated variance : } \quad & { \operatorname{e}}\left [ \frac{1}{2 } \bigl\ { { \widehat}{c}_n^{(1 ) } ( { \boldsymbol{v } } ) - { \widehat}{c}_n^{(2 ) } ( { \boldsymbol{v } } ) \bigr\}^2 \right ] ; \\ \text{integrated mean squared error : } \quad & { \operatorname{e}}\left [ \frac{1}{2 } \sum_{j=1}^2 \bigl\ { { \widehat}{c}_n^{(j ) } ( { \boldsymbol{v } } ) - c ( { \boldsymbol{v } } ) \bigr\}^2 \right].\end{aligned}\ ] ] to see these identities , first condition on @xmath334 and then use the fact that @xmath332 and @xmath333 are independent of @xmath334 and are independent random copies of @xmath330 . to compute the above three expectations , we rely on monte carlo simulation . for a given large integer @xmath337 , we simulate the independent random vectors @xmath338 where @xmath339 . the random vectors @xmath340 are sampled from @xmath4 and the random vectors @xmath334 are sampled from the uniform distribution on @xmath30^d$ ] . then we compute the copula estimator under consideration based on the samples @xmath341 and @xmath342 and evaluated at @xmath343 , yielding @xmath344 and @xmath345 , respectively . we also compute the true copula , @xmath4 , at @xmath343 . then we compute the desired function of @xmath344 , @xmath345 and @xmath346 as in the three expectations above . finally , we average over the @xmath337 samples to obtain a monte carlo estimate of the desired performance measure . 99 baker , r. ( 2008 ) . an order - statistics - based method for constructing multivariate distributions with fixed marginals , _ journal of multivariate analysis , * 99 * , 23122327 . bcher , a. and dette , h. ( 2010 ) . a note on bootstrap approximations for the empirical copula process , _ statistics and probability letters _ , * 80 * , 19251932 . bcher , a. and volgushev s. ( 2013 ) . empirical and sequential empirical copula processes under serial dependence , _ journal of multivariate analysis _ , * 119 * , 6170 . carley , h. and taylor , m. d. ( 2002 ) . a new proof of sklar s theorem , in : cuadras , c. m. , fortiana , j. and rodrguez - lallena , j. a. ( eds . ) ( 2002 ) . _ distributions with given marginals and statistical modelling _ , p.p . 2934 , kluwer academic publishers , dordrecht . deheuvels , p. ( 1979 ) . la fonction de dpendence empirique et ses proprits , un test non paramtrique dindpendance , _ bulletin de la classe des sciences , acadmie royale de belgique , 5e srie _ , * 65 * , 274292 . devore , r. a. and lorentz , g. g. ( 1993 ) . _ constructive approximations _ , springer , berlin - heidelberg . fermanian , j .- d . , radulovi , d. and wegkamp , m. j. ( 2004 ) . weak convergence of empirical copula processes . _ bernoulli _ , * 10 * , 847860 . li , x. , mikusiski , p. , sherwood , h. and taylor , m. d. ( 1997 ) . on approximation of copulas , in : bene , v. and tpn , j. ( eds . ) ( 1997 ) . _ distributions with given marginals and moment problems _ , p.p . 107116 , kluwer academic publishers , dordrecht . marius hofert , ivan kojadinovic , martin maechler and jun yan ( 2015 ) . _ copula : multivariate dependence with copulas . _ r package version 0.999 - 14 . url http://cran.r-project.org/package=copula . janssen , p. , swanepoel , j. and veraverbeke , n. ( 2012 ) . large sample behavior of the bernstein copula estimator , _ journal of statistical planning and inference _ , * 142 * , 11891197 . joe , h. ( 1997 ) . _ multivariate models and dependence concepts _ , chapman and hall , london . lorentz , g. g. ( 1986 ) , _ bernstein polynomials _ , 2nd ed . , chelsea publishing company , new york . mai , j .- f . and scherer , m. ( 2012 ) . _ simulating copulas : stochastic models , sampling algorithms , and applications _ , imperial college press , london . nelsen , r. b. ( 2006 ) . _ an introduction to copulas _ , 2nd ed . , springer - verlag , new york . r core team ( 2016 ) . _ r : a language and environment for statistical computing . _ r foundation for statistical computing , vienna , austria . url https://www.r - project.org/. sancetta , a. and satchell , s. ( 2004 ) . the bernstein copula and its applications to modeling and approximations of multivariate distributions , _ econometric theory _ , * 20 * , 535562 . segers , j. ( 2012 ) . asymptotics of empirical copula processes under nonrestrictive smoothness assumptions , _ bernoulli _ , * 18 * , 764782 . sklar , m. ( 1959 ) . fonctions de rpartition n dimensions et leurs marges , _ publ . statist . paris _ , * 8 * , 229231 . tsukahara , h. ( 2005 ) . semiparametric estimation in copula models , _ canadian journal of statistics _ , * 33 * , 357375 [ erratum : _ canadian journal of statistics _ , * 39 * , 734735 ( 2011 ) ] . van der vaart , a. w. and wellner , j. a. ( 1996 ) . _ weak convergence and empirical processes : with applications to statistics _ , springer - verlag , new york .

in this paper we describe a new uv - initiative _ hst _ project ( go-13297 ) that will complement the existing f606w and f814w database of the acs globular cluster ( gc ) treasury by imaging most of its clusters through uv / blue wfc3/uvis filters f275w , f336w and f438w . this `` magic trio '' of filters has shown an uncanny ability to disentangle and characterize multiple - population ( mp ) patterns in gcs in a way that is exquisitely sensitive to c , n , and o abundance variations . combination of these passbands with those in the optical also gives the best leverage for measuring helium enrichment . the dozen clusters that had previously been observed in these bands exhibit a bewildering variety of mp patterns , and the new survey will map the full variance of the phenomenon . the ubiquity of multiple stellar generations in gcs has made the formation of these cornerstone objects more intriguing than ever ; gc formation and the origin of their mps have now become one and the same problem . in the present paper we will describe the data base and our data reduction strategy , as well as the uses we intend to make of the final photometry , astrometry , and proper motions . we will also present preliminary color - magnitude diagrams from the data so far collected . these diagrams also draw on data from go-12605 and go-12311 , which served as a pilot project for the present go-13297 . # 1_#1

1410.4564

it was once thought that we understood how globular clusters ( gcs ) formed in the early universe , but recent discoveries of their multiple populations ( mps ) have unraveled our long - held view of gcs as `` simple stellar populations '' formed in a single burst . it now appears that virtually _ all _ of them host multiple stellar generations , with secondary populations characterized by puzzling chemical compositions . after almost a decade since the first breakthrough ( bedin et al.2004 ) , our current understanding can be summarized as follows : * ubiquity . almost all gcs that have been adequately studied are composed of distinct stellar populations ( piotto et al . 2012 ) , from two in number up to six or more ( @xmath0 cen , bellini et al . even clusters not yet sufficiently probed photometrically exhibit the telling na - o anti - correlation in the spectra of their giant stars ( carretta et al . an exception is rup106 . villanova et al . ( 2013 ) have analyzed spectra of nine rgb stars , and concluded that this gc hosts a single stellar population . in addition , walker et al . ( 2011 ) claimed that ic4499 does not show evidence of multiple populations . in both cases , higher precision stellar photometry is needed for a definitive conclusion on the nature of the stellar populations in these two gcs . * enrichment . while the first - generation ( 1 g ) stars have a composition that can be ascribed to the proto - galactic interstellar matter out of which they formed , the second - generation ( 2 g ) stars can be depleted in c and o , enhanced in n and na ( e.g. , carretta et al . 2009 , marino et al . 2008 , 2011a 2011b , 2012 2014 ) , and enhanced in he ( milone et al . 2012a , 2012b , 2014a ) . in some cases the he enhancement can be quite strong , close to y=0.40 ( e.g. , dantona and caloi 2004 , norris 2004 , piotto et al . 2005 , 2007 , king et al . 2012 , bellini et al . 2013a ) , while in others it is still measurable but small ( e.g. , @xmath10.01 as in the cases of ngc 6397 , milone et al . 2012b , and ngc 288 , piotto et al . 2013 ) . this chemical pattern indicates that the material out of which gcs formed was exposed to proton - capture processes at high temperatures and must have come from relatively massive 1 g stars . with a few notable exceptions , i.e. , omega centauri ( norris & da costa 1995 ) , m22 ( marino et al . 2009 , 2011c ) , m54 ( carretta et al . 2010 ) , ngc 1851 ( yong and grundahl 2008 ) , m2 ( yong et al . 2014 ) , ngc5824 ( da costa et al . 2014 ) , and terzan 5 ( ferraro et al . 2009a ) , 2 g stars have the same abundances of iron and other heavy - elements as the 1 g stars , indicating that their material was somehow not tainted by 1 g supernova products . even in the case of clusters exhibiting multiple iron abundances , the differences are relatively small , indicating that only a tiny fraction of the supernova products from 1 g were incorporated in 2 g stars ( renzini 2013 ) . it is not clear at present whether the clusters with iron abundance variations are part of a different class of gcs . because of the large number of gcs sampled , go-13297 will surely help to address this idea . * variety . thus far , we have found no two clusters that manifest the mp phenomenon in the same way . the relative proportion of 1 g to 2 g stars differs enormously from one gc to another , and in several cases the 2 g stars outnumber 1 g stars and are more centrally concentrated ( as in @xmath0 centauri , bellini et al . 2010 , and 47 tuc , milone et al . 2012a , see also lardo et al . 2011 , but based on lower quality sdss data . ) , becoming quite dominant at the very center . in other cases , large differences are also seen in the degree of he enhancement and in p - capture elements . * discreteness . one crucial property of multiple populations is that in a large number of gcs the populations can be separated into quite distinct sequences within each cmd and/or in appropriate two - color plots , as opposed to a continuous spread , observational results that any formation scenario must be able to account for . different progenitors of the 2 g have been proposed . intermediate - mass asymptotic giant branch ( agb - m ) stars ( dantona et al . 2002 ) , fast - rotating massive stars ( frms ) ( decressin et al . 2007 ) , or interacting massive binary stars ( de mink et al . 2009 , vanbeveren et al . 2012 ) may eject materials with composition similar to that of 2 g stars , i.e. , somewhat helium enriched and exposed to proton capture reactions at high temperatures , but with sizable differences from one case to another . + the agb - m and the frms models are those for which all the different aspects of the study of multiple populations have been more extensively investigated . although these models are based on different sources of processed gas , in both models the amount of gas available for 2 g formation is only a small fraction of the initial 1 g . this has the important implication that both these models require the 1 g star cluster to be initially more massive than it is now ( see e.g. decressin et al . 2007 , renzini 2008 , 2013 , dercole et al.2008 , ventura et al . 2014 for estimates of the possible initial 1 g cluster masses ) , and must have undergone a significant loss of 1 g stars ( see , e.g. , dercole et al . 2008 , decressin et al . 2010 ) leading to the large fraction of 2 g stars currently observed in many gcs ( renzini 2008 ) . this large loss of 1 g stars also implies that gcs must have significantly contributed to the formation of the galactic halo ( see e.g. vesperini et al . 2010 , schaerer & charbonnel 2011 ) . another requirement shared by these models is that , not only ejected material from the 1 g stars is necessary . it may have been somehow diluted with gas with the same composition as the 1 g in order to reproduce the na - o anticorrelation ( as advocated by decressin et al.2007 , dercole et al . 2008 , 2011 ) , though the modalities of such dilution remain obscure . the origin of this pristine gas is still a matter of investigation and debate . curiously , recent surveys of young ( 10 myr - 1 gyr ) extragalactic massive clusters ( bastian et al.2013a ) do not show any evidence of presence of gas and/or ongoing star formation , though the facts are not quite clear on this ( vink et al.2009 ) . recently bastian et al . ( 2013b ) , following an earlier suggestion by dantona et al . ( 1983 ) , proposed a model which would not require a larger 1 g cluster mass and is based on the accretion of enriched material released from interacting massive binaries and rapidly rotating stars onto their circumstellar disks , and ultimately onto the young pre - main sequence stars . this model requires some very ad hoc assumptions ( see cassisi & salaris 2014 , salaris & cassisi 2014 , dantona et al . 2014 ) both regarding the lifetimes of the pre - main sequence accretion disk and regarding the he content of the accreting material . most importantly , the accretion scenario has serious difficulties to account for the discreteness of the sequences we observe in the cmds . in summary , the series of events that led from massive gas clouds in the early universe to the gcs we see today remains an intriguing puzzle . clearly , gc formation was a far more complex process than ever imagined before . clues and insight needed to solve this puzzle can come only from homogeneous mapping of a large sample of gcs with appropriate multi - band photometry . a photometric survey can map a large sample of stars , at all evolutionary phases . such a survey will enable us to identify 2 g stars even in the cases where they represent a minor fraction of the entire cluster population . multi - band photometry allows us to infer chemical properties of stars in a complementary way to spectroscopy ( milone et al . 2012a ) , and to do this for a large sample of gcs . this is the subject of the go-13297 hst large treasury project ( pi g. piotto ) approved for cycle 21 . this paper will describe in detail the project , the selected sample of clusters ( see table 1 ) , the observational strategy , the data - reduction strategy , and the intended use of the data by the team . we will also present preliminary results from the data acquired thus far . previous studies , based on spectroscopy and ground - based photometry have shown that stremgren and ultraviolet photometry does a good job in separating multiple populations along the rgb on nearbly gcs ( e.g. yong et al.2008 ; marino et al.2008 ) . our wfc3/uvis observations in cycles 1820 have opened a spectacular new window by showing that uv / optical colors separate the various sub - populations much better than the traditional optical colors alone ( bellini et al.2010 , 2013a , milone et al . 2010 , 2012a , 2012b , piotto et al . 2012 , 2013 ) and do so across the _ entire _ color - magnitude diagram , from the main sequence ( ms ) all the way to the horizontal branch ( hb ) . figure [ magicuvf ] shows the power of the uv photometry for the case of ngc 6752 . it reproduces the simulated spectra of rgba ( n - poor , he - poor red line , first stellar generation ) and rgbc ( n - rich , he - rich blue line , third stellar generation ) of ngc 6752 ( milone et al . 2010 ) . the reason why f275w , f336w , and f438w work so well is quite simple . milone et al . ( 2012a ) emphasized that the f275w passband includes an oh molecular band , f336w an nh band , and f438w ( or f435w for acs ) cn and ch bands , as illustrated in fig . 1 . this property of the hst filter system is at the basis of the project we are presenting in this paper . the 1 g stars , which are oxygen- and carbon - rich and nitrogen - poor , are relatively faint in f275w and f438w , but bright in f336w . conversely , 2 g stars , whose material has been cno - cycle processed , are oxygen- and carbon - poor but nitrogen - rich . as a consequence , they are relatively bright in f275w and f438w but faint in f336w . therefore , 1 g stars are bluer than 2 g stars in one color ( f336w @xmath2 f438w ) , but redder in another ( f275w @xmath2 f336w ) , and this _ inversion _ is seen in fig . [ magicuvf ] milone et al . ( 2013 ) defined a pseudo - color @xmath3 = ( @xmath4)@xmath2(@xmath5 ) , which maximizes the virtue of both f336w @xmath2 f438w and f275w @xmath2 f336w , and has proven to be quite efficient in separation of multiple sequences . figure [ n6352 ] shows an example of the application of go-13297 data to ngc 6352 . left panels show the power of the pseudo - color @xmath3 in separating the two mss . the middle panels show that in the @xmath6 vs. @xmath7 color - magnitude diagrams ( cmds ) , 1 g stars are redder than 2 g ones , but they become bluer in the @xmath6 vs. @xmath8 cmd ( right panels ) . thanks to f275w , f336w , and f438w data , the two stellar generations are clearly distinguished and are seen to wind , in a distinct but intertwined fashion , from the ms through the subgiant branch ( sgb ) and red - giant branch ( rgb ) , to the hb . the uv passbands are thus able to probe the cno content of the cluster stars . note that the nuclear processes that produce the light - element ( anti-)correlations ( e.g. the na - o or the mg - al anticorrelations ) also produce helium , and measuring the he enhancement of each sub - population is particularly important to help pinpoint what kind of stars may have produced the 2 g material . the photometric impact of he is mainly on the optical bands ( through the stellar temperature ) . after population - tagging stars according to their uv colors , we can make fine distinctions ( @xmath10.01 mag in color ) in the ms ridge lines of the generations in the optical bands , and thus infer variations of only a few per cent in he ( milone et al . 2010 , 2012ab ) . in summary , the combination of uv and optical colors allows us to estimate the content of c , n , o , and he , even for stars that are far too faint to pursue with spectroscopy . ( see milone et al . 2012ab , 2013 for applications of the method . ) the combination of uv / blue filters that we have selected for go-13297 is also ideally suited for studies of hb stars , blue stragglers stars ( bsss ) , and compact binaries . for hot hb stars , optical colors quickly become degenerate , but uv colors are able to follow the entire hb extension . this is particularly interesting for those gcs ( even some metal - rich ones , rich et al . 1997 , busso et al . 2007 , bellini et al . 2013a ) that have very extended hbs and often exhibit the most complex mps . enhanced helium also distorts the hb morphology : when @xmath9 is increased from 0.23 to 0.38 the blue hb becomes 1 mag brighter in f275w , while the ehb becomes 0.5 mag fainter . thus the uv band is crucial for determining the physical parameters of hb stars ( including @xmath10 ) and their helium content ( rood & crocker 1989 , dantona and caloi 2008 ) . bsss , compact binaries , and high - energy sources are also best studied with filters of shorter wavelength . detailed studies , made possible largely by _ hst _ , have revealed remarkable variety and complexity . as mentioned above , among the dozen or so gcs in which multiple photometric sequences have been resolved thus far , the fraction of 2 g stars ranges from a few per cent to well over 50% . each gc has its own individuality , thus pointing to a great deal of variance in the formation process itself . yet the sheer range of variance remains to be established , and can not be properly mapped without the widest possible survey of galactic gcs . furthermore , since spectroscopy has shown that nearly all gcs host mps , it is all the more appropriate to move from studying a few `` odd ducks '' to a systematic survey of the whole flock . the largest homogeneous set of _ hst _ data on gcs until now is the acs treasury database of f606w and f814w images for 65 gcs ( go-10775 , pi sarajedini , anderson et al . 2008 ) ; however , its optical - wavelength baseline offers sub - optimal leverage for mp studies . the new go-13297 data set represents what is needed : to augment the existing database with uv and blue observations through the `` magic trio '' of wfc3/uvis filters , f275w , f336w , and f438w , whose power in population separation we have just illustrated . our photometry will characterize the populations in each cluster : how many stellar generations are present , and in what relative proportions even for small minority populations that might easily escape spectroscopic surveys . the uv / blue wfc3/uvis photometry , combined with visible photometry from the acs gc treasury , will also enable a systematic survey and detailed intercomparisons of the populations of ehb stars , bsss , and compact binaries in a large sample of galactic gcs , as outlined in section [ legacy ] . figures 1 and 2 show that multiple populations of many clusters can be readily disentangled by means of a two - color diagram made from f275w , f336w , and f438w . in order to effect a clean separation of the populations with the cmds and the pseudo - color diagrams , our observational goal was a color precision of about 0.02 magnitude or better , which implies a @xmath11 of 50 or better in f275w ( the most photon - starved filter ) . we did not try to follow the multiple sequences all the way down the main sequence for every cluster that would have been prohibitively expensive orbit - wise . rather , our goal was simply to maintain this precision in f275w to just below the turn - off . that will naturally give us a population separation for the evolved populations as well . given that we need a minimum of 4 exposures in each band to make a robust catalog , two orbits represent the minimum set of observations for each cluster . fifteen clusters are observed with that allocation ( see table 1 for a detailed description of the observing time allocated for each cluster and for the different filters ) . four clusters need 3 orbits each , eight clusters need 4 orbits , three need 5 orbits , and two clusters need 6 orbits each . ( one of the 6-orbit clusters is ngc 6715 [ m54 ] , which is too far away to get 0.02-magnitude precision below the turnoff in f275w . but it is likely to have a larger ms spread , and therefore not need a @xmath11 as high as 50 . furthermore , understanding its upper sequences with a mp - sensitive filter system will be important . ) the remaining 14 clusters are so far away or too reddened that we can not hope to resolve ms splits with f275w ; for these clusters we allot two orbits , which are adequate for the sgb , rgb , and hb populations . we observe each gc in two visits , with orientations separated by roughly 90 degrees , to allow a direct validation of the charge - transfer efficiency ( cte ) corrections ( see below ) . for most gcs , we take two exposures per filter per visit . these exposures are dithered to cover the gap , so that most stars will be found in four exposures in each band , thus allowing a robust empirical estimate of the photometric errors . in summary ( see table 1 ) , the go-13297 gc sample includes 47 gcs and a very old ( @xmath12 gyr , king et al . 2005 ) , open cluster , ngc 6791 , which has been included in the sample because of its many peculiarities ( bedin et al . 2008 , geisler et al . the program has been allocated 131 orbits . originally , the program also included ngc 4147 . unfortunately , the cluster could not be scheduled ( for guiding star problems ) and has been substituted with m4 . before go-13297 we had a pilot project , go-12605 ( pi piotto , 22 orbits ) , where we began our collection of wfc3/uvis f275w , f336w , and f438w images for seven gcs ( m3 , m13 , m15 , m80 , ngc 288 , ngc 362 , and ngc 2808 ) , extending previous f606w and f814w data from go-10775 . in an even earlier project , go-12311 ( pi piotto , 17 orbits ) , we had collected wfc3/uvis f275w images in the go-10775 fields of five more clusters ( 47tuc , m4 , m22 , ngc 1851 , and ngc 6752 ) . the go-12311 data set is now complemented by f336w and f438w wfc3/uvis images for ngc 1851 , m4 , and m22 . some of the gcs from the previous go-10775 treasury program were not included in the go-13297 target list either because they are too poorly populated for a meaningful search for multiple sequences , or because they are highly reddened or extremely distant , which would have made them impossible to observe in the uv in a reasonable number of orbits . in summary , completion of the go-13297 program gives us , across all programs , 57 gcs with a homogeneous set of images in the five photometric bands f275w , f336w , f438w , f606w , and f814w . table 1 lists these 57 clusters and summarizes their wfc3/uvis and parallel acs images . the final five columns in the table give the galactic @xmath13 coordinates , the reddening , and the distance modulus ( from the harris 1996 catalog , 2010 edition ) . gcs observed within go-12311 are marked by an asterisk ; gcs from go-12605 by two asterisks . figure [ gcdist ] shows the distribution of the clusters in the galactic @xmath14 and @xmath15 planes . wfc3/uvis has been on board _ hst _ for only a few years , yet cte losses have become a concern for it much more quickly than they were for acs . the uvis team has traced this to the fact that cte losses are particularly bad when the background is low ( see mackenty & smith 2012 ) . in order to minimize cte losses , we determined the anticipated background for each exposure and supplemented it with postflash to ensure a minimum of 12 e@xmath16 . in addition , to the extent possible , we observed each cluster at two different orientations 90 degrees apart so that any remaining cte issues can be dealt with properly . we will postflash our images to ensure a background of a least 12 e@xmath16 . the main focus of our program is uv imaging of the central field of each cluster , and our observing strategy was optimized for that field . even so , we could still afford to take @xmath14 medium - length exposures of a parallel field with acs ( more for the clusters with more than two orbits ) . our general strategy has been to take one orbit at one orientation and another at a different orientation . this resulted in placing the acs / wfc camera on two different fields 6 from the cluster center . for large he differences among the different populations , even these few parallel observations will allow us to detect split mss , as was successfully done in the case of ngc 2808 ( piotto et al . the mapping of the outskirts of the clusters , albeit at lower photometric accuracy , adds a large baseline to the spatial aspects of our mp study , and will allow us to measure the 1g/2 g radial gradient . the relative concentration of the 2 g stars provides critical information about the formation process ( see bellini et al . 2009 and dercole et al . 2008 ) . when possible , we have chosen an orientation that allows the parallel fields to overlap previous observations , so that we can also measure proper motions for the outer field . at a minimum , we can use the wfpc2 parallels taken during the go-10755 program , but many clusters have even better archival data . note that the overlaps with previosly collected go-10775 images provide us with a @xmath17-year temporal baseline for relative proper - motion measurement . the outer - field observations should be deep enough to allow us also to measure absolute proper motions for the clusters , relative to background galaxies . the ultimate goal of this program is to provide comprehensive catalogs to the community in order to enable the many studies that this 57-cluster data set makes possible . it will not be possible to produce the best possible catalogs immediately . once we have the entire data set in hand , we will be able to do an exquisite characterization of the psfs and distortion solutions for each filter , as well as empirical corrections for any remaining cte issues . therefore the final catalogs will not be available until about a year after the last data are taken . there are , however , many studies that will not require a definitive catalog , and we are making preliminary products available to the community as soon as possible . in addition , team members will be performing special processing on individual clusters to produce early science results . we describe in the following subsections the three different reductions we envision for this project : the early - release preliminary reduction , the specialized reductions for early science , and the final comprehensive reduction . we have created a web page ( http://groups.dfa.unipd.it/espg/treasury.html ) where interested persons can follow the progress of the project and retrieve the released photometric and astrometric data . a mirror site will also be available in the mast archive , at stsci . much of the progress in understanding the mp phenomenon will be possible only by combining photometric observations of larger numbers of stars with targeted spectroscopic studies of individual stars . in an effort to allow spectroscopists to identify stars and plan their runs as soon as possible , we will provide an early reduction of the data shortly after the publication of this paper . we will update the archive as more data come in . the preliminary reduction will focus on measuring the stars in the simplest way possible and collating them with the existing go-10775 catalogs in f606w and f814w . to this end , we run on each exposure a version of img2xym_wfc3uv ( similar to the one - pass star finding and measuring for acs described in acs / isr-2006 - 01 by anderson & king , but adapted to the wfc3/uvis detector ) . this is a one - pass finding routine that goes through the image pixel by pixel , identifying as a potential star any pixel that has more than 100 counts within its inner 3x3 pixels and has no brighter pixels within a radius of 4 pixels . on each identified source we then use an empirical library psf to perform psf - based 5@xmath175-pixel aperture photometry and determine an accurate position . it then uses the wcs information to transform the measured positions into the go-10775 catalog frame and matches up each star list to the stars in the catalog . once this has been done for all the available exposures , we collate the observations of each star in each filter and produce a catalog that has the following elements : first , we provide a 2013 - 2014 position that can be compared against the 2006 position from go-10775 . since we have not optimized the psfs and astrometric transformations , this position has only 0.1-pixel precision , but that should be sufficient for cluster - field separation . we next provide the f275w , f336w , and f438w photometry for all the stars in the go-10775 catalog that could be found by our one - pass routine . we provide the vega - mag - calibrated average magnitude , the rms deviation of the individual observations about this average , and the number of images that contributed to the average . finally , we also produce stacked images in each of the three new bands , in the frame of the original f606w products . the early - release catalog described above has been constructed with a library psf and a one - pass finding approach that treats each star as if it is isolated . the final reduction will improve on several of these limitations , but since it will be more than a year before the final reductions are available , in the meantime several members of the team have been running supplemental reductions in order to start the detailed scientific analysis of several clusters . these reductions will be described in more detail in the individual papers ( for example , see milone et al . for m2 , submitted to apj ) , but many of them will make use of photometry software ( such as that described in bellini et al . 2010 ) that allows for the fact that the psfs in individual exposures often differ slightly from the library psf . the specialized reductions will also include differential - reddening corrections . both of these improvements are necessary to enable the best possible sequence separation for study of the multiple populations . other early - science projects will focus on the internal proper motions . the one - size - fits - all preliminary products will not do the careful transformations and multi - epoch fitting that is necessary to measure accurate motions and their errors . for the clusters of particular interest , we will reduce all the available epochs and explore the internal motions in papers before the final catalogs are available . these data are available for specific projects under request to the first author of this paper , and will be stored on the dedicated web page at the department of physics and astronomy , of the university of padova ( http://groups.dfa.unipd.it/espg/treasury.html ) . the preliminary photometry coming from this process is at the base of the cmds we present in this paper . the final reduction for this program will be possible only after all of the data are collected . the full data set will enable us to construct an exquisite model of both the average psf and the average distortion solution in each of the three filters . we will also construct a tailored psf for each exposure to reflect the fact that breathing causes it to vary slightly from the average . whereas for the preliminary reduction ( described in section [ dirty ] , [ presentred ] ) we reduced each exposure separately with a simple one - pass approach and combined the starlists with the existing go-10775 catalog , for the final reduction we will construct an entirely new catalog using an approach that analyzes all exposures simultaneously in a multi - pass , artifact - avoiding , neighbor - subtracting algorithm similar to that used to reduce the original go-10775 data set ( anderson et al . this new approach will involve all five filters and will produce proper motions , source - subtracted stacks ( which anyone can search for remaining objects of any sort ) , and a variety of artificial - star tests . in addition to reducing the go-10775 and go-13297 data in a uniform way , we will also reduce all of the existing wfpc2 , acs , and wfc3 data for the clusters and cross - identify the stars with our catalog , producing time - series photometry and astrometry that can be used for high - precision proper motions and variable - star studies . from these many individual observations of each star , we will measure a variability index and a high - precision proper motion , along with an error estimate of thr latter . along with each photometric observation , we will also provide a differential - reddening correction , so that accurate cmds can be constructed for multi - population analysis . the data products described in section [ magicred ] ( the new f275w , f336w , and f438w photometry for the go-13297 star catalog and new stacked images ) are being made available via the http://groups.dfa.unipd.it/espg/treasury.html web page and the stsci mast archive ( via the high - level product link in the archive listing for the proposal ) . we will continually update the files as new observations come in . once all the data are in hand , we will perform the specialized uniform reductions described in section [ magicred ] . in addition to outputting those reductions we will also produce for each cluster several other high - level products : * a unified star list for the central field for all filters . we will have at a minimum f275w , f336w , f438w , f606w , and f814w for all clusters . we will also link up other relevant photometry from other archival data , such as sbc and fuv wfpc2 observations , which are particularly relevant to hot populations . for each filter , we will provide an average flux and its error , based on internal agreement . we will also provide the flux measured for each star in each exposure , so that users can search for variability . * a differential - reddening map for each cluster , and a correction for each measurement of each star . * proper motions for most of the stars in the catalogs . these motions will be focused on the uv data set , so they will extend from the brightest giants to a few magnitudes below the turnoff , so that they will be particularly useful for comparison with radial velocities and spectra from the ground . * artificial - star tests done for stars inserted along the fiducial sequences . this will allow users to quantify the extent to which crowding may broaden the sequences . * stacked images of the field in all filters , co - registered and with 2mass - calibrated wcs headers . this includes data sets from the archive and any other data we can link - up ( including chandra images ) . we will also provide subtracted stacks , so that the community users can determine whether there might be any stars that escaped our finding algorithm . * a specialized website portal to serve up the data in a way that is most useful to globular - cluster researchers . it will allow users to zoom in and out of images and select stars for dynamic display on cmds and pm diagrams . stars will be flagged by evolutionary stage and 1g/2 g designation . we will also indicate which stars have spectra available in the archive that we will maintain . we will , of course , also deliver our catalogs and images to mast for a more standard dissemination . while there is much science we are eager to do on this data set , the primary focus of the program is the generation of a legacy data set for the entire community . that is why we are providing the preliminary reductions as quickly as possible , and are equally intent on getting the final high - level products to the community within a year of the last observation . at the time of the submission of the present paper , about 95% of the go-13297 observations have been collected . though it will take some time before we can perform the final reduction as described in section [ magicred ] , we are constantly monitoring the data acquisition , and all data are reduced following the procedures described in section [ presentred ] . a more detailed analysis of the multiple stellar population for some specific , particularly interesting clusters is ongoing , and corresponding papers will be submitted in the next few months . here , we want to make public the first cmds , as they come out from our preliminary photometry . the data reduction process is still ongoing , and the purpose of the cmds we are presenting here is just to show the potentiality of the go-13297 database . in figs . [ cmd1]-[cmd20 ] we present the f275w vs. @xmath18 and the f275w vs. f275w @xmath2 f814w diagrams . magnitudes are instrumental magnitudes . instrumental magnitudes are defined as the @xmath19 of the sum of the detected photo - electrons within a circular aperture of 10 pixels , in the reference exposures of each filter , chosen as the one with the longest exposure time . a careful calibration will need appropriate aperture correction tailored to the adopted psfs , which differs frame by frame . we will eventually do so , but , at the moment , in order to avoid future confusion , we prefer to show the cmds using the instrumental magnitudes as above defined . the figures show the diagrams for the go-13297 clusters observed ( even partially , i.e. , not all orbits collected ) by august , 2014 . ( the figures also include the cmds from the two pilot projects go-12605 and go12311 . more specifically , f275w , f336w , f438w data for ngc 288 , ngc 362 , ngc 5272 , ngc 6093 , ngc 6205 , and ngc 7078 come from go-12605 . f275w photometry for ngc 104 , ngc 1851 , ngc 6121 , ngc 6656 , and ngc 6752 is from go-12311 . additional f275w and f336w data for ngc 104 and ngc 6752 come from hst archive ( see milone et al . 2012a , b for details ) . the f814w photometry comes from anderson et al . ( 2008 ) and is based on go-10775 data . photometry for ngc 6791 comes from go10265 archive data . for completeness , we also added the cmd of @xmath0 cen from bellini et al . ( 2010 ) . the cmds are based on a preliminary selection of the best measured stars . there are margins for an improved photometry , and a more complete star counts . a detailed analysis of the individual cmds is beyond the purposes of this paper and will be deferred to future papers . still it seems appropriate to emphasize that all gcs in this sample show multiple sequences or at least broadened sequences . this confirms the idea that the presence of multiple stellar populations in gcs is a widespread phenomenon , with single - stellar - population gcs being an exception ( if , indeed , single - stellar - population gcs exist at all ) . the central theme of the go-13297 project is to explore as fully as possible the mp phenomenon in gcs , which is clearly the key to understanding star formation and feedback in the cluster environment . however , our database will become a gold mine for a large number of other projects and will be a general resource for decades to come . in the following , we briefly describe the research projects our group intends to carry out using the go-13297 astrometric and photometric database . _ the identification of multiple stellar populations . _ this is of course our main goal . we will search for mps in all evolutionary branches , in all clusters , following the procedure described in milone et al . ( 2013 ) for ngc 6752 , and in milone et al . ( 2012a , b ) for 47tuc and ngc 6397 . the different populations will be chosen via different combinations of the photometry from the five bandpasses . we will do this by selecting individual stars belonging to different photometric sequences and calculating all their colors from combining the various passbands . in the final catalog each star will be assigned to a particular sub - population , in order to enable spectroscopic follow - up aimed at a precise chemical tagging of each sub - population . the catalogs and finding charts resulting from this effort will allow the entire community to contribute to this critical aspect of mp studies . _ modeling multiple stellar populations . _ the peculiarities characterizing the chemical properties of the various sub - populations hosted by galactic gcs require the computation of extended sets of models for low - mass stars properly accounting for these chemical anomalies . this is a crucial step in order to investigate and trace the distribution of the distinct sub - populations in the various evolutionary sequences in the cmd from the faintest portion of the ms ( milone et al . 2012a , b ) to the hb ( dantona & caloi 2008 , dalessandro et al . 2011 , and references therein ) . the photometric appearance of multiple populations strongly depends on the adopted photometric bands ( sbordone et al . 2011 , cassisi et al . 2013 ) . in particular , since light - element anti - correlation strongly affects the stellar spectra at wavelengths shorter than 400 nm , when using uv and far - uv passbands it is crucial to use appropriate color-@xmath10 relations properly accounting for the various chemical anomalies . basically , we will follow the method described in milone et al . ( 2013 and references therein ) . _ measurement of he content from multiband photometry . _ multi - wavelength photometry provides unique information on the chemical composition of large samples of stars belonging to the different stellar populations hosted in gcs . as described in section [ magicuvs ] , the combination of filters centered at different wavelengths allows us to discriminate between various effects , like differences * in * he , c , n , o , na content . exploiting the sensitivity of various colors to different properties of stellar populations , and combining the photometric information with grids of synthetic spectra constructed with chemical composition constrained from spectroscopic observations will allow us to better isolate the specific role of the helium abundance . this technique has been successfully applied to a few gcs , and differences in helium among different stellar populations have been measured in 47 tuc , ngc 6397 , ngc 288 and ngc 6752 ( milone et al . 2013 , and references therein ) . we plan to extend these measurements to the whole sample of clusters observed for this survey . the knowledge of the he variations within a gc will be used to the extend investigation on how this observable impact on the second parameter on hb morphology ( see milone et al . 2014b ) . _ cluster structural parameters from star counts . _ the appropriate combination of the uv - survey data and complementary wide - field observations will allow us to construct a new generation of gc radial star density profiles based on resolved star counts ( see ferraro et al . 2003 , miocchi et al . this is indeed the most robust and reliable way to derive cluster structural parameters , since surface brightness can be biased by the presence of sparse , bright stars . this data base will allow a complete re - characterization of the structural and dynamical properties of gcs . _ internal proper motion analysis . _ bellini et al . ( 2014 ) recently compiled the first large sample of gc proper - motion ( pm ) catalogs , using heterogeneous data from the _ hst _ archive ( see also bellini et al . this has shown that it is possible to reach the level of precision and accuracy required to study in detail the internal pm kinematics of gcs . among other things , this will allow us to measure separately the kinematical properties of each of the different stellar populations photometrically identified ( as in e.g. , anderson & van der marel 2010 ; richer et al . each gc observed as part of our go-13297 program was also observed in 2006 by the go-10775 treasury program . moreover , a good fraction of these gcs was also observed by other _ hst _ programs ( in particular using the state - of - the - art detectors that have been available on _ hst _ since 2002 ) . this will allow us to measure exquisite pms with time baselines between 7 and 14 years . this will extend the work of bellini et al . ( 2014 ) to a larger number of clusters , higher pm accuracy , and bluer stars . among other things , this will allow us to measure separately the kinematical properties of each the different stellar populations photometrically identified . any kinematical differences thus identified would strongly constrain formation mechanisms for the multiple population phenomenon . _ internal dynamical model and central intermediate - mass black holes . _ the new pm catalogs to be derived from our data will allow several new state - of - the - art studies of gc dynamics . for example , we will obtain direct measurements of the radial velocity dispersion profiles , the velocity dispersion anisotropy , and possible rotation of many gcs . dynamical modeling will constrain the mass profiles of the gcs , including the possible presence of intermediate - mass ( @xmath20 @xmath21 ) black holes ( van der marel & anderson 2010 ) . the existence of such black holes continues to be debated in the literature ( e.g. , noyola et al . 2010 ; lutzgendorf et al . 2013 ; lanzoni et al . 2013 ) , but this can be tightly constrained using pms measured in the central arcseconds of the cluster . we will also be able to measure the velocity dispersion as a function of main sequence mass , and this will allow us to establish the extent to which gcs are in energy equipartition ( trenti & van der marel 2013 ) . _ absolute proper motions based on an extragalactic reference frame . _ where possible , in our observing strategy we chose an orientation that allowed the parallel fields to overlap previous observations , so that we can also measure proper motions in the outer fields . at a minimum , we can use the wfpc2 parallels taken during the sarajedini program , but many clusters have even better archival data . the outer - field observations are generally deep enough to allow us to measure absolute proper motions for the clusters with respect to background galaxies . the central fields will be observed in uv / blue filters , so , although we will explore them , we do not expect to find many background galaxies bright enough to be used as absolute points of reference . absolute proper motions will be combined with radial velocities from the literature or from measurement of archival spectra , to derive the space velocities of the clusters ; the position - velocity phase space will then be examined for possible groupings of clusters with similar multiple stellar population properties . _ parallel fields and ground - based ancillary data for the radial distributions of multiple stellar populations . _ astro - photometric catalogs and atlases in each of the filters will be electronically provided , also for the acs parallel fields . when possible , proper - motion membership , and a sub - population flag will also be provided . ground - based wide - field multiband photometry is already available for a large number of the target clusters . merging ground - based with the new _ hst _ data will help connect stellar populations in the cluster centers with those in the external regions by using similar color combinations ( @xmath22 is analogous to @xmath3 ) , and therefore investigate their radial distribution over the whole extent of the cluster . moreover , identifying stars with available high - resolution spectra will allow us to connect photometric and spectroscopic properties of the different stellar populations . observations with ground - based facilities ( typically 8 - 10 m size telescope ) have already been acquired , and additional observations requested for a subsample of the go-13297 clusters _ modeling the radial distribution of multiple stellar populations . _ theoretical models and simulations predict that 2 g stars should initially be more concentrated than 1 g stars ( see , e.g. , dercole et al . 2008 , bekki 2011 , bastian et al . 2013b , and references therein ) , and that many clusters should still preserve , at least in part , some memory of the initial differences in the 1g/2 g spatial distribution ( vesperini et al . 2013 ; for observational studies that show radial gradients in the fraction of 2 g stars see , e.g. , sollima et al . 2007 , bellini et al . 2009 , lardo et al . 2011 , milone et al . we will carry out a systematic observational and theoretical study aimed at addressing a number of key questions concerning this issue . by studying the presence and extent of radial gradients in clusters with different dynamical properties and at different phases of their dynamical evolution we will be able a ) to further test the general validity of models predicting the initial spatial segregation of the different stellar populations , b ) to make progress in the observational identification of an evolutionary sequence in the spatial mixing of 1 g and 2 g stars , and c ) to shed light on the relationship between the global 2 g fraction and its local value measured at different distances from the cluster center . _ distances . _ knowing the distances to milky way gcs is fundamental for several astrophysical fields , including stellar evolution , cluster dynamics , and the cosmological distance ladder . to date , gc distances have generally been obtained mostly through isochrone - fitting techniques and the rr lyrae variables . by comparing high - precision _ hst _ pms ( at the level of mas / yr ) with existing line - of - sight velocity data ( in km / s ) , it becomes possible to obtain entirely independent kinematical distance estimates ( either from the assumption of velocity isotropy or from detailed dynamical models ) . such a study is currently in progress ( watkins et al . , in prep . ) using the pm catalogs of bellini et al . ( 2014 ) . with the new pm catalogs to be derived from the go-13297 data , we plan to extend this work to a larger number of clusters , and higher distance accuracies . these distances will be completely independent from , and in some ways complementary to gaia measurements . _ rgb bump . _ we plan to measure the rgb bump brightness and its extension in magnitude of the various sub - populations hosted by each cluster in our sample . comparison with models will allow us to estimate he - abundance differences among them . these differences can be compared with the he abundance measurements obtained with other methods ( see previous sections ) . at the same time , the relative brightness of the rgb bump with respect other cmd features , such as the hb luminosity and the msto , will provide useful tests for stellar evolution models ( e.g. , cassisi et al . 2011 ) . _ hb morphology . _ because optical colors become degenerate at high temperature , our use of uv bands will enable an accurate characterization of the full hb morphology , including the extreme horizontal branch ( ehb ) . the hb morphology will yield additional insight into any sub - populations enhanced in helium . at higher @xmath9 , the hb generally extends to hotter temperature , and also exhibits significant variations in luminosity , with higher luminosity on the blue hb and lower luminosity on the ehb . it will also be possible to establish to what extent the long - standing issue of the `` second parameter '' can be accounted for by helium - enriched second generations . some of us have already shown that the multi - modal hb of ngc 2808 ( dantona & caloi 2004 , dalessandro et al . 2011 ) , as well as of the metal - rich gcs ngc 6388 and ngc 6441 ( busso et al . 2007 , caloi & dantona 2007 , bellini et al . 2013a ) is consistent with multiple stellar populations with different helium abundance . this connection is supported by recent studies , based on spectroscopy of hb stars , which have shown that distinct stellar populations , with different content of light elements and helium , occupy different regions along the hb ( marino et al . 2011 , 2013 , 2014 ; villanova et al . 2012 ; gratton et al . 2011 , 2012 , 2013 ) . additional examples of the possible effects of helium enhancement on the hb morphology for many other clusters can be found in brown et al . ( 2010 ) and dalessandro et al . ( 2013a ) . with the new go-13297 we can extend the investigation of milone et al . ( 2014b ) on the nature of the gc global and local parameters and their effect on the hb morphology . _ modeling hbs . _ the hb morphology is very sensitive to the main parameters of the multiple populations , and in addition , to mass loss on the red giant branch . we need to build synthetic hbs , based on the computation of large grids of models , to check and quantify the role of all the parameters . new grids are being computed ( tailo et al.2014 , in preparation ) , including models populating the highest @xmath23 s of the hb . these structures are the outcome of late helium core flash " ignition ( dcruz et al . 1996 , brown et al.2001 ) . a crucial outcome of the comparison with observations will be to verify whether the very hot hb stars result from a high - helium population . as a byproduct of the population synthesis for the hb , we also predict how large a population of helium white dwarfs is expected in clusters hosting extreme hb stars , as found by bellini et al . ( 2013c ) in @xmath0 cen . _ analysis of rgb+hb+agb star counts ( searching for agb manqu objects ) . _ spectroscopic studies have revealed that the agbs of some gcs host only cn - weak / na - poor stars , at odds with what is observed along the rgb , where both cn - weak and cn - strong stars are present ( e.g. , norris et al . 1981 , pilachowski et al . 1996 , campbell et al . 2010 , 2013 ) . this has been interpreted as the presence of multiple stellar populations with different nitrogen and helium abundances . the helium - rich / nitrogen - rich stars would populate the hottest part of the hb , and then evolve without ascending the agb , thus making the agb populated by cn - weak stars only ( norris et al.1981 ) . we plan to analyze star counts of rgb , hb , and agb stars to further investigate the connection between other stellar populations and agb - manqu objects . _ globular cluster ages . _ relative ages of 64 globular clusters were measured in the acs treasury program of galactic globular clusters ( marin - franch et al . single stellar populations were assumed , however , and no dependence on chemical abundances was taken into account ; these limitations were due to lack of composition - dependent stellar - evolution libraries and to having only two - color photometry . the go-13297 data set now adds three more uv / optical bands , and the resulting set of five photometric bands is much better able to disentangle multiple stellar populations , which are also distinguished on the basis of their cno content . we are also computing a new stellar - evolution library , including varying chemical compositions , which will enable us to interpret our new photometry in terms of mps in gcs . in particular the new models include variation of cno content , which has a strong impact on age estimates ( cassisi et al . 2008 , ventura et al . the final product will be a new estimate of relative ages for the different populations , this will be done with a precision that has never been attained before . _ binaries in multiple stellar populations . _ an accurate study of the cmd can provide information on the population of binary systems in gcs , such as the fraction of ms ms binaries , the distribution of mass ratios , and the radial distribution ( see , e.g. , romani & weinberg 1991 , sollima et al . 2007 , milone et al . these are fundamental ingredients for understanding the dynamical evolution of gcs and their population of exotic stellar objects , such as bsss , cataclysmic variables , millisecond pulsars , and low - mass x - ray binaries . we plan to exploit the large number of cmds that can be obtained from our data set to refine our previous investigations on binary fraction , binary mass distribution , and binary radial distribution ( milone et al . 2012c ) , which was based on f606w and f814w photometry only . moreover , in clusters with multiple mss , we will estimate the fraction of binaries in the different stellar populations . _ blue straggler stars . _ blue straggler stars are crucial probes of the internal dynamics of globular clusters ( e.g. , ferraro et al.2012 ) . since they are hotter than 60007000 k , they are best detectable at uv wavelengths , so that we can use the uv survey data to obtain complete samples of bsss in the central regions of gcs , with a significant improvement in completeness over that of previous efforts ( e.g. , piotto et al . 2004 ) . detailed star counts and complete luminosity functions will be obtained , possible sub - structure unveiled ( such as the double sequences in m30 and ngc 362 ; ferraro et al.2009b , dalessandro et al . 2013b ) , and extended radial distributions will be derived from combination with wider - field ground - based observations . _ stellar exotica . _ the uv survey will be used to search for exotic objects that came from the evolution of binary systems in the cores of gcs . among these , of particular interest is the identification of the companion stars in binary millisecond pulsars ( ferraro et al . 2001 , 2003b ; pallanca et al . 2010 ) , since it would allow us to quantify the occurrence of dynamical interactions and to constrain the mass of neutron stars , with invaluable consequences for the equation of state of matter at equilibrium densities that are near nuclear . _ cross correlation of x - ray sources with hst uv / optical data . _ more generally , studies of the nature , origin , and evolution of binary stars and their progeny will be aided by cross - correlation of the go-13297 data with imaging at x - ray wavelengths . as repositories of energy , binaries play a central role in the dynamics of globular clusters . with sub - arcsecond positions provided by the chandra x - ray observatory , whose archive already includes more than half of the survey clusters , optical and/or uv counterparts can be identified for a wide variety of systems . these include cataclysmic variables , quiescent low - mass x - ray binaries , millisecond pulsars , red stragglers , and systems with active coronae ( e.g. , rs , cvn , and by dra stars ) . proper motions measured as a part of this program will make the additional contribution of separating binaries and stellar exotica from field stars that occupy the same regions of the cmd . _ luminosity and mass functions of different stellar populations . _ go-10775 allowed us to extract the stellar mass functions for 17 gcs ( paust et al . 2010 ) . with the new data we plan to extend this survey ; this will have important implications , in particular for the effect of the galactic gravitational potential ( djorgovski et al.1993 ) on the gc stellar mass function . moreover , from cmds that now include uv colors , with which multiple mss are better distinguished , we should be able to obtain the mass functions of the different stellar populations in the same cluster , as milone et al . ( 2012d ) did for ngc 2808 . _ galactic bulge globular clusters . _ multi - populations have been detected in a few massive galactic - bulge gcs , but so far no systematic search for them has been undertaken . about half a dozen clusters in the go-13297 target list are located in the inner bulge or at the edge of the bulge / halo transition zone , and may be dynamically connected to the bulge . the bulge clusters have different properties from the classical halo clusters , and there is evidence that they formed in an early epoch ( barbuy et al . 1998 , cot 1999 ) . the metal - rich clusters there might show a high helium abundance . + a few other clusters show a blue horizontal branch and a metallicity [ fe / h ] @xmath24 . therefore these gcs are either very old or else they host at least two stellar populations . _ spectral energy distributions of multiple stellar populations . _ we will use the multi - band photometry to calculate integrated cluster uv / optical colors by summing the flux of all measured stars in order to obtain an integrated - color database that can be used for the study of extragalactic gcs . more generally , gcs are widely used as template stellar populations for the calibration of stellar evolutionary sequences and of the synthetic stellar populations that are constructed with them . these are then used to interpret the light from very distant galaxies , including the rest - frame uv . the multiband database resulting from this treasury program will constitute the main benchmark for this indispensable calibration . _ link between globular cluster uv properties and extragalactic sources . _ with these data it will be possible to calibrate integrated cluster uv / optical colors for the diagnostics of mps in unresolved stellar systems , such as gcs in galaxies well beyond the local group , exploring to what extent the formation of gcs has proceeded in a similar ( or different ) fashion in different galactic environments , e.g. , in ellipticals vs. spirals . indeed , the suspicion exists that several massive gcs in the giant elliptical m87 may harbor helium - enriched populations as revealed by uv excess , likely due to the presence of very hot 2 g hb stars ( e.g. , kaviraj et al . 2007 ) . _ differential reddening analysis and reddening maps . _ the interstellar medium exhibits complex structure , on scales ranging from a few astronomical units to several kiloparsecs ( e.g. , schlegel et al . we plan to determine for each cluster a high - resolution map ( typically around 10 square arcsec ) of differential reddening ( as in milone et al . 2012b , sect . 3 ) correcting photometry for differential reddening is a crucial step for identifying multiple stellar populations in the cmd of any gc . _ uv photometric standards . _ to date there is no large space - based network of uv photometric standards . we propose to use hst uv archival material to improve the existing absolute photometric calibration of wfc3/uvis in f275w and f336w bands , and to build a network of photometric standards that will remain a fundamental reference for years to come . go-13297 gcs are rather evenly distributed in the sky and contains hundreds of bright and relatively isolated ( at least in uv images ) stars that can be used to calibrate other uv instruments , and the ground - based u - band observations . this is particularly valuable taking into account that after _ hst _ , no other large space - based observatory with uv capability is expected to fly for at least 10 years . _ serendipity . _ we are certain that in addition to those mentioned above , this multi - band legacy database for galactic gcs will offer unlimited possibilities for serendipitous discoveries and an extremely wide variety of other scientific investigations , and will contribute to our understanding of the formation of our galaxy itself . aa , sc , sh , mm , and gp recognize partial support by the iac ( grant p301031 ) and the ministry of competitiveness and innovation of spain ( grant aya2010 - 16717 ) . ja , ac , irk , as , and ev acknowledge support from stsci grant go-13297 . apm acknowledges support by the australian research council through discovery project grant dp120100475 . mz acknowledges support by proyecto fondecyt regular 1110393 , by the basal center for astrophysics and associated technologies pfb-06 , and by project ic120009 `` millennium institute of astrophysics ( mas ) '' of iniciativa cientfica milenio by the chilean ministry of economy , development and tourism . frf , bl and ed acknowledge the support from the cosmic - lab project ( web site : http://www.cosmic-lab.eu ) funded by the european research council , under contract erc-2010-adg-267675 . anderson , j. , sarajedini , a. , bedin , l. r. , king , i. r. , piotto , g. , reid , i. n. , siegel , m. , majewski , s. r. , paust , n. e. q. , aparicio , a. , milone , a. p. , chaboyer , b. , & rosenberg , a. 2008 , , 135 , 2055 carretta , e. , bragaglia , a. , gratton , r. g. , lucatello , s. , catanzaro , g. , leone , f. , bellazzini , m. , claudi , r. , dorazi , v. , momany , y. , ortolani , s. , pancino , e. , piotto , g. , recio - blanco , a. , & sabbi , e. 2009 , , 505 , 117 ferraro , f. r. , lanzoni , b. , dalessandro , e. , beccari , g. , pasquato , m. , miocchi , p. , rood , r. t. , sigurdsson , s. , sills , a. , vesperini , e. , mapelli , m. , contreras , r. , sanna , n. , & mucciarelli , a. 2012 , , 492 , 393 kaviraj , s. , schawinski , k. , devriendt , j. e. g. , ferreras , i. , khochfar , s. , yoon , s .- j . , yi , s. k. , deharveng , j .- m . , boselli , a. , barlow , t. , conrow , t. , forster , k. , friedman , p. g. , martin , d. c. , morrissey , p. , neff , s. , schiminovich , d. , seibert , m. , small , t. , wyder , t. , bianchi , l. , donas , j. , heckman , t. , lee , y .- w . , madore , b. , milliard , b. , rich , r. m. , & szalay , a. 2007 , , 173 , 619 marn - franch , a. , aparicio , a. , piotto , g. , rosenberg , a. , chaboyer , b. , sarajedini , a. , siegel , m. , anderson , j. , bedin , l. r. , dotter , a. , hempel , m. , king , i. , majewski , s. , milone , a. p. , paust , n. , & reid , i. n. 2009 , , 694 , 1498 marino , a. f. , sneden , c. , kraft , r. p. , wallerstein , g. , norris , j. e. , da costa , g. , milone , a. p. , ivans , i. i. , gonzalez , g. , fulbright , j. p. , hilker , m. , piotto , g. , zoccali , m. , & stetson , p. b. 2011c , , 532 , a8 marino , a. f. , milone , a. p. , przybilla , n. , bergemann , m. , lind , k. , asplund , m. , cassisi , s. , catelan , m. , casagrande , l. , valcarce , a. a. r. , bedin , l. r. , corts , c. , dantona , f. , jerjen , h. , piotto , g. , schlesinger , k. , zoccali , m. , & angeloni , r. 2014 , , 437 , 1609 milone , a. p. , piotto , g. , bedin , l. r. , king , i. r. , anderson , j. , marino , a. f. , bellini , a. , gratton , r. , renzini , a. , stetson , p. b. , cassisi , s. , aparicio , a. , bragaglia , a. , carretta , e. , dantona , f. , di criscienzo , m. , lucatello , s. , monelli , m. , & pietrinferni , a. 2012a , , 744 , 58 milone , a. p. , piotto , g. , bedin , l. r. , aparicio , a. , anderson , j. , sarajedini , a. , marino , a. f. , moretti , a. , davies , m. b. , chaboyer , b. , dotter , a. , hempel , m. , marn - franch , a. , majewski , s. , paust , n. e. q. , reid , i. n. , rosenberg , a. , & siegel , m. 2012c , , 540 , a16 milone , a. p. , marino , a. f. , bedin , l. r. , piotto , g. , cassisi , s. , dieball , a. , anderson , j. , jerjen , h. , asplund , m. , bellini , a. , brogaard , k. , dotter , a. , giersz , m. , heggie , d. c. , knigge , c. , rich , r. m. , van den berg , m. , & buonanno , r. 2014a , , 439 , 1588 paust , n. e. q. , reid , i. n. , piotto , g. , aparicio , a. , anderson , j. , sarajedini , a. , bedin , l. r. , chaboyer , b. , dotter , a. , hempel , m. , majewski , s. , marn - franch , a. , milone , a. , rosenberg , a. , & siegel , m. 2010 , , 139 , 476 vink , j. , srneczky , k. , balog , z. , immler , s. , sugerman , b. e. k. , brown , p. j. , misselt , k. , szab , g. m. , csizmadia , s. , kun , m. , klagyivik , p. , foley , r. j. , filippenko , a. v. , csk , b. , & kiss , l. l. 2009 , , 695 , 619

we study the relationship between galaxy colour , stellar mass , and local galaxy density in a deep near - infrared imaging survey up to a redshift of using the goods nicmos survey ( gns ) . the gns is a deep near - infrared hubble space telescope survey imaging a total of 45 arcmin of the goods fields , reaching a stellar mass completeness limit of at . using this data we measure galaxy local densities based on galaxy counts within a fixed aperture , as well as the distance to the 3 , 5 and 7 nearest neighbour . we compare the average rest - frame colour and fraction of blue galaxies in different local densities and at different stellar masses . we find a strong correlation between colour and stellar mass at all redshifts up to . massive red galaxies are already in place at at the expected location of the red - sequence in the colour - magnitude diagram , although they are star forming . we do not find a strong correlation between colour and local density , however , there may be evidence that the highest overdensities are populated by a higher fraction of blue galaxies than average or underdense areas . this could indicating that the colour - density relation at high redshift is reversed with respect to lower redshifts ( ) , where higher densities are found to have lower blue fractions . our data suggests that the possible higher blue fraction at extreme overdensities might be due to a lack of _ massive _ red galaxies at the highest local densities . galaxies : evolution galaxies : high - redshift

1011.4846

it is now well - established that the most massive galaxies formed very early in the history of the universe and are mainly in place in their present form by a redshift of @xmath8 ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , when the universe was only about half its present age . these massive galaxies largely stopped forming stars at redshifts @xmath9 , and their stellar populations subsequently for the most part evolved and reddened passively . the bulk of the star formation activity in the universe shifts to lower and lower mass galaxies as cosmic time proceeds ( see e.g. * ? ? ? * ; * ? ? ? this mass dependent shift in star formation activity might be the basis of the observed bimodality in colour - magnitude space , where we see a population of old , passive , red galaxies following a relatively tight correlation between colour and magnitude , the so - called red sequence , and a population of young , starforming , blue galaxies , which occupy a rather diffuse area in colour - magnitude space , known as the blue cloud . the presence of a strong correlation between galaxy colour and stellar mass is supported by numerous observational studies in the local universe ( see e.g. * ? ? ? * ; * ? ? ? * ) and up to intermediate redshifts of @xmath10 ( see e.g. * ? ? ? * ; * ? ? ? however , the colour - magnitude relation ( and related quantities ) not only depends on mass , but also on a galaxy s environment . the preference of red galaxies for denser local environments , first noticed by @xcite , and confirmed by @xcite , is now well - studied in the local universe ( see e.g. * ? ? ? * ; * ? ? ? * ) . in the early universe , however , the evidence is controversial . while some studies find that the colour - density relation at @xmath10 is mainly due to a bias in stellar mass selection and only persists for low - mass galaxies ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , others argue that a strong colour - density relation is already in place at @xmath11 ( see e.g. * ? ? ? * ; * ? ? ? * ) even at fixed stellar mass @xcite . we know that the bulk of star formation occurs in the early universe with half of the currently existing stellar mass already in place by @xmath12 @xcite . however , it is still unclear at which redshift the star formation rate density peaks and how it evolves at higher @xmath13 . it is also unknown where most of the star formation happens in the early universe . does the local sfr - density relation turn around by @xmath8 , as reported by some authors @xcite , or is the local morphology - density relation already present early in the history of the universe ( e.g. * ? ? ? we also know that massive red galaxies exist at very high redshifts , but what are their typical environments ? are they already found preferentially at higher local densities like most red galaxies in the local universe ( see e.g. * ? ? ? can we witness their transformation from heavily star - forming to passively evolving galaxies , i.e. the build - up of the red sequence in the colour - magnitude relation ? if so , when and where does it happen ? does the red sequence grow `` inside - out '' , from the densest to the least dense regions , or is it the stellar mass that predominantly determines the evolutionary path of a galaxy ? in this study we intend to answer some of the above questions by investigating the relationship between galaxy colours , stellar mass and local densities in the early universe at @xmath14 . recent high-@xmath13 studies of the environmental influence on galaxy evolution have focused on investigating the presence of bright neighbours and the incidence of major merging ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . to study the influence of local environment on a wider range of scales , including also other environment - related effects like minor merging or galaxy harassment , the depth of the survey is critical . additionally , we want to disentangle the influence of environment and stellar mass , with a sample that is volume limited with the same stellar mass completeness limit at all colours and redshifts . the sample we use in this study is drawn from the goods nicmos survey ( gns ) , a large hubble space telescope survey of unprecedented depth , reaching galaxies @xmath15200 times less luminous than available from ground based near - infrared surveys . the gns samples over two orders of magnitude in stellar mass with a stellar mass completeness limit of @xmath16 at @xmath0 . the paper is structured as follows : section 2 presents the gns survey and data , the sample selection and measurement of colours and stellar masses . section 3 describes how we quantify galaxy environment , and section 4 presents the results : the colour - magnitude relation as well as rest - frame colour and blue fraction as a function of stellar mass and local density . section 5 summarises our conclusions . we use standard cosmological parameters ( h@xmath17 km s@xmath18 mpc@xmath18 , @xmath19 , and @xmath20 ) and the @xmath21 magnitude system throughout the paper . the goods nicmos survey ( gns ) is a 180 orbit hubble space telescope survey consisting of 60 pointings with the nicmos-3 near - infrared camera . each pointing is centred on a massive galaxy ( @xmath22 ) in the redshift range @xmath23 , selected by their optical - to - infrared colours ( see * ? ? ? * ) from the goods ( great observatories origins deep survey ) fields @xcite . goods is designed to compile deep observational data over two fields of about 320 arcmin@xmath1 and a wide range of the electromagnetic spectrum , ranging from chandra and xmm - newton x - ray data , over hubble space telescope ( hst - acs ) optical imaging to spitzer irac and mips observations in the infrared . the extensive imaging data is complemented by spectroscopic surveys using the very large telescope ( vlt ) spectrographs fors2 and vimos @xcite for the goods south field , and deimos at the keck observatory to cover goods north @xcite . the uv - to - infrared photometric and spectroscopic data available in the goods south field are compiled by the fireworks project @xcite . in this study we use the optical hst - acs imaging data @xcite and spectroscopic redshifts from all available spectroscopic surveys . the positions of the 60 gns pointings were optimised to contain as many massive galaxies as possible and are partly overlapping , covering a total area of about 45 arcmin@xmath1 @xcite . the field of view of the nicmos-3 camera is @xmath24 arcsec@xmath1 with a pixel scale of 0.203 arcsec / pixel . the single exposures were dithered to obtain a higher resolution of @xmath150.1 arcsec / pixel . the psf has a width of @xmath150.3 arcsec fwhm . the data reduction is described in detail by magee , bouwens & illingworth ( 2007 ) . a full description of the survey as well as the target selection will be given in @xcite . the gns is also described in @xcite , @xcite , @xcite and bauer et al . ( submitted ) . sources are detected in the @xmath25-band images using sextractor @xcite . we use a 2 sigma threshold above the background noise and a minimum number of 3 adjacent pixels with values above this threshold . magnitudes are measured using the mag_auto output parameter , which measures the flux in a kron - like elliptical aperture , where the aperture is not fixed but depends on the light distribution of the galaxy . the flux within the aperture is then converted to magnitudes using the magnitude zeropoint of 25.17 . the limiting magnitude reached at 5@xmath26 is @xmath27 = 26.8 . this is significantly deeper than ground based near - infrared imaging of the goods fields done with isaac on the vlt , reaching a 5@xmath26 depth of @xmath28 @xcite . stars , and obvious spurious detections , were removed from the catalogue , yielding a total of 8298 galaxies in the final full catalogue . to obtain photometric redshifts , stellar masses and rest - frame colours , the nicmos @xmath25-band sources were matched to a catalogue of optical sources in the goods acs fields . photometry in the @xmath29 , @xmath30 , @xmath31 and @xmath13 bands is available for sources down to a limiting magnitude of @xmath32 . the matching is done within a radius of @xmath33 , however the mean separation between optical and @xmath25-band coordinates is much better with @xmath34 , roughly corresponding to the nicmos resolution . we identify 1076 complete drop - outs in our @xmath25-band selected catalogue , i.e. , galaxies that are detected in @xmath25-band , but have no counterpart in any of the acs bands . this corresponds to a drop - out rate of 13% . the nature of these sources and a full analysis on the likelihood of some of them being @xmath35 galaxies is presented in @xcite . the photometric redshifts were obtained by fitting template spectra to the @xmath36 photometric data points . the degeneracy in color - redshift space is problematic , especially when few filters are available . to cope with this effect we used two different approaches : the standard @xmath37 minimisation procedure , using hyperz @xcite and the bayesian approach using bpz @xcite . the synthetic spectra used by hyperz are constructed with the bruzual & charlot evolutionary code @xcite representing roughly the different morphological types of galaxies found in the local universe . we use 5 template spectra corresponding to the spectral types of e , sa , sc and i m as well as a single burst scenario . the reddening law is taken from @xcite . the photo - z code then computes the most likely redshift solution in the parameter space of age , metallicity and reddening . the best fit redshift and corresponding probability are then output together with the best fit parameters of spectral type , age , metallicity , @xmath38 and secondary solutions . the bayesian approach uses a similar template fitting method , apart from using empirical rather than synthetic template seds . the main difference though is that it relies not only on the maximum likelihood of the redshift solution in the parameter space as described above . instead it uses additional empirical information about the likelihood of a certain combination of parameters , also called prior information . the redshift solution with the maximum likelihood is determined after weighting the probability of each solution by the additional probability determined from the prior information . in our case the prior is the distribution of magnitudes for the different morphological types as a function of redshift , obtained from the hubble deep field north ( hdf - n ) data @xcite . in short , the code not only determines the best fit redshift and spectral type , but takes into account how likely it is to find a galaxy of that spectral type and magnitude at the given redshift . to assess the reliability of our photometric redshifts we compare them to available spectroscopic redshifts in the goods fields . spectroscopic redshifts of sources in the goods - n field were compiled by @xcite , whereas the goods - s field spectroscopic redshifts are taken from the fireworks compilation @xcite . we matched the two catalogues to our photometric catalogue with a matching radius of @xmath33 , obtaining 537 spectroscopic redshifts for our sources in goods - n and 369 in goods - s , adding up to 906 redshifts in total . the mean separation between photometric and spectroscopic sources is @xmath39 arcsec in the goods - n field and @xmath40 arcsec in the goods - s field . the reliability of photometric redshift measures is usually defined as @xmath41 . in the following we compare the median offset from the one - to - one relation between photometric and spectroscopic redshifts , @xmath42 , and the rms scatter around this relation , @xmath43 . we find good agreement between photometric and spectroscopic redshifts for both codes , however , hyperz is slightly better at identifying outliers by assigning low probabilities to their redshifts . the comparison between photometric and spectroscopic redshifts is shown in figure [ fig1 ] . the photometric redshifts of hyperz ( left ) and bpz ( right ) are plotted against the spectroscopic redshifts available in the literature . high probability @xmath44 are circled in red . the lower panel shows the @xmath45 dependence on redshift @xmath46 : there is no clear trend with redshift , with perhaps a slight trend to underestimate @xmath13 at high redshifts . we obtain the following results for hyperz : when considering only high probability photometric redshifts ( @xmath47 ) , the median offset is @xmath48 , with a scatter of @xmath49 ( 356 out of 906 galaxies with @xmath47 ) . using all redshifts regardless of their probability , the median offset decreases to @xmath50 , while the scatter rises slightly to @xmath51 . bpz gives comparable offsets , but slightly higher scatters : @xmath42 = 0.026 with a scatter of @xmath52 for high probability redshifts ( 792 out of 906 ) and @xmath53 and @xmath54 for all galaxies . bpz identifies more redshifts with a high probability than hyperz , however , they are not necessarily better than the low probability redshifts , as shown by the comparable scatter . overall , the performance of hyperz is slightly better than that of bpz . figure [ fig2 ] shows the dependence of @xmath45 on @xmath25-band magnitude . hyperz and bpz results are plotted as circles and crosses respectively . only galaxies in the redshift range of @xmath55 are shown in this plot . the median offset and rms scatter are computed in each magnitude bin and are plotted as big symbols ( with corresponding errorbars ) in red for the hyperz output and in blue for the bpz output . the figure shows the slightly better performance of hyperz , which is also visible in the fraction of catastrophic outliers with @xmath56 ( bottom panel of figure [ fig2 ] ) . the redshift error is stable up to faint magnitudes of @xmath57 , as is the fraction of outliers . in the following we use the hyperz photometric redshifts , regardless of their probability . we use all probability redshifts since this gives us a much larger sample of galaxies with only a slightly higher scatter . we now investigate the performance of hyperz at different redshifts , at low redshift ( @xmath58 ) and in the redshift range of @xmath55 , which is the redshift range of the galaxy sample we use in this study . these are the values that are used in the monte carlo simulations described in section [ mc ] to account for the photometric redshift errors . for the high redshift sample we obtain an average offset @xmath59 and a rms of @xmath60 , with a fraction of catastrophic outliers of @xmath61 . as above catastrophic outliers are defined as galaxies with @xmath56 , which corresponds to @xmath15 3 times the rms scatter . galaxies below @xmath62 show a slightly lower , but still comparable scatter of @xmath63 , however the outlier fraction decreases dramatically to only @xmath64 . the comparison of our results with photometric redshifts already available for the brighter part of our sample shows that our photometric redshifts are of comparable quality . photometric redshifts taken from the fireworks compilation have a median offset of @xmath65 and a rms scatter of @xmath66 , compared to @xmath50 and @xmath51 for our full sample . on magnitude._top panel : _ @xmath45 vs. @xmath25-band magnitude for hyperz ( red circles ) and bpz ( blue crosses ) . median values in each magnitude bin ( bin size 1 mag ) are plotted as big symbols with errorbars representing the scatter in each bin . the dashed lines show the limit for catastrophic outliers at @xmath56,scaledwidth=48.0% ] . _ bottom panel : _ fraction of catastrophic outliers @xmath56 as a function of @xmath25-band magnitude . only galaxies in the redshift range @xmath67 are used in this plot . [ fig2 ] the stellar masses and rest - frame colours we use are measured by multicolour stellar population fitting techniques , based on the same catalogue used for the photometric redshift measurements . spectroscopic redshifts are used if available . a large set of synthetic spectral energy distributions ( seds ) is constructed from the stellar population models of ( * ? ? ? * bc03 ) , assuming a salpeter initial mass function . the star formation history is characterised by an exponentially declining model with various ages , metallicities and dust extinctions . these models are parametrised by an age of the onset of star formation , and by an e - folding time such that @xmath68 where the values of @xmath69 are randomly selected from a range between 0.01 and 10 gyr , while the age of the onset of star formation ranges from 0 to 10 gyr . the metallicity ranges from 0.0001 to 0.05 ( bc03 ) , and the dust content is parametrised by @xmath70 , the effective v - band optical depth for which we use values @xmath71 . the model seds are then fit to the observed photometric datapoints of each galaxy using a bayesian approach . for each galaxy a likelihood distribution for the stellar mass , age and absolute magnitude at all star formation histories described above is computed . we chose to compute rest - frame @xmath6 colours , since the wavelength range of the @xmath72 and @xmath29 bands is covered best by the observed optical and h bands . the peak of the likelihood distribution is then adopted as the galaxy s stellar mass and absolute @xmath72 and @xmath29-band magnitude , while the uncertainty of these values is given by the width of the distribution . the rest - frame @xmath6 colours are obtained by subtracting the two absolute magnitudes . while parameters such as age , e - folding time and metallicity are not accurately fit due to various degeneracies , the stellar masses and colours are robust . from the width of the probability distributions we determine typical errors for our stellar masses of 0.2 dex . there are additional uncertainties from the choice of the imf and due to photometric errors , resulting in a total random error of our stellar masses of @xmath150.3 dex , roughly a factor of two . we do not probe with our filters redder than roughly the rest - frame @xmath29-band at @xmath0 in our measurements of stellar masses . the reason for this is similar to our reasons for not using longer wavelength data for the photometric redshifts - essentially there is no data redder than the @xmath73band which has the same fidelity and depth as the @xmath36 bands we use in this paper . the k - band data available from the ground is no where near as deep as the @xmath73band nicmos data . while we have irac data for our sources , we do not use these data due to the psf issues and contamination from neighbouring galaxies . furthermore , the rest - frame @xmath29-band gives us a good anchor for measuring stellar masses , as is shown by e.g. , @xcite at lower redshifts , and @xcite for higher redshifts . there is also a question as to whether or not our stellar masses are overestimated by the choice of stellar population models used . it has been argued by @xcite , among others , that the @xcite models we use here can result in stellar masses that are too high by a factor of a few due to how thermal - pulsating agb stars are treated . while we consider random uncertainties of a factor of two in our stellar masses , it is worth investigating whether or not our stellar masses could suffer from a systematic bias . @xcite have determined that galaxy stellar masses computed with an improved treatment of tp - agb stars are roughly 50 - 60% lower . however , the effect of tp - agb stars is less important at the rest - frame wavelengths we probe than it is at longer wavelengths , especially in the rest - frame ir . our sample is selected in the @xmath25-band , which corresponds roughly to rest - frame @xmath74-band at @xmath75 and approximately rest - frame @xmath29-band at @xmath0 , where the effects of tp - agb stars are negligible , as shown in previous work using similar data and the same stellar mass code @xcite . in this earlier study the effect of tp - agb stars was tested on a sample of massive galaxies by using the newer bruzual and charlot ( in preparation ) models , which include a new tp - agb star prescription . from this we find that stellar masses are on average 0.07 dex smaller than the ones calculated with the bc03 models . this systematic error is however much smaller than both the random error we assume ( 0.3 dex ) , and the cosmic variance uncertainties , and thus we conclude it is not a significant factor within our analysis . to investigate the different behaviour of red and blue galaxies we split the sample into red and blue populations by their location in the colour - magnitude diagram . due to the low number of galaxies in our sample and the lack of accurate spectroscopic redshifts for the majority of objects , we do not attempt to fit the red sequence of the colour - magnitude relation in this study . instead we evolve the red sequence found at lower redshift back in time as expected from the passive colour evolution of an old stellar population . this procedure is valid due to the high formation redshift ( @xmath76 ) of red sequence galaxies in stellar population models . after the star formation has stopped , the stellar populations of these galaxies are supposedly ageing ( = reddening ) quickly and subsequently evolving passively @xcite . we take the red sequence of galaxies in the deep2 redshift survey at @xmath10 @xcite , converted to ab magnitudes : @xmath77 and evolve it back with redshift according to @xcite . the separation between the narrow red sequence and the diffuse blue cloud was found to be present up to at least @xmath11 @xcite , with a separation between the two populations of 0.25 magnitudes . we adopt this colour limit of 0.25 mag blueward of the red sequence to separate between red and blue galaxies . ) with redshift @xmath13 for red galaxies ( red big circles ) and blue galaxies ( blue small circles ) . the adopted completeness limit of @xmath78 ( at @xmath3 ) is shown as dashed line . the limits of the redshift range used in this study are marked with dotted lines . the resulting window in redshift and stellar mass containing the galaxy sample we use in the following is given by the thick solid lines.[fig3],scaledwidth=48.0% ] we compute the expected completeness limits for red and blue galaxies from the 5@xmath26 magnitude limit of the gns ( see section [ sourcedetection ] ) and the mass - to - light ratios ( m / l ) of simple stellar populations ( ssp ) assuming a galaxy as old as the universe at each redshift of interest who formed all of their stars in one initial burst . for this purpose m / l at the rest - frame wavelength corresponding to the observed @xmath25-band at @xmath0 is determined for different ssps using the models of @xcite . to obtain the limiting stellar mass , the m / l is then multiplied with the luminosity corresponding to a @xmath3 galaxy at the detection limit of @xmath27 = 26.8 . to compute the completeness limit for red galaxies we use an evolved ssp with an age of 2 gyr , corresponding to the age of the universe at @xmath0 . a young ssp with an average age of 1 gyr is used to obtain the limiting stellar mass for the blue galaxy population . we obtain limiting stellar masses of @xmath79 for blue galaxies and @xmath80 for red galaxies respectively at @xmath81 . in figure [ fig3 ] we plot stellar mass versus redshift @xmath13 for blue and red galaxies , as defined above . the dashed magenta line shows the completeness limit of @xmath82 at the upper redshift limit of @xmath3 ( dotted line ) we adopted in this study . the completeness of blue galaxies at a given stellar mass is higher than for red galaxies , which are more difficult to detect at high @xmath13 due to the shape of their sed . however , since we want to compare red and blue galaxy populations we have to make sure that our sample is not biased towards blue galaxies at high @xmath13 , and we therefore use the conservative stellar mass cut of @xmath82 . the colour dependent completeness limits and mass functions of red and blue galaxies in the gns are further investigated by mortlock et al . ( submitted ) , who find similar completeness limits . in this study we will focus on the redshift range of @xmath67 , which provides a co - moving survey volume of @xmath83 mpc@xmath84 , minimising the effects of cosmic variance . the cosmic variance associated with a certain co - moving volume depends on the clustering strength of the objects and can be roughly estimated from the average galaxy number density and the expected variance of dark matter halos at a certain redshift @xcite . for the co - moving volume of the gns within the redshift range of @xmath67 , and at the average stellar mass of galaxies in the survey , we expect the influence of cosmic variance to be less than 10% . the final galaxy catalogue we use in the following comprises 1289 galaxies down to a stellar mass of @xmath82 within the redshift range of @xmath67 . in this study we investigate the environment of each galaxy by computing projected local number densities . we use two different methods to derive local densities : ( 1 ) the aperture density , based on galaxy counts in a fixed physical aperture and ( 2 ) the nearest neighbour density , based on the distance to the 3@xmath4 , 5@xmath5 and 7@xmath5 nearest neighbour . for both methods we use a redshift interval of @xmath85 to minimise the contribution of foreground or background objects . as for the stellar masses and rest frame colours , spectroscopic redshifts are used to determine these properties , if available . the two methods are described in more detail in the following section . the fixed aperture densities are measured by counting neighbouring galaxies within a fixed physical radius ( or aperture ) and redshift interval . the fixed physical radius was chosen to be 500 @xmath86 kpc , which is roughly the size of a single gns pointing at @xmath87 . the number of objects within 500 @xmath86 mpc co - moving radius at a galaxy s distance and within @xmath85 is computed for each galaxy , yielding the surface density @xmath88 in galaxies per mpc@xmath1 . edge effects are a major issue due to the design of the gns , since it does not have a continuous survey area , and edges are encountered frequently . the radius of 500 @xmath86 kpc is not covered for all galaxies , especially for galaxies close to the edges of isolated , non - overlapping pointings , since the field of view of a single pointing covers about 500 kpc at @xmath89 . to account for the area lost due to survey edges we approximate the surveyed area around each galaxy by the number of image pixels within the aperture , which are actually covered by the observations . @xmath88 is then normalised by the number of pixels within the aperture . this procedure effectively accounts for differences in aperture area due to galaxies being close to survey edges . the normalisation however does not account for variations with redshift due to differences in depth and resolution . to account for those differences we divide each galaxy number density by the median density ( denoted by the angled brackets ) of all galaxies within a redshift slice of @xmath85 centred on the galaxy . this then gives the relative overdensity @xmath90 where @xmath91 is the overdensity , i.e. the density excess over the median density . using the logarithm of @xmath92 divides the sample in under- and over - dense environments relative to the average density , where by definition @xmath93 . nearest neighbour densities are widely used in the literature to characterise local galaxy environment . the best choice of the number of neighbours to count , @xmath94 , is open to debate , and depends on the characteristics of the sample and the survey . values from @xmath95 @xcite up to @xmath96 @xcite have been used in different studies , depending on the size and mass of the structure that is investigated . a high @xmath94 is suited for high - mass and high density areas like galaxy clusters , whereas in low density environments a high @xmath94 measures the distance to other structures rather than characterising the local galaxy density , and therefore a smaller @xmath94 should be used . however , @xcite argue that the choice of @xmath94 does not change the resulting densities significantly . in this study we measure third , fifth and seventh nearest neighbour densities and compare the results in section [ results : densities ] . to account for edge effects a similar technique as used for the aperture densities described above is employed . first , the distance to the third , fifth and seventh nearest galaxy , @xmath97 , @xmath98 and @xmath99 , within the redshift interval of @xmath100 is computed for each galaxy . then , we count the number of image pixels within the area @xmath101 to approximate the covered area around each galaxy . since the number of pixels within @xmath102 ( @xmath103 ) is directly proportional to the covered area , it can be used instead of @xmath101 to compute the surface density @xmath104 . this gives @xmath105 in arbitrary units , which is ideal for our purposes , since we are only interested in relative densities and @xmath105 is normalised by the median value in each redshift slice , @xmath106 . the nearest neighbour density in units of a relative overdensity is then given by : @xmath107 where @xmath108 itself is the overdensity and @xmath94 has the values of 3 , 5 and 7 respectively . analogue to the nearest neighbour density we use @xmath109 to distinguish between underdense ( @xmath110 ) and overdense ( @xmath111 ) environments . to estimate the reliability of the local density estimates we perform a set of monte carlo simulations . the main source of uncertainty for the local densities are the uncertainties of the photometric redshifts ( see e.g. * ? ? ? * ) , leading to a blurring and loss of information along the line of sight . we therefore randomise the photometric redshift input according to the @xmath112 error obtained by the comparison with available secure spectroscopic redshifts in section [ photoz ] and repeat the computation of the local density with the new photometric redshift input . for this purpose we take the typical photometric redshift error @xmath113 in our redshift range of @xmath55 and assume a gaussian distribution of errors , where the width of the distribution @xmath26 corresponds to @xmath112 . for each galaxy a random value is selected within this distribution , which is then added to the measured photo - z . galaxies outside our redshift range of interest are included in the randomisation to account for scattering in and out of the redshift range we use . those galaxies obtain an added value according to the respective @xmath112 in their redshift range . we compute an average @xmath114 for galaxies with redshifts lower than our range of interest ( @xmath115 ) and an average @xmath116 for all galaxies with redshifts higher than the range we use in this study ( @xmath76 ) . catastrophic outliers are accounted for by randomly adding much larger offsets to the original redshifts of the percentage of galaxies corresponding to the fraction of catastrophic outliers , which is calculated from the comparison with secure spectroscopic redshifts . galaxies with @xmath117 are treated as catastrophic outliers here . the offsets are randomly picked from the interval @xmath118 and added or subtracted from the original redshift . following the above procedure we obtain a randomised photometric redshift input catalogue accounting for the average redshift uncertainty within the redshift range of interest as well as for the respective fraction of catastrophic outliers . it also accounts for scattering galaxies in and out of the redshift range we use in this study ( @xmath67 ) . the catalogue of randomised photometric redshifts is then used to recompute the local densities , aperture and nearest neighbour densities , as described above . the whole procedure is repeated 100 times , obtaining 100 randomised local density estimates , from which we then compute the average local density uncertainty . the local density error for each individual galaxy is given by the standard deviation of all monte carlo runs . these individual errors are then averaged to obtain an average uncertainty of the respective local density estimator . we obtain the following average uncertainties for the different local density estimates : @xmath119 , @xmath120 , @xmath121 for the three nearest neighbour densities and @xmath122 for the aperture density . these values of around 0.2 dex show the we can measure local densities with roughly the same or even slightly better accuracy than stellar masses ( see section [ stellar mass and colours ] ) . the results of the monte carlo simulations are also discussed in the context of the relations between colour density and stellar mass in the following sections , and are shown in the respective figures . in the following we first discuss the correlations between the local galaxy densities @xmath109 and @xmath123 . we then investigate the colour - magnitude relation in bins of stellar mass and bins of local density , and show the behaviour of rest - frame colour @xmath6 and blue fraction of galaxies as a function of both local density ( @xmath124 ) and stellar mass ( @xmath125 ) . additionally , we test differences between galaxy properties in the high and low quartiles of the local density and stellar mass distributions respectively . we focus on the redshift range of @xmath55 , which has a reasonable surveyed volume and stellar mass completeness down to @xmath126 . as described above two different methods were used to estimate local galaxy densities , the fixed aperture density , @xmath92 , and the @xmath127 , @xmath128 and @xmath129 nearest neighbour density , @xmath130 , @xmath131 and @xmath132 . figure [ fig4 ] , left panel , shows the distribution of the different measures of relative overdensities : @xmath123 is plotted as a black dashed - dotted line and @xmath109 as a green solid line ( @xmath95 ) , red dotted line ( @xmath133 ) and blue dashed line ( @xmath134 ) . the figure shows the different dynamical range of the four methods : the nearest neighbour densities @xmath109 cover a wider range of relative densities , especially at the high density end . the @xmath127 nearest neighbour density has the widest range , since it is most sensitive to density variations on very small scales . the standard deviation of the @xmath123 distribution is @xmath135 , while @xmath136 . the good overall agreement between the four methods is shown in the right panel of figure [ fig4 ] , where @xmath109 is plotted against @xmath123 . the different symbols used are green crosses ( @xmath95 ) , red circles ( @xmath133 ) and blue triangles ( @xmath134 ) , while the one - to - one relation is shown as a black solid line . the four density estimates agree well on average , with an average spread of @xmath137 dex . this scatter reflects the average uncertainty of the local density measurements of @xmath138 dex obtained by the monte carlo simulations described in section [ mc ] . however , it also shows that the different methods trace galaxy environment on different scales and are not necessarily comparable . the different sensitivity of the different values of @xmath94 is visible at the low density end , where the higher @xmath94 traces the average densities on larger scales , rather than local underdensities , which we are interested in . at densities above @xmath139 the four methods agree best on average , while the higher sensitivity of densities measured with lower @xmath94 towards galaxy concentration on small scales becomes visible : high relative overdensities measured with @xmath95 ( green crosses in figure [ fig4 ] ) are partly recovered by @xmath133 but not detected using @xmath134 or the fixed aperture density . the @xmath140 nearest neighbour density is by construction a better tracer of the extremes in the local density distribution , since it uses an adaptive area and therefore is sensitive to galaxy concentrations at very small scales , that the fixed aperture can not probe . the lower the value of @xmath94 , the more sensitive the density measure is to local galaxy concentrations on small scales . these very local galaxy concentrations however are potentially influencing the evolution of a galaxy dramatically via merging or galaxy interactions . in the following we use the @xmath127 nearest neighbour density @xmath124 as a local density estimator , since - although it has slightly larger uncertainties than the other density estimates ( see section [ mc ] ) - it has a wider dynamical range and higher sensitivity to the most underdense and overdense environments and is therefore a better probe of the local environment . in the following we investigate the colour - magnitude relation of the gns sample in different redshift ranges . we investigate if the red sequence is already present at @xmath141 and if it is populated by high or low - mass galaxies . additionally we examine if the build - up of the red sequence preferentially takes place in a certain environment . for this purpose the sample is split up in four stellar mass bins and four local density bins . figure [ fig5 ] shows the @xmath6 - @xmath142 colour - magnitude relation ( cmr ) in three different redshift bins : @xmath143 , @xmath144 and @xmath145 . in the three top panels the sample is split in 4 stellar mass bins with a bin size of 0.5 dex ; the symbols are shaped and sized according to the galaxy s stellar mass : larger symbols represent higher stellar mass galaxies . in the three lower panels the sample is split in bins of local density with a bin size of 0.6 dex . here , larger symbols correspond to galaxies located in higher overdensities . the red solid lines indicate the expected location of the red sequence observed at @xmath10 , and evolved passively back in time according to the models of @xcite , as described in section [ completeness - colour ] . the blue long - dashed lines is the border to the blue cloud , defined as being 0.25 magnitudes bluer than the red sequence ( see section [ completeness - colour ] ) . the different slices in stellar mass are clearly separated in colour - magnitude space . the separation is perpendicular to the red sequence , and the distance from the expected red sequence increases with decreasing stellar mass . the different slices in local density on the other hand are not well separated in colour - magnitude space . the relation between colour and local density is much weaker than the relation between colour and stellar mass . this subject is further discussed in section [ colour - mass - density ] . , scaledwidth=48.0% ] as can be seen , massive red galaxies are already present at a redshift of @xmath146 , when the universe was only @xmath152 gyrs old . the location of these galaxies in the colour - magnitude diagram is consistent with the location expected from passively evolving back the red sequence found at @xmath10 ( see above ) . at @xmath147 this region in colour - magnitude space is populated mainly by the most massive galaxies ( @xmath148 ) , while at lower @xmath13 the range of galaxies on or close to the red sequence extends towards lower mass galaxies . note that these most massive galaxies are mostly star - forming ( bauer et al . , submitted ) and not red and dead " as their location in the colour - magnitude diagram close to the red sequence might suggest . galaxies close to or on the location of the red sequence span a wide range of local densities , and there is no clear preference for a certain environment , however , we do not see red galaxies uniquely located in the most overdense regions . this is opposite to what was found at lower redshifts up to @xmath12 ( see e.g. * ? ? ? * ) and might hint at an inversion of the colour - density relation at high @xmath13 , as discussed later in this paper ( section [ colour - mass - density ] and section [ highlow ] ) . clear evolution in @xmath142 is seen for the galaxies on or close to the expected red sequence , while the blue cloud remains basically unchanged . part of this evolution , as well as the lack of low - mass galaxies at high @xmath13 may be due to the fact that faint red galaxies become more difficult to detect with high @xmath13 than faint blue galaxies and we might miss some low surface brightness , low - mass red galaxies in our sample . however , the mass limit we adopt in this study is very conservative and our sample should not be affected by colour - dependent incompleteness . a more detailed discussion on the stellar mass completeness and stellar mass functions for blue and red galaxies will be presented in mortlock et al . ( submitted ) . a strong relation between rest - frame colour and stellar mass and the apparent lack of a strong relation between colour and local density was found in the colour - magnitude relation in the previous section . we now investigate these correlations in more detail . the same redshift ranges as already used for the analysis of the colour - magnitude relation are used here : @xmath143 , @xmath144 and @xmath145 . to quantify the correlation between two variables we compute spearman rank correlation coefficients @xmath74 . the value of @xmath74 ranges between @xmath149 , where @xmath150 ( @xmath151 ) means that two variables are perfectly correlated ( anti - correlated ) by a monotonic function . completely uncorrelated variables result in @xmath152 . taking into account the sample size in each bin we estimate the significance of the value of @xmath74 using the conversion from correlation coefficient to z - score @xcite . figure [ fig6 ] , shows rest - frame @xmath6 colour ( top ) and the fractions of blue and red galaxies ( bottom ) as a function of stellar mass in three redshift bins . the correlation between @xmath6 colour and stellar mass @xmath125 is highly significant at all redshifts . spearman rank correlation coefficients in the three redshift bins range between 0.7 - 0.8 , corresponding to a significance of @xmath153 at all redshifts . the colour - mass relation also does not appear to evolve strongly with redshift . the cross - over mass , i.e. the stellar mass at which there is an equal number of red and blue galaxies , stays roughly constant at @xmath154 . this is the same cross - over mass found at lower redshifts @xmath155 @xcite . note that in the computation of blue and red galaxy fractions we account for the passive reddening of galaxy colours with redshift ( see section [ completeness - colour ] ) . the fraction of blue galaxies over all stellar masses evolves with redshift from @xmath156 at @xmath0 to @xmath157 at @xmath7 . notice that the fading and reddening expected from passive evolution of a galaxy s stellar population is already accounted for in the definition of red and blue galaxies at each redshift . combining this with the above considerations of almost no evolution in the colour - stellar mass relation , the colour evolution is mainly taking place at low stellar masses of @xmath158 . the population of more massive galaxies already comprises a similar fraction of red galaxies at high redshifts up to @xmath0 compared to @xmath7 . the left panel of figure [ fig7 ] shows the relation between @xmath6 colour ( top row ) and the fraction of blue and red galaxies ( bottom row ) as a function of relative overdensity @xmath124 . we do not find a significant correlation between colour and overdensity in the data there is , however , a possible trend for a higher fraction of blue galaxies ( @xmath159 ) at the highest overdensities ( @xmath160 ) at all redshifts , where as the blue fraction at intermediate and low densities is around 80 - 90% . the trend of a lack of red galaxies at high overdensities noticed in the cmr in figure [ fig5 ] is visible , but not significant due to the low number of galaxies at high overdensities . to further check the significance of the higher fraction of blue galaxies at high overdensities we use monte carlo simulations . we perform 100 monte carlo runs based on the randomised photometric redshift input as described in section [ mc ] . the scatter between the individual runs represents the uncertainties in the local densities due to the photometric redshift error . the right panel of figure [ fig7 ] shows the result of the simulations . the average colour ( top panels ) and fraction of blue and red galaxies ( bottom panels ) as a function of local density in each monte carlo run is plotted . the large scatter in the high density end due to the low number of galaxies in extreme overdensities becomes visible . although , above densities of @xmath160 , and at all redshifts , the vast majority of monte carlo runs show a blue fraction of 100% , whereas at lower densities the blue fractions vary between 85 - 90% . the average and rms of all monte carlo runs at @xmath161 shows an average blue fraction of 85@xmath1622% in the two lower @xmath13 bins and 90@xmath1622% in the high @xmath13 bin . at the highest densities the average blue fraction is 95@xmath16213% , however , the distribution of average blue fractions at the highest overdensities is not gaussian and the standard deviation does not represent the distribution adequately . to estimate the probability that the difference between high and low density is caused by chance , we consider the number of times a monte carlo run gives a blue fraction at the highest densities which is lower than the average value plus 3@xmath26 at lower densities . in the lowest redshift bin about 90% of monte carlo runs result in @xmath163 , giving a @xmath164 probability that the difference between highest and lower densities is caused by chance . the intermediate and high redshift bin have a probability of @xmath165 . we conclude that the colour - density relation at @xmath141 is practically not existent or very weak . if a trend with local density persists it must be very minor , with a variation of the average @xmath6 colours of less than @xmath150.1 magnitudes between relative local densities that differ by a factor of 100 . there could be an environmental influence on the blue fractions of galaxies in the most extreme overdense environments . this difference in blue fractions amounts to about 10% and is not detectable at a statistically significant level with the low number of galaxies in our sample located at the highest overdensities ( 33 galaxies at @xmath160 between @xmath67 ) . in the following we compare the properties of galaxies in the low and high quartiles in the distribution of local density and stellar mass , i.e. 25% of galaxies at each extreme respectively . the high quartile of local density corresponds to galaxies located in overdense regions of @xmath166 , while the low quartile corresponds to underdense regions of @xmath167 . the stellar mass distribution has its low quartile at @xmath168 and the high quartile at @xmath169 . after dividing the sample into low and high quartiles a kolmogorov - smirnov ( k - s ) test between the different quartiles is used to investigate the statistical difference of the two subsamples . to increase the number statistics we consider the full redshift range of our sample at @xmath55 , instead of binning into the three redshift ranges used above . the stellar mass limit is kept at @xmath126 , as above . the k - s test gives the probability @xmath170 that the galaxy properties in the low and high quartile come from the same parent distribution . the statistical differences we find over the whole redshift range ( @xmath55 ) can be summarised as : + ( 1 ) galaxies in the low and high quartile of the local density distribution have marginally different colours ( @xmath171 , i.e. @xmath172 ) but not statistically different stellar masses ( @xmath173 , i.e. @xmath174 ) ; and + ( 2 ) galaxies in the low and high quartile of the stellar mass distribution have statistically different colours ( @xmath175 , i.e. @xmath176 ) , as well as marginally different local densities ( @xmath177 , i.e. @xmath178 ) . although the differences between the low and high stellar mass quartiles are much clearer than the differences in the local density quartiles , there is a difference at the 2 @xmath26 confidence level between the colours of galaxies in over and under - dense environments . we further investigate the reliability of this result by using the monte carlo simulations described in section [ mc ] . in figure [ fig8 ] and figure [ fig9 ] we plot galaxy properties in the low and high quartiles of local density and stellar mass distribution , respectively , as a function of redshift . the results of each of the 100 monte carlo runs are plotted as a single line . figure [ fig8 ] shows the median @xmath6 colour , blue fraction and median stellar mass in each redshift bin ( @xmath179 ) in the low quartile ( blue ) and high quartile ( red ) of local densities . figure [ fig9 ] shows median @xmath6 colour , and two different densities ( fixed aperture density @xmath123 and third nearest neighbour density @xmath124 ) in the low ( blue ) and high quartile ( red ) of stellar mass . the average of all monte carlo runs in each redshift slice is overplotted as big symbols , blue circles for the low quartile and red triangles for the high quartile . figure [ fig8 ] shows that there is barely any difference in colour , blue fraction or stellar mass between the low and high quartile of local density . we may see evidence for a weak trend of a lower blue fraction at high densities emerging at @xmath180 , as well as a correspondent peak in the average stellar mass in the high density quartile , which is probably responsible for the positive correlation found from the k - s test on the data described above . however , we conclude that we lack the number statistics to reliably confirm the presence of a difference between low and high density quartile and that our data is consistent with there being no difference . note however that the lack of a clear separation between the low and high density quartiles does not contradict the results from the colour - density relation in section [ colour - mass - density ] , where we possibly see evidence for a higher blue fraction in the most overdense environments above @xmath181 . the high density quartile includes all galaxies at densities above @xmath182 and does not comprise the extreme overdensities only , in which we see the higher blue fraction . since there is no relation with local density at densities below @xmath181 , we do not see any difference between the colours of galaxies in the low and high quartile of the density distribution . we conclude that in the redshift range of @xmath55 the highest local overdensities ( @xmath160 , i.e. , an overdensity of roughly a factor of 5 ) possibly show a higher fraction of blue galaxies . at densities below @xmath181 there is no correlation between rest - frame colours or fraction of blue galaxies with local density . this finding is different from the colour density relation found at lower redshifts ( up to @xmath11 ) , where higher density environments show a higher fraction of red galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . figure [ fig9 ] shows two different local densities @xmath124 and @xmath123 and @xmath6 colour in the low and high quartiles of the stellar mass distribution . the high quartile includes galaxies with @xmath169 ( red ) , while the low quartile corresponds to @xmath168 ( blue ) . there are strong differences in the rest - frame colour distributions of low- and high - mass galaxies at all redshifts . as already investigated in the previous section , high - mass galaxies have consistently redder colours ( @xmath183 magnitudes ) . these differences are highly significant . the colour difference of @xmath150.4 mag between the high- and low - mass quartile appears constant with redshift up to @xmath184 . the magenta line in the colour panel of figure [ fig9 ] shows the distinction between red and blue galaxies calculated from the red sequence at @xmath10 evolved back in time assuming passive evolution of the stellar populations ( see section [ completeness - colour ] ) . it corresponds to the position of the blue line at @xmath185 in the colour - magnitude diagrams in figure [ fig5 ] . at a redshift of @xmath184 the median colour of the high mass quartile decreases slightly towards the region occupied by the blue cloud however , a similar decrease is seen at redshifts below @xmath58 and is probably caused by the target selection of the gns , which favours the inclusion of very massive red galaxies between @xmath23 in our sample . this could slightly bias the average colour of massive galaxies towards redder colours within this redshift range . we see slight evidence for a mass segregation emerging at @xmath186 , where more massive galaxies tend to be located preferentially in high density areas ( see figure [ fig9 ] ) . this tentative result is consistent with what is found at lower and intermediate redshifts , where more massive galaxies tend to be located in high density environments ( see e.g. * ? ? ? recent studies at intermediate redshift ( @xmath10 , see e.g. @xcite ) have suggested that the observed colour - density relation is mainly caused by the strong correlation between rest - frame colour and stellar mass . this also requires the presence of a stellar mass - density relation , such that higher mass galaxies are preferentially located in regions of higher overdensity . we indeed find a weak trend for this behaviour , emerging at @xmath187 , at a significance of @xmath188 ( see figure [ fig9 ] ) . this is roughly consistent with the extrapolation of a similar weak positive correlation between stellar mass and local density found at @xmath7 by @xcite , which seems to get steeper and more significant with decreasing redshift . another question is if the higher fraction of blue galaxies in high overdensities ( see section [ colour - mass - density ] ) is due to a trend in stellar mass . to answer this question we investigate the colour - stellar mass relation and the colour - density relation in quartiles of local density and stellar mass respectively . figure [ fig10 ] , left panel , shows the colour - stellar mass relation in high and low density quartiles . the right panel of figure [ fig10 ] shows the colour - density relation in high and low stellar mass quartiles . the same high and low quartiles of local density and stellar mass are used as described above ( section [ extreme ] ) . as in figures [ fig8 ] and [ fig9 ] we plot the results of each monte carlo run as well as the average and rms of all runs instead of the measured data . the whole redshift range of @xmath67 is shown in figure [ fig10 ] . to compare our result with the colour - density and colour - stellar mass relations at lower redshift we use the data of ( * g10 hereafter ) . this sample is based on a deep near - infrared survey ( the powir survey , @xcite ) and spectroscopic redshifts from the deep2 redshift survey @xcite . the local densities and colours are measured in a similar way as in the present study . we split the g10 sample in high and low quartiles of local density and stellar mass as described above . the data covers the redshift range @xmath189 . the median @xmath6 colours of this sample are overplotted as big symbols : dark red boxes are used for the high quartile and dark blue pentagons for the low quartile of the powir sample data - points respectively . the left panel of figure [ fig10 ] shows that the strong correlation between @xmath6 and @xmath125 is present in both , high ( red ) and low ( blue ) density quartiles . the relation is very similar in both density quartiles . the colour - stellar mass relation at lower redshifts of @xmath190 ( squares and pentagons in figure [ fig10 ] ) is remarkably similar to the colour - stellar mass relation in the gns data ( triangles and circles ) , with a similar slope and a slightly larger colour difference between low and high local density quartiles ( @xmath191 mag ) . at the high mass end low @xmath13 galaxies have redder average colours than galaxies of the same stellar mass at high @xmath13 . the colours of @xmath192 galaxies at @xmath193 are indistinguishable from the colours of similar mass galaxies at @xmath189 . the right panel of figure [ fig10 ] shows the correlation between colour and local density for galaxies in the low and high stellar mass quartile . a clear colour offset between low- and high - mass galaxies of @xmath194 mag is present at all densities . interestingly , there is a trend that mean colours of galaxies in the high - mass quartile are bluer at higher overdensities ( @xmath195 ) than at low and average local densities . for galaxies in the low - mass quartile we do not see a strong correlation between colour and local density . combined with the results of the colour - density relation in figure [ fig7 ] , this would point towards high - mass galaxies being responsible for the higher fraction of blue galaxies at the highest overdensities . note however that the rms scatter between the single monte carlo simulations is larger than the decrease in colour . this large spread is due to the low numbers of objects at the highest overdensities ( 33 galaxies at @xmath196 . the colour difference between the gns data - points and the @xmath7 data - points from g10 is mainly due to the differences in the stellar mass distribution in the two surveys , leading to different limits of the low and high quartiles . the gns reaches much lower stellar masses down to @xmath197 , whereas the completeness limit of the powir survey is @xmath198 . additionally , the g10 sample has a larger number of high mass galaxies with @xmath199 than the gns sample , leading to redder average colours , as expected from the colour - mass relation . if the colour - density relation we see for high - mass galaxies is real , then it is reversed with respect to what is found at lower and intermediate redshifts up to @xmath200 , where galaxies in higher local densities tend to be redder ( see e.g .. * ? ? ? * ; * ? ? ? * ; * ? ? ? figure [ fig10 ] shows that the high and low mass quartiles of the g10 sample ( up to @xmath10 ) show a weak correlation between colour and local density , such that the average colour increases with higher local density . on the other hand , it was argued by some studies that the colour - density relation and colour - sfr relation at @xmath10 might be reversed @xcite . note , however , that @xcite argue that the reversed sfr - density relation ( were galaxies in denser environments have higher sfrs ) is mainly driven by starforming _ low_-mass galaxies , whereas our data indicate that the reversed colour - density relation is due to on average bluer _ high_-mass galaxies . in other worlds , we see a lack of red high - mass galaxies in regions of high relative overdensities . the on average bluer high - mass galaxies , however , are on average still redder than low - mass galaxies at comparable densities , but bluer than their high - mass counterparts at average and low local densities . note that the bluer colour of high mass galaxies does not necessarily imply that there is more ongoing star formation in these galaxies , but that there is possibly _ less dust - attenuated star formation _ at highest overdensities . massive galaxies in our sample show an overall very high dust content relative to low mass galaxies ( bauer et al . submitted ) . the behaviour of the sfr - density relation using dust corrected star formation rates from bauer et al . ( submitted ) will be investigated in a forthcoming paper . recently , a study of the colour - density relation at redshifts up to @xmath201 was performed by chuter et al . ( submitted ) using the ukidss ( ukirt infrared deep sky survey ) ultra deep survey ( uds ) data . they find that the local colour - density relation , with redder galaxies located in regions of higher local density , is present up to a redshift of @xmath202 . above this redshift the distinction between the environments of red and blue galaxies can not be confirmed at a statistically significant level . the authors also report a possible reversal of the colour - density relation at even higher redshift , however , the currently available uds data release ( dr3 ) does not allow for reliable conclusions at high @xmath13 . the results we present in this study are consistent with extrapolating the results of chuter et al . ( submitted ) , suggesting a gradual disappearance of the colour density relation with redshift and a possible reversal of the colour - density relation at the most extreme overdensities at @xmath141 . in the following we investigate how , and if , the environments of the most massive galaxies ( @xmath148 ) differ from the rest of the galaxy population and if there is a correlation between local density and galaxy size . the subsample of massive galaxies for which a size measurement is available comprises 57 galaxies from the selection of massive galaxies in the goods fields on which the gns pointings were centred @xcite . the measured local densities of those galaxies are very reliable since they are in the centre of their respective pointing and should not be affected by survey edges , as discussed in section [ density ] . the left panel of figure [ fig11 ] plots the local densities of the massive galaxy sample ( red dots ) and the rest of the sample ( blue circles ) against redshift . a k - s test between the two subsamples shows that they are not statistically different . to examine a possible environmental effect on the sizes of massive galaxies in our sample , we use the galaxies effective radii measured by @xcite . they argue that the measured effective radii of these massive galaxies are consistently smaller than the typical effective radii of galaxies of comparable stellar mass in the local universe . different scenarios have been proposed to account for this size evolution , one of them being dry merging . in this scenario repeated minor merging of gas poor galaxies would not trigger the formation of new stars , but could possibly increase the size of a galaxy by dynamical friction and the injection of angular momentum @xcite . this scenario would then suggest a connection between galaxy sizes and local density , since it requires the presence of numerous companions for galaxies to grow in size . the right panel of figure [ fig11 ] shows the effective radii @xmath203 of massive galaxies as a function of their local density @xmath124 . a spearman rank correlation test finds no significant correlation between @xmath203 and local density , although all galaxies in the most underdense environments ( @xmath204 ) have small @xmath203 ( @xmath205 3 kpc ) . however , due to the small number of objects in our sample this effect probably arises by chance . the lack of a correlation between galaxy size and local density is probably due to the different effects that the environment might have on galaxies . high local densities may not only increase the size of a galaxy via minor mergers as suggested by e.g. @xcite . high density environments might as well tidally truncate galaxies ( see e.g. * ? ? ? * ) , although tidal truncation is most effective on low - mass galaxies . another possibility is that the size evolution , does not occur through frequent minor merging of satellite galaxies , but through internal processes like agn feedback @xcite . in this study we investigate the influence of stellar mass and local density on galaxy rest - frame colour and the fraction of blue galaxies at redshifts between @xmath206 based on observational data from a deep hst @xmath25-band survey of unprecedented depth , the goods nicmos survey ( gns ) , reaching a stellar mass completeness limit of @xmath207 at @xmath3 . we find the following results : 1 . galaxy colour depends strongly on galaxy stellar mass at all redshifts up to @xmath0 . the colour - stellar mass relation does not evolve with redshift below @xmath0 . the stellar mass where the blue and red fractions cross over is roughly constant at @xmath208 in the redshift range between @xmath55 , which is remarkably similar to the cross - over mass found at lower redshifts between @xmath190 . the strong colour - stellar mass relation is very similar across all local environments and does not evolve strongly with redshift . the colour - stellar mass relation at @xmath67 has a slope similar to the relation at lower redshift ( @xmath190 ) , which also has a larger offset in colour between low and high local densities . at the same stellar mass , massive galaxies at high redshifts ( @xmath67 ) are bluer than at low redshifts ( @xmath190 ) , whereas the average colour of lower mass galaxies ( @xmath192 ) does not vary strongly with redshift . we do not find a strong influence of local environment on galaxy colours . if the colour - density relation persists at @xmath141 it must be very weak . we determine an upper limit to the possible change in the average @xmath6 colour between high and low relative densities of @xmath209 magnitudes . however , the most overdense regions ( @xmath210 times overdense ) may be populated by a higher fraction of blue galaxies than average and underdense regions . the difference in @xmath211 is @xmath212 . we find possible evidence that this higher blue fractions at the most extreme overdensities could be caused by a lack of massive red galaxies at the highest local densities . + 4 . we do not find a significant correlation between galaxy sizes ( effective radii ) and relative overdensity , although we do not find any galaxies with large effective radii ( @xmath213kpc ) at very low densities ( @xmath214 . however , this might be due to the small number of objects we find in low density environments and is probably a chance effect . + to summarise , our data suggests that stellar mass is the most important factor in determining the colours of galaxies in the early universe up to @xmath0 . local density might have a small additional effect , but only at the most extreme overdensities , which are populated by a higher fraction of blue galaxies . these results are consistent with studies at lower and intermediate redshift that suggest a gradual weakening of the environmental influence with higher redshift . a possible interpretation for this is that the environmental processes that alter the properties of galaxies are proceeding slowly over cosmic time . some of the most influential high density environments like galaxy clusters are still in the process of being build - up at @xmath141 and yet can not change galaxy colours via e.g. , ram pressure stripping or strangulation . if the trend for higher blue fractions at the highest local densities is real , it would suggest that we are witnessing the epoch of high star formation in more massive galaxies and that the local environment contributes to this epoch by triggering sf through galaxy interactions . note however , that the bluer colours at high local densities could as well be due to lower dust attenuation in these environments . the star formation rates of galaxies in the gns are currently investigated by bauer et al . ( submitted ) and will be the focus of a subsequent study of the relation between sfr , stellar mass and local density ( grtzbauch et al . , in preparation ) . future deep and wide surveys such as the ukidss uds and vista will be better able to address the environment vs. stellar mass issue in more detail in the coming years . rg would like to thank jess varela for many useful discussions and very helpful suggestions . we acknowledge funding from the uk science and technology facilities council ( stfc ) . we also thank the anonymous referee for their suggestions , which greatly improved the content of this paper .

certain bacteria form filamentous colonies when the cells fail to separate after dividing . in _ bacillus subtilis _ , _ bacillus thermus _ , and _ cyanobacteria _ , the filaments can wrap into complex supercoiled structures as the cells grow . the structures may be solenoids or plectonemes , with or without branches in the latter case . any microscopic theory of these morphological instabilities must address the nature of pattern selection in the presence of _ growth _ , for growth renders the problem nonautonomous and the bifurcations dynamic . to gain insight into these phenomena , we formulate a general theory for growing elastic filaments with bending and twisting resistance in a viscous medium , and study an illustrative model problem : a growing filament with preferred twist , closed into a loop . growth depletes the twist , inducing a twist strain . the closure of the loop prevents the filament from unwinding back to the preferred twist ; instead , twist relaxation is accomplished by the formation of supercoils . growth also produces viscous stresses on the filament which even in the absence of twist produce buckling instabilities . our linear stability analysis and numerical studies reveal two dynamic regimes . for small intrinsic twist the instability is akin to euler buckling , leading to solenoidal structures , while for large twist it is like the classic writhing of a twisted filament , producing plectonemic windings . this model may apply to situations in which supercoils form only , or more readily , when axial rotation of filaments is blocked . applications to specific biological systems are proposed . supercoil , growth , elastic filaments , filamentous bacteria 87.17.-d , 87.16.ka , 05.45.-a , 46.70.de

cond-mat0312563

many structures in the biological world grow so slowly that they adopt a shape that can be considered as a minimizer of some configurational energy associated solely with the _ internal _ structure . the logarithmic spiral of the nautilus shell is an example . it enlarges through a process of differential growth whereby its shape represents the accumulated history of identical events , save for scale changes @xcite . thus , the microscopic rules of growth are essentially unchanging as the three - dimensional form develops , and the properties of the external environment do not fundamentally determine the form . environmental effects on the development of biological forms are well known . consider the formation of tendril perversions in climbing vines @xcite . a perversion is a junction between regions of opposite helix handedness that forms as an initially straight tendril first attaches to a support structure and then , through the activation of tension - sensitive receptors , undergoes a helical instability . the constraint of fixed ends enforces the formation of a structure with zero net twist , consisting of concatenated regions of opposite chirality joined by a transition region the perversion . thus , the interaction with the external environment fundamentally alters the pattern formation through the tension induced by a point contact . even more complex phenomena occur for those structures whose very process of growth induces an _ external _ force , as with motion through an environment that offers viscous resistance . these new forces can dramatically alter the ultimate configuration , and the present paper is a case study in such phenomena . we focus on a system in which the formation of patterns occurs through a finite - wavelength instability in which the process of growth introduces an intrinsic time dependence to the control parameters . the resulting bifurcation problem is nonautonomous and can exhibit a rich phenomenology as the intrinsic time scale of growth competes with those of the various modes of instability . this competition places the problem among the class of so - called dynamic bifurcations , " of which many examples are of continuing interest . these include instabilities in directional solidification in which the initial acceleration of the interface from rest provides the nonautonomous character @xcite , fingering instabilities of magnetic fluids under the influence of time - dependent magnetic fields @xcite , and separatrix crossing in hamiltonian systems viewed as models for stellar fission @xcite . motivation for this focus comes from the phenomenon of supercoiling exhibited by filamentous colonies of _ bacillus subtilis _ and other bacteria . _ b. subtilis _ cells are rod - shaped bacteria , typically four microns in length and slightly over one - half micron in diameter . wild - type rod - shaped bacteria grow by extending along the cylindrical axis of symmetry , and then dividing and separating in the middle @xcite . under certain conditions , the cells of some mutant forms fail to separate upon replication , leading to a long chain of cells . other species have also been observed to form chains , including _ escherichia coli _ @xcite , _ cyanobacteria _ @xcite , _ myxococcus xanthus _ @xcite , and _ mycobacterium tuberculosis _ @xcite . under certain growth conditions , strains of _ b. subtilis _ , _ bacillus stearothermophilus _ @xcite , _ thermus _ @xcite , and _ mastigocladus laminosus _ @xcite form complex braided structures . helical ( or , in the jargon of dna biophysics , solenoidal " ) morphologies have also been observed @xcite . of these examples , the supercoiled structures of _ b. subtilis _ have been studied most extensively @xcite . recent experimental work @xcite has indicated that the cell wall of _ b. subtilis _ contains helical protein structures . these may supply the molecular imprinting responsible for this morphological development , in a manner analogous to the way microtubules control macroscopic handedness in certain plants @xcite , but as of yet no successful microscopic theory for the formation of these supercoiled structures exists . @xmath0 m . figure courtesy of n. mendelson ; see @xcite . , height=288 ] throughout much of the development of complex structures in _ b. subtilis _ , the length @xmath1 of the elongating chain , a single cell thick , grows exponentially in time , @xmath2 , with the growth rate @xmath3 s@xmath4 . moreover , material cross - sections of the cells rotate relative to each other @xcite , so that the angle describing the relative orientation of any two material cross - sections is proportional to their exponentially increasing separation . in one of the earliest observations of supercoiling in _ b. subtilis _ , the ends of the fiber adhered to a spore coat , prohibiting axial rotation of the ends @xcite . mendelson supposed that this blocked rotation leads to torsional stress , eventually causing supercoiling @xcite ( see fig . [ expt_loop ] ) . since this discovery , it has been shown that adhesion is not required for supercoiling @xcite , and that many different factors such as temperature , ph , and the concentration of ions such as magnesium and ammonium affect the morphology of the coils @xcite . the evolution of the coils after the formation of the first braid is remarkable . the first plectonemic braid , essentially a filament that is two cells thick , continues to grow and eventually reaches a critical length of order 100 @xmath0 m , after which it supercoils to form another braided structure which is four cells thick . this process can repeat many times , leading to a hierarchy of braids and eventually a macroscopic object . in this paper , we study in detail mendelson s blocked rotation mechanism of supercoiling in a growing closed loop , and focus on the formation of the initial braid . while not yet explaining the microscopic physical origin of the coiling instability , we elucidate the rich dynamics that occurs when growth competes with blocked rotation , and thereby help constrain more detailed theories . with this focus we exploit several simplifying assumptions in our analysis . first , the chain of cells is treated as an elastic filament with uniform properties along its length . there is evidence that this assumption holds until times comparable to a few doubling times ( @xmath5 ) , but may be violated later . for example , when the chirality of nutrient molecules in the growth medium is reversed , the supercoils unwind and even begin to wrap up in the opposite handedness , but the hairpin bends from the original braid remain @xcite . thus , some of the deformation of the growing filament becomes permanent . this phenomenon is reminiscent of the morphological development of plant tendrils , in which young and flexible tendrils age with time , becoming woody and locked in a fixed shape @xcite . on the time scales that will concern us , @xmath6 , single fibers have been shown to behave like elastic rods , with a bending modulus @xmath7 erg - cm @xcite , as discussed further in section [ constit ] . a second major simplification we introduce is to treat the growth rate as constant in time , independent of stress and filament geometry . again , the permanent hairpin bends in the chirality reversal experiment show that this assumption can not hold everywhere along the filament for all times after the first braid forms . together with the observation that viscous effects dominate inertial effects in the low reynolds number environment of the growing fibers , these assumptions lead to the model studied in this paper : an elastic ring with intrinsic twist suspended in a viscous fluid and lengthening at an exponential rate . our work is complementary to that of other investigators . for example , shelley and ueda studied the euler - like buckling of a growing liquid crystal filament , using a local drag model for the linear stability of a growing loop @xcite , and incorporating nonlocal stokesian hydrodynamics to study the pattern formation @xcite . drasdo studied similar patterns in the context of the growth of single - cell - layer tissue sheets @xcite . klapper has studied inertial writhing instabilities of _ open _ rods subject to exponential growth , as well as the relaxation to equilibrium of twisted rings in the absence of growth @xcite . goriely and tabor @xcite introduced the idea of _ twist depletion _ as a possible mechanism driving buckling in _ b. subtilis_. our analysis begins with a generalization of the kinematics and dynamics of slender filaments to account for growth . section [ grow_elas_loop ] treats the growing elastic loop , beginning with a qualitative discussion of the instability . the linear stability analysis is greatly simplified by the use of the natural frame , so in this section we include a self - contained summary of the properties of the natural frame . we present a quasi - analytic treatment of the linear stability of the loop , and then present numerical simulations of the full nonlinear problem . section [ conclude ] is the conclusion . in this section we extend the standard kinematics and dynamics of elastic rods to allow for growth . let @xmath8 denote the arclength measured at time @xmath9 from one end of an open rod , or from a fixed material point for a rod closed to form a loop . then @xmath10 is the position in space of the centerline of the rod with arclength coordinate @xmath8 at time @xmath9 . since material points on the rod centerline are convected along the rod by growth , fixed values of @xmath8 do not correspond to fixed material points . we choose to label the material points of the centerline at all times by the arclength parameterization @xmath11 at a fixed time @xmath12 . we will study exponential growth , for which @xmath13 . note that the partial derivatives with respect to @xmath8 and @xmath9 commute because @xmath8 and @xmath9 are independent variables : @xmath14 however , @xmath15 and @xmath16 do not commute since @xmath17 by the chain rule . the velocity of a material point is the time derivative of position at fixed @xmath11 , @xmath18 it is convenient to develop the equations of motion in terms of @xmath8 rather than @xmath11 , so that @xmath19 to avoid confusion , we will explicitly denote which variables are fixed when finding the partial derivatives with respect to @xmath9 . however , since partial derivatives with respect to @xmath8 will always be taken at fixed @xmath9 , we will write @xmath20 . the first term of eq . ( [ convect1 ] ) corresponds to the velocity of the centerline in the absence of growth , and the second term arises from growth - induced convection . henceforth we will specialize to exponential growth , for which @xmath21 and @xmath22 where @xmath23 is the unit tangent vector of the centerline . although the velocity of a material point may seem from eq . ( [ convect2 ] ) to depend on the arbitrary choice of origin for @xmath8 , eq . ( [ v - eqn ] ) shows that the velocity is manifestly independent of this choice . note that eq . ( [ convect2 ] ) implies that @xmath24 . is tangent to the rod centerline . the other members of the orthonormal frame , @xmath25 and @xmath26 , point to material points on the rod s surface.,height=192 ] in kirchhoff rod theory @xcite , the configuration of a rod is completely specified by the orientation of the material orthonormal frame @xmath27 . the vectors @xmath25 and @xmath26 point to material points on the rod surface ( fig . [ mat_frame ] ) . as the rod bends and twists , the positions of these material points change , causing the material frames to rotate @xcite : @xmath28 the vector @xmath29 describes the bending and twisting strain at a given instant , and @xmath30 is the angular velocity of a material frame at a given material point @xmath11 . in general , @xmath29 and @xmath30 depend on the choice of material frame . once the choice is made for a given configuration , say the stress - free state , then the choice is specified for all configurations . in the classical rod theory without growth , it is natural to align @xmath25 and @xmath26 with the principal axes of the cross - section . if the cross - section is circular , as we henceforth assume , and if the rod is straight in the absence of stress , then there are many equivalent natural choices . for example , if the rod aligns along the @xmath31-axis when it is stress - free , then @xmath32 is natural . any uniform rotation of this frame about @xmath33 is equally convenient ; all these choices lead to @xmath34 in the absence of stress . if the rod is curved in the stress - free state , then the direction of curvature breaks the rotational symmetry of the circular cross - section and provides a natural choice for the directions of @xmath25 and @xmath26 . _ e.g. _ , if the rod has a helical stress - free state , then we may take @xmath35 and @xmath36 , where @xmath37 and @xmath38 are the unit normal and binormal of the frenet - serret frame @xcite : @xmath39 where @xmath40 is the curvature and @xmath41 the torsion . thus , @xmath42 for this frame in the absence of stress , where @xmath43 is the curvature and @xmath44 is the torsion of the helix . when @xmath45 , the helix of the last example degenerates to a straight rod with spontaneous twist @xmath44 . we will show below that if @xmath45 and the cross - section is circular , then @xmath44 can be eliminated from the equations of motion for an inextensible ( non - growing ) rod , and therefore does not affect the rod shape and dynamics . however , @xmath44 has physical meaning for a _ growing _ rod , even if @xmath45 and the cross - section is circular . figure [ growth_fig ] illustrates two kinematic possibilities for growth . each sub - figure shows the stress - free configuration of a growing rod at two different times . in all cases , the left end of the rod has a fixed position and orientation . two growth schemes are shown ; in the growth scheme of fig . [ growth_fig]a and [ growth_fig]b , the material frames are carried to greater values of @xmath31 by growth and have no angular velocity . a line of material points parallel to the @xmath31-axis at time @xmath9 remains parallel to the @xmath31-axis at time @xmath46 ( fig . [ growth_fig]a ) . however , the pitch of a _ helical _ line of material points increases as the filament grows ( fig . [ growth_fig]b ) . since the angular velocity is zero , the number of helical turns is constant . figures [ growth_fig]c and [ growth_fig]d illustrate a growth scheme in which the cross - sections rotate with an angular velocity that increases with arclength . in this case , a line of material points parallel to the @xmath31-axis at time @xmath9 wraps around the rod at time @xmath46 ( fig . [ growth_fig]c ) . as the rod grows , the material frames are carried to greater values of @xmath31 but also _ rotate _ relative to the fixed frame at @xmath47 . in this paper , we will study the growth model of fig . [ growth_fig]c and d since it describes the relative rotation of the cross - sections of the _ b. subtilis _ fibers @xcite . define @xmath48 as the angle between the orientation of the ( zero - stress ) material frames at @xmath11 and @xmath47 . we will suppose that @xmath49 increases linearly with @xmath11 and exponentially in time with rate @xmath50 : @xmath51 since @xmath52 and @xmath53 increase in time with the same exponential rate , a helical material line on the rod surface with pitch @xmath54 remains a helical line with the same pitch as time passes ( fig . [ growth_fig]d ) . thus , the natural choice for the material frame in the stress - free state is @xmath55 note that @xmath56 and @xmath57 for this frame ; @xmath58 since the material frames must _ rotate _ to maintain zero stress as the rod grows ( in contrast to the growth model of fig . the parameter @xmath44 not only characterizes the configuration of the natural material frame ( as in the inextensible helix example ) but _ also _ the mode of growth . geometry relates the strain vector @xmath29 and the angular velocity @xmath30 . to see how , consider fig . [ frames_diagram ] . the lower curve represents the centerline of the rod at time @xmath9 , and the upper curve represents the centerline of the rod at time @xmath59 . the labelled points on the upper curve have the same material coordinate as the corresponding points on the lower curve ; thus , the arclength parameter for @xmath60 is @xmath61 , while the arclength parameter for @xmath62 is @xmath63 . let @xmath64 be the rotation matrix carrying the frame at @xmath65 to @xmath60 , @xmath66 the rotation matrix carrying the frame at @xmath60 to @xmath62 , @xmath67 the rotation matrix carrying the frame at @xmath65 to @xmath68 , and @xmath69 the rotation matrix carrying the frame at @xmath68 to @xmath62 . furthermore , let @xmath70 and @xmath71 denote the infinitesimal rotation matrices associated with the rotation vectors @xmath29 and @xmath30 respectively ( _ e.g. _ @xmath72 , where @xmath73 is the alternating symbol ) . from the definitions of the rotation matrices , @xmath74 , or @xmath75 \big[{\mathsfsl{i}}+\mathrm{d}t{\mathsfsl{k}}(s , t)\big]\nonumber\\ \approx\big[{\mathsfsl{i}}&+&\mathrm{d}t{\mathsfsl{k}}(s+\mathrm{d}s , t)\big ] \big[{\mathsfsl{i}}+\mathrm{d}s{\mathsfsl{j}}(s , t)\big],\label{rotations}\end{aligned}\ ] ] where @xmath76 is the identity matrix . expanding eq . ( [ rotations ] ) to @xmath77 , we find @xmath78\label{rotations2a}\\ { \partial{\mathsfsl{k}}\over\partial s}&=&\left . { \partial{\mathsfsl{j } } \over\partial t}\right|_{s_0 } + \sigma{\mathsfsl{j}}+[{\mathsfsl{j}},{\mathsfsl{k } } ] , \label{rotations2}\end{aligned}\ ] ] where @xmath79 $ ] is the commutator . in terms of components in the material frame , @xmath80 these compatibility relations show how strain changes in time due to non - uniform rotation rates ( the first term in each of eqs . ( [ comp1])([comp3 ] ) ) , growth ( the second term in each of eqs . ( [ comp1])([comp3 ] ) ) , and the geometric coupling between twisting and bending ( the last two terms in each of eqs . ( [ comp1])([comp3 ] ) ) . the compatibility relation eq . ( [ comp3 ] ) will be used below to determine the dynamics of the twist strain @xmath81 . note that eq . ( [ comp3 ] ) can be re - written in a form valid for arbitrary growth laws @xcite @xmath82 the kirchhoff constitutive relations for a rod with intrinsic twist relate the moment on a cross - section of the rod to the strain @xcite : @xmath83 note that @xmath84 . equation ( [ khirch_const ] ) implies that in the stress - free state ( defined by @xmath85 ) , @xmath86 and @xmath26 rotate around @xmath87 with rate @xmath88 , and @xmath87 is constant . the force and moment balances for a growing rod are the same as the balances for an inextensible rod : @xmath89 where @xmath90 is the force the internal elastic stresses exert through the cross section at @xmath8 on the portion of the rod with arclength less then @xmath8 . the external force per unit length @xmath91 and moment per unit length @xmath92 are measured per unit arclength . in an alternative but equivalent formulation , the elastic force and moment per unit length arise from variational derivatives of the energy @xmath93 where @xmath94 is the square of the curvature , and @xmath95 is the lagrange multiplier associated with the constraint of prescribed length , @xmath96 @xcite . the external force per unit length @xmath97 consists of a viscous drag force per unit length and an artificial short - ranged repulsive force that prevents self - crossing : @xmath98 . the external moment per unit length is purely viscous : @xmath99 . the repulsive force takes the form @xmath100 where @xmath101 is a short - distance cutoff , @xmath102 simulates the repulsive part of a leonard - jones potential , and @xmath103=180 is sufficient to keep the filament from self crossing . the viscous force per unit length @xmath104 depends on the velocity field of the ambient fluid , which in turn is coupled to the motion of the filament . for simplicity , we do not solve the full hydrodynamic problem , but instead use resistive force theory for slender bodies @xcite . resistive force theory amounts to the leading terms in an expansion in the aspect ratio @xmath105 of slender - body theory , which has a nonlocal relation between force and velocity due to incompressibility . to leading order , the nonlocality can be neglected , leading to the local drag law , @xmath106 likewise , we take the viscous moment to be proportional to the tangential component of the angular velocity of the material frames , @xmath107 the friction coefficients in eqs . ( [ resistive_force ] ) and ( [ resistive_moment ] ) are @xmath108 where @xmath109 is the viscosity of the ambient fluid , @xmath1 is the total contour length of the rod , and @xmath110 is the rod radius . for simplicity , we disregard the anisotropy and define @xmath111 , where @xmath112 . note that we have kept only the leading order terms in the logarithm of the initial aspect ratio . thus , @xmath113 . these assumptions lead to qualitative differences with the exact theory . for example , the neglect of hydrodynamic interactions implicit in the local drag approximation of eqs . ( [ resistive_force ] ) and ( [ resistive_moment ] ) will affect the time - dependence of the shape of the rod near self - contact points just before contact ( see _ e.g. _ also , the assumption of isotropy implies that the center of mass of a deforming closed loop remains fixed @xcite , whereas in the exact theory the center of mass can move . these limitations of the simplified hydrodynamic theory do not prevent it from capturing the essential physics of the phenomena we wish to study , such as the onset of buckling instabilities and the subsequent evolution of complex shapes . it is convenient for the numerical calculations of section [ numeric ] to write the equations of motion in terms of position @xmath114 and twist strain @xmath81 . to this end , substitute the constitutive relation ( [ khirch_const ] ) into the moment balance equations ( [ mbal ] ) to find the force @xmath115 on a cross - section @xmath116 and the balance of the tangential components of the moment per unit length , @xmath117 the unknown function @xmath118 occurs in eq . ( [ feqn ] ) because the moment balance equation ( [ mbal ] ) does not determine the tangential component of @xmath115 . combining eq . ( [ feqn ] ) with the force balance equation ( [ fbal ] ) and the expressions for @xmath97 yields @xmath119-{\partial\over\partial s } \left[\lambda{\partial\mathbf{r}\over\partial s}\right ] + \mathbf{f}_\mathrm{self}. \label{r - eqn}\ ] ] equation ( [ r - eqn ] ) has the same form as the corresponding equation for the overdamped dynamics of an inextensible rod , but in eq . ( [ r - eqn ] ) the @xmath8-domain ( length ) depends on @xmath9 . to determine @xmath95 , evaluate @xmath120 using eq . ( [ r - eqn ] ) : @xmath121 one can show that the function @xmath95 in eq . ( [ lambdaeqn ] ) is identical to the lagrange multiplier function of eq . ( [ energy ] ) . to complete the determination of the dynamical equations , the torque balance ( [ tan_mom ] ) and compatibility relation ( [ comp3 ] ) yield @xmath122 where the twist diffusion constant @xmath123 . since resistive force theory includes the leading order terms in the expansion in @xmath105 of the hydrodynamic drag force and torque , our equations are asymptotically consistent . the boundary conditions which accompany eqs . ( [ r - eqn][twistdyn ] ) depend on the situation . for a filament with free ends , the appropriate conditions are @xmath85 and @xmath124 . for a closed loop , the variables @xmath114 , @xmath125 , @xmath30 , @xmath115 , and @xmath126 must be periodic in @xmath11 with period @xmath127 . we close this section by returning to the claim of section [ frame_growth_model ] that the spontaneous twist @xmath44 does not affect the shape of an inextensible rod with circular cross - section and vanishing spontaneous curvature @xmath45 . consider such a rod with @xmath56 in the stress - free state . we can eliminate @xmath44 from the problem using the freedom to redefine the material frame . if @xmath128 then @xmath129 and @xmath130 under this change of basis , @xmath131 for any @xmath132 . choosing @xmath133 fixes @xmath134 in the stress - free state . thus , @xmath135 the parameter @xmath44 has been eliminated from the constitutive relation . note that our argument up to this point holds for a growing rod as well . now consider the effect of the transformation ( [ erot1][erot2 ] ) on the compatibility relations . once again , even in the presence of exponential growth , the compatibility equations take the same form , _ e.g. _ @xmath136 for an inextensible rod , @xmath137 , since @xmath138 if @xmath139 . therefore , the change of basis ( [ erot1][erot2 ] ) does not affect the rotational drag or translational drag equations , and we conclude that @xmath44 is not a physical parameter for an inextensible rod with circular cross - section and no spontaneous curvature . however , we expect the opposite conclusion for a growing rod , since we saw in section [ frame_growth_model ] that @xmath44 has physical meaning . in fact , once @xmath44 is eliminated from the constitutive relation using eqs . ( [ erot1][erot2 ] ) , a new @xmath44-dependent term appears in the torque balance equation ( [ tan_mom ] ) : @xmath140 therefore , @xmath44 can not be eliminated from the equations of motion for this mode of growth . in this section we treat the problem of a growing elastic loop with preferred twist @xmath44 . as described in the introduction , this example is motivated by the growth of _ b. subtilis _ filaments from a spore . sometimes the ends of the growing filament stick to the spore coat , leading to a closed loop @xcite . the filament lengthens , depleting the twist , but the closed geometry prevents the rotation of cross - sections normally seen in unconstrained filaments . thus , twist stress builds up , and the filament eventually writhes and coils . although the model does not address the writhing and coiling of unconstrained filaments , it displays some of the features exhibited by _ b. subtilis _ loops . to study the stability of an exponentially growing circular loop with preferred twist @xmath44 , we begin with the unperturbed solution . the unperturbed loop lies in the @xmath141 plane and has radius @xmath142 . since each material point on the filament moves radially outward with fixed @xmath31 in unperturbed growth , the angular velocity vanishes , @xmath143 , where we use the superscript , @xmath144 , to denote the unperturbed value . the bending part of the energy ( [ energy ] ) decreases exponentially as the loop grows because @xmath145 . however , the twist energy density _ increases _ because the closed geometry prevents the cross - sections from rotating with the rate @xmath146 required to attain the twist state of zero energy . if we assume for simplicity that @xmath147 , then the exponentially increasing length leads to an exponentially decreasing twist density , @xmath148 . to summarize , the moment on a cross - section takes the form @xmath149 note that it is much more convenient to express @xmath150 in terms of the cylindrical coordinate unit vectors @xmath151 instead of the material frame vectors @xmath152 , since @xmath153 and @xmath154 continuously rotate about @xmath155 as @xmath8 increases . a choice of frame which does not rotate about the tangent vector as arclength increases is a _ frame @xcite . we will use the natural frame extensively in the linear stability analysis of section [ lin_stab_anal ] . since the filament is simultaneously bent and twisted , moment balance ( [ mbal ] ) implies a component of force in the @xmath31-direction , whereas force balance ( [ fbal ] ) leads to a tangential component proportional to the growth rate @xmath50 : @xmath156 the tangential force on the cross - section is compressive and grows exponentially in time , eventually leading to an euler - like buckling instability when @xmath157 , or @xmath158^{1/4}$ ] ( _ cf . _ @xcite ) . the euler buckling time scales as @xmath159 $ ] ( when @xmath160 , the loop begins to buckle at @xmath12 ) . in section [ lin_stab_anal ] , we will refine this estimate using our linear stability calculation and see that the appropriate buckling time at small @xmath50 actually scales as @xmath161 , since in this regime , a growing perturbation does not become noticeable until long after the perturbation begins to grow . however , our numerical calculations of section [ numeric ] reveal that the correct picture is even more complicated : for sufficiently small @xmath162 , a secondary instability arises and grows before the linear instability has significant amplitude . in addition to the euler buckling , there is also a writhing instability . the magnitude of the twist moment increases as @xmath163 as the filament lengthens , leading to a writhing instability ( like that of a twisted ring @xcite ) when @xmath164 , or @xmath165 $ ] , ( assuming @xmath166 ) . note that the critical time for writhing is @xmath167 $ ] for @xmath168 , and @xmath169 for @xmath170 . the sense of rotation of the cross - sections of unconstrained filaments determines the handedness of the coils that form after the instability of the ring : positive @xmath44 ( counter - clockwise rotation when viewed from the direction of increasing @xmath8 ) leads to right - handed plectonemic braids . note that drag is the ultimate cause for the euler buckling instability ; once growth ceases , the buckled filament relaxes back to the unperturbed circular shape . since the writhing instability arises not from drag but instead from the frustration of growth - induced twist stress , the braided post - instability shape remains after growth ceases . for small enough @xmath162 , we will see in section [ numeric ] that the small intrinsic twist biases the euler buckling , leading to solenoidal shapes , which relax to plectonemes if growth halts after a sufficiently long time . observations of the growing fibers suggest that writhing is the dominant mechanism in the instability of a closed loop @xcite . using @xmath7 erg - cm @xcite , @xmath171 s@xmath4 , @xmath172 cm , @xmath173 cm , and @xmath174 erg - s/@xmath175 ( see eqn . ( [ zetas ] ) ) leads to @xmath176^{1/4}\approx150 $ ] @xmath0 m , which is much larger than the observed critical radius and implies that growth - induced euler buckling is not important . thus , we can use the critical radius of the twisted ring to estimate @xmath44 ; assuming @xmath44 is comparable to or smaller than @xmath177 , where @xmath178 is the initial radius , and using @xmath179 cm as the experimentally observed critical radius @xcite , we find @xmath180 cm@xmath4 . this twist rate corresponds to a few turns per cell . it is intriguing to note that the corresponding length scale is close to the pitch of helical filaments of mbl , a recently discovered protein which resides near the cell wall and plays a role in maintaining the shape of _ b. subtilis _ cells @xcite . before studying the evolution of small perturbations of the growing circular shape , we consider the choice of representation . since only the shape and twist are of interest , it is convenient to use an intrinsic representation , such as the material frame . however , we saw in the solution of the unperturbed problem of section [ buck - writhe ] that the natural frame leads to further simplification . the advantages of the natural frame over the material frame are even greater for the linear stability analysis , and more generally for the full nonlinear problem @xcite . in a natural orthonormal frame @xmath182 the instantaneous rate of rotation of @xmath183 about @xmath184 is zero @xcite : @xmath185 a rotation of @xmath186 about @xmath184 by uniform ( arclength - independent ) angle leaves the condition ( [ nat_frame1 ] ) invariant ; every space curve has a family of natural frames , the members of which are related to each other by rotation through a uniform angle . to construct a natural frame from the material frame @xmath187 , rotate the material frame at @xmath8 by minus the accumulated rotation angle @xmath188 : @xmath189 the natural frame is nonlocal in the sense that deformations of the filament centerline in the region @xmath190 affect the natural frame at @xmath8 . our formulas can be further simplified with complex notation . for example , if @xmath191 , then eqs . ( [ rot_frame1][rot_frame2 ] ) become @xmath192 the vector @xmath193 is a complex normal vector . note that @xmath194 ; therefore , @xmath195 . furthermore , the defining property @xmath196 implies @xmath197 . thus , @xmath198 is proportional to @xmath184 . in analogy with the frenet - serret equation @xmath199 , we define the complex curvature @xmath200 via @xmath201 where @xmath202 also , the rate of change along @xmath8 of the unit tangent vector is the complex curvature times the complex normal vectors , @xmath203 note that rotation of a natural frame about @xmath184 by a uniform angle leads to a constant shift in the phase of @xmath200 . for example , the natural frame arising from the construction ( [ rot_frame1][rot_frame2 ] ) applied to the frenet - serret normal and binormal has @xmath204 since torsion is the rate at which @xmath205 and @xmath206 twist around the tangent vector . the corresponding complex curvature is @xmath207 the ratio of @xmath200 and @xmath208 is a constant phase . to complete the specification of the kinematics of the natural frame , we introduce the complex angular velocity describing the rate of change with time of the unit tangent vector : @xmath209 in terms of the natural frame variables , the compatibility relations ( [ comp1][comp3 ] ) become @xmath210 the integral in eq . ( [ psi - comp ] ) reflects the nonlocality of the natural frame , and arises from the temporal rate of change of @xmath211 : @xmath212\,\mathrm{d}s^\prime ; \label{theta_t3}\end{aligned}\ ] ] eq . ( [ psi - comp ] ) follows from eqs . ( [ theta_t3 ] ) , ( [ comp1][comp3 ] ) , and the identity @xmath213 in the natural frame variables , the moment @xmath126 obeys @xmath214 just as in section [ constit ] , moment balance ( [ mbal ] ) determines the perpendicular component of the force on a cross - section @xmath215 recall that @xmath216 is not determined by moment balance since only the perpendicular components of @xmath115 enter eq . ( [ mbal ] ) ; @xmath216 is determined by the condition @xmath217 . defining the tangential and perpendicular components @xmath218 and @xmath219 of the force per unit length @xmath220 it follows from eqs . ( [ d - ep ] ) , ( [ d - e3 ] ) , ( [ f - nat ] ) , and ( [ fperp - nat ] ) that @xmath221 and @xmath222.\label{f - perp - nat}\ ] ] note @xmath223 ; in terms of @xmath218 and @xmath219 the condition @xmath217 becomes @xmath224 finally , the force per unit length determines the complex angular velocity through the relations @xmath225 and @xmath226 , or @xmath227 likewise , the tangential component of the angular velocity is given by eq . ( [ tan_mom ] ) . in summary , the equations of motion for the filament in the natural frame are eqs . ( [ psi - comp][om - comp ] ) , ( [ f - par - nat][pifperp ] ) , and ( [ tan_mom ] ) . we now return to the stability analysis of a growing circular ring . we write the shape as @xmath228 and work to first order in @xmath229 and @xmath230 . it is convenient to find the equations of motion for @xmath231 and @xmath232 first , and then express these equations in terms of @xmath229 and @xmath230 . to this end , we choose @xmath233 to make @xmath234 . expanding ( [ psi - def2 ] ) to first order leads to @xmath235 since the unperturbed loop has zero torsion , @xmath236 . to express @xmath40 and @xmath41 in terms of @xmath229 and @xmath230 , use the frenet - serret equations ( [ f - s1][f - s2 ] ) and @xmath237 ( to leading order ) to find @xmath238 and @xmath239 thus , @xmath240 , with @xmath241 as described in section [ buck - writhe ] , the unperturbed angular velocities vanish : @xmath242 , @xmath243 ; furthermore , eq . ( [ fzero ] ) together with ( [ f - par - nat][f - perp - nat ] ) imply @xmath244 and @xmath245 . expanding the equations of motion to first order , we find @xmath246 where @xmath247 and @xmath248 are determined by expanding ( [ f - par - nat][f - perp - nat ] ) to first order . inspection of ( [ psi - comp - pert][pifperp - pert ] ) and ( [ f - par - nat][f - perp - nat ] ) reveals that @xmath249 , @xmath250 , @xmath251 , and @xmath252 , with @xmath253 a positive integer . this choice of origin for @xmath8 and eq . ( [ tan_mom ] ) imply @xmath254 . the perturbation @xmath255 corresponds to a translation of the ring in the @xmath141 plane ; likewise , the perturbation @xmath256 corresponds to a rotation of the ring about an axis in the @xmath141 plane . thus , the @xmath257 perturbations are rigid motions , leading to no change in curvature : @xmath258 ( see eqs . ( [ xi - eqn][eta - eqn ] ) ) . inserting the perturbations into ( [ psi - comp - pert],[om - comp - pert],[pifperp - pert ] ) and using ( [ sigma - zeta - pert ] ) to eliminate @xmath259 yields a linear system of differential equations for @xmath260 , @xmath261 , and @xmath262 : @xmath263\hat r^{(1)}\nonumber\\&&-{c\over\zeta r^3}(\omega_3^{(0 ) } -\tau_0){n^3(n^2 - 1)\over n^2 + 1}\hat z^{(1)}\label{growth - eq1}\\ \hat z^{(1)}_t&=&-{c\over\zeta r^3}(\omega_3^{(0)}-\tau_0)n(n^2 - 1)\hat r^{(1)}+\bigg[\sigma n^2-{a\over\zeta r^4}n^2(n^2 - 1)\bigg]\hat z^{(1)}\nonumber\\ & & -{c\over\zeta r^2}n\hat\omega_3\label{growth - eq2}\\ \hat\omega_{3t}&=&{c\over\zeta r^3}(\omega_3^{(0)}-\tau_0)n^2(1-n^2){\hat r^{(1)}\over r^2 } + \bigg[{an^2\over\zeta r^4}-\sigma\bigg]n(1-n^2){\hat z^{(1)}\over r^2}\nonumber\\ & & -\bigg[\sigma+{dn^2\over r^2}+{cn^2\over\zeta r^4}\bigg]\hat\omega_3.\label{om - t}\label{growth - eq3}\end{aligned}\ ] ] equations ( [ growth - eq1][growth - eq2 ] ) hold for @xmath264 , and equation ( [ growth - eq3 ] ) holds for all integer @xmath265 . for @xmath257 , the shape drops out of eq . ( [ growth - eq3 ] ) since the @xmath257 shape perturbations are rigid motions . note that @xmath44 enters ( [ growth - eq1][growth - eq3 ] ) in the combination @xmath266 ; @xmath44 drops out of the equations when @xmath139 , in accord with the general arguments of section [ change_of_basis ] . we can simplify the linear system ( [ growth - eq1][growth - eq3 ] ) by exploiting the large ratio of relaxation time scales for twisting and bending modes : @xmath267 for @xmath268 @xmath0 m and the _ b. subtilus _ parameters of section [ buck - writhe ] , @xmath269 s and @xmath270 s. as time passes and the filament lengthens , @xmath271 and @xmath272 increase exponentially , but @xmath273 for all time therefore , twist perturbations @xmath274 relax immediately , and ( [ growth - eq1][growth - eq3 ] ) may be simplified by setting @xmath274 to zero . using the initial radius for the bending relaxation time ( @xmath275 ) and assuming @xmath276 for simplicity , ( [ growth - eq1][growth - eq2 ] ) reduce to @xmath277 where @xmath278 and @xmath279.\label{linearop}\end{aligned}\ ] ] since @xmath280 depends on time , the system ( [ lineq ] ) is non - autonomous , and classical modal analysis @xcite does not apply . note also that @xmath280 is not a normal operator : @xmath281\ne0 $ ] . therefore , the eigenvectors of @xmath280 are not perpendicular , which in general signals the possibility of transient or algebraic growth of perturbations @xcite . however , @xmath280 is only weakly non - normal , since the eigenvectors are almost perpendicular . a similar conclusion applies to the problem of a ring with twist but no growth ( @xmath139 ) in a viscous fluid ( _ cf . _ therefore , we do not expect the phenomenon of transient growth of perturbations in our problem . nevertheless , the methods developed to study non - normal linear problems are well - suited to non - autonomous problems such as ( [ lineq ] ) @xcite . we will characterize the growth of perturbations by the amplification of the magnitude of @xmath282 . the optimal amplification @xmath283 is defined by maximizing this factor over all initial conditions : @xmath284 ( a more physical choice for the optimal growth rate would be to use the bending energy to second order in @xmath285 and @xmath286 instead of @xmath287 ; it turns out that either choice yields essentially the same @xmath283 . ) to compute @xmath283 , recast eqn . ( [ lineq ] ) as an equation for the propagator matrix @xmath288 : @xmath289 where @xmath290 and @xmath291 . thus , the optimal growth factor is the rayleigh quotient @xmath292 or @xmath293 , where @xmath294 is the largest eigenvalue of @xmath295 @xcite . when @xmath280 is a @xmath9-independent normal matrix , then @xmath296 , where @xmath297 is the largest eigenvalue of @xmath280 . note that since @xmath283 is maximized at each @xmath9 over all initial conditions @xmath282 , the maximum amplitudes at two different times may correspond to two different initial conditions @xmath282 . we computed @xmath283 by using standard runge - kutta techniques to solve for @xmath298 @xcite . . solid line : @xmath299 . dashed line : @xmath300 . dotted line : @xmath301 . note that the intermediate regime of rapid growth of @xmath283 is most apparent for @xmath301 , especially for @xmath302.,height=384 ] inspection of the diagonal components of @xmath280 ( [ linearop ] ) reveals that for sufficiently rapid growth , @xmath303 $ ] , the loop deforms away from its circular shape as soon as it begins to grow . when the rate of growth of the ring is sufficiently slow , @xmath304 , bending stiffness stabilizes the circular shape for @xmath305 and perturbations decay roughly as @xmath306 . thus , the growth factor @xmath307 decays extremely rapidly with increasing @xmath253 at early times . as time passes , the loop lengthens and eventually buckles , with the nature of the buckling dependent on the magnitude of @xmath44 . for @xmath308 , the distortion is the three - dimensional analog of the in - plane euler buckling studied by shelley and ueda @xcite and drasdo @xcite . for @xmath309 , the off - diagonal elements of @xmath280 are large [ see ( [ linearop ] ) ] , writhing dominates the nature of the initial distortion , and @xmath283 increases roughly as @xmath310 in the intermediate regime @xmath311 . in the late - time regime @xmath312 , @xmath313 for any value of @xmath162 or @xmath314 . these results are summarized in fig . [ taudep ] . for optimal perturbation to regain initial amplitude as a function of @xmath315 for @xmath316 ( solid line ) , @xmath317 ( dashed line ) , @xmath318 ( dotted line ) , @xmath319 ( dashed - dotted line ) , and @xmath320 ( solid line with small dots ) . the large filled dots are the results of numerical simulations of the full nonlinear equations with @xmath317 and the open squares are with @xmath318 ; see section [ numeric].,height=336 ] there are two different times which may be chosen to represent the time of the instability . the estimate for the buckling time of section [ buck - writhe ] amounts to the time @xmath321 at which @xmath283 reaches its minimum value and starts to increase . however , since the rate of decay of @xmath283 in the stable period can be very different from the rate of growth of @xmath283 in the unstable period , the time at which a perturbation becomes noticeable may be significantly greater than @xmath321 . thus , it is natural to define the time for the onset of the instability to be the time @xmath322 at which the amplitude of the optimal perturbation regains its initial value : @xmath323 . the graphs of fig . [ taudep ] suggest that @xmath322 and @xmath321 are comparable whenever @xmath315 is large enough for the intermediate growth regime of rapid growth discussed above to be present . however , if @xmath315 is small enough that this intermediate regime is absent , then @xmath322 will be much greater than @xmath321 when the extensional growth rate is slow , @xmath324 . figure [ tcrit ] shows how dramatic this difference can be . for @xmath168 and @xmath325 , @xmath326 whereas @xmath327 . when @xmath170 , both @xmath321 and @xmath322 scale as @xmath328 ( see fig . [ tcrit ] ) . thus , for small @xmath329 , there is a sharp transition in the onset time @xmath322 as a function of @xmath315 . in section [ numeric ] we will see how this prediction of the linear theory captures the early - time dynamics for @xmath170 , but that nonlinearities intervene before @xmath330 for @xmath168 . the curves for the onset time @xmath322 of fig . [ tcrit ] were computed from the linearized equations ( [ lineq ] ) using the adiabatic theorem . if @xmath324 , and if @xmath280 were normal , then the adiabatic theorem @xcite would imply that @xmath331 where @xmath332 are the dyads formed from the eigenvectors @xmath333 of @xmath334 . since @xmath280 is not normal , eqn . ( [ adiabatic ] ) is in error by an amount governed by @xmath335 , which is never more than about @xmath336 and is often much smaller . note also that if @xmath337 , then the off - diagonal elements of @xmath280 are small compared to the diagonal elements , causing the equations for @xmath285 and @xmath286 to decouple and leading to @xmath338 where @xmath339 is the largest eigenvalue of @xmath334 . thus , in both extremes @xmath337 and @xmath325 , @xmath340 . this result is especially useful in the limit of small growth rate @xmath325 , since the rapid relaxation and growth of bending modes makes it difficult to solve for @xmath288 numerically . and @xmath341 , @xmath342 , @xmath336.,height=384 ] figure [ optimal ] shows the optimal growth factors for the first two modes ( @xmath343 and @xmath302 ) for @xmath341 , @xmath344 , @xmath336 and @xmath301 . for both the intermediate writhing regime and the large-@xmath345 asymptotic regime , the rate of increase of the growth factor increases with mode number @xmath253 . as @xmath329 increases , the time at which the growth factor for the @xmath302 mode overtakes that of the @xmath343 mode decreases because the instability of each mode occurs at earlier times . for the larger values of @xmath329 , the @xmath302 mode overtakes the @xmath343 mode before the amplitudes have grown large enough for nonlinearities to come into play . thus , we expect to see double - stranded plectonemic braids with two hairpin turns for small @xmath346 , and braids with three or more hairpin turns when @xmath329 is large . the numerical computations of section [ numeric ] confirm these expectations . we solved the closed set of equations ( [ r - eqn],[lambdaeqn],[twistdyn ] ) using a pseudospectral method @xcite for the backbone dynamics ( [ r - eqn ] ) , direct integration of ( [ lambdaeqn ] ) at each time step using finite differences to find @xmath95 , and a crank - nicholson routine for the twist dynamics ( [ twistdyn ] ) . we used initial conditions such that the backbone of the loop is perturbed from circular shape with @xmath347 by a few small - amplitude modes ( @xmath343@xmath348 ) . depending on the simulation , @xmath81 ranged from somewhat less than @xmath44 to @xmath44 . [ energy_shapes ] shows a time series of the shape of the growing loop with @xmath349 , @xmath350 , and @xmath351 . for early times ( fig . [ energy_shapes]-1 ) , the circular loop is stable and perturbations decay . as the loop grows , @xmath352 increases exponentially and @xmath81 decreases . , @xmath353 , and @xmath354 at ( 1 ) @xmath355 , ( 2 ) @xmath356 , ( 3 ) @xmath357 , and ( 4 ) @xmath358 . , height=288 ] at a critical value of @xmath81 and @xmath352 ( fig . [ energy_shapes]-2 ) , the loop begins to buckle and wrap about itself . for sufficiently large @xmath162 , the loop takes on the conformation of a _ plectoneme _ , initially forming a figure - eight structure ( fig . [ energy_shapes]-3 ) and then wrapping into a braided form ( fig . [ energy_shapes]-4 ) . [ loope ] shows the twist energy , @xmath359 , and bend energy , @xmath360 , for the growing loop depicted in fig . [ energy_shapes ] . note that the total energy is not fixed in our model since growth acts to inject energy into the system . at point 1 in fig . [ loope ] , growth along the filament axis leads to a decrease of twist in time and thus an increase in the twist energy . at the same time , backbone perturbations die away and the curvature decreases exponentially , leading to a decrease in the bend energy . at point 2 , the circular loop becomes observably unstable . the bend energy begins increasing as perturbations in the filament grow , and the twist energy continues increasing ( see fig . [ loope ] ) . at the inflection point , point 3 , the filament forms a figure - eight pattern . note that a figure - eight shape of a closed loop which is not growing is a minimum of the total energy for a range of twist . at later times ( such as point 4 ) , the filament wraps into a braided structure . the bend energy increases as more braids are added . the twist energy also increases ; however , writhing motions act to decrease the twisting stress imposed by growth , leading to a twist energy that grows sub - exponentially ( see fig . [ loope ] ) . and label important times during growth ( see text for further explanation ) . , height=260 ] the dynamic equations were solved for a range of @xmath361 and @xmath329 . when @xmath361 was larger than about 410 ( the actual value depended on the value of @xmath346 ) , we found plectonemes . smaller values of @xmath362 yielded solenoids . figure [ tau0_loop ] shows how the morphology of the loop depends on the value of @xmath361 at fixed @xmath329 . in the region with plectonemes , the shapes form as a result of a writhing instability ; the euler - like buckling instability @xcite plays little role . in particular , the braided shapes remain after growth ceases . thus , the shapes are qualitatively similar to the minimizers of the elastic energy without growth . however , as discussed in section [ lin_stab_anal ] , the rate of growth affects the shape since the number of branches increases with @xmath329 . in the region with solenoids , the euler - like buckling instability comes into play since the small value of @xmath162 delays the onset of the writhing instability . solenoids are not minimizers of the elastic energy without growth ; when growth ceases the solenoids relax . therefore , the solenoids are the three - dimensional analogs of the two - dimensional shapes of references @xcite . the plectonemic , three - armed , and solenoidal morphologies we obtain are similar to many of the supercoiled patterns that are observed in _ b. subtilis _ ; in the conclusion we discuss the relation between our results and the experiments . . solenoidal morphologies occur when @xmath363 . branched plectonemes are observed for @xmath364 and unbranched plectonemes are observed when @xmath365 . @xmath366 for all figures . , height=376 ] in these calculations , we defined the time to the onset of the instability as the point where the bend energy begins increasing ( point 2 in fig . [ loope ] ) . this time should correspond roughly to @xmath322 defined in section [ lin_stab_anal ] , since the perturbations must be noticeably large before they affect the bending energy . just as in the linear stability analysis , increasing @xmath361 or @xmath50 reduces the time to onset of the instability . for values of @xmath367 , the time to onset found numerically was in quantitative agreement with the linear stability analysis ( see fig . [ tcrit ] ) . at smaller values of @xmath361 the onset time from the simulations was earlier than that predicted by the linear analysis , with the deviation getting larger for decreasing @xmath329 . the deviation is due to a secondary instability : before the growing @xmath343 mode becomes observably large , it is overtaken by a higher order mode which quickly dominates the shape of the loop . the value of the mode that dominates depends on the value of @xmath329 and @xmath361 . as @xmath361 is decreased at constant @xmath329 , the mode that dominates increases . as @xmath329 is increased at constant @xmath361 , the mode that dominates decreases . figure [ loop_phase ] shows the phase diagram and examples of plectonemes and solenoids . as a check , we verified the secondary instability by discretizing the full nonlinear equations ( [ r - eqn],[lambdaeqn],[twistdyn ] ) , and integrating with the matlab routine ode15s ( a variable order , variable time step method for stiff problems ) . and @xmath362 . for @xmath361 greater than roughly 10 , the linear stability analysis predictions are valid and the @xmath343 and @xmath302 modes dominate the pattern formation . at lower values of @xmath362 , a secondary instability drives pattern selection and higher order modes dominates . insets show the morphology of the loop in different regions of the diagram . , height=356 ] the three essential elements of any growth process appear in our model for _ b. subtilis _ fibers in a simple form : the mode of growth is exponential extension with rotation , the material properties are described by the bending and twisting elasticity of a slender filament , and the interaction with the external environment is governed by resistive - force theory . thus , our model is ideal for illustrating the basic phenomena of the physics of growth . we have assumed that the bacterial filaments can be treated as perfectly elastic and that the growth rates are uniform and independent of stress . the degree to which these simplifying assumptions hold remains an outstanding experimental problem . furthermore , our model does not lead to plectonemic structures for the case of bacterial fibers with free ends for the measured values of the growth parameters and elastic moduli . nevertheless , we can draw some general conclusions from our model and use our results to suggest new mechanisms for pattern formation in the presence of biological growth . first , even though the ultimate microscopic mechanism for the supercoiled patterns of _ b. subtilis _ fibers is unknown , the blocked rotation mechanism we study here must play a role in the supercoiling of closed fibers . for example , since the closed loops form supercoils at lengths which are much shorter than the length at which the open fibers supercoil @xcite , we expect that the blocked rotation mechanism dominates over whatever mechanism causes the open fibers to supercoil . thus , we can directly compare our calculations with the experiments on the supercoiling of closed filaments . recalling the parameters of section [ buck - writhe ] , we estimate @xmath368 , so that we expect unbranched plectonemes when @xmath369 , branched plectonemes when @xmath370 , and solenoids when @xmath371 . typical observations of the filaments yield unbranched plectonemes , but branched plectonemes and solenoids also arise , depending on the growth medium . using the observed buckling radius for an unbranched plectoneme , in section [ buck - writhe ] we were able to estimate @xmath180 cm@xmath4 , suggesting the presence of a structure in the cell wall with pitch @xmath372 cm . the observations of helical actin - like polymers in the cell wall with comparable pitch @xcite support our estimate , and further suggests a starting point for a theory for the microscopic mechanism of the supercoiling . the second major conclusion of our work is the dynamical nature of the pattern selection . for large @xmath361 , the plectonemes remain once growth ceases , and are qualitatively similar to the minimizers of the elastic energy , although the rate of growth plays an important role in determining the shape . for small @xmath361 , the solenoids are transient structures which relax away when growth halts . since the curvature of the bacterial fibers can become permanent @xcite , the solenoidal shapes may act as a template for patterns which remain in the absence of growth . this mechanism of pattern formation via the hardening of transient structures formed from the interplay of flexibility and external friction may apply to other biological systems , such as those studied in @xcite . to help justify , refine , or rule out our model , we suggest three basic experiments . first , the shape of a growing loop as function of time should be measured precisely enough to compare with our theory . although our numerical results ( fig . [ energy_shapes ] ) are qualitatively similar to the experimentally observed shapes ( fig . [ expt_loop ] ) , the lack of detailed information about the evolution in time of a single loop in @xcite prevents a stringent test of our theory . second , the change in elastic properties with time during growth should be quantitatively measured . for example , to what degree and how long must a fiber be bent to develop a permanent curvature ? finally , the nature of the twist stress in a growing fiber with free ends should be determined . is there a twist moment on the cross sections of the growing fibers with free ends , leading to a writhing mechanism qualitatively similar to the blocked rotation mechanism for closed loops , or is the mechanism for open fibers completely different ? future progress toward understanding the pattern formation of _ b. subtilis _ , both at the microscopic level of the structure of the cell wall and the more macroscopic level treated here , depends critically on new experiments such as these . we thank k. breuer , d. coombs , a. goriely , m. kim , and c.h . wiggins for useful discussions . we are particularly grateful to n.h . mendelson for discussions and the micrograph of fig . [ expt_loop ] , and a. goriely for alerting us to references @xcite and @xcite . cww was partially supported by the nsf postdoctoral fellowship in microbial biology . reg acknowledges support from nsf dmr-9812526 ; trp acknowledges support from nsf cms-0093658 . j.a . warren and j.s . , * 47 * ( 1993 ) 2702 . dickstein , s. erramilli , r.e . goldstein , d.p . jackson , and s.a . , * 261 * ( 1993 ) 1012 . jackson , r.e . goldstein , and a.o . cebers . , * 50 * ( 1994 ) 298 n.r . lebovitz and a.i . * 55 * ( 1995 ) 1117 n.h . , 46(3 ) ( 1982 ) 341 . , 66(9 ) ( 2000 ) 3657 . richard e. maxwell and violet s. nickel . , * 120 * ( 1954 ) 270 . neil h. mendelson , patrick shipman , darshan roy , liling chen , and john j. thwaites . the dynamic behavior of bacterial macrofibers growing with one end prevented from rotating : shaft rotation along the fiber s length , and supercoil migration on a solid surface toward the constrained end . , 2003 . shelley and t. ueda . the nonlocal dynamics of stretching , buckling , filaments , in _ advances in multi - fluid flows _ , papgeorgiou , y.y . renardy , a.v . coward , s .- sun ( eds . ) , siam , philadelphia , pa , 1996 .

the rosat wide field camera ( wfc ) survey of the extreme ultraviolet ( euv ) has provided us with evidence for the existence of a previously unidentified sample of hot white dwarfs in unresolved , detached binary systems . these stars are invisible at optical wavelengths due to the close proximity of their much more luminous companions ( spectral type k or earlier ) . however , for companions of spectral typea5 or later the white dwarfs are easily visible at far - ultraviolet ( far - uv ) wavelengths , and can be identified in spectra taken by iue . sixteen such systems have been discovered in this way through rosat , euve and iue observations , including four identified by us in paper i ( burleigh , barstow and fleming 1997 ) . in this paper we report the results of our continuing search during the final year of iue operations . one new system , re j0500 ( daf6/7v ) , has been identified . this star appears to lie at a distance of betweenpc , making it one of the most distant white dwarfs , if not the most distant , to be detected in the euv surveys . the very low line - of - sight neutral hydrogen volume density to this object could place a lower limit on the length of the cma interstellar tunnel of diffuse gas , which stretches away from the local bubble in a similar direction to re j0500 . in this paper we also analyse a number of the stars observed where no white dwarf companion was found . some of these objects show evidence for chromospheric and coronal activity . finally , we present an analysis of the previously known wdactive f6v binary hd27483 ( bhm - vitense 1993 ) , and show that , at t,000k , the white dwarf may be contributing significantly to the observed euv flux . if so , it is one of the coolest such stars to be detected in the euv surveys . -1_td-1 _ stars : binaries stars : white dwarfs x - ray : stars ultra - violet : stars ism : general .

astro-ph9806269

the vast majority of the @xmath52000 known white dwarfs ( mccook and sion 1998 ) are isolated stars discovered at optical wavelengths by virtue of their photometric colours or proper motions . in either case , there is a strong bias against detecting any white dwarfs in unresolved binary systems . a companion star of type k or earlier will completely dominate the optical spectrum of the white dwarf , effectively rendering it invisible . indeed , sirius b , the first white dwarf to be discovered , would never have been resolved from the a1 dwarf sirius were it not for the close proximity of the system to earth ( 2.64pc ) . prior to the rosat and extreme ultraviolet explorer ( euve ) surveys , a small number of unresolved white dwarf / main sequence binaries had been discovered serendiptously . for example , the white dwarf in the v471 tauri system was found as a result of an eclipse by its active k2v companion . in addition , a number of white dwarfs have been accidently discovered during various far - ultraviolet ( far - uv ) observations of the normal stellar companions by the international ultraviolet explorer ( iue ) , e.g. @xmath6 cap ( bhm - vitense 1980 ) , 56 peg ( schindler et al . 1982 ) and 4 @xmath7@xmath8 ori ( johnson and ake 1986 ) , although shipman and geczi ( 1989 ) systematically studied the then existing iue archive for white dwarf companions to g , k and m stars , and found no further examples . now , however , the extreme ultraviolet ( euv ) surveys of the rosat wide field camera ( wfc , pounds et al . 1993 , pye et al . 1995 ) and euve ( bowyer et al . 1994 , 1996 ) have provided evidence for the existence of a substantial sample of these previously unknown white dwarfs , through the detection of euv radiation . sixteen new systems have been discovered in this way , including @xmath3 crateris ( a2iv@xmath2wd , fleming et al . 1991 ) , kw aur c ( f4v@xmath2da , hodgkin et al . 1993 ) , hd18131 ( k0iv@xmath2da , vennes et al . 1995 ) , re j0357@xmath2283 ( k2v@xmath2da jeffries , burleigh and robb 1996 ) , and the latest two , re j0702@xmath2129 ( k0ve@xmath2da , vennes et al . 1997 ) , and hr2875 ( b5v@xmath2wd , vennes , berghoefer and christian 1997 , burleigh and barstow 1998 ) . detailed studies of groups of these objects have been undertaken by barstow et al . ( 1994 ) and burleigh , barstow and fleming ( 1997 , paper i ) . positive identifications have been made in each case through follow - up observations in the far - uv with iue , since for companions later than @xmath0a5 the hot white dwarf is easily visible at these wavelengths ( the companion to the b5v star hr2875 was identified through an euve spectrum , since the b star still dominates the spectrum at far - uv wavelengths ) . it is well established that over half of all stars are members of binary or multiple systems , yet the overwhelming majority of catalogued white dwarfs are isolated objects . the new population of optically hidden white dwarfs emerging from the euv surveys has profound implications , therefore , for our knowledge of the white dwarf luminosity function , space density and formation rate ( e.g. fleming , liebert and green 1986 ) . observations of white dwarfs in detached binary systems also allow us to place constraints on their evolutionary models ( e.g. de kool and ritter 1993 ) . in this paper we report the results of a continuing search for more of these binaries in the rosat wfc catalogue , during the final year of iue operations . we have discovered one new unresolved white dwarf@xmath2main sequence star binary ( re j0500@xmath1362 ) . in the cases where no white dwarf was detected , a number of the target stars show evidence for chromospheric activity . we also present an analysis of the previously known white dwarf@xmath2active star system hd27483 ( bhm - vitense 1993 ) . we show that the white dwarf is contributing to the euv flux and should be included in the growing list of euv - emitting optically - hidden hot white dwarfs in binary systems . the rosat wfc euv all - sky survey was conducted between july 1990 and january 1991 . two broad band filters were utilised ( designated s1 and s2 ) , and most of the count rates quoted in this paper ( see table 1 ) are taken from the 2re catalogue ( pye et al . this revised list contains 479 euv sources , as compared with 383 in the original bright source catalogue ( pounds et al . the 2re catalogue was compiled from the original survey data with improved methods for source detection , background screening etc . the resulting count rates are equivalent to on - axis , at - launch values . the rosat pspc x - ray survey was conducted simultaneously with the wfc . the soft ( [email protected] ) and hard ( 0.4 - 2.4kev ) band count rates ( table 1 ) were obtained via the world wide web from the on - line all - sky survey bright source catalogue , maintained by the max - planck institute in germany ( voges et al . 1996 ) . all of the x - ray flux from hot white dwarfs is expected to lie within the soft band . the euve all - sky survey was conducted in four pass bands between july 1992 and january 1993 . these count rates are also given in table 1 , and are mainly taken from the second euve source catalog ( bowyer et al . 1996 ) which , like the rosat wfc 2re catalogue , includes better source detection algorithms and improved reliability . hot white dwarfs in unresolved binaries with companions of spectral type k or earlier are virtually impossible to discern from optical spectra alone . however , it is possible to unambiguously identify these hidden white dwarfs in far - uv spectra taken by iue . the problem is how to select likely candidates from just their euv and soft x - ray count rates . most of the hot white dwarfs discovered by rosat have very soft spectra compared to normal stars , particularly where the interstellar hydrogen column density is low . the ratio of the wfc s2 to s1 count rates can often exceed a factor of two . additionally , no photons are usually detected from a white dwarf above the 0.28kev carbon edge of the rosat pspc . all other x - ray and euv - emitting astronomical objects generally have spectra extending to higher energies . thus , many of the stars originally selected for observation by iue were relatively bright euv sources with very similar colours to known hot white dwarfs ( e.g. kw aur c , hodgkin et al . the success rates of the early searches by , for example , barstow et al . ( 1994 ) were very high , and appeared to represent the tip of an iceberg . however , many of the white dwarfs in the rosat wfc survey are relatively faint euv sources , indistinguishable by count rate ratios alone from coronally active objects . further selection criteria , in addition to simple euv colour and brightness , needed to be applied . in paper i ( burleigh , barstow and fleming 1997 ) we conducted a search for fainter , less obvious examples of these binary systems , with a @xmath040% success rate . candidates were selected , in particular , from normal stars that had been observed in the wfc optical identification programme ( mason et al . 1995 ) , and where little or no evidence of activity had been found . in some cases , e.g. hd2133 ( re j0024@xmath1741 ) , a hidden white dwarf was indeed detected in an iue swp spectrum . in others , e.g. hr2468 ( re j0637@xmath1613 ) , the iue observations revealed evidence of chromospheric and coronal emission that had eluded the optical identification team . in this paper we report the results of a further search for hidden white dwarfs during the last year of iue operations ( 1995/96 ) . thirteen candidates were selected and observed , including many that were detected for the first time in the reprocessed rosat wfc 2re data . these are , in general , faint euv sources . in most cases , the target stars were not known to be active , and had s2/s1 count rate ratios @xmath52 . in a separate optical programme , a number of the unidentified 2re fields were observed to try to determine the counterpart to the euv source . in some cases , e.g. re j0500@xmath1364 , 2re j0222@xmath2503 and 2re j0232@xmath1025 , only one relatively faint star ( v@xmath510 ) was visible in the field . although a few euv sources ( mainly red dwarfs ) are active stars with v@xmath510 ( e.g. proxima cen , m5ve , [email protected] ) , most are much brighter than this . thus , for these sources , a hidden hot white dwarf companion was a feasible explanation for the euv radiation , and they were added to the iue target list . a log of all the far - uv observations is given in table 2 . the white dwarf companion to hd27483 was serendipitously discovered in october 1992 by bhm - vitense ( 1993 ) , independently of the rosat survey , during an iue survey of hyades f stars . re j0500@xmath1364 was first observed in a short half - hour exposure in november 1995 ( swp56217 ) , and a faint white dwarf was detected . a follow up spectrum , with an 8 hour exposure to achieve the required signal - to - noise , was obtained a month later . the programme was unfortunately cut short in february 1996 when a gyro failed on iue , limiting observations to objects for which a bright ( v@xmath108 ) guide star was available . this was not the case for any of our remaining targets . two spectra were affected by the so - called ` 159dn anomaly ' ( see table 2 ) . a number of pixels were assigned the flux value 159dn when they should in fact have had a higher value . however , on close inspection it was found that only a small fraction of each of these two spectra was contaminated , and their scientific usefulness was not compromised . all of the spectra have been processed with newsips ( new spectral image processing system ) . newsips spectra contain a number of significant geometrical and photometric corrections which enhance spectral signal - to - noise and improve the photometric reliability of the data ( e.g. a correction for the degradation of the detector with time , bohlin and grillmair 1988 ) . recent analysis by garhart ( 1997 ) shows that , since 1993 , the swp camera sensitivity has in fact degraded at a rate greater than predicted , so that the current newsips calibration underestimates the swp fluxes by up to 5% . archival iue data from 1993 - 96 are now being reprocessed to include a new degradation correction . however , this data was not available to us when we were preparing this paper , although we would expect the effect of the new calibration to be relatively minimal on the determination of any white dwarf parameters ( for example , the errors on the flux values for the re j0500@xmath1364 iue swp data are typically of order @xmath010% anyway ) . the method used to analyse the far - uv and euv data for the hidden white dwarfs and the active stars detected and observed during this search has been discussed in detail in paper i ( burleigh , barstow and fleming 1997 ) . a summary is given here . in binary systems like these , it is not possible to use the h balmer line profiles to measure temperature and gravity . however , the iue swp spectra can be used to estimate these parameters by matching the observed lyman @xmath11 profile and the uv continuum ( the region uncontaminated by the companion ) with synthetic spectra . unfortunately , it is not possible to get an unambiguous determination of t and log g from a single spectral line . instead , a range of possible models is determined by stepping through values of the surface gravity , from log [email protected] to 9.0 , and finding the best fit temperature and normalisation at each point . we compare the observed far - uv data with fully line blanketed , homogeneously mixed h@xmath2he , lte model atmospheres , spanning a temperature range from 20,000k to 100,000k , and kindly supplied by detlev koester ( e.g. koester 1991 ) . in this analysis , we assume the white dwarfs have pure hydrogen envelopes . the spectral fitting is conducted with the xspec programme ( schafer et al . 1991 ) , which calculates a chi - squared statistic for the fit between the data and the model , and which is then minimised by incremental steps in the free parameters . there is no need to take into account any interstellar component in the analysis of the lyman @xmath11 profile , because for columns greater than a few @xmath1210@xmath13 the white dwarf is unlikely to be detected in euv surveys . the white dwarf radius and mass are calculated using the temperature - dependent evolutionary models of wood ( 1995 ) , which assume thick ( 10@xmath14m@xmath15 ) h layers , and the radii are then used to estimate the distance from the model normalisation parameter ( which is actually the solid angle of the star , equivalent to [ radius / distance]@xmath16 ) . the distances to the primaries in each case can be calculated from the spectral type and v magnitude , and are given in table 3 . the range of white dwarf temperatures and gravities that give the best match to the distance of the primary can then be estimated . the v magnitude of each white dwarf is estimated from the model flux at 5500 , and given in table 5 . once the temperature and gravity of each white dwarf has been estimated , the rosat euv and soft x - ray fluxes can give an indication of the level of photospheric opacity in the stellar atmosphere , by comparing them with predicted values for a pure h model . the rosat data is fitted independently from the iue data since contamination from elements heavier than h and he only has an effect at euv and soft x - ray wavelengths . we fit the data from the two wfc filters , and the integrated count rate in the [email protected] kev pspc band , within which all the white dwarf soft x - ray flux is expected to lie . it is possible that some euv and x - ray emission might also originate from the normal stars in each system . any detected pspc flux at energies above 0.4 kev is an indication of the presence of an active companion ( table 6 ) . once again we utilise detlev koester s fully line blanketed h@xmath2he models . these assume a homogenous distribution of hydrogen and helium , under lte conditions , in the range @xmath18@xmath10loghe / h@xmath10@xmath13 . the temperature , gravity and normalisation , estimated from the fit to the iue data , are frozen during the modelling , but the he / h ratio is allowed to vary . the interstellar hi column density is also estimated by assuming that the local ism is not highly ionised ( i.e. there is minimal heii absorption ) and that the hei / h ratio is cosmic ( 0.1 ) . the fitting is again conducted using the xspec programme . we consider a good fit to the data to correspond to the probability that a particular value of the reduced @xmath17@xmath16 ( @xmath17@xmath18@xmath16 ) can occur by chance to be 0.1 or greater ( i.e. 90% confidence ) , and a bad fit 0.01 or less ( 99% confidence ) . the fits in between might not be very good , but can not be ruled out with high confidence . in this analysis a good fit requires @xmath17@xmath18@[email protected] , but until it exceeds 6.63 a model can not be excluded with any certainty . consequently , we list all model fits to the data for which @xmath17@xmath18@[email protected] ( table 6 ) . a number of the stars observed with iue where no white dwarf was detected are probably coronally and chromospherically active . some of these stars are also comparatively hard x - ray sources ( i.e. they are detected in the [email protected] kev band of the rosat pspc ) , and in some of the iue lwp spectra chromospheric mgii emission was seen at 2798 . in these cases , estimates of the l@xmath19/l@xmath20 and l@xmath21/l@xmath20 ratios have been made as an indicator of the level of activity ( table 7 ) . we adopt the methods outlined by jeffries ( 1995 ) and fleming et al . ( 1995 ) to estimate these parameters . where chromospheric mgii emission was seen the line fluxes above the continuum level were measured using a simple gaussian profile , fitted to each line , after the continuum had first been subtracted ( the continuum flux was represented by a low degree polynomial ) . note that in each case the mgii emission actually fills - in an underlying absorption dip . however , it is impossible to estimate the depth of this absorption line , and thus the measurements presented here are only lower limits . the measured mgii fluxes are listed in table 7 , and have been used to estimate l@xmath22/l@xmath20 . the field of this euv source ( see figure 1 ) was surveyed by mason et al . ( 1995 ) during the wfc optical identification programme . no evidence of activity was found in the cores of ca ii h & k in the spectrum of a @xmath013th magnitude star located at the centre of the source error box . various stars were also examined outside the error box , but none could plausibly account for the euv emission . the field was also examined on 1995 september 29/30 with the 2.3 m steward observatory telescope on kitt peak , as part of a programme to search for the optical counterparts to unidentified sources in the rosat wfc 2re catalogue . a spectrum ( figure 2 ) of the central star in the source error box was obtained with the boller and chivens spectrometer and 800@xmath121200 blue sensitive ccd . a 2.5 " slit and 600 lines / mm grating blazed at 3658 were used , and the data were reduced with standard iraf routines . again , there was no evidence for activity . one untested proposition was that this central object could be hiding a hot white dwarf companion . therefore , the star was added to our iue target list for observations during 1995/96 , and a faint , @xmath418th magnitude hot white dwarf companion was discovered ( figure 3 ) . by comparing the relative line strengths and widths of the primary s optical spectrum with spectra in the atlas of jacoby , hunter and christian ( 1984 ) , we conclude that it most closely resembles an f6/7v . it should be noted that the spectrum appears to be deficient in flux at the blue end , below @xmath04500 . as the star was observed at a relatively high air mass ( sec [email protected] ) , this deficiency in counts was probably caused by differential atmospheric dispersion . the spectral identification is strengthened when the iue lwp spectrum ( lwp31729 , figure 4 ) is compared to spectra in the iue spectral atlas ( wu et al . the data are very noisy , and have been binned up as a consequence , but most closely match stars in the range f5v@xmath1g0v . there also appears to be a slight excess of flux in the iue swp spectrum , at wavelengths @xmath51850 , above the level expected from the white dwarf ( figure 3 ) . if the excess is real , then it must attributable to the primary . as @xmath013th magnitude mid - late g and k stars are not bright enough to be detected by iue in this region of the spectrum , this also indicates that the companion is probably a late f. taking the guide star catalogue magnitude of [email protected] , the system lies between 755 ( f7v ) and 870 ( f6v ) parsecs away , although if we consider possible errors on the gsc magnitude ( @xmath230.3 mags ) , and that the absolute magnitudes may be in error by up to @xmath230.5 magnitudes , the system could be as close as 520 parsecs or as distant as 1250 parsecs . is this a true binary or a chance alignment ? given that the source of the euv radiation is almost certainly the hot white dwarf , we can calculate the probability of a random 13th magnitude star also falling within the iue aperture . according to allen ( 1973 ) , the number of stars per square degree at the galactic latitude of re j0500@xmath1364 , brighter than v@xmath913 , is 87.1 . the iue large aperture is a 10``@xmath1220 '' oval , hence we calculate a probability of @xmath41/250 of a chance alignment . ( note that the point spread function of the stellar image of the 13th magnitude star on the digitised sky survey plate is @xmath015 " in radius , see figure 1 ) . assuming this is indeed a true binary , then we believe re j0500@xmath1364 to be one of the most distant white dwarfs , if not the most distant , to be identified in the euv surveys ( for a comparison , see the distance estimates for the white dwarfs detected by the rosat pspc by fleming et al . 1996 , and those detected by euve by vennes et al . . a model fit to the iue spectrum at log [email protected] gives t@xmath938,260k and [email protected]@xmath15 , corresponding to a distance of 830 parsecs . a fit at log [email protected] corresponds to the minimum distance estimate to the system , but gives a higher mass , [email protected]@xmath15 and higher temperature , t@xmath945,000k ( and brings the star into the regime where its atmosphere may be contaminated by elements heavier then he ) . the inferred v magnitude from these models is 17.9 , making this one of the faintest white dwarfs to be detected by rosat . the faintest , re j0616@xmath1649 ( [email protected] ) , is a rare magnetic da ( jordan 1997 ) , but there are few other hot white dwarfs detected fainter than 17th magnitude . this source is not included in the rosat pspc all - sky survey bright source catalogue ( voges et al . 1996 ) , but it has been observed and detected in x - rays during a pointed observation by the rosat high resolution imager ( hri ) . the target was observed as part of programme to identify possible isolated neutron star candidates in the rosat surveys ( pi wang ) . using the point source search programme ( pss ) within the starlink asterix x - ray analysis package ( see pye et al . 1995 ) to analyse the hri image , the position of the x - ray source was found to be coincident with the wfc and euve detections ( figure 1 ) , and has a flux of [email protected] counts per 1000 sec@xmath24 . the source shows no evidence for time variability . although the hri has very limited spectral response ( the spectral properties of the hri are known to vary with detector position and time in an ill - defined manner ) , it is possible to obtain a crude hardness ratio . in this case , the hri data can then be used to test whether there is any hard x - ray emission from the f6/7v companion star , since no emission is expected from the white dwarf photosphere above 0.4 kev . figure 5 shows the hri data plotted as a function of pulse height distribution . the exact energy of each channel varies temporarily and spatially , but channels 1 to 5 are roughly equivalent to the soft ( [email protected] kev ) pspc band , and channels 6@xmath116 are equivalent to the hard ( [email protected] kev ) band . as with the pspc , a hardness ratio can be defined by ( hard@xmath1soft)/(hard@xmath2soft ) . in this case , the hardness ratio @[email protected] confirming , as can be readily seen in figure 5 , that this is a very soft source . we conclude that all of the euv and soft x - ray emission originates from the hot white dwarf . unfortunately , there is no reliable detector matrix for the hri , and it did not prove possible to use the x - ray count rate in the subsequent analysis . therefore , since this left only the two rosat wfc data points to try to constrain the interstellar hydrogen column density and white dwarf atmospheric parameters , we assumed that the white dwarf photosphere was essentially pure hydrogen ( a reasonable assumption for t@xmath1040,000k , and for many hot das t@xmath1050,000k ) , and allowed only the neutral hydrogen column density to vary in the subsequent fitting of the euv data . we found that the wfc s2 photometric data point is well matched by a model fit at log [email protected] , although the s1 flux was predicted to be slightly lower than observed . the hydrogen column density given by this model is n@[email protected]@xmath1210@xmath26 atoms @xmath27 . at 830 parsecs distance , this translates into a neutral hydrogen line - of - sight volume density of only 0.0029 atoms @xmath28 , well below the average local volume density within @xmath080pc of the sun of 0.05 atoms @xmath28 ( warwick et al . if the system is closer ( 530 parsecs ) and the log [email protected] model is applied , then the h column density is higher ( n@[email protected]@xmath1210@xmath13 atoms @xmath27 ) and the line - of - sight volume density is also higher at 0.0146 atoms @xmath28 . again , though , this is well below the average local volume density . notably , this system ( galactic co - ordinates [email protected]@xmath29 , b@[email protected]@xmath29 ) lies in a similar direction to the known exceptionally low column densities towards the b1 giant @xmath3 cma ( l@xmath9226@xmath29 , b@xmath9@xmath114@xmath29 , welsh 1991 ) , and in particular to two other rosat - discovered hot white dwarfs , re j0457@xmath1281 and re j0503@xmath1289 ( [email protected]@xmath29 , b@[email protected]@xmath29 , barstow et al . 1993 ) . at @xmath4200pc distance , @xmath3 cma is known to exist in a rareified `` interstellar tunnel '' of very low neutral gas density , which is itself an extension of the region surrounding the sun called the local bubble ( welsh 1991 ) . the features of the bubble were first mapped out by frisch and york ( 1983 ) , and welsh ( 1991 ) speculates that the @xmath3 cma tunnel may extend for _ at least _ 300pc in that direction away from the sun . welsh ( 1991 ) also estimates the tunnel to be about 50pc in diameter ; the white dwarfs re j0457@xmath1281 and re j0503@xmath1289 ( which are much closer at @xmath490pc ) may then possibly exist in a southward extension of this tunnel , or more likely lie in the foreground , within the local bubble itself . table 8 details the known column densities and volume densities to these three stars , and also to @xmath30 cma ( b2 ii , [email protected]@xmath29 , b@[email protected]@xmath29 ) , the brightest euv source in the sky ( cassinelli et al . 1995 ) , which lies in a similar direction . the line - of - sight neutral hydrogen volume density we measure for re0500@xmath1364 compares favourably with the average volume density to these four stars ( 0.0030 atoms @xmath28 ) . this suggets that re j0500@xmath1364 might also lie within the @xmath3 cma extension of the local bubble . if re j0500@xmath1364 really does lie as far as @xmath0500 or more parsecs away , then it presents a possible lower limit to the size of any neutral gas - free corridor stretching away from the local bubble in that direction . the region is bounded on three sides by several ob associations : the orion nebula ( 450pc away ) , the cma ob1 association ( 800pc distant ) , and the gum nebula ( 290pc away ) . welsh ( 1991 ) hypothesises that this tunnel may have been evacuated by a number of supernova explosions in the last few @xmath1210@xmath31 years . the injection of driven , heated , rarefied gas into an older ( @xmath010@xmath32 years old ) low density cavity in the local interstellar medium would produce the large region of very low density neutral gas that we now see . the hyades system hd27483 consists of two active f6v stars orbiting each other with a period of 3.05 days . although an euv source was detected originating from the direction of this system in the rosat wfc all sky survey in 1990 , the hot white dwarf component was identified independently and serendipitously by bhm - vitense ( 1993 ) during an iue swp observation as part of a survey of hyades f stars ( bhm - vitense 1995 ) . bhm - vitense ( 1993 ) derived atmospheric parameters for the white dwarf of 23000@xmath231000k and [email protected]@xmath15 . however , in the analysis of the white dwarf spectrum the author utilised the unblanketed models of wesemael et al . ( 1980 ) , assuming log [email protected] , to fit the far - uv spectrum at just two points . since the white dwarf might be hot enough to be contributing to the observed euv flux , it could be argued that the system should be included on the list of euv - detected hidden white dwarfs , and thus we have decided to re - analyse the far - uv and rosat data here , in the same manner as the other recently discovered systems . the rosat wfc source re j0420@xmath2138 , associated with hd27483 , was listed in the original bright source catalogue ( pounds et al . 1993 ) , with a count rate of [email protected] counts ksec@xmath24 ( it was not detected in s2 ) , but was not subsequently detected in the reprocessed 2re survey ( pye et al . the significance of the s1 detection in the 2re survey was 4.2 ; sources had to exceed a significance of 5.5 in a combination of both bands to be included in the catalogue . even so , at t@xmath423,000k the white dwarf may be contributing to this small euv flux , despite the fact that the two f star companions are known to be active themselves . there is significant contamination in the iue swp spectrum ( see figure 6 ) at the long wavelength end from the two f6v star companions , but this falls to zero by 1600 . therefore , we were able to use the continuum flux up to this point . bhm - vitense estimated a distance of 47.6 parsecs to this system . we can further constrain this figure with the recently published _ hipparcos _ parallaxes ( esa 1997 ) , where the measured value for hd27483 is [email protected] , corresponding to a distance of [email protected]@[email protected] parsecs . the spectral model which best matches this distance has log [email protected] and t@xmath922,000 , and the corresponding stellar mass is [email protected]@xmath15 . the white dwarf age is then 1.4@xmath1210@xmath34 years , in comparison with the age of the hyades cluster , @xmath07@xmath1210@xmath34 years ( bhm - vitense 1993 ) . given that the white dwarf might have a higher mass than is the average for these stars ( @xmath40.6m@xmath15 , marsh et al . 1997 ) , we can estimate its progenitor mass and place a possible lower limit on the maximum mass for white dwarf progenitor stars in the hyades cluster . from wood ( 1992 ) : @xmath35@xmath9aexp(b@xmath12@xmath36 ) where [email protected]@xmath15 and [email protected]@xmath15@xmath24 . we find , for @[email protected]@xmath15 , @[email protected]@xmath15 . the main sequence lifetime of a 6.7m@xmath15 star is in fact significantly shorter than 560 million years ( the difference between the hyades age and the white dwarf cooling age , e.g. schaller et al . 1992 ) . this suggests that the white dwarf is , in reality , probably lower in mass than 0.94m@xmath15 . however , in order to unambiguously and tightly constrain the fundamental parameters of this star , we will need to obtain a spectrum of the h lyman series with an instrument such as the forthcoming _ fuse _ mission ( see also section 6 ) . the wfc source is coincident with a rosat pspc x - ray source , with a total count rate of [email protected] counts ksec@xmath24 , including a detection in the upper band . this confirms that at least one of the two f6v companion stars is active , as the hard x - rays could not have originated from the white dwarf . the rosat data points can not be matched with any of the white dwarf models ( which assume a homogeneous atmospheric mixture of h and he ) . this implies that the active star(s ) must be providing a significant fraction of the euv flux , since little or no heavy element contamination is expected in the white dwarf photosphere in this cool temperature regime . it is even possible that there is no flux at all from the white dwarf at these wavelengths . the contribution of the white dwarf to the s1 count rate can , however , be estimated . another hot wd@xmath2ms binary , v471 tauri , is detected by rosat in the hyades cluster ( barstow et al . 1992 ) . after subtracting the contribution from the active k2v companion , marsh et al . ( 1997 ) use the wfc count rates to estimate the h column density to this system ( 8.52@xmath1210@xmath26 atoms @xmath27 ) . adopting the same column density to hd27483 , assuming a pure h atmosphere , and using the parameters derived from the log [email protected] model , the white dwarf is predicted to contribute 5.4 counts ksec@xmath24 to the s1 flux ( i.e. @xmath41/3 of the 15 counts ksec@xmath24 detected ) . how is this count rate affected by uncertainties in the h column density ? in fact , from euve spectra , dupuis et al . ( 1997 ) derived a much lower h column density to v471 tauri of 1.5@xmath1210@xmath26 . using this value , in the log [email protected] model , the white dwarf contributes 7.6 counts ksec@xmath24 to the s1 flux ( i.e. @xmath41/2 of the observed flux ) . this hot degenerate companion to hd27483 could itself , then , be regarded as a real euv source . at t@xmath422,000k , this would make it one of the coolest white dwarfs to be detected in the euv surveys . the combined x - ray luminosity of the two f6v stars in the hd27483 system can also be estimated , by subtracting the contribution of the white dwarf to the rosat pspc lower band flux . the pspc count rates are 98@xmath2314 counts ksec@xmath24 in the softer [email protected] band , and 33@xmath239 counts ksec@xmath24 in the harder [email protected] band ( voges et al . 1996 ) . in the log [email protected] model , assuming a column density of 8.52@xmath1210@xmath26 atoms @xmath27 , the white dwarf flux in the [email protected] band is found to be 36.0 counts ksec@xmath24 . eliminating this from the total pspc lower band rate and following the method detailed by fleming et al . ( 1995 ) , we find l@[email protected]@xmath1210@xmath37 ergs sec@xmath24 , and l@xmath21/l@[email protected]@xmath1210@xmath38 . this unclassified star was only observed by the swp camera ( swp56333 ) , and there was no flux visible above the background . if bd@xmath249@xmath29646 is a g or k star , then we probably would not detect it in this waveband anyway . no emission features are visible in the uv spectrum , but the euv source is coincident with a rosat pspc hard x - ray source , and the possibility must remain that bd@xmath249@xmath29646 is coronally active . alternatively , the euv / x - ray source might be another object in the field , or it is possible that the target was missed altogether in the iue observation . as with the wd@xmath2ms binary re j0500@xmath1364 ( discussed above ) , the field of this wfc source was originally observed in 1995 with the 2.3 m steward observatory telescope at kitt peak , as part of a programme to try to identify the remaining optical counterparts to unknown euv sources in the rosat wfc catalogues . in the absence of any plausible euv source , the 15th magnitude g - type central star in the error box may be hiding a hot white dwarf companion . the far - uv spectra obtained with iue ( swp56272 and lwp31800 ) were very noisy and showed no evidence for a hot white dwarf . there was some flux above the background longwards of @xmath02700 in the lwp spectrum , which may have been due to a g star , but it is possible that the target was missed altogether , and this flux was due to scattered solar light which effects the lwp camera sporadically . at [email protected] this star is unlikely to be the source of the euv flux : a 15th magnitude main sequence mid - g star would require l@xmath19/l@[email protected] to produce the count rate seen in the wfc s2 filter ( 45 counts ksec@xmath24 ) , far in excess of the saturation level for coronal emission ( l@xmath19/l@xmath20@xmath010@xmath39 , mathioudakis et al . 1995 ) . prior to the publication of the 2re catalogue ( pye et al . 1995 ) , the 9th magnitude f6v star sao150508 was not thought to be active , although there are no published optical observations which might offer evidence one way or another . the iue spectra ( swp 56195 and lwp31700 , figure 7 ) show no evidence for a hot white dwarf companion . sao150508 is , however , coincident with a rosat pspc x - ray source . therefore , estimates of the x - ray and euv luminosities , assuming sao150508 is active and the true source of the euv and x - ray flux , are presented in table 7 . there is no evidence in the literature that the 8th magnitude g2v star hd36869 is active . the iue spectra ( swp56169 and lwp 31701 , figure 8) also show no evidence for activity , although the star is coincident with a rosat pspc source . given that 8th magnitude stars are comparatively rare , it is still possible that hd36869 is coronally active , and thus we provide estimates of the x - ray and euv luminosities in table 7 . gl216b ( k2v ) is part of a nearby ( @xmath48 parsecs ) triple system , and was chosen as a candidate white dwarf binary on the basis of the s2/s1 count rate ratio . the iue spectra ( swp56194 & lwp31699 , figure 9 ) show no evidence for a hot white dwarf . however , mg ii 2798 is visible in emission ( with a line flux of [email protected]@xmath1210@xmath40 ergs @xmath27 s@xmath24 above the continuum ) . the emission feature in the swp spectrum at @xmath01720 is probably spurious , since it does not coincide with any commonly seen line . from observations made in the optical , de strobel et al . ( 1989 ) concluded that this is a young , active star . if it is the only source of the euv flux , then we determine l@xmath19/l@[email protected]@xmath1210@xmath41 . active stars have a characteristic euv to mg ii flux ratio . for example , jewell ( 1993 ) shows that the [email protected] kev euv flux is 1@xmath110 @xmath12 the mgii flux . for gl216b , though , the euv to mg ii flux ratio is only @xmath40.8 . schmitt et al . ( 1990 ) observed the entire gl216 system in an einstein hri pointing , and found that the nearby f7v star gl216a was 8 times brighter in x - rays than gl216b . thus hodgkin and pye ( 1994 ) concluded that all of the euv radiation in fact comes from gl216a . however , we only targeted gl216b with iue since this is the object associated with the euv source in the 2re catalogue , and at the time of the observation we were unaware of hodgkin and pye s conclusion . the low euv to mg ii flux ratio for gl216b does , though , support these earlier conclusions that the major source of the euv radiation is actually gl216a . estimates of the x - ray and euv luminosities are given in table 7 assuming a ) all the flux comes from gl216a and b ) all the flux comes from gl216b . this g5v star was not known to be active prior to the rosat survey , but subsequently it has been studied in detail by jeffries and jewell ( 1993 ) and is almost certainly the euv source . it is also an x - ray source , and measurements of the x - ray and euv luminosities are given in table 7 . no obvious emission features are visible in the iue lwp and swp spectra ( figure 10 ) ; the feature longwards of 1800 in the swp spectrum is probably spurious , as there is no commonly seen line at this wavelength . there are no references in the literature to this being an active star . however , mg ii is clearly seen in emission at 2800 in the iue lwp spectrum ( figure 11 ) , and there is a suggestion of civ in emission at 1550 in the short wavelength region . measurements of the x - ray and euv luminosities are given in table 7 using the g0 classification given by simbad , and assuming the star is on the main sequence . the euv to mg ii flux ratio ( @xmath48.0 ) strongly suggests that this star is active and the true source of the euv radiation . note that the ratios l@xmath19/l@xmath20 and l@xmath42/l@xmath20 are significantly larger than for any of the other stars in this sample , approaching the saturated level for coronal emission ( @xmath010@xmath39 ) . this suggests that the star may be rapidly rotating . no references are given in the literature to this being an active star . there is clearly no white dwarf visible in the iue swp spectrum ( figure 12 ) , and the emission line at @xmath01800 is probably spurious , perhaps due to a cosmic ray hit . the euv source is not coincident with an x - ray source . the star needs to be examined optically to search for any evidence of chromospheric activity . again , this star was not known to be active prior to the rosat survey . the iue lwp spectrum ( figure 13 ) reveals mg ii in emission at 2800 . measurements of the euv and x - ray luminosities ( this star is also a pspc source ) are given in table 7 , assuming the k0 spectral type given by simbad , and that the star is on the main sequence . the euv to mg ii flux ratio ( @xmath48.25 ) strongly suggests that this star is active and the true source of the euv radiation . with s1 and s2 count rates of 52@xmath237 and 83@xmath239 counts ksec@xmath24 , this is a relatively bright euv source in comparison with most of the targets in this paper . the soft x - ray and euv photometric colours are also characteristic of a hot white dwarf , and it is not detected in the pspc hard band . therefore , it was selected as a potential hidden white dwarf binary . the [email protected] star in the centre of the field , hd70907 ( f3iv / v ) , was observed in both the iue swp and lwp cameras ( figure 14 ) . there is no evidence for a white dwarf companion or emission features indicative of an active star . a nearby v@xmath411 star was also observed and again there was no evidence for a white dwarf , although , in the absence of any flux in the lwp camera that could be attributed to a stellar source , it seems possible that the star was not in the lwp slit . mason et al . ( 1995 ) report that this fainter object is indeed active , although they give no indication of the size of any emission features seen in the optical . they also do not give a spectral type for this star , and thus we have not been able to determine the euv and x - ray luminosities . whether this star is active enough to be the true euv source remains unclear , and the suspicion remains that there is indeed an unresolved hot white dwarf hiding in this field . it is highly unusual to detect an a star in euv or x - ray surveys ( fleming et al . observations by the einstein and exosat observatories failed to find any convincing detections other than the nearby quadruple a star system castor ( pallavicini et al . therefore hr4646 , an am star coincident with rosat wfc 2re and pspc sources , was selected as a potential hidden white dwarf binary . it should be noted that am stars do not possess significant magnetic fields and they are slow rotators , but they almost always appear to lie in close binary systems ( abt 1961 ) , and indeed margoni , munari and stagni ( 1992 ) found that hr4646 is a spectroscopic binary with a period of 1.27 days . the iue swp spectrum ( figure 15 ) shows no evidence for a hot white dwarf companion . it is extremely likely , therefore , that hr4646 has a coronally active cooler companion ( f5 or later ) . a search for unresolved white dwarfs in binary systems with optically brighter normal stellar companions has been conducted with iue during its final year of operation . targets were chosen among the fainter and still un - identified euv sources in the rosat 2re catalogue ( pye et al . one new system was discovered ( re j0500@xmath1364 , da@xmath2f6/7v ) , which appears to lie in a direction of low interstellar neutral hydrogen volume density . if this star is really as distant as @xmath0500 - 1000pc , then it may represent a lower limit to the size of the @xmath3 cma tunnel of low density neutral gas stretching away from the local bubble in that direction . including the independently identified hd27483 ( da@xmath2f6v , bhm - vitense 1993 ) , this new discovery brings the total number of such systems found in the rosat and euve surveys to nineteen . these two satellites have , therefore , been very successful in helping us to identify these kinds of binary systems , which have never been seen optically and could only have been identified through such satellite surveys . however , these stars still represent @xmath1020% of the hot white dwarfs identified in the euv . the demise of iue , which suffered a gyro failure in february 1996 , limiting it to observe only targets with bright guide stars ( the mission was finally terminated in september 1996 ) , effectively cut short our programme . the loss of iue means that this particular method of searching for these important systems is no longer available . although hst can observe the same wavelength region , it is of course much harder to obtain the time required for this kind of search . it is likely , therefore , that very few new examples of these systems will be discovered in the near future . the results presented in this paper seem to suggest that this search , using the euv catalogues as a basis from which to identify potential systems , has been fairly exhaustive ( we have identified only one new system from thirteen targets ) . however , the identification by vennes et al . ( 1997 ) of a hot white dwarf companion to a previously catalogued euv - bright active star , re j0702@xmath2129 ( k0ve ) , instead suggests that , in fact , some systems may have been completely overlooked . re j0702@xmath2129 was classified as active by mason et al . ( 1995 ) in their optical follow - up programme to identify the euv sources found in the rosat wfc survey , and was not , therefore , included on any of our target lists for the iue search . this raises the question of how many of the @xmath5200 ` active ' stars in the wfc and euve catalogues really are the source , or at least the only source , of the euv radiation . until each star has been observed and analysed in detail , the suspicion remains that there are more of these binary systems in the survey waiting to be identified . for example , the rosat wfc catalogue includes @xmath4120 isolated white dwarfs , and @xmath460 in some kind of binary , e.g. classic sirius - type systems , cvs , non - interacting da@xmath2 dm systems , and visual binaries . conservatively , then , we might assume that we have already identified the majority of the white dwarf binaries to be found in the survey . however , if as many as 80% of stars reside in binary or multiple systems , another 30 might be awaiting discovery . these could include double degenerate systems as well as further examples of sirius - type systems ( e.g. the apparantly isolated da re j0512@xmath1007 has a mass , [email protected]@xmath15 , too low for it to be the result of single star evolution , and may have a degenerate companion ) . what is really needed to try to find more sirius - type binaries is an all - sky uv survey . none has been undertaken since the td-1 survey of 1972/73 , which originally appeared to have only detected sirius b among these systems . subsequently , though , landsman , simon and bergeron ( 1996 ) found white dwarf companions to 56 persei and hr3643 as a result of the uv excess detected by td-1 . the recently approved _ galex _ mission ( bianchi 1998 ) , which will survey the sky at uv wavelengths and follow - up some targets spectroscopically , may reveal many more of these binaries . in the meantime , follow - up observations of these nineteen new euv - bright systems are required . in particular , it is important to determine whether these systems are wide , or close enough that they must have undergone common envelope evolution . this information will help to place constraints on theoretical models of binary evolution ( e.g. de kool and ritter 1993 ) . detailed studies of the normal stellar companions may also reveal evidence of past interaction ( e.g. jeffries and stevens 1996 ) , and stars with possible abundance anomalies that may be the progenitors of the barium and carbon giants ( e.g. jeffries and smalley 1996 ) . in addition , the forthcoming _ fuse _ mission ( far ultra - violet spectroscopic explorer ) will , for the first time , allow us to unambiguously determine the temperatures and surface gravities of the white dwarfs in these systems ( and hence their masses and radii ) , through modelling of the h lyman absorption series down to 912 . mrb and mab acknowledge the support of pparc , uk . we wish to thank detlev koester ( kiel ) for the use of his white dwarf model atmosphere grids , and jim collins ( arizona ) and nigel bannister ( leicester ) for their help in obtaining and reducing the optical spectrum of re j0500@xmath1364 . mrb wishes to thank the staff at iue vilspa for their help and co - operation in carrying out this programme . we also thank the referee , rob jeffries , for useful suggestions to improve this paper . this research has made use of the simbad database operated at cds , strasbourg , france , and data obtained from the leicester database and archive service ( ledas ) at the department of physics and astronomy , leicester university , uk . 99 abt h.a . , 1961 , apjs , 6 , 37 allen c. , 1973 , _ astrophysical quantities _ , athlone press , london barstow m.a . , schmitt j.h.m.m . , clemens j.c . , pye j.p . , denby m. , harris a.w . , pankiewicz g.s . , 1992 , mnras , 255 , 369 barstow m.a . et al . , 1993 , adspr , 131 , 281 barstow m.a . et al . , 1994 , mnras , 270 , 499 bianchi l. , 1998 , in : ultraviolet astrophysics @xmath1 beyond the iue final archive , esa publication sp@xmath1413 , 797 bhm - vitense e. , 1980 , apj , 239 , l79 bhm - vitense e. , 1993 , astron j. , 106 , 1113 bhm - vitense e. , 1995 , aj , 110 , 228 bohlin r.c . , grillmair c.j . , 1988 , apjs , 66 , 209 bowyer s. et al . , 1994 , apjs 93 , 569 bowyer s. et al . , 1996 , apjs , 102 , 129 burleigh m.r . , barstow m.a . , 1997 , in : white dwarfs , eds . j. isern , m. hernanz and e. garcia - berro , kluwer , 329 burleigh m.r . , barstow m.a . , 1998 , mnras , 295 , l15 burleigh m.r . , barstow m.a . , fleming t.a . , 1997 , mnras , 287 , 381 ( paper i ) cassinelli j.p . et al . , 1995 , apj , 438 , 932 cassinelli j.p . et al . , 1996 , apj , 460 , 949 dupuis j. , vennes s. , chayer p. , cully s. , rodriguez - bell t. , 1997 , in : white dwarfs , eds . j. isern , m. hernanz and e. garcia - berro , kluwer , 277 esa , 1997 , _ the hipparcos catalogue _ , esa sp@xmath11200 fleming t.a . , liebert j. , green r.g . , 1986 , apj , 308 , 176 fleming t.a . , schmitt j.h.m.m . , barstow m.a . , mittaz j.p.d . , 1991 , a&a , 246 , l47 fleming t.a . et al . , 1995 , apjs , 99 , 701 fleming t.a . , snowden s.l . , pfefferman e. , briel u. , greiner j. , 1996 , a&a , 316 , 147 frisch p.c . , york d.g . , 1983 , apj , 271 , l59 garhart m. , 1997 , iue newsletter ( gsfc ) vol . 5 , no . 5 hodgkin s.t . , barstow m.a . , fleming t.a . , monier r. , pye j.p . , 1993 , mnras , 263 , 229 hodgkin s.t . , pye j.p . , 1994 , mnras , 267 , 840 koester d. , 1991 , iau symposium 145 , evolution of stars : the photospheric abundance connection , eds . g. michaud and a. tutukov , kluwer dordrecht , 435 de kool m. , ritter h. , 1993 , a&a , 267 , 397 jacoby j.h . , hunter d.a . , christian c.a . , 1984 , apjs , 56 , 257 jeffries r.d . , jewell s.j . , 1993 , mnras , 264 , 106 jeffries r.d . , burleigh m.r . , robb r.m . , 1996 , a&a , 305 , l45 jeffries r.d . , stevens i.r . , 1996 , mnras , 279 , 180 jeffries r.d . , smalley b. , 1996 , a&a , 315 , l19 jewell s.j . , 1993 , phd thesis , university of birmingham johnson h.r . , ake t.b . , 1986 , esa sp-263 , 395 jordan , s. , 1997 , in : white dwarfs , eds . j. isern , m. hernanz and e. garcia - berro , kluwer , 397 landsman w. , simon t. , bergeron p. , 1996 , pasp , 108 , 250 margoni r. , munari u. , stagni r. , 1992 , a&as , 93 , 545 marsh m.c . et al . , 1997 , mnras , 286 , 369 mason k.o . et al . , 1995 , mnras , 274 , 1194 mathioudakis m. , fruscione a. , drake j.j . , mcdonald k. , bowyer s. , malina r.f . , a&a , 300 , 775 mccook g.p . and sion e.m . , 1998 , apjs , in preparation pallavicini r. , tagliaferri g. , pollock a.m.t . , schmitt j.h.m.m . , rosso c. , 1990 , a&a , 227 , 483 pounds k.a . et al . , 1993 , mnras , 260 , 77 pye j.p . et al . , 1995 , mnras , 274 , 1165 schindler m. , stencel r.e . , linsky j.l . , basri g.s . , helfand d.j . , 1982 , apj , 263 , 269 schafer r.a . et al . , 1991 , esa tm-09 schaller g. , schaerer d. , meynet g. , maeder a. , 1992 , a&as , 96 , 269 schmitt j.h.m.m . , collura a. , sciortino s. , vaiana g.s . , harnden f.r . , rosner r. , 1990 , apj , 365 , 704 shipman h.l . , geczi j. , in ` white dwarfs ' , ed . g. wegner , springer - verlag , 134 vennes s. , mathioudakis m. , doyle j.g . , thorstensen j.r . , byrne p.b . , 1995 , a&a , 299 , l29 vennes s. , berghoefer t. , christian d.j . , 1997 , apj , 491 , l85 vennes s. , christian d.j . , mathioudakis m. , doyle j.g . , 1997 , a&a , 318 , l9 vennes s. , thejll p.a . , genova galvan r. , dupuis j. , 1997b , apj , 480 , 714 voges w. , et al . , 1996 , iau circ . 6420 warwick r.s . , barber c.r . , hodgkin s.t . , pye j.p . , 1993 , mnras , 262 , 289 welsh , b.y . , 1991 , apj , 373 , 556 wesemael f. , auer l.h . , van horn h.m . , savedoff m.p . , 1980 , apjs , 43 , 159 wood m.a . , 1992 , apj , 386 , 539 wood m.a . , 1995 , in proc . of the 9th european workshop on white dwarfs , eds . d. koester and k. werner , springer , 41 wu c .- c . . , 1992 , ultraviolet spectral atlas of selected astronomical objects , nasa reference publication 1285 figure 1 : field of the rosat wfc euv source re j0500@xmath1364 ( 6@xmath126 arcmins ) . the circles are the 90% error boxes , centred on the source co - ordinates , of ( in order of decreasing size ) euve , the wfc , and the rosat hri . the wfc error box is centred on co - ordinates ( j2000 ) 05 00 03.9 @xmath136 24 02 . the da+f6/7v binary lies near the centre of each error box . the arrows indicate other stars examined by mason et al . ( 1995 ) and ourselves during optical searches for the euv and x - ray counterpart . none of these objects was particularly remarkable . + figure 2 : optical spectrum of re j0500@xmath1364 obtained with the steward observatory 2.3 m telescope on kitt peak . using this spectrum , we classify the star as f6/7v . + figure 3 : far - ultraviolet iue swp spectrum of re j0500@xmath1364 ( swp56217 ) , obtained in december 1995 ( 8 hour exposure ) , clearly showing the white dwarf companion . also shown is a pure h model atmosphere fit for log [email protected] , t@xmath938,260k and [email protected]@xmath15 . + figure 4 : far - ultraviolet iue lwp spectrum of re j0500@xmath1364 ( lwp31729 ) . the data are very noisy and have been binned up , but the spectral shape most closely matches stars in the range f5v@xmath1g0v , consistent with the optical data . + figure 5 : pulse height distribution of the x - rays from re j0500@xmath1364 detected by the rosat hri . although the exact energies of each channel are known to vary spatially and temporally , it can be seen that this is a soft source , and there are no hard counts . + figure 6 : far - ultraviolet iue swp spectrum of hd27483 ( swp45940 ) . the hot white dwarf can clearly be seen emerging from the glare of the two f6v companions shortwards of @xmath01700 . also shown is a pure h model atmosphere fit for log [email protected] , t@xmath922,000k and [email protected]@xmath15 . + figure 7 : far - ultraviolet iue spectrum of sao150508 ( g5v , swp56195@xmath2lwp31700 ) . clearly , there is no white dwarf companion to this star . + figure 8 : far - ultraviolet iue spectrum of hd36869 ( g2v , swp56169@xmath2lwp31701 ) . no white dwarf companion is visible in the short wavelength spectrum . + figure 9 : far - ultraviolet iue spectrum of gl216b ( k2v , swp56194@xmath2lwp31699 ) . mg ii 2798 is visible in emission . the feature at @xmath01720 ( inset ) is probably due to a cosmic ray hit . + figure 10 : far - ultraviolet iue spectrum of hr2225 ( g5v , swp56206@xmath2lwp31715 ) . the emission feature at @xmath01800 ( inset ) is probably spurious . + figure 11 : far - ultraviolet iue spectrum of hd295290 ( g0v , swp56211@xmath2lwp31726 ) . mgii is clearly seen in emission at 2798 . civ 1549 emission is also suggested in the swp spectrum ( inset ) . + figure 12 : far - ultraviolet iue spectrum of hd54402 ( k0 , swp56193@xmath2lwp31698 ) . the feature at @xmath01800 is probably due to a cosmic ray hit . + figure 13 : far - ultraviolet iue spectrum of sao135659 ( k0 , swp56273@xmath2lwp31801 ) . mgii is visible in emission at 2798 . + figure 14 : far - ultraviolet iue spectrum of hd70907 ( f3iv / v , swp56344@xmath2lwp31836 ) . there is no evidence for activity . + figure 15 : far - ultraviolet iue swp spectrum of hr4646 ( a5 m , swp55658 ) . clearly , there is no white dwarf companion to this star .

for a full shift with symbols and for a non - positive potential , locally proportional to the distance to one of disjoint full shifts with symbols , we prove that the equilibrium state converges as the temperature goes to 0 . the main result is that the limit is a convex combination of the two ergodic measures with maximal entropy among maximizing measures and whose supports are the two shifts where the potential is the flattest . in particular , this is a hint to solve the open problem of selection , and this indicates that flatness is probably a / the criterion for selection as it was conjectured by a.o . lopes . as a by product we get convergence of the eigenfunction at the log - scale to a unique calibrated subaction .

1201.5452

in this paper we deal with the problem of ergodic optimization and , more precisely with the study of grounds states . ergodic optimization is a relatively new branch of ergodic theory and is very active since the 2000 s . for a given dynamical system @xmath3 , the goal is to study existence and properties of the invariant probabilities which maximize a given potential @xmath4 . we refer the reader to @xcite for a survey about ergodic optimization . ground states are particular maximizing measures which can be reached by freezing the system as limit of equilibrium states . namely , for @xmath5 , which in statistical mechanics represents the inverse of the temperature , we consider the / an equilibrium state associated to @xmath6 , that is a @xmath7-invariant probability whose free energy @xmath8 is maximal ( where @xmath9 is the kolmogorov entropy of the measure @xmath10 ) . then , considering an equilibrium state @xmath11 , it is easy to check ( see @xcite ) that any accumulation point for @xmath11 as @xmath12 goes is the inverse of the temperature , thus @xmath13 means @xmath14 , which means we freeze the system . ] to @xmath15 is a maximizing measure for @xmath16 . the first main question is to know if @xmath11 converges . it is known ( see @xcite ) that for an uniformly hyperbolic dynamical systems , generically in the @xmath17-topology , @xmath16 has a unique maximizing measure . therefore , convergence of @xmath11 obviously holds in that case . nevertheless , generic results do not concern all the possibilities , and it is very easy to build examples , which at least for the mathematical point of view are meaningful , and for which the set of ergodic maximizing measures is as wild as wanted . for these situations , the question of convergence is of course fully relevant . cases of convergence or non - convergence are known ( see @xcite ) , but the general theory is far away of being solved . in particular , no criterion which guaranties convergence ( except the uniqueness of the maximizing measure or the locally constant case ) is known , say for the lipschitz continuous case . the second main question , and that is the one we want to focus on here , is the problem of the selection . assuming that @xmath16 has several ergodic maximizing measures and that @xmath11 converges , what is the limit . in other words , is there a way to predict the limit from the potential , or equivalently , what makes the equilibrium state select one locus instead of another one ? inspired by a similar study for the lagrangian mechanics ( @xcite ) , it was conjectured by a.o . lopes that flatness of the potential would be a criterion for selection and that the equilibrium state always selects the locus where the potential is the flattest . in @xcite it is actually proved that the conjecture is not entirely correct . authors consider in the full 3-shift a negative potential except on the two fixed points @xmath18 and @xmath19 , where it vanishes but is sharper in @xmath19 than in @xmath18 . then , they prove that the equilibrium state actually converges , but not necessarily to the dirac measure at @xmath18 . the first part of the conjecture is however not ( yet ) invalided and the question to know if flatness is a criterion for selection is still relevant . here , we make a step in the direction of proving that flatness is indeed a criterion for selection . precise statements are given in the next subsection . we consider in the full shift with @xmath0 symbols a potential , negative everywhere except on @xmath1 bernoulli subshifts with @xmath2 symbols . figure [ fig - shift1 ] illustrates the dynamics . then , flatness is ordered on these @xmath1 bernoulli subshifts : the potential is flatter on the first one than on the second one , then , it is flatter on the second one than on the third one , and then so on ( see below for complete settings ) . any of these bernoulli shifts has a unique measure of maximal entropy , and the set of ground states is contained in the convex hull of these @xmath1-measures . we show here that the equilibrium state converges and selects a convex combination of the two ergodic measures with supports in the two flattest bernoulli subshifts . we emphasize that this result is absolutely not in contradiction with @xcite . indeed , in @xcite it is proved that the equilibrium state converges to a convex combination of the dirac measures at @xmath18 and @xmath19 , which are obviously the two flattest loci ! we consider the full - shift @xmath20 with @xmath0 symbols , with @xmath1 and @xmath2 two positive integers . we also consider @xmath0 positive real numbers , @xmath21 , @xmath22 , @xmath23 and @xmath24 . we assume @xmath25 we set @xmath26 , @xmath27 , @xmath28 . for simplicity the last letter @xmath0 is denoted by @xmath29 . the letter @xmath30 will be denoted by @xmath31 . the set of letters defining @xmath32 is @xmath33 . hence , @xmath34 , and a word admissible for @xmath32 is a word ( finite or infinite ) in letters @xmath31 . the length of a word is the number of digit ( or letter ) it contains . the length of the word @xmath35 is denoted by @xmath36 . if @xmath37 and @xmath38 are two finite words , we define the concatenated word @xmath39 . this is easily extended to the case @xmath40 . if @xmath41 is a finite - length admissible word for @xmath32 , @xmath42 $ ] denotes the set of points starting with the same @xmath43 letters than @xmath41 and whose next letter is not in @xmath44 . the distance in @xmath20 is defined ( as usually ) by @xmath45 where @xmath46 is a fixed real number in @xmath47 . this distance is sometimes graphically represented as in figure [ fig : distance ] . ( 12,4.5)(-2,0 ) ( -1,1.5)@xmath48 ( 0,2)(1,0)5 ( 5,2)(1,1)1 ( 5,2)(1,-1)1 ( 5,1)(0.1,3)(4.5,4.3)@xmath49 ( 4,0)@xmath50 ( 6,3)(1,0)3 ( 9.5,3)@xmath51 ( 6,1)(1,0)3 ( 9.5,1)@xmath52 we emphasize here , that contrarily to @xcite we have not chosen @xmath53 in view to get the most general result as possible . indeed , in @xcite it was not clear if some results where independent or not of @xmath46 s value . moreover , this also means that we are considering all hlder continuous functions , and not only the lipschitz ones , because in @xmath20 , a hlder continuous function can be considered as a lipschitz continuous , up to a change of @xmath46 s value . the potential @xmath54 is defined by @xmath55\\ a(x)=-{\alpha } , \text { if } x\in[u].\\ \end{array}\right.\ ] ] the potential is negative on @xmath20 but on each @xmath32 where it is constant to 0 . the transfer operator , also called the ruelle - perron - frobenius operator , is defined by @xmath56 it acts on continuous functions @xmath57 . we refer the reader to bowen s book @xcite for detailed theory of transfer operator , gibbs measures and equilibrium states for lipschitz potentials . the eigenfunction is @xmath58 and the eigenmeasure is @xmath59 . they satisfy : @xmath60 the eigenmeasure and the eigenfunction are uniquely determined if we required the assumption that @xmath61 is a probability measure and @xmath62 . _ the gibbs state _ @xmath11 is defined by @xmath63 . the measure @xmath11 is also the equilibrium state for the potential @xmath64 : it satisfies @xmath65 and this maximum is @xmath66 and is called the _ pressure _ of @xmath67 . @xmath68 is also the spectral radius of @xmath69 . it is a single dominating eigenvalue . in each @xmath32 we get a measure of maximal entropy @xmath70 . as each @xmath32 is a subshift of finite type , @xmath70 is again of the form @xmath71 where @xmath72 and @xmath73 are respectively the eigenfunction and the eigen - probability associated to the transfer operator in @xmath32 for the potential constant to 0 . note that , as @xmath12 goes to @xmath15 , @xmath11 has only @xmath1 possible ergodic accumulation points , which are the measures of maximal entropy in each @xmath32 , @xmath70 . our results are * theorem 1 * _ the eigenmeasure @xmath61 converges to the eigenmeasure @xmath74 for the weak * topology as @xmath12 goes to @xmath15 . _ * theorem 2 * _ the gibbs measure @xmath11 converges to a convex combination of @xmath75 and @xmath76 for the weak * topology as @xmath12 goes to @xmath15 . this combination depends in which zone @xmath77 ( see figure [ fig - mu ] ) the parameters are : _ 1 . for parameters in @xmath78 , @xmath11 converges to @xmath79 . 2 . for parameters in @xmath80 , @xmath11 converges to @xmath75 . 3 . for parameters in @xmath81 @xmath11 converges to @xmath82 for some @xmath83 locally constant in @xmath84 , @xmath85 and @xmath86(see equalities page , page and page ) . and @xmath76 ] zone @xmath88 corresponds to @xmath89 and @xmath90 . zone @xmath91 corresponds to @xmath92 and @xmath93 . we emphasize that @xmath91 exists if and only if @xmath94 . as a by product of theorem 1 and theorem 2 we get the exact convergence for the eigenfunction to a unique subaction ( see section [ sec - peierls ] for definition ) : * corollary 3 * _ the calibrated subactions are all equal up to an additive constant . moreover , the eigenfunction @xmath95 converges at the log - scale to a single calibrated subaction : @xmath96 _ the question of convergence and uniqueness of a subaction seems to be important for the theory of ergodic optimization . it appeared very recently in @xcite . we point out that convergence of the eigenfunction to a subaction is related to the study of a large deviation principle for the convergence of @xmath11 ( see @xcite and @xcite ) . nevertheless , in these two papers , lopes _ _ always assume the uniqueness of the maximizing measure , which yields the uniqueness of the calibrated subaction ( up to a constant ) . here we prove convergence to a unique subaction without the assumption of the uniqueness of the maximizing measure . it it thus allowed to hope that we could get a more direct proof of a large deviation principle , without using the very indirect machinery of dual shift ( see @xcite ) . the present work is part of a work in progress . the situation described here is far away from the most general case and our goal is to prove the next conjecture . in @xcite , garibaldi _ et al . _ introduce the set of non - wandering points with respect to a hlder continuous potential @xmath54 , @xmath97 . this set contains the union of the supports of all optimizing measures ( here we consider maximizing measures , there they consider minimizing measures ) . the set @xmath97 is invariant and compact . under the assumption that it can be decomposed in finitely many irreducible pieces , it is shown that calibrated subactions are constant on these irreducible pieces and their global value is given by these local values and the peierls barrier ( see section [ sec - peierls ] below ) . we believe that , under the same hypothesis , it is possible to determine which irreducible component have measure at temperature zero : * conjecture*. _ for @xmath98 hlder continuous , if @xmath97 has finitely many irreducible components , @xmath99 , then , @xmath100 goes to @xmath101 if @xmath102 is not one of the two flattest loci for @xmath54 . _ we emphasize that this conjecture does not mean that there is convergence `` into '' the components . it may be ( as in @xcite ) that an irreducible components has several maximizing measures and that there is no selection between these measures . this conjecture is for the moment far of being proved , in particular because several notions are not yet completely clear . in particular the notion of flatness has to be specified . moreover , the components are not necessarily subshifts of finite type , which is an obstacle to study their ( for instance ) measures of maximal entropy . the work presented here , is for a specific form of potential for which flatness is easily defined . the dynamics into the irreducible components and also the global dynamics are easy . we believe that the main issue here is to identify flatness as a criterion for selection . the next step would be to release assumptions on the dynamics ; in particular we would like that theses components are not full shift and that the global dynamics is not a full shift . it is also highly probable that the conjecture should be adapted after we have solve this case . distortion into the dynamics could perhaps favor other components . the last step would be to get the result for general ( or as general as possible ) potential . nevertheless , and even if the present work is presented as a work in progress and an intermediate step before a more general statement , we want to moderate the specificity of the potential we consider here . for a uniformly hyperbolic system @xmath3 and for any hlder function @xmath16 , there exists two hlder continuous @xmath103 and @xmath104 such that @xmath105 , where @xmath106 and @xmath103 is non - negative and vanishes only on the aubry set . this means that , up to consider a cohomologous function , the assumption on the sign of @xmath54 is free . now , if we would consider a very regular potential ( say at least @xmath107 ) on a geometrical dynamical systems , the fact that @xmath103 vanishes on the aubry set means that close to that aubry set , @xmath108 is proportional to the distance between @xmath52 and the aubry set ( with coefficient related to the derivative of @xmath103 ) . consequently , the potential @xmath54 we consider here is a kind of discrete version for the symbolic case of a @xmath107 potential on a manifold . in section [ sec - peierls ] we prove that the pressure behaves like @xmath109 for some specific @xmath110 and some sub - exponential function @xmath111 . the real number @xmath110 is obtained as an eigenvalue for the max - plus algebra ( see proposition [ prop2-maxplus ] ) . in section [ sec - f ] , we define and study an auxiliary function @xmath112 ; this function gives the asymptotic both for the eigenmeasure and for the gibbs measure . in section [ sec - proof - th1 ] we prove theorem 1 and in section [ sec - proof - th2 ] we prove theorem 2 . as a by - product we give an asymptotic for the function @xmath113 ( in the pressure ) . in the last section , section [ sec - proof - coro ] we prove corollary 3 . part of this work was done as i was visiting e. garibaldi at campinas ( brazil ) . i would like to thank him a lot here for the talks we get together and the attention he gave me to listen and correct some of my computations . the main goal of this section is proposition [ prop2-maxplus ] where we prove that @xmath66 converges exponentially fast to @xmath114 . the exponential speed is obtained as an eigenvalue for a matrix in max - plus algebra . [ lem2-aneis ] the eigenfunction @xmath95 is constant on the cylinder @xmath115 $ ] . it is also constant on cylinders of the form @xmath42 $ ] , where @xmath41 is an admissible word for some @xmath32 . furthermore , if @xmath116 is another admissible word for the same @xmath32 with the same length than @xmath41 , then @xmath95 coincide on both cylinders @xmath42 $ ] and @xmath117 $ ] . the eigenfunction @xmath95 is defined by @xmath118 for @xmath52 and @xmath119 in @xmath42\cup[m'*]$ ] , if @xmath120 is a preimage for @xmath52 then @xmath121 is a preimage for @xmath119 and @xmath122 . the same argument works on @xmath115 $ ] . this lemma allows us to set @xmath123 thus , for @xmath52 in @xmath124 $ ] we get @xmath125 [ lem - hb - const - aubry ] the function @xmath95 is constant on each @xmath32 : for any @xmath52 in @xmath32 , @xmath126 the function @xmath95 is continuous on the compact @xmath32 . it thus attains its minimum and its maximum . let @xmath127 and @xmath128 be two points in @xmath32 where @xmath95 is respectively minimal and maximal . the transfer operator yields : @xmath129 by definition , for each @xmath130 , @xmath131 . this yields @xmath132 similarly we will get @xmath133 . as the potential is lipschitz continuous , the pressure function @xmath66 is analytic and decreasing ( @xmath54 is non - positive ) . then @xmath134 . this shows @xmath135 . let @xmath52 be any point in @xmath32 . equality @xmath136 yields @xmath137 for @xmath138 , @xmath139 , and as @xmath95 is constant on @xmath32 we get @xmath140 the family of functions @xmath141 is uniformly bounded and equi - continuous ; any accumulation point @xmath142 for @xmath143 as @xmath144 goes to @xmath15 ( and for the @xmath17-norm ) is a _ calibrated subaction _ , see @xcite . in the following , we consider a calibrated subaction @xmath142 obtained as an accumulation point for @xmath145 . note that the convergence is uniform ( on @xmath20 ) along the chosen subsequence for @xmath12 . for simplicity we shall however write @xmath146 . a direct consequence of lemma [ lem - hb - const - aubry ] is that the subaction @xmath142 is constant on each @xmath32 . actually , it is proved in @xcite that this holds for the more general case and , moreover , that any calibrated subaction satisfies @xmath147 where @xmath148 is the _ peierls barrier _ and @xmath149 is any point in @xmath32 . it is thus important to compute what is the peierls barrier here . [ lem2-pierlsbar ] let @xmath149 be any point in @xmath32 . the peierls barrier satisfies @xmath150 . for simplicity we shall set @xmath151 for @xmath152 . let @xmath52 be in @xmath20 . recall that @xmath151 is defined by @xmath153 as we consider the limit as @xmath154 goes to 0 , we can assume that @xmath155 . now , to compute @xmath151 , we are looking for a preimage of @xmath52 , which starts by some letter admissible for @xmath32 ( because @xmath155 ) and which maximizes the birkhoff sum of the potential `` until @xmath52 '' . as the potential is negative , this can be done if and only if one takes a preimage of @xmath52 of the form @xmath156 , with @xmath41 a @xmath32 admissible word . moreover , we always have to chose the letter @xmath157 to get the maximal @xmath158 possible . in other word we claim that for every @xmath159 for every @xmath160 and for every word @xmath41 of length @xmath161 , @xmath162 let assume @xmath163 , the maximal admissible word for @xmath32 of the form @xmath164 has length @xmath165 . this yields -0.5@xmath166 then , lemma [ lem2-pierlsbar ] and equality yield @xmath167 [ lem2-subac-1234 ] for every @xmath168 , @xmath169 by lemma [ lem2-aneis ] , the eigenfunction @xmath95 is constant on rings @xmath42 $ ] with @xmath170 , hence @xmath171 . now , inequalities @xmath172 show that @xmath173 is exponentially bigger than all the other terms as @xmath12 goes to @xmath15 . here we use the max - plus formalism . we refer the reader to @xcite ( in particular chapter 3 ) for basic notions on this theory . some of the results we shall use here are not direct consequence of @xcite ( even if the proofs can easily be adapted ) but can be found in @xcite . [ prop2-maxplus ] let @xmath174 then , there exists a positive sub - exponential function @xmath111 such that @xmath175 . we have seen ( lemma [ lem - hb - const - aubry ] ) that @xmath95 is constant on the sets @xmath32 . this shows that @xmath176 is also constant of the @xmath32 . for simplicity we set @xmath177 . this is a point in @xmath32 . now we have @xmath178 note that the results we get concerning the subaction @xmath142 are actually true for any calibrated subaction . we point out that we can first chose some subsequence of @xmath12 such that @xmath179 converges , and then take a new subsequence from the previous one to ensure that @xmath180 also converges . we thus consider @xmath176 and @xmath181 . at that moment we do not claim that @xmath110 has the exact value set in the proposition . it is only an accumulation point for @xmath182 . the convergence of @xmath182 will follow from the uniqueness of the value for @xmath110 . then and lemma [ lem2-subac-1234 ] yield for every @xmath168 , @xmath183 consider the @xmath184 rows and @xmath185 columns matrix @xmath186 then , using the max - plus formalism we get @xmath187 now , consider the @xmath185 rows and @xmath184 columns matrix @xmath188 then can be written as @xmath189 combining and , we get that @xmath190 is an eigenvalue for the matrix @xmath191 ( for the max - plus algebra ) and @xmath192 is an eigenvector . let us compute the matrix @xmath193 . let us consider the row @xmath194 for @xmath195 and the column @xmath168 for @xmath196 . assume @xmath197 . we have to compute the maximum between the sum of the @xmath198 term of the row and the @xmath198 term of the column . all the terms in the column are equal to @xmath199 except the @xmath200 which is @xmath201 . this term is added to @xmath202 ( the @xmath200 term of the column ) , and this addition gives @xmath199 . therefore , this term is the maximum ( any other term is that one plus something negative ) . assume now that @xmath203 . then , the @xmath200 term of the column is added to @xmath204 , hence disappears . now , we just have to compute the maximum of all the terms respectively equal to a negative term minus @xmath205 . this means that we just have to take the maximal term in the row and to subtract @xmath205 . finally , the coefficient @xmath206 of @xmath207 is @xmath208 to compute the eigenvalue for this matrix , we have to find the `` basic loop '' with biggest mean value . a basic loop is a word in @xmath209 where no letter appear several times . then we compute the mean value of the costs of the transition @xmath210 given by the coefficient @xmath206 of the matrix for the letters of the basic loop . @xmath211 inequalities @xmath212 yields that any basic loop of length greater than 2 gives a lower contribution than the length 2-loop @xmath213 . this contribution is @xmath214 @xmath211 we claim that every basic loop of length 1 gives a smaller contribution than the first one . the claim is easy if @xmath215 . in that case we have @xmath216 if @xmath217 the claim is also true : @xmath218 @xmath219 and this last inequality holds because @xmath220 and finally , if @xmath221 , @xmath222 is non - positive and @xmath223 is non - negative , and we let the reader check that this yields @xmath224 this shows @xmath225 in particular , @xmath179 has a unique accumulation point , hence converges . then , there exists a sub - exponential function @xmath113 such that @xmath226 the pressure is convex and analytic ( the potential is lipschitz continuous ) and always bigger than @xmath114 . it is decreasing because its derivative is @xmath227 and @xmath11 gives positive weight to any open set and @xmath54 is negative except on the empty interior sets @xmath32 . this proves that @xmath113 is positive . [ rem - gammaal1 ] we emphasize @xmath228 . in this subsection we want to state an exact expression for @xmath110 depending on the values for the parameters . we have seen @xmath229 now , @xmath230 if and only if @xmath231 ( which is possible only for @xmath94 ) . obviously , @xmath232 means @xmath233 . ] finally , @xmath234 means @xmath235 . note that for @xmath92 , @xmath236 [ lem - prodxi1xi2 ] let @xmath237 be @xmath2 positive real numbers ( @xmath238 ) . let us set * @xmath239 , for @xmath240 , * @xmath241 , * @xmath242 . then , if @xmath243 goes first to @xmath15 and then @xmath144 goes to @xmath15 , @xmath244 where @xmath245 is bounded in absolute value by a term of the form @xmath246 for some universal constant @xmath247 and @xmath248 is bounded in absolute value by a term of the form @xmath249 for some universal constant @xmath250 . first we write @xmath251 and @xmath252 note that @xmath253 decreases in @xmath168 . thus we can compare this later sum with an integral @xmath254 let @xmath255 and @xmath256 respectively denote the integral from left hand side and the right hand side . first , we focus the study on @xmath255 . in order to study @xmath255 , let us set @xmath257 . then we have @xmath258 and we split this last integral in two pieces @xmath259 and @xmath260 . we remind that @xmath243 is supposed to go first to @xmath15 and then @xmath12 goes to @xmath15 . hence , @xmath261 is close to 0 . for @xmath29 close to 0 , @xmath262 is non - positive and bigger than a term of the form @xmath263 for some universal constant @xmath247 . this shows that the integral @xmath264 converges ( the function has a limit as @xmath29 goes to @xmath101 ) and @xmath265 where @xmath266 . for the other part we get : @xmath267_{{\beta}}^{1}+\int_{{\beta}}^{1}\frac{\sum_{i=2}^{p}e^{-\eta_{i}u}}{1+\sum_{i=2}^{p}e^{-\eta_{i}u}}\log u\,du\\ & = & -\log\left(1+\frac{\sum_{i=2}^{p}\eta_{i}e^{-\eta_{i}{\beta}}-1}p)\right)\log { \beta}+\int_{{\beta}}^{1}\frac{\sum_{i=2}^{p}e^{-\eta_{i}u}}{1+\sum_{i=2}^{p}e^{-\eta_{i}u}}\log u\,du\\ & = & \log p\log{\beta}-\log\left(1+\sum_{i=2}^{p}\eta_{i}e^{-\eta_{i}{\beta}})\right)\log { \beta}+\\ & & \int_{+{\infty}}^{1}\frac{\sum_{i=2}^{p}e^{-\eta_{i}u}}{1+\sum_{i=2}^{p}e^{-\eta_{i}u}}\log u\,du+ \int_{{\beta}}^{+{\infty}}\frac{\sum_{i=2}^{p}e^{-\eta_{i}u}}{1+\sum_{i=2}^{p}e^{-\eta_{i}u}}\log u\,du.\end{aligned}\ ] ] now , both terms @xmath268 and @xmath269 are bounded from above by some @xmath270 . the computation for @xmath256 is similar except that borders have to be exchanged . namely @xmath261 in @xmath259 has to be replaced by @xmath271 which improves the estimate , and @xmath12 in @xmath260 has to be replaced by @xmath272 . this produce an upper bound of the form @xmath273 instead of @xmath274 . this concludes the proof of the lemma . [ def - fun - f ] we define the auxiliary function @xmath275 by @xmath276 for an integer @xmath277 , @xmath278 denotes the truncated sum to @xmath277 : @xmath279 [ prop - equil - f ] let @xmath237 be @xmath2 positive real numbers ( @xmath238 ) . we re - employ notations from lemma [ lem - prodxi1xi2 ] . then , as @xmath12 goes to @xmath15 @xmath280 @xmath281 where @xmath282 goes to 0 as @xmath12 goes to @xmath15 . let @xmath283 be a positive real number such that @xmath284 . we set @xmath285 note that @xmath286 goes to @xmath15 as @xmath12 goes to @xmath15 . furthermore , @xmath287 and @xmath288 goes to @xmath101 if @xmath289 and @xmath144 goes to @xmath15 . the proof has three steps . the function @xmath112 is defined as a sum over @xmath243 , for @xmath159 . in the first part , we give bounds for a fixed @xmath144 , and for the sum for @xmath289 . this quantity is a trivial bound from below for the global sum . in the second step we prove that the sum for @xmath290 goes to @xmath101 as @xmath12 goes to @xmath15 . this allows to conclude the proof for the case @xmath291 . in the last step , we use the computations of the second step to conclude the proof for the case @xmath292 . * first step*. remember that @xmath293 , where @xmath113 is a positive and sub - exponential function in @xmath12 . then lemma [ lem - prodxi1xi2 ] yields for @xmath289 , @xmath294 as we only consider @xmath289 and @xmath12 goes to @xmath15 , we can replace @xmath245 and @xmath295 by @xmath282 . doing the sum over @xmath243 , only the terms @xmath296 have to be summed . we thus get @xmath297 now , @xmath298 . both @xmath113 and @xmath286 are sub - exponential in @xmath12 , hence the numerator goes to 1 as @xmath12 goes to @xmath15 . the denominator behaves like @xmath299 . then , yields @xmath300 * second step . * all the @xmath301 are bigger than @xmath302 . we thus trivially get @xmath303 for the rest of the proof , we set @xmath304 and @xmath305 . the sequence @xmath306 decreases in @xmath243 , and we can ( again ) compare the sum with the integral . we get @xmath307 now we get @xmath308.\\ & = & \frac1d\left(1-\theta^{-n({\beta})\frac{d}{\log\theta}}\right)+\frac1{\log\theta}\sum_{k=1}^{+{\infty}}\frac{e^{k}}{k!}\frac1{k-\frac{d}{\log\theta}}\left[\theta^{n({\beta})(k-\frac{d}{\log\theta})}-1\right].\end{aligned}\ ] ] as we shall consider @xmath12 close to @xmath15 we can assume that @xmath12 is big enough such that @xmath309 . then we have @xmath310 let us first study the last integral . @xmath311.\end{aligned}\ ] ] remember that @xmath312 and @xmath313 . hence , @xmath314 and @xmath315 go to 0 as @xmath12 goes to @xmath15 . then @xmath316&= & \frac1d\left[\left(\frac12\right)^{-\frac{d}{\log\theta}}-1\right]+\frac1d\left[1-\theta^{-n({\beta})\frac{d}{\log\theta}}\right]\\ & = & \frac{\log 2}{\log\theta}+n({\beta})+o(d).\end{aligned}\ ] ] let us now study @xmath317 . @xmath318.\end{aligned}\ ] ] remember that @xmath319 . therefore we finally get @xmath320 for some universal constant can be assume to be smaller than 4 if @xmath12 is chosen sufficiently big . ] @xmath321 . the term in the right hand side in goes to 0 as @xmath12 goes to @xmath15 . in particular , and show that the result holds if @xmath291 because the sum for @xmath322 is negligible with respect to the sum for @xmath323 . * third step . * we assume @xmath292 . let @xmath277 be sufficiently big such that @xmath324 . then , note that we get @xmath325 again we have @xmath326 this last term @xmath327 goes to @xmath328 as @xmath144 goes to @xmath15 . the term @xmath329 goes to the constant @xmath330 , and @xmath331 behaves ( at the exponential scale ) like @xmath332 . this proves that the second term in the right hand side of behaves like @xmath333 . now , the finite sum @xmath334 terms which are exponentially small with respect to @xmath335 if @xmath12 goes to @xmath15 . hence , this finite sum behaves like the biggest term , namely like @xmath336 . this concludes the proof of the lemma . we set @xmath337 $ ] . this is the set of points whose first digit is one of the @xmath31 s of the alphabet @xmath44 . [ prop - value - nube ] for every @xmath168 , @xmath338 let @xmath41 be an admissible word for @xmath32 with length @xmath243 . let @xmath339 . then , @xmath340)=e^{-n{{\mathcal p}}({\beta})-\sum_{j=0}^{n-1}\alpha_{m_{j}}\theta^{n - j}}\nu_{{\beta}}([k]).\ ] ] this yields @xmath341)\\ & = & \sum_{n=1}^{+{\infty}}\left(e^{-n{{\mathcal p}}({\beta})}\prod_{j=1}^{n}\left(e^{-\alpha_{1}{\beta}\theta^{j}}+\ldots + e^{-\alpha_{p}{\beta}\theta^{j}}\right)\right)e^{-{{\mathcal p}}({\beta})}(1-\nu_{{\beta}}(o_{j}))\\ & = & f({{\mathcal p}}({\beta}),{\alpha}_{1}{\beta},\ldots,{\alpha}_{p}{\beta})e^{-{{\mathcal p}}({\beta})}(1-\nu_{{\beta}}(\sqcup_{i=1}^{p}[i]))\end{aligned}\ ] ] therefore we get @xmath342 we also have an explicit value for @xmath343)$ ] : we write @xmath344=[uu]\bigsqcup_{i=1}^{kp}[u\,i],\ ] ] and use conformity to get @xmath345)=e^{-{{\mathcal p}}({\beta})-{\alpha}{\beta}}.\ ] ] in particular this quantity goes to 0 as @xmath12 goes to @xmath15 . this shows that only the @xmath1 subshifts @xmath32 can have positive measure as @xmath144 goes to @xmath15 . now , propositions [ prop2-maxplus ] and [ prop - equil - f ] show that @xmath346 behaves like @xmath347 as @xmath12 goes to @xmath15 , and this quantity diverges to @xmath15 . it also shows that for every @xmath348 , @xmath349 goes exponentially fast to @xmath101 . this shows that @xmath350 goes to 1 . the speed of the convergence is given by @xmath351 which goes exponentially fast to @xmath101 . as a by - product , we also immediately get from that for any @xmath352-admissible word @xmath41 with @xmath353 and for any @xmath354 , @xmath355)=e^{-n{{\mathcal p}}({\beta})-\sum_{j=0}^{n-1}\alpha_{m_{j}}\theta^{n - j}}\nu_{{\beta}}([k])\rightarrow_{{\beta}\to+{\infty}}0,\ ] ] as @xmath356)$ ] goes to @xmath101 if @xmath12 goes to @xmath15 . hence we get [ prop - limitnube - s1 ] any accumulation point of @xmath61 is a probability measure with support in @xmath352 . the measure of maximal entropy @xmath75 is the product of the eigenfunction and the eigenmeasure both associated to the transfer operator for @xmath357 and the constant potential zero . hence , the eigenmeasure @xmath74 is characterized by the fact that every @xmath352-admissible words with a fixed length @xmath243 have the same measure @xmath358 . we have already seen above that any accumulation point for @xmath61 is a measure , say @xmath359 , such that @xmath360 . our strategy to prove that @xmath61 converges to @xmath74 is now to prove that for any two @xmath352-admissible words @xmath41 and @xmath361 with the same length , @xmath362)}{\nu_{{\beta}}([{\widehat}m])}$ ] goes to 1 as @xmath12 goes to @xmath15 . let us thus consider two @xmath352-admissible words @xmath41 and @xmath361 with length @xmath243 . in the following , @xmath116 is a word ( possibly the empty word ) admissible for @xmath352 . we get @xmath363)&= & \sum_{m ' } \nu_{{\beta}}(mm'*)\nonumber\\ & = & \sum_{k=0}^{+{\infty}}e^{-n{{\mathcal p}}({\beta})}e^{s(m){\beta}\theta^{k}}e^{-k{{\mathcal p}}({\beta})}\prod_{j=1}^{k}(e^{-{\alpha}_{1}\theta^{j}{\beta}}+\ldots+e^{-{\alpha}_{p}\theta^{j}{\beta}})(1-\nu_{{\beta}}(o_{1}))\nonumber\\ & & \hskip 1.5 cm \text{where } s(m):=-\sum_{l=0}^{n-1}{\alpha}_{m_{l}}\theta^{n - l}\nonumber\\ & = & e^{-n{{\mathcal p}}({\beta})}\left(e^{s(m){\beta}}+ \sum_{k=1}^{+{\infty}}e^{s(m){\beta}\theta^{k}}e^{-k{{\mathcal p}}({\beta})}\prod_{j=1}^{k}(e^{-{\alpha}_{1}\theta^{j}{\beta}}+\ldots+e^{-{\alpha}_{p}\theta^{j}{\beta}})\right)(1-\nu_{{\beta}}(o_{1})).\nonumber\\ & & \label{equ1-nube - topo}\end{aligned}\ ] ] the series in the right hand side of this last equality is almost the same than the one defining @xmath364 up to the extra term @xmath365 . replacing @xmath41 by @xmath361 we get a similar formula for @xmath366)$ ] . the quantities @xmath367 and @xmath368 are negative , hence @xmath369 is lower than 1 for @xmath370 . now , remember the definition of @xmath371 ( see p. ) . step 2 of the proof of proposition [ prop - equil - f ] shows that @xmath372 goes to @xmath101 as @xmath12 goes to @xmath15 , whereas step 1 of the proof of proposition [ prop - equil - f ] shows that the tail @xmath373 diverges ( exponentially fast ) to @xmath15 . note that for @xmath374 , @xmath375 . therefore we get @xmath376)}{\nu_{{\beta}}([{\widehat}m])}\le \frac{1+f_{n({\beta})}({{\mathcal p}}({\beta}),{\alpha}_{1}{\beta},\ldots , { \alpha}_{p}{\beta})+{\displaystyle}\sum_{k > n({\beta})}e^{-k{{\mathcal p}}({\beta})}\prod_{j=1}^{k}(e^{-{\alpha}_{1}\theta^{j}{\beta}}+\ldots+e^{-{\alpha}_{p}\theta^{j}{\beta}})}{e^{\frac{s({\widehat}m)}{{\beta}}}{\displaystyle}\sum_{k > n({\beta})}e^{-k{{\mathcal p}}({\beta})}\prod_{j=1}^{k}(e^{-{\alpha}_{1}\theta^{j}{\beta}}+\ldots+e^{-{\alpha}_{p}\theta^{j}{\beta}})}.\ ] ] doing @xmath12 goes to @xmath15 in this last inequality we get @xmath377)}{\nu_{{\beta}}([{\widehat}m])}\le 1 $ ] . exchanging @xmath41 and @xmath361 we also get @xmath378)}{\nu_{{\beta}}([m])}\le 1 $ ] , which means @xmath379)}{\nu_{{\beta}}([{\widehat}m])}=1.\ ] ] in other words , any accumulation point for @xmath61 is a probability measure with support in @xmath352 which gives the same weight to all the cylinders of same length . there exists only one such measure , it is @xmath74 . this finishes the proof of theorem 1 . we recall that @xmath286 was defined ( see equation [ equ - def - nbeta ] ) by @xmath382 its main properties are that @xmath286 goes to @xmath15 as @xmath12 goes to @xmath15 and @xmath383 goes to @xmath101 . we recall that for every @xmath168 and for every @xmath32-admissible word @xmath41 with finite length we have @xmath384 the main result in this subsection is the following proposition , which gives an expression for @xmath381 . we employ notations from lemma [ lem - prodxi1xi2 ] and proposition [ prop - equil - f ] ; we remind @xmath385 and that @xmath386 was defined there . [ prop - value - mubeoj ] for every @xmath168 , let us set @xmath387 . then @xmath388,\ ] ] where @xmath282 goes to 0 as @xmath12 goes to @xmath15 . as @xmath286 is proportional to @xmath389 , the term @xmath390 goes to 0 as @xmath12 goes to @xmath15 . the importance of the formula is that , either @xmath391 and the second term goes to @xmath15 , or the possible convergence occurs at the sub exponential scale . in particular , it will show that only the components with @xmath392 sufficiently small can have weight as @xmath12 goes to @xmath15 . in view to prove proposition [ prop - value - mubeoj ] , let us first start with some technical lemmas . [ lem - fjtauj ] for every @xmath393 , @xmath394 . for a by lemma [ lem2-aneis ] @xmath95 is constant on rings . setting @xmath395 we get @xmath396 we set @xmath397 . note that @xmath398 and remember @xmath399 then , multiplying both sides of equation by @xmath400 and adding @xmath401 we get @xmath402 we get , by induction , a relation between @xmath403 , @xmath243 and @xmath404 . now , remember @xmath405 , and we finally get @xmath406+e^{-{{\mathcal p}}({\beta})}\tau_{j}(1+f_{n}({{\mathcal p}}({\beta}),{\alpha}_{p(j-1)}{\beta},\ldots , { \alpha}_{jp}{\beta})).\ ] ] now , let @xmath243 goes to @xmath15 in . the term @xmath407 converges to @xmath408 ( see lemma [ lem - hb - const - aubry ] ) and the term @xmath409 is the general term of a converging series , thus goes to @xmath101 . then we get @xmath410 this holds for any @xmath168 . [ lem - hbe - length - n ] for any @xmath168 and for any integer @xmath243 , @xmath411 equation also yields @xmath412 by induction we get @xmath413h_{{\beta}}(u_{1j}^{n+i}*)+e^{-{{\mathcal p}}({\beta})}\tau_{j}(1+\sum_{l=1}^{i}e^{-{{\mathcal p}}({\beta})}\prod_{r=1}^{l}\rho(j , n+r,{\beta}))\\ & = & \left[e^{-i{{\mathcal p}}({\beta})}\prod_{l=1}^{i}\rho(j , n+l,{\beta})\right]h_{{\beta}}(u_{1j}^{n+i}*)+e^{-{{\mathcal p}}({\beta})}\tau_{j}(1+f_{i}({{\mathcal p}}({\beta}),{\alpha}_{p(j-1)}{\beta}\theta^{n},\ldots , { \alpha}_{jp}{\beta}\theta^{n})).\end{aligned}\ ] ] as above , as @xmath130 goes to @xmath15 , @xmath414 converges to @xmath408 ( see lemma [ lem - hb - const - aubry ] ) and the term @xmath415 is the general term of a converging series , thus goes to @xmath101 . [ lem - form - mum ] @xmath416 . we pick some @xmath168 . in the following @xmath41 is a generic @xmath32-admissible word with finite length . @xmath417&=&\sum_{m,|m|=l}h_{{\beta}}(m*)\nu_{{\beta}}(m*)\nonumber\\ & = & \sum_{m,|m|=l}h_{{\beta}}(m*)e^{-lp+{\beta}.s_{l}(a)(m*)}(1-\nu_{{\beta}}(o_{j}).\nonumber\\ & = & h_{{\beta}}(u_{1j}^{l}*)e^{-l{{\mathcal p}}({\beta})}\prod_{i=1}^{l}\left(e^{-{\alpha}_{(j-1)p+1}{\beta}\theta^{i}}+\ldots+e^{-{\alpha}_{jp}{\beta}\theta^{i}}\right)(1-\nu_{{\beta}}(o_{j}),\label{equ1-mubering - l}\end{aligned}\ ] ] equality and lemma [ lem - hbe - length - n ] yield @xmath417 & = & h_{{\beta}}(u_{1j}^{l}*)e^{-l{{\mathcal p}}({\beta})}\prod_{i=1}^{l}\left(e^{-{\alpha}_{(j-1)p+1}{\beta}\theta^{i}}+\ldots+e^{-{\alpha}_{jp}{\beta}\theta^{i}}\right)(1-\nu_{{\beta}}(o_{j})\\ & = & e^{-{{\mathcal p}}({\beta})}\tau_{j}\left(f({{\mathcal p}}({\beta}),{\alpha}_{(j-1)p+1}{\beta},\ldots , { \alpha}_{jp}{\beta})-f_{l-1}({{\mathcal p}}({\beta}),{\alpha}_{(j-1)p+1}{\beta},\ldots , { \alpha}_{jp}{\beta})\right)(1-\nu_{{\beta}}(o_{j}))\end{aligned}\ ] ] now , it is an usual exercise that the sum of the tail of a series of positive terms @xmath418 is equal to @xmath419 . we split the sum in the formula of lemma [ lem - form - mum ] in two pieces , the one for @xmath420 and the one for @xmath421 . for the part for @xmath420 , we use inequality , and @xmath422 for the sum for @xmath421 , we use equality . we have to `` update '' it and replace @xmath423 by @xmath424 . in other words , we are computing the formal power series @xmath425 with @xmath426 . it is the derivative of the power series @xmath427 . hence we get @xmath428 again , @xmath429 goes to @xmath328 as @xmath12 goes to @xmath15 and @xmath430 behaves like @xmath299 . for simplicity we set @xmath431 , @xmath432 and @xmath433 . we remind that @xmath282 means a function going to 0 as @xmath12 goes to @xmath15 . this corresponds to zone @xmath80 ( see figure [ fig - mu ] ) . we emphasize that , if @xmath168 is such that @xmath435 , then propositions [ prop - value - mubeoj ] and [ prop - equil - f ] show that @xmath381 behaves like @xmath436 . [ lem - mubeoj - jgrand ] under the assumption @xmath437 , @xmath381 goes to 0 exponentially fast as @xmath12 goes to @xmath15 . let us assume that @xmath168 is such that @xmath435 . @xmath438 this shows that @xmath439 is exponentially bigger ( in @xmath12 ) than @xmath381 , and thus @xmath381 goes exponentially fast to @xmath101 . the same holds if we only assume @xmath437 , because for the equality case , we have just to replace the term @xmath282 by some sub exponential quantity ( which does not necessarily goes to 0 ) . however , this does not eliminate the exponential ratio in the computation . [ rem - measuo2 ] furthermore , this shows that @xmath440 can have a positive accumulation point only if @xmath441 . @xmath442 [ lem - o2oj ] assume @xmath441 . then , for every @xmath443 , @xmath444 . we copy the previous computation with @xmath445 instead of @xmath446 . note that @xmath447 goes to @xmath101 as @xmath12 goes to @xmath15 . we first assume @xmath448 then we have @xmath449 where @xmath282 may be replaced by a sub exponential function ( if @xmath450 ) . note that @xmath451 and @xmath452 as @xmath13 . then , @xmath381 is exponentially smaller than @xmath453 . if @xmath391 , then @xmath454 is up to a sub - exponential multiplicative ratio equal to @xmath455 . now , we remind @xmath456 [ lem - o1-win - gamma ] if @xmath457 , then @xmath458 . the result holds if @xmath459 . let us thus assume @xmath441 . then equation is still valid , provide that we replace @xmath282 by @xmath460 ( following proposition [ prop - value - mubeoj ] ) . hence we get @xmath461 for some sub - exponential function @xmath462 . our assumption in the lemma means that this last quantity diverges exponentially fast to @xmath15 as @xmath12 goes to @xmath15 , which means that @xmath439 is exponentially bigger than @xmath440 as @xmath12 goes to @xmath15 . hence , @xmath440 goes to @xmath101 . lemma [ lem - o2oj ] shows that @xmath381 goes to @xmath101 for every @xmath463 , thus @xmath439 goes to 1 . this corresponds to zone @xmath78 ( see figure [ fig - mu ] ) . we emphasize that in that case , @xmath465 always holds . therefore only @xmath446 and @xmath445 can have weight for @xmath13 . moreover , @xmath466 implies @xmath467 which yields @xmath468 then , proposition [ prop - equil - f ] shows @xmath469 it also shows that for every @xmath443 , @xmath470 is of order @xmath471 . it is thus exponentially smaller than @xmath472 . [ lem - f1f2 ] under these hypothesis , @xmath473 and @xmath474 . equation can be rewritten under the form : @xmath475).\ ] ] this yields a linear system @xmath476)f_{1 } \\e^{-{{\mathcal p}}({\beta})}\nu_{\beta}([u])f_{2 } \\\vdots \\e^{-{{\mathcal p}}({\beta})}\nu_{\beta}([u]f_{n})\end{array}\right).\ ] ] we remind that @xmath470 goes to @xmath101 as @xmath12 goes to @xmath15 for @xmath477 , and @xmath478 goes to @xmath15 . we compute the dominating term of the determinant of the @xmath479 matrix in the left hand side of the last equality . developing this determinant with respect to the first row , we left it to the reader to check that the determinant is of the form @xmath480 now we compute the cofactors for the terms of the first column . again , we left it to the reader to check that the cofactor of the term in position @xmath481 is of the form @xmath482 now , remind equality @xmath343)={\displaystyle}e^{-{{\mathcal p}}({\beta})-{\alpha}{\beta}}$ ] . therefore , equality and the property @xmath483 ( for @xmath443 ) yield @xmath484 we remind that with our values of the parameter , @xmath485 . now , @xmath478 behaves ( at the exponential scale ) like @xmath486 . this shows that the numerator in goes to @xmath101 as @xmath12 goes to @xmath15 . therefore the denominator also goes to 0 and the first part of the lemma is proved as @xmath487 . let us now replace @xmath478 and @xmath447 by their values . following proposition [ prop - equil - f ] and we get @xmath488 as @xmath489 goes to @xmath490 , we get @xmath474 . we can now finish the proof of theorem 2 item _ ( 1)_. we remind that we get @xmath491 and @xmath492 we also remind * @xmath493 ( lemma [ lem - fjtauj ] ) * @xmath494 and @xmath495 , * @xmath474 ( lemma [ lem - f1f2 ] ) , * @xmath496 . then we get @xmath497 this corresponds to zone @xmath84 ( see figure [ fig - mu ] ) . the situation is very similar than the previous one . we rewrite equality : @xmath499 again we claim that we get @xmath500 . nevertheless , and contrarily to the previous case , the numerator is equal to @xmath501 let @xmath502 be any accumulation for @xmath503 . then we get @xmath504 note that @xmath505 , then solving the equation we get only one non - negative solution . hence @xmath506 now , copying what is done above we get @xmath507 we set @xmath508 this corresponds to zone @xmath85 ( see figure [ fig - mu ] ) . [ lem - f2-z4 - 3 ] the quantity @xmath447 is exponentially bigger than every @xmath470 for @xmath443 . we remind that @xmath92 . then , proposition [ prop - equil - f ] shows that @xmath447 behaves , at the exponential scale , like @xmath510 and these two quantities are equal . now , for @xmath443 , @xmath470 is lower than @xmath511 , thus exponentially smaller than @xmath447 . again , we rewrite equality . lemma [ lem - f2-z4 - 3 ] shows that @xmath512 is again of the form @xmath513 : @xmath499 [ lem - lim - bergbe - z4 - 3 ] the quantity @xmath489 goes to @xmath490 as @xmath12 goes to @xmath15 and @xmath514 converges as @xmath12 goes to @xmath15 . the limit is denoted by @xmath502 . remember that @xmath478 behaves like @xmath515 , and our assumption yields that the numerator in the last expression for @xmath350 goes exponentially fast to @xmath101 . hence , the denominator also goes to @xmath101 and ( again ) @xmath489 goes to @xmath490 . let @xmath502 be any accumulation point for @xmath514 ( in @xmath516 ) . proposition [ prop - equil - f ] shows that @xmath478 behaves like @xmath517 and @xmath447 behaves like @xmath518 . this yields equality @xmath519 and doing @xmath13 we get @xmath520 now , the function @xmath521 is increasing . thus there exists a unique @xmath52 such that its value is @xmath490 . this proves that @xmath514 has unique accumulation point , thus converges . now , copying what is done above we get @xmath522 we set @xmath523 this corresponds to zone @xmath86 ( see figure [ fig - mu ] ) . the situation is very similar than the previous one . lemma [ lem - f2-z4 - 3 ] still holds as we just used inequalities @xmath525 for @xmath443 . hence , @xmath500 . the main difference with the previous case is that writing equality : @xmath526 the numerator does not necessarily goes to @xmath101 . namely it behaves like and @xmath478 behaves like @xmath527 . ] @xmath528 . nevertheless , lemma [ lem - lim - bergbe - z4 - 3 ] still holds . indeed , equality has just to be replaced by @xmath529 again , the function @xmath530 is increasing and there is a unique value for which the function is equal to @xmath490 . now , copying what is done above we get @xmath531 we set @xmath532 this concludes the proof of theorem 2 . in the proof of proposition [ prop2-maxplus ] we showed that only two basic loops can have a maximal weight ( which value is @xmath110 ) . these two loops are @xmath533 or @xmath534 . following the max - plus formalism ( see @xcite ) , in both cases there is a unique _ maximal strongly connected subgraph _ ( m.s.c.s . in abridge way ) which is either @xmath533 or @xmath535 . now , theorem 3.101 in @xcite shows that , in both cases , there is a unique eigenvector for @xmath207 , up to an additive constant ( added to all the coordinates ) , which is given by the first column of the associated matrix @xmath536 ( where @xmath537 is the sum for the max - plus formalism , and @xmath538 is computed for the product of the max - plus formalims ) . the proof is done by contradiction . assume that @xmath541 is an accumulation point for @xmath542 ( for @xmath52 in @xmath352 ) . let consider any accumulation point @xmath142 such that @xmath543 ( this is always possible up to consider a subsequence of @xmath12 s ) . then , @xmath142 is hlder continuous ( all the @xmath544 are equip - continuous with a upper bounded hlder norm ) , and @xmath545 on some neighborhood of @xmath52 . let us consider some cylinder @xmath247 in @xmath352 , such that for every @xmath546 , @xmath547 . by theorem 2 , @xmath548 converges to a positive value as @xmath12 goes to @xmath15 ( each cylinder in @xmath352 has positive @xmath75-measure ) . similarly , by theorem 1 , @xmath549 converges to a positive value as @xmath12 goes to @xmath15 . now , @xmath550 which yields that @xmath548 is of order @xmath551 as @xmath12 goes to @xmath15 . convergence of @xmath549 and @xmath548 and @xmath541 yield a contradiction .

we proved that there are infinitely many cousin primes .

math0510329

if @xmath0 and @xmath1 are both primes , then this prime pair is called cousin primes . hardy and littlewood conjectured that cousin primes have the same asymptotic density as twin primes@xcite . is there infinitely cousin primes ? it is an old unsolved conjecture@xcite . we will prove it as theorem [ theorem ] . [ theorem ] there are infinitely many cousin primes . let p=@xmath2 be the primes not exceeding @xmath3 , then the number of primes not exceeding @xmath4 @xcite is , @xmath5 for simplicity , we can write it as , @xmath6 \left[1- \frac{1}{p_2}\right ] \cdots \left[1- \frac{1}{p_v}\right]\\ & = ( \pi(\sqrt{n})-1)+n\prod_{i=1}^{v}\left[1- \frac{1}{p_i}\right ] , \\ \end{array}\ ] ] where @xmath7=n-\left\lfloor\frac{n}{p_i}\right\rfloor$ ] , @xmath8\left[1- \frac{1}{p_j}\right]=n-\left\lfloor\frac{n}{p_i}\right\rfloor -\left\lfloor\frac{n}{p_j}\right\rfloor+\left\lfloor\frac{n}{p_ip_j}\right\rfloor$ ] . the operator @xmath9 $ ] will leave the items which are not multiples of @xmath10 . in this paper , @xmath11 is the floor function , @xmath12 the ceiling function of @xmath13 . the integral operator ` @xmath14 $ ] ' , which is not the floor function in this paper , has meanings only operating on ( real ) number : @xmath15=[m]=\lfloor m\rfloor$ ] . let @xmath17 be a natural arithmetic progression , @xmath18 be its accompanying arithmetic progression , so that @xmath19 . there are @xmath4 such pairs . @xmath20 if we delete all the pairs in which one or both items @xmath21 and @xmath22 are composite integers of the primes @xmath23 , then the pairs left are all cousin primes . because after deleted the composite integers in sets @xmath24 and @xmath25 , the items in the middle set @xmath26 are all composite integers : if @xmath27 is a prime , then @xmath28 or @xmath29 , we have either @xmath30 or @xmath31 , so this item has already been deleted . for a given @xmath10 , first we delete the multiples of @xmath10 in set @xmath24 , or the items of @xmath32 , @xmath33=\left\lfloor\frac{n}{p_i}\right\rfloor.\ ] ] secondly we delete the multiples of @xmath10 in @xmath25 , i.e. the items of @xmath34 , or @xmath35 ( or @xmath36 for @xmath37 , hereafter ) , @xmath38=\left\lfloor\frac{n+4{{\bf mod \/}}p_i}{p_i}\right\rfloor.\ ] ] because @xmath39 when @xmath40 , the items deleted by @xmath41 and @xmath42 are not the same . for @xmath43 , @xmath44 , we need only delete the items in set @xmath24 : @xmath45 eq . ( [ equ : y1 ] ) will delete all pairs with @xmath32 in set @xmath24 , and eq . ( [ equ : y2 ] ) will delete all pairs with @xmath46 in set @xmath24 . @xmath47=n\left\lfloor\frac{1}{p_i}\right\rfloor+\delta,\ ] ] where @xmath48 with , @xmath49 the pairs left , after deleted the multiples of @xmath10 in both @xmath24 and @xmath25 , have , @xmath50.\ ] ] the operator @xmath51 $ ] , when operating on @xmath4 , will leaves the items having not those of @xmath52 . when the multiples of the primes @xmath53 in both @xmath24 and @xmath25 have been deleted , the pairs left will be prime pairs ( cousin primes ) and have , @xmath54 \left[1-\frac{1}{p_2}-\frac{\widetilde{1}}{p_2}\right ] \cdots\left[1-\frac{1}{p_{v}}-\frac{\widetilde{1}}{p_{v}}\right]\\ & = n\left[1-\frac{1}{p_1}\right ] \prod_{i=2}^v\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right],\\ \end{array}\ ] ] the meaning is as follows , @xmath55=\left\lfloor\frac{n}{p_i}\right\rfloor\\ \\ n\left[\frac{\widetilde{1}}{p_i}\right]=\left\lfloor\frac{n+4{{\bf mod \/}}p_i}{p_i}\right\rfloor\\ \\ n\left[\frac{1}{p_i}\right]\left[\frac{1}{p_j}\right ] = n\left[\frac{1}{p_ip_j}\right ] = \left\lfloor\frac{n}{p_ip_j}\right\rfloor\\ \\ n\left[\frac{\widetilde{1}}{p_i}\right]\left[\frac{\widetilde{1}}{p_j}\right ] = n\left[\frac{\widetilde{1}}{p_ip_j}\right ] = \left\lfloor\frac{n+4}{p_ip_j}\right\rfloor\\ \\ n\left[\frac{1}{p_i}\right]\left[\frac{\widetilde{1}}{p_j}\right ] = n\left[\frac{1}{p_i}\frac{\widetilde{1}}{p_j}\right ] = \left\lfloor\frac{n+\theta_{i , j}}{p_ip_j}\right\rfloor = \left\lfloor\frac{n+p_ip_j-\lambda_{i , j}}{p_ip_j}\right\rfloor,\\ \end{array } \right.\ ] ] where @xmath56 is the position of the first item with @xmath57 and @xmath58 , @xmath59 let @xmath60 , @xmath61 be the position of the first item with @xmath62 , then @xmath63 , @xmath64\left[\frac{\widetilde{1}}{p_j}\right ] = \left\lfloor\frac{n+p_ip_j-\lambda_{i , j}}{p_ip_j}\right\rfloor = \left\lfloor\frac { \left\lfloor\frac{n}{p_i}\right\rfloor + p_j-\lambda_{j } } { p_j}\right\rfloor.\\ \end{array}\ ] ] if there is no such @xmath65 in @xmath24 , i.e. , @xmath66 , then the last item in eq . ( [ equ : expandp ] ) will equal zero . because @xmath67 , so @xmath68 . the total number of cousin primes in @xmath4 is , @xmath69 \prod_{i=2}^v\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] + d(\sqrt{n})-d_1,\\ \end{array}\ ] ] where @xmath70 is the number of cousin primes @xmath71 $ ] when @xmath72 , where @xmath73 and , @xmath74 @xmath75 is the number of pairs which are not cousin primes and have not been deleted by the process . when @xmath76 , some pairs which are not cousin primes may not be deleted . if @xmath77 then @xmath78 . because if @xmath79 , then @xmath80 , if @xmath81 , as any prime must has the form @xmath82 , where @xmath83 is an integer , so @xmath84 , this item has already been deleted . the only item which may not be deleted is the one @xmath85 . but after deleted the items of @xmath86 in set @xmath24 , the items left in set @xmath24 must have the form @xmath87 and set @xmath25 have the form @xmath88 correspondingly . but @xmath89 , @xmath90 ( not 1 ) , so the pair including this item has also been deleted . certainly , the pair ( 1,5 ) has been deleted by @xmath91 . if @xmath92 then @xmath93 . because the pair of ( 1,5 ) is not one of cousin primes and has not been deleted . besides , we should note that if @xmath94 , then one number ( 11 ) in the cousin primes ( 7,11 ) is not less than @xmath95 , and if we delete this pair , then @xmath96 for @xmath43 . later , we will suppose that @xmath77 , and so @xmath78 for ` large ' number @xmath4 . let @xmath97 , then @xmath98 , @xmath99 $ ] , @xmath100 $ ] . from eq . ( [ equ : d0n ] ) , ( [ equ : lambda2 ] ) , @xmath101 \left[1-\frac{1}{3}-\frac{\widetilde{1}}{3}\right ] \left[1-\frac{1}{5}-\frac{\widetilde{1}}{5}\right]\\ & = 48\left[1-\frac{1}{2}-\frac{1}{3}+\frac{1}{6 } -\frac{\widetilde{1}}{3}+\frac{1}{2}\frac{\widetilde{1}}{3 } -\frac{1}{5}+\frac{1}{10}+\frac{1}{15}-\frac{1}{30 } + \frac{1}{5}\frac{\widetilde{1}}{3}-\frac{1}{10}\frac{\widetilde{1}}{3 } \right.\\ & { \ \ } \left . -\frac{\widetilde{1}}{5}+\frac{1}{2}\frac{\widetilde{1}}{5 } + \frac{1}{3}\frac{\widetilde{1}}{5 } -\frac{1}{6}\frac{\widetilde{1}}{5 } + \frac{\widetilde{1}}{15}-\frac{1}{2}\frac{\widetilde{1}}{15}\right]\\ & = 48-\left\lfloor\frac{48}{2 } \right\rfloor -\left\lfloor\frac{48}{3 } \right\rfloor+\left\lfloor\frac{48}{6 } \right\rfloor -\left[\frac{\widetilde{48}}{3 } \right ] + \left[\frac{48}{2}\frac{\widetilde{1}}{3}\right]\\ & { \ \ } -\left\lfloor\frac{48}{5 } \right\rfloor+\left\lfloor\frac{48}{10 } \right\rfloor + \left\lfloor\frac{48}{15 } \right\rfloor -\left\lfloor\frac{48}{30 } \right\rfloor + \left[\frac{48}{5}\frac{\widetilde{1}}{3}\right ] -\left [ \frac{48}{10}\frac{\widetilde{1}}{3}\right]\\ & { \ \ } -\left[\frac{\widetilde{48}}{5 } \right ] + \left[\frac{48}{2}\frac{\widetilde{1}}{5 } \right ] + \left[\frac{48}{3}\frac{\widetilde{1}}{5 } \right ] -\left[\frac{48}{6}\frac{\widetilde{1}}{5 } \right ] + \left[\frac{\widetilde{48}}{15 } \right ] -\left[\frac{48}{2}\frac{\widetilde{1}}{15 } \right]\\ & = 48 - 24 -16 + 8 -\left\lfloor\frac{48 + 4{{\bf mod \/}}3}{3 } \right\rfloor + \left\lfloor\frac { \left\lfloor\frac{48}{2}\right\rfloor + 3 - 1 } { 3}\right\rfloor\\ & { \ \ } -9 + 4 + 3 -1 + \left\lfloor\frac { \left\lfloor\frac{48}{5}\right\rfloor + 3 - 1 } { 3}\right\rfloor -\left\lfloor\frac { \left\lfloor\frac{48}{10}\right\rfloor + 3 - 2 } { 3}\right\rfloor\\ & { \ \ } -\left\lfloor\frac{48 + 4}{5 } \right\rfloor + \left\lfloor\frac { \left\lfloor\frac{48}{2}\right\rfloor + 5 - 3 } { 5}\right\rfloor + \left\lfloor\frac { \left\lfloor\frac{48}{3}\right\rfloor + 5 - 2 } { 5}\right\rfloor -\left\lfloor\frac { \left\lfloor\frac{48}{6}\right\rfloor + 5 - 1 } { 5}\right\rfloor + \left\lfloor\frac{48 + 4}{15 } \right\rfloor -\left\lfloor\frac { \left\lfloor\frac{48}{2}\right\rfloor + 15 - 13 } { 15}\right\rfloor\\ & = 16 -16 + 8\\ & { \ \ } -3 + 3 - 1\\ & { \ \ } -10 + 5 + 3 - 2 + 3 - 1\\ & = 5 . \end{array}\ ] ] we can check it directly , @xmath102 $ ] @xmath103 $ ] @xmath104 $ ] @xmath105 $ ] @xmath106 $ ] . so @xmath107 . it is the same as before . from @xmath108 for @xmath109 , @xmath110 for @xmath111 . so @xmath112 . the set : @xmath113 . the five cousin primes are @xmath114 and @xmath115 . [ def:3 ] let @xmath116 be an ( any ) integer set , then @xmath117 : = \sum\limits _ { x_k{{\bf mod \/}}p_i\ne0,p_i-4{{\bf mod \/}}p_i}1 = t-\left[\frac{t+\phi'_i}{p_i}\right]-\left[\frac{t+\phi''_i}{p_i}\right ] \geq0 \end{array}\ ] ] is the number of items left after deleted the items of @xmath118 . usually @xmath119\left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \ne \left(m\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\right ) \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right]$ ] , we can express it as @xmath120\left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] = \left(m\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\right)\left[1-\frac{1'}{p_j}-\frac{1''}{p_j}\right ] , \end{array}\ ] ] is the number of items left when we first delete those @xmath121 from set @xmath24 , and then delete those @xmath122 from set @xmath123\}$ ] , where set @xmath124 is no longer an arithmetic sequence . in general , @xmath125 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] = \left(m\prod_{i=1}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\right ) \left[1-\frac{1'}{p_j}-\frac{1''}{p_j}\right ] . \end{array}\ ] ] @xmath126\left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] = \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right]\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] . \end{array } \ ] ] @xmath127 & = \left[\frac{m_1+m_2}{p_i}\right ] = \left[\frac{m_1}{p_i}\right]+\left[\frac{m_2}{p_i}\right]+\left[\frac{m_1 { { \bf mod \/}}p_i+m_2 { { \bf mod \/}}p_i}{p_i}\right ] . \end{array}\ ] ] @xmath128 = ap_i\left(1-\frac{2}{p_i}\right)+b\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \mbox{\quad for $ m = ap_i+b$}. \end{array}\ ] ] @xmath129 -m_1\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] -m_2\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\leq 1.\\ \end{array}\ ] ] let @xmath130 @xmath131 -m_1\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] -m_2\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\nonumber\\ = m_1+m_2-\left\lfloor\frac{m_1+m_2}{p_i}\right\rfloor-\left\lfloor\frac{m_1+m_2 + 4{{\bf mod \/}}p_i}{p_i}\right\rfloor -m_1+\left\lfloor\frac{m_1}{p_i}\right\rfloor+\left\lfloor\frac{m_1 + 4{{\bf mod \/}}p_i}{p_i}\right\rfloor\\ \ \ -m_2+\left\lfloor\frac{m_2}{p_i}\right\rfloor+\left\lfloor\frac{m_2 + 4{{\bf mod \/}}p_i}{p_i}\right\rfloor\nonumber\\ \\ = -\left\lfloor\frac{\alpha+\beta}{p_i}\right\rfloor-\left\lfloor\frac{\alpha+\beta+4{{\bf mod \/}}p_i}{p_i}\right\rfloor + \left\lfloor\frac{\alpha+4{{\bf mod \/}}p_i}{p_i}\right\rfloor + \left\lfloor\frac{\beta+4{{\bf mod \/}}p_i}{p_i}\right\rfloor\nonumber.\\ \end{array}\ ] ] the minimum : @xmath132 , when @xmath133 , @xmath134 , @xmath135 , and @xmath136 . the maximum : @xmath137 , when @xmath138 , @xmath139 , @xmath140 ( it is true for @xmath37 ) . we can represent it as @xmath141 = m_1+m_2-\left[\frac{m_1+m_2}{p_i}\right]-\left[\frac{m_1+m_2 + 4{{\bf mod \/}}p_i}{p_i}\right]\\ & = & m_1-\left[\frac{m_1}{p_i}\right]-\left[\frac{m_1 + 4{{\bf mod \/}}p_i}{p_i}\right ] + m_2-\left[\frac{m_1 { { \bf mod \/}}p_i+m_2}{p_i}\right]-\left[\frac{m_2+(m_1 + 4{{\bf mod \/}}p_i){{\bf mod \/}}p_i}{p_i}\right]\\ & = & m_1\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] + m_2\left[1-\frac{1'}{p_i}-\frac{1''}{p_i}\right].\\ \end{array}\ ] ] @xmath142 $ ] will delete the items of @xmath143 in the sequence of @xmath144 , and @xmath145 $ ] will delete the items of @xmath146 in the sequence of @xmath147 or the items of @xmath148 in set @xmath124 . for any @xmath149 , @xmath150 \leq m_2 $ ] , so , @xmath151 \geq m\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \mbox{\quad for \ $ m'\geq m$}. \end{array}\ ] ] [ lem:1 ] for @xmath152 , @xmath153 \geq \left\lceil m\left(1-\frac{3}{p_i}\right)\right\rceil.\\ \end{array}\ ] ] @xmath154 [ lem:3 ] for @xmath155 , @xmath156 , @xmath157 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \geq \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right].\\ \end{array}\ ] ] @xmath158 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \geq \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_i}\right]\mbox{\quad for $ p_i=2$}. \end{array}\ ] ] let @xmath159 , @xmath160 , @xmath161 , because @xmath162 , so @xmath163 . for simplicity , we write @xmath164 as 4 except for @xmath37 , or @xmath165 . from eq . ( [ equ : p2 ] ) , ( [ equ : p3 ] ) , @xmath166 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] -\left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\\ & = sp_ip_j\left(1-\frac{2}{p_i}\right)\left(1-\frac{2}{p_j}\right ) -sp_ip_j\left(1-\frac{2}{p_i}\right)\left(1-\frac{3}{p_j}\right)\\ & \ \ + t-\left\lfloor\frac{t}{p_i}\right\rfloor-\left\lfloor\frac{t+4}{p_i}\right\rfloor -\left\lfloor\frac{t}{p_j}\right\rfloor+\left\lfloor\frac{t}{p_ip_j}\right\rfloor + \left\lfloor\frac{t+\theta_{j , i}}{p_ip_j}\right\rfloor -\left\lfloor\frac{t+4}{p_j}\right\rfloor+\left\lfloor\frac{t+\theta_{i , j}}{p_ip_j}\right\rfloor + \left\lfloor\frac{t+4}{p_ip_j}\right\rfloor\\ \\ & \ \ -\left(t-\left\lfloor\frac{3t}{p_j}\right\rfloor\right ) + \left\lfloor\frac{t-\left\lfloor\frac{3t}{p_j}\right\rfloor}{p_i}\right\rfloor + \left\lfloor\frac{t+4-\left\lfloor\frac{3t}{p_j}\right\rfloor}{p_i}\right\rfloor\\ & = s(p_i-2)\\ & \ \ + \left\lfloor\frac{3t}{p_j}\right\rfloor -\left\lfloor\frac{t}{p_j}\right\rfloor-\left\lfloor\frac{t+4}{p_j}\right\rfloor + \left\lfloor\frac{t}{p_ip_j}\right\rfloor + \left\lfloor\frac{t+\theta_{j , i}}{p_ip_j}\right\rfloor + \left\lfloor\frac{t+\theta_{i , j}}{p_ip_j}\right\rfloor + \left\lfloor\frac{t+4}{p_ip_j}\right\rfloor\\ \\ & \ \ -\left\lfloor\frac{\left\lfloor\frac{3t}{p_j}\right\rfloor}{p_i}\right\rfloor -\varepsilon_1 -\left\lfloor\frac{\left\lfloor\frac{3t}{p_j}\right\rfloor}{p_i}\right\rfloor -\varepsilon_2\\ & = s(p_i-2)\\ & \ \ + 3a+\left\lfloor\frac{3b}{p_j}\right\rfloor -a - a-\left\lfloor\frac{b+4}{p_j}\right\rfloor + 0 + \left\lfloor\frac{ap_j+b+\theta_{j , i}}{p_ip_j}\right\rfloor + \left\lfloor\frac{ap_j+b+\theta_{i , j}}{p_ip_j}\right\rfloor + \left\lfloor\frac{ap_j+b+4}{p_ip_j}\right\rfloor\\ & \ \ -2\left\lfloor\frac{3a+\left\lfloor\frac{3b}{p_j}\right\rfloor}{p_i}\right\rfloor -\varepsilon_1 -\varepsilon_2\\ & = s(p_i-2)+a+\left\lfloor\frac{3b}{p_j}\right\rfloor + \left\lfloor\frac{ap_j+b+\theta_{j , i}}{p_ip_j}\right\rfloor + \left\lfloor\frac{ap_j+b+\theta_{i , j}}{p_ip_j}\right\rfloor + \left\lfloor\frac{ap_j+b+4}{p_ip_j}\right\rfloor\\ & \ \ -\varepsilon_1 -\varepsilon_2 - 2\varepsilon_3-\varepsilon_4,\\ \end{array}\ ] ] where @xmath167 @xmath168 1 . if @xmath169 , then @xmath170 2 . if @xmath171 , then 1 . if @xmath172 , then @xmath173 2 . if @xmath174 , then @xmath175 , @xmath176 , so @xmath177 3 . if @xmath178 , then 1 if @xmath179 , then @xmath180 2 . if @xmath181 , then @xmath182 , @xmath183 , so @xmath184 3 . if @xmath185 : if @xmath186\geq1 $ ] then @xmath187 , @xmath188 , so @xmath189-\delta \varepsilon\geq 3 + 1 + 1 - 5=0 $ ] ; else for @xmath186=0 $ ] then @xmath190 , @xmath176 , so @xmath191 . 4 . if @xmath192 , then @xmath193 , @xmath194 , so @xmath195 5 . if @xmath196 : if @xmath186\geq1 $ ] then @xmath189-\delta \varepsilon\geq 3 + 3 + 1 - 7=0 $ ] ; else for @xmath186=0 $ ] then @xmath197 , @xmath188 , so @xmath198 . 4 . if @xmath37 , then @xmath199 , + @xmath200 + @xmath201 1 . if @xmath202 , then @xmath203 , 1 . if @xmath204 , then @xmath205 , so @xmath206 . if @xmath174 then @xmath207 . else for @xmath208 , because @xmath209 , @xmath210 . so @xmath211 . else for @xmath212 , from @xmath213 , we have @xmath214 . if b=0 we follow the process above . for b=1 , @xmath215 . @xmath216 * for a=0 , @xmath217 ; * for a=2 , @xmath218 ( it comes from one of 5,10,15 ) , @xmath219 ; * for a=1 , if @xmath220 then @xmath221 ; else for s=1 , from @xmath222 , and @xmath223 , we have @xmath192 . so this case is impossible . 2 . if @xmath224 ( note @xmath225 ) , then @xmath226 1 . if @xmath220 then @xmath227 . 2 . if @xmath228 , i.e. , @xmath222 , so @xmath212 . from @xmath223 , we have @xmath192 . let us consider @xmath229 , @xmath230 , with the condition @xmath231 i.e. @xmath232 , @xmath233 . @xmath234 , therefore , @xmath235 + besides , @xmath236 1 . if @xmath237 , then @xmath238 . 2 . if @xmath239 , because , + @xmath240 + so @xmath241 and @xmath242 + therefore , @xmath243 + if @xmath244 then @xmath245 . + else for @xmath208 , because @xmath246 , @xmath247 . so @xmath248 . in summary , @xmath249 for all @xmath250 . when @xmath43 , @xmath251\equiv0 $ ] , @xmath252 , @xmath253 , @xmath254 , @xmath163 . @xmath255 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] -\left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{2}\right]\\ & = 2sp_j\left(1-\frac{1}{2}\right)\left(1-\frac{2}{p_j}\right ) -2sp_j\left(1-\frac{1}{2}\right)\left(1-\frac{3}{p_j}\right)\\ & \ \ + t-\left\lfloor\frac{t}{2}\right\rfloor -\left\lfloor\frac{t}{p_j}\right\rfloor+\left\lfloor\frac{t}{2p_j}\right\rfloor -\left\lfloor\frac{t+4{{\bf mod \/}}p_j}{p_j}\right\rfloor+\left\lfloor\frac{t+\theta_{i , j}}{2p_j}\right\rfloor \\ & \ \ -\left(t-\left\lfloor\frac{3t}{p_j}\right\rfloor\right ) + \left\lfloor\frac{t-\left\lfloor\frac{3t}{p_j}\right\rfloor}{2}\right\rfloor\\ & = \ \ s+\left\lfloor\frac{3t}{p_j}\right\rfloor -\left\lfloor\frac{t}{p_j}\right\rfloor + 0-\left\lfloor\frac{t+4{{\bf mod \/}}p_j}{p_j}\right\rfloor + \left\lfloor\frac{t+\theta_{i , j}}{2p_j}\right\rfloor -\left\lfloor\frac{\left\lfloor\frac{3t}{p_j}\right\rfloor}{2}\right\rfloor -\varepsilon_1\\ & = s+3a+\left\lfloor\frac{3b}{p_j}\right\rfloor -a - a-\left\lfloor\frac{b+4{{\bf mod \/}}p_j}{p_j}\right\rfloor + \left\lfloor\frac{ap_j+b+\theta_{i , j}}{2p_j}\right\rfloor -\left\lfloor\frac{3a+\left\lfloor\frac{3b}{p_j}\right\rfloor}{2}\right\rfloor -\varepsilon_1\\ & = s+\left\lfloor\frac{3b}{p_j}\right\rfloor + \left\lfloor\frac{ap_j+b+\theta_{i , j}}{2p_j}\right\rfloor -\varepsilon_1-\varepsilon_3-\varepsilon_4,\\ \end{array}\ ] ] where @xmath256 @xmath257 1 . if @xmath258 , then 1 . if @xmath213 , then @xmath259 , @xmath260 + if @xmath181 , then @xmath261 , @xmath262 , @xmath263 ; + else for @xmath185 , @xmath264 . @xmath265 , @xmath266 . + @xmath267 + @xmath268 + so @xmath269 2 . if @xmath270 , then @xmath271 : + if @xmath181 , then @xmath259 , @xmath272 ; + else for @xmath185 , then @xmath261 , @xmath273 . 3 . if @xmath274 , then @xmath275 . if @xmath165 , we can follow the process of @xmath225 , only note that ` 4 ' should be changed to @xmath276 and @xmath277 , i.e. @xmath278 , @xmath279 . in summary , @xmath249 for all @xmath280 . this lemma means that , from equation ( [ equ : l1 ] ) and ( [ equ : p5 ] ) , we can let @xmath281= \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil+t , t\geq 0 $ ] , and operated by @xmath51 $ ] to get the items which have no multiples of @xmath10 and @xmath282 in @xmath24 and@xmath25 . @xmath283 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] = \left ( \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil+t\right ) \left[1-\frac{1'}{p_i}-\frac{1''}{p_i}\right]\\ \\ = \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] + t\left[1-\frac{1'}{p_i}-\frac{1''}{p_i}\right]\\ \\ \geq \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] . \end{array}\ ] ] [ lem:4 ] @xmath284 \left[1-\frac{1}{3}-\frac{\widetilde{1}}{3}\right ] \geq\left\lfloor\frac{m}{6}\right\rfloor.\\ \end{array}\ ] ] for @xmath285 , let @xmath286 , @xmath287 \left[1-\frac{1}{3}-\frac{\widetilde{1}}{3}\right ] = 6s\left[1-\frac{1}{2}\right ] \left[1-\frac{1}{3}-\frac{\widetilde{1}}{3}\right ] + t\left[1-\frac{1}{2}\right ] \left[1-\frac{1}{3}-\frac{\widetilde{1}}{3}\right]\\ \\ \geq 6s\left(1-\frac{1}{2}\right)\left(1-\frac{2}{3}\right ) = s=\left\lfloor\frac{m}{6}\right\rfloor . \end{array}\ ] ] for the residual class of modulo 6 , @xmath288 , @xmath289 , there are 4 elements @xmath290 of multiples of 2 or 3 . for the other elements @xmath291 , there is at least one with @xmath292 , i.e. , @xmath293 , and @xmath294 . the cousin primes have at least @xmath295 items , or @xmath296\left[1-\frac{1}{3}-\frac{\widetilde{1}}{3}\right ] \geq\left\lfloor\frac{m}{6}\right\rfloor$ ] . [ lem:5 ] for @xmath297 , @xmath298 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \geq \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i=1}^{i_m}\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right],\\ \end{array}\ ] ] where @xmath299 for @xmath43 . suppose that for @xmath300 , @xmath301 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] = \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]+t,\\ \end{array}\ ] ] where @xmath302 . it means that the effect of the operator @xmath303 $ ] when operating on @xmath304 is that dividing @xmath305 into two effective sets @xmath306 and @xmath307 . from equation ( [ equ : d32 ] ) , ( [ equ : d33 ] ) , ( [ equ : l1 ] ) , ( [ equ : l3 ] ) , and ( [ equ : l5 t ] ) , we have @xmath308 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right]\\ = m\prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \left[1-\frac{1}{p_{r-1}}-\frac{\widetilde{1}}{p_{r-1}}\right]\\ = \left ( \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]+t \right ) \left[1-\frac{1'}{p_{r-1}}-\frac{1''}{p_{r-1}}\right]\\ \geq \left(\left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\right ) \left[1-\frac{1'}{p_{r-1}}-\frac{1''}{p_{r-1}}\right]\\ = \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r-1}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right].\\ \end{array}\ ] ] or @xmath309 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \left[1-\frac{1}{p_{r-1}}-\frac{\widetilde{1}}{p_{r-1}}\right]\\ = m\prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] -m\prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \left[\frac{1}{p_{r-1}}\right]\\ { \ \ } -m\prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] \left[\frac{\widetilde{1}}{p_{r-1}}\right]\\ = \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] + t\\ { \ \ } -\left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[\frac{1}{p_{r-1}}\right ] -t\left[\frac{1'}{p_{r-1}}\right]\\ { \ \ } -\left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[\frac{\widetilde{1}}{p_{r-1}}\right ] -t\left[\frac{1''}{p_{r-1}}\right]\\ = \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[1-\frac{1}{p_{r-1}}-\frac{\widetilde{1}}{p_{r-1}}\right ] + t\left[1-\frac{1'}{p_{r-1}}-\frac{1''}{p_{r-1}}\right]\\ \geq \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i = r-1}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] . \end{array}\ ] ] in fact , if @xmath310 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] < \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \prod_{i=1}^{i_m}\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] , \end{array}\ ] ] then for any @xmath311 , before deleting the multiples of other primes , it must have + @xmath312 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] < \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] . \end{array}\ ] ] which contradicts lemma [ lem:3 ] . so this lemma is true . if @xmath313 , then @xmath314\prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right ] \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right]\\ \geq \left\lceil m\left(1-\frac{3}{p_j}\right)\right\rceil \left[1-\frac{1}{p_{r-1}}\right ] \prod_{i = r}^{i_m}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right].\\ \end{array}\ ] ] with lemma [ lem:3 ] , the operator @xmath303 $ ] can be represented by @xmath315 , and other operator @xmath51 $ ] can operate on this inequality unchanged . let @xmath316 , then for all @xmath317 , @xmath318 $ ] is effective to a natural sequence @xmath319 whose number is not less than @xmath320 . the reason is as follows . when a natural sequence is deleted by the items of @xmath321 , the sequence is subtracted by @xmath322+\left[\frac{m+4{{\bf mod \/}}p_j}{p_j}\right]$ ] . we can arrange the m items in a table of @xmath282 rows ( table 1 ) . @xmath322 $ ] will delete the @xmath323 row , and @xmath324 $ ] will delete the @xmath325 row . thus there are @xmath326 rows left in which each item @xmath327 . [ table1 ] @xmath328 but every @xmath10 items ( @xmath329 ) in any row of the first @xmath330 columns consist in a complete systems of residues modulo @xmath10 , because @xmath331 and @xmath332 are both complete system of residues modulo @xmath10 , where @xmath333 is an any ( row or column ) constant . there are @xmath326 such rows or @xmath334 items left . these items are effective to a nature sequence when deleting the items of @xmath335 . @xmath336 \left[1-\frac{1}{p_j}-\frac{\widetilde{1}}{p_j}\right ] = sp_i(p_j-2)\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]=s(p_i-2)(p_j-2).$ ] let @xmath160 , @xmath337 , @xmath338 . @xmath339 $ ] will delete at most @xmath340 items . if we add these items by removing those from the end of sequence then the sequence is again effective to a nature sequence , the sequence left has at least , @xmath341-[a+(a+1)]\\ & \geq sp_i(p_j-3)+t+sp_i-4a-2 \\ & = sp_i(p_j-3)-3\left\lfloor\frac{t}{p_j}\right\rfloor+t+(sp_i - a-2)\\ & \geq sp_ip_j(1-\frac{3}{p_j})+\left\lceil t-\frac{3t}{p_j}\right\rceil+(sp_i - a-2)\\ & = \left\lceil(sp_ip_j+t)(1-\frac{3}{p_j})\right\rceil+(sp_i - a-2)\\ \\ & = \left\lceil m(1-\frac{3}{p_j})\right\rceil+(sp_i - a-2 ) . \end{array}\ ] ] for @xmath342 or @xmath343 , we have @xmath344 , @xmath345 . for @xmath346 and @xmath347 , the items of t have @xmath282 rows , @xmath348 columns and some @xmath349 items . in each of the first @xmath349 rows , there are exact @xmath10 items which consist in a complete system of residues modulo @xmath10 , and these items can be considered as an effective nature sequence when deleting the multiples of @xmath10 ( @xmath335 ) . the other items have at most @xmath282 rows and @xmath348 columns where the multiples of @xmath282 have at most @xmath350 . as before , we can add these items to make the @xmath351 as an effective nature sequence , therefore , @xmath352 - 2(p_i-1)\\ & \geq \left\lceil m(1-\frac{3}{p_j})\right\rceil+(sp_i - a-1)\geq \left\lceil m(1-\frac{3}{p_j})\right\rceil . \end{array}\ ] ] thus for any @xmath353 , the original sequence of @xmath155 , after deleted the items of @xmath321 from @xmath24 , is effective to reconstruct a new nature sequence having at least @xmath354 items . the analysis above is for @xmath204 . for @xmath212 , we can reach at the same conclusion . @xmath355 , @xmath356 , @xmath357 . for @xmath212 , after deleted the item of @xmath358 , it becomes + @xmath359 . we can rearrange these items as @xmath360 . the first @xmath361 items can be taken as an effective nature sequence ( @xmath362 ) from the original one when deleting the items of @xmath363 . the other sequence @xmath364 , having at least zero item when deleting the multiples of all primes , will be neglected in further process . for a given @xmath4 , consider the possible cousin primes of @xmath365 $ ] in set @xmath366 , @xmath367 . @xmath376\prod_{i=2}^v\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\\ & \geq \left\lceil n\left(1-\frac{3}{p_v}\right)\right\rceil\left[1-\frac{1}{2}\right ] \prod_{i=2}^{v-1}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\\ & \geq \left\lceil\left\lceil n\left(1-\frac{3}{p_v}\right)\right\rceil \left(1-\frac{3}{p_{v-1}}\right)\right\rceil \left[1-\frac{1}{2}\right ] \prod_{i=2}^{v-2}\limits\left[1-\frac{1}{p_i}-\frac{\widetilde{1}}{p_i}\right]\\ & \geq \cdots\\ & \geq \left\lceil n\prod_{i=3}^{v}\limits\left(1-\frac{3}{p_i}\right)\right\rceil \left[1-\frac{1}{2}\right ] \left[1-\frac{1}{p_2}-\frac{\widetilde{1}}{p_2}\right]\\ & \geq \ \left\lfloor\frac{n}{6}\prod\limits_{i=3}^{v}\left(1-\frac{3}{p_i}\right)\right\rfloor , \end{array}\ ] ] therefore , @xmath398 , and @xmath399 so that @xmath400 . it contradicts the supposition of eq . ( [ equ : tp1 ] ) . therefore ` there are infinitely many pairs of cousin prime ' . from eq . ( [ equ : tp3 ] ) and ( [ equ : tp4 ] ) , @xmath401 approaches infinity as n grows without bound . the proof is completed . figure [ fig : fig1 ] shows the actual pairs of cousin prime @xmath404 ( solid line ) in the range of @xmath405 $ ] , and its simplified formula @xmath406(dashed line ) from eq . ( [ equ : f1 ] ) . it clearly shows that @xmath403 , and @xmath407 has no up bound for enough large @xmath4 .

under ideal conditions , quantum metrology promises a precision gain over classical techniques scaling quadratically with the number of probe particles . at the same time , no - go results have shown that generic , uncorrelated noise limits the quantum advantage to a constant factor . in frequency estimation scenarios , however , there are exceptions to this rule and , in particular , it has been found that transversal dephasing does allow for a scaling quantum advantage . yet , it has remained unclear whether such exemptions can be exploited in practical scenarios . here , we argue that the transversal - noise model applies to the setting of recent magnetometry experiments and show that a scaling advantage can be maintained with one - axis - twisted spin - squeezed states and ramsey - interferometry - like measurements . this is achieved by exploiting the geometry of the setup that , as we demonstrate , has a strong influence on the achievable quantum enhancement for experimentally feasible parameter settings . when , in addition to the dominant transversal noise , other sources of decoherence are present , the quantum advantage is asymptotically bounded by a constant , but this constant may be significantly improved by exploring the geometry .

1411.0716

high - precision parameter estimation is fundamental throughout science . quite generally , a number of probe particles are prepared , then subjected to an evolution which depends on the quantity of interest , and finally measured . from the measurement results an estimate is then extracted . when the particles are classically correlated and non - interacting , as a consequence of the central limit theorem , the mean - squared error of the estimate decreases as @xmath0 , where @xmath1 is the number of particles ( probe size ) . this best scaling achievable with a classical probe is known as the _ standard quantum limit _ ( sql ) @xcite . quantum metrology aims to improve estimation by exploiting quantum correlations in the probe . in an ideal setting without noise , it is well known that quantum resources allow for a quadratic improvement in precision over the sql @xcite ; i.e , the mean - squared error of the estimate after a sufficient number of experimental repetitions can scale as @xmath2 , yielding the the so - called _ heisenberg limit_. realistic evolution , however , always involves noise of some form , and although quantum metrology has been demonstrated experimentally , e.g. , for atomic magnetometry @xcite , spectroscopy @xcite , and clocks @xcite , there is currently much effort to determine exactly when , and by how much , quantum resources allow estimation to be improved in the presence of decoherence @xcite . it is known that for most types of uncorrelated noise ( acting independently on each probe particle ) the asymptotic scaling is constrained to be sql - like @xcite . specifically , when estimating a parameter @xmath3 , the mean - squared error obeys @xmath4 , where @xmath5 is the number of repetitions and @xmath6 is a constant which depends on the evolution . if the evolution , which each probe particle undergoes , is independent of @xmath1 , the scaling is constrained to be sql - like . however , for frequency estimation this is not necessarily the case . in frequency estimation scenarios , such as those of atomic magnetometry @xcite , spectroscopy @xcite , and clocks @xcite , there are two relevant resources , the total number of probe particles @xmath1 and the total time @xmath7 available for the experiment @xcite . the experimenter is free to choose the interrogation time @xmath8 and , in particular , @xmath9 may be adapted to @xmath1 . in this case , the time over which unitary evolution and decoherence act is different for each @xmath1 and thus the evolution is not independent of @xmath1 . schematically , the no - go results for noisy evolution in this case become @xmath10 with @xmath11 . thus , if for some optimal choice of @xmath12 the coefficient @xmath13 decreases with @xmath1 , although the no - go results may hold for any fixed evolution time , the bound does not imply sql - like scaling . note that the bound eq . ( [ eq.sqllike ] ) is always achievable in the many - repetitions limit @xmath14 @xcite , which corresponds to @xmath15 @xcite . although without noise it is optimal to take @xmath9 as large as possible , i.e , @xmath16 , for any noisy evolution the optimal @xmath9 becomes finite because of noise dominating at large times . so the many - repetitions regime can always be ensured by considering sufficiently large @xmath7 @xcite . in frequency estimation scenarios , for the asymptotic scaling to be superclassical , @xmath13 must vanish as @xmath17 , which is only possible if the evolution is such that decoherence can be neglected at short time scales , and the no - go theorems then do not apply @xcite . this is also necessary for error - correction techniques , which utilise ancillary particles not sensing the parameter @xcite or employ correcting pulses during the evolution @xcite , to surpass the sql @xcite . without such additional resources considering just interrogation - time optimisation the possibility of superclassical scaling has been demonstrated for non - markovian @xcite evolutions ( for which the effective decoherence strength vanishes as @xmath18 ) , as well as for dephasing directed along a direction perpendicular to the unitary evolution @xcite . in the latter case , it was shown that an optimal variance scaling of @xmath19 can be obtained by choosing @xmath20 @xcite . this result was based on numerical analysis of the _ quantum fisher information _ ( qfi ) @xcite and was shown to be saturable by greenberger - horne - zeilinger ( ghz ) states @xcite . however , ghz states of many particles are not easily generated in practice , and the fisher information approach does not explicitly provide the required measurements . thus , the question of whether the scaling is achievable in practically implementable metrology was left open . in this paper , we argue that the transversal - noise model applies to atomic magnetometry , in particular the experimental setting of @xcite , and study the quantum advantage attainable with use of _ one axis - twisted spin - squeezed states _ ( oatsss ) @xcite and _ ramsey - interferometry - like measurements _ @xcite , both of which are accessible with current experimental techniques . we explicitly show that the setup geometry plays an important role for the achievable quantum enhancement . a suboptimal choice leads to a constant factor of quantum enhancement , while superclassical precision scaling can be maintained for a more appropriate choice . we study the enhancement achievable with the numbers of the experiment @xcite and demonstrate the advantage of modifying the geometry . we further consider the case of noise which is not perfectly transversal and find that , although the asymptotic precision scaling is then again sql - like , the precision may be substantially enhanced by optimising the geometry . as the previous results @xcite were based on numerics , we also provide an analytical proof of the scaling for ghz states in appendix [ app.ghz_states ] . we consider a scheme in which @xmath1 two - level quantum systems are used to sense a frequency parameter @xmath3 in an experiment of total duration @xmath7 , divided into rounds of interrogation time @xmath9 . we keep in mind that this can correspond to atomic magnetometry , in which the particles then represent the atoms with a spin precessing in a magnetic field at a frequency proportional to the field strength . as in ref . @xcite , we describe the noisy evolution by a master equation of lindblad form @xmath21 here , @xmath22 $ ] is the unitary part of the evolution that encodes the parameter dependence . the hamiltonian is given by @xmath23 where @xmath24 is a pauli operator acting on the @xmath25th particle ( qubit ) . the liouvillian @xmath26 describes the noise , which is uncorrelated on different qubits , so that @xmath27 , and for a single qubit we have @xmath28\ ! , \label{lindblad}\ ] ] where @xmath29 is the overall noise strength and @xmath30 with @xmath31 . for @xmath32 , eq . ( [ lindblad ] ) describes dephasing along the direction of the unitary , while @xmath33 ( or equivalently @xmath34 ) corresponds to the transversal - dephasing noise . for @xmath35 , we have an isotropic depolarizing channel . under this model , interrogation - time optimisation leads to a quantum scaling advantage for transversal ( @xmath36 ) but not for parallel ( @xmath37 ) noise . this can be understood by looking at how the coefficient @xmath13 in eq . ( [ eq.sqllike ] ) behaves in the two cases . for short times , one can obtain bounds of the form eq . ( [ eq.sqllike ] ) with @xcite @xmath38 from this , we see that for parallel dephasing , interrogation - time optimisation can not prevent asymptotic sql - like scaling , because @xmath39 is bounded from below by the non - zero factor of @xmath40 . however , for perpendicular noise @xmath41 as @xmath42 , and hence in this case , optimisation may allow for superclassical scaling . in ref . @xcite , it was found that taking @xmath43 leads to @xmath44 and that this bound is achievable with the ghz input states . which induces a level splitting between the magnetic sublevels . the atoms are used to sense a weak field @xmath45 in the plane perpendicular to @xmath46 , which rotates in this plane with a frequency matched to the larmor precession induced by @xmath46 . we consider two cases for the state preparation and readout . scenario ( a ) corresponds to the geometry of the experiment @xcite . all atoms are initially pumped to an extreme magnetic sublevel @xmath47 creating a coherent spin state aligned with @xmath46 . the state is then squeezed to make it more sensitive to the evolution induced by @xmath45 . in a frame rotating around @xmath46 at the larmor frequency , the state can be depicted as shown in the lower part . @xmath45 points then along @xmath48 and induces a rotation around the @xmath48 axis . the state is squeezed in @xmath49 and @xmath45 is estimated from a measurement of the collective spin component @xmath50 . scenario ( b ) is similar , but the state is initially perpendicular to @xmath46 . in the rotating frame , it is squeezed in @xmath51 and @xmath52 is measured . the dominant noise in both cases comes from individual atomic motion causing variations in the effective magnetic field and hence the energy splitting . this results in uncorrelated dephasing noise in the direction of @xmath46 ; the impact on the collective spin is schematically illustrated by the inner prolate spheroids . importantly , the noise preserves the spin along @xmath46 but shrinks it in the perpendicular directions . ] to see that the model is relevant in practice , we consider the atomic magnetometry experiment of ref . @xcite illustrated in fig . [ fig.magnetosetup ] . in this experiment , entanglement was demonstrated to enhance the sensitivity , but the precision scaling with @xmath1 was not studied . the relevant magnetometer consists of a vapour of caesium atoms , which is subject to a strong external dc magnetic field @xmath46 and used to sense a weak radio - frequency field @xmath45 perpendicular to @xmath46 ( note that in ref . @xcite two separate ensembles were used ; this is not important for the present argument ) . the atoms are optically pumped into an extreme magnetic sublevel and may be treated as effective two - level systems with an energy splitting determined by @xmath46 . with @xmath53 , the dominant noise is due to small variations in the dc magnetic field seen by different atoms ( e.g. , due to field inhomogeneities and atomic motion ) , which leads to fluctuations of the individual energy splittings . this corresponds to a dephasing noise which acts on each atom independently and is characterised by the spin - decoherence time @xmath54 @xcite . as the experiment is conducted at a time scale much shorter than the ones of spontaneous emission and @xmath46-field fluctuations , other noise sources are suppressed . in particular , the spin - relaxation time @xmath55 can be taken infinite ( see appendix [ app.master ] for discussion ) and collective noise can be neglected . the larmor frequency of the strong field @xmath46 is matched to the frequency of the weak field @xmath45 , and it is then convenient to describe the system in a _ rotating frame _ ( rf ) . if @xmath56 is the state of a single atom in the non - rotating frame and @xmath46 is directed along the @xmath51 axis ( see fig . [ fig.magnetosetup ] ) , the state in the rotating frame reads @xmath57 , where @xmath58 and @xmath59 is the coupling strength to the magnetic field . in such a larmor - precessing frame , the master equation for the evolution may be written as ( see also appendix [ app.master ] ) @xcite @xmath60 - \frac{1}{t_{2}}\left(\rho_{\textrm{rf } } - \hat\sigma_{x}\rho_{\textrm{rf } } \hat\sigma_{x}\right ) , \ ] ] where the first term can be understood as the effective free hamiltonian in the rotating frame with the @xmath45 field pointing along @xmath48 . the dephasing noise is directed along @xmath46 and parametrised by @xmath54 . since eq . ( [ eq.master_magnetometry ] ) is exactly of the eq . ( [ eq.master ] ) form , it is clear that this experimental setting is captured by the previously stated model with @xmath61 and transversal noise @xmath62 , @xmath63 , @xmath64 . we note that @xmath65 is important for the noise to be transversal , which may imply that @xmath29 is large relative to @xmath3 . in particular , this is the case in ref . @xcite , as we show below . in ref . @xcite , superclassical precision was demonstrated by initially aligning the collective spin of the atomic ensemble _ along _ @xmath46 and reducing fluctuations of its component in the direction perpendicular to both @xmath46 and @xmath45 via spin squeezing [ fig . [ fig.magnetosetup](a ) ] . below , we study such a geometry along with another setting , in which the collective spin is initially _ perpendicular _ to both @xmath46 and @xmath45 and its component along @xmath46 is squeezed [ fig . [ fig.magnetosetup](b ) ] . in principle , scenario ( b ) can be obtained from ( a ) by applying a @xmath66 pulse to the atomic ensemble before the evolution . in both cases , @xmath45 is estimated from a measurement of a component of the collective spin ( in the rotating frame ) , read out , e.g. , via the scheme of ref . @xcite , which resembles a standard ramsey - interferometry @xcite measurement . we show that in scenario ( b ) , for one axis - twisted spin - squeezed states , a superclassical scaling @xmath67 of the mean - squared error can be maintained , thus demonstrating that a scaling quantum advantage is possible with feasible states and measurements . at the same time , we find that in the case of ( a ) the quantum advantage is limited by a constant , which matches the bound for parallel dephasing @xcite . as a consequence , for an atomic ensemble size and parameters matching the experiment ref . @xcite , ( b ) may considerably outperform ( a ) . as an aside , we note that when the true value of the estimated parameter is zero , the bound of eq . ( [ eq.prevbound ] ) vanishes . this does not mean that the precision is unbounded , but indicates that the bound gives no information in such a limit . one may then speculate whether the scaling can be further improved if @xmath3 can be made arbitrarily small in an adaptive manner . we discuss this issue in appendix [ app.omega_zero ] . to obtain results for the precision achievable within the above scenarios , we make use of error propagation and apply it for adequate choices of squeezed states and collective - spin - observable measurements . generally , when a parameter @xmath68 is estimated based on the average of measuring an observable @xmath69 , and when the prior knowledge of @xmath68 is sufficiently tight , fluctuations in the estimate can be linearly related to the fluctuations in @xmath69 . thus , for a system in a state @xmath56 , in such a local estimation regime the mean - squared error of the estimate may be quantified as @xmath70 if the measurement is repeated , @xmath71 will additionally decrease inverse proportionally to the number of repetitions @xmath5 , which also ensures the above local regime as @xmath14 and thus that eq . ( [ eq.errprop ] ) always holds . here , we are interested in frequency estimation over a total time @xmath7 with a single - round duration @xmath9 , such that @xmath72 . we therefore write the overall mean - squared error of the @xmath3 estimate as @xcite @xmath73 which , like eq . ( [ eq.sqllike ] ) , is valid in the @xmath15 regime , i.e , for sufficiently large @xmath7 . the expectation values in eq . ( [ eq.freqerrprop ] ) can be evaluated by computing the expectation value of either the static operator in the time - evolved state or the time - evolved operator in the input state ( analogously to the usual schrdinger and heisenberg pictures for unitary dynamics ) . specifically , in terms of the kraus representation of the evolution , one has @xmath74 = \sum_\mathbf{s } { \textrm{tr } } \left [ k_\mathbf{s}^\dagger(t ) \hat{o } k_\mathbf{s}(t ) \rho_0 \right ] , \ ] ] where @xmath69 is the time - independent observable , @xmath75 is the input state , and @xmath76 are the kraus operators of the global channel . for independent channels acting on each qubit , @xmath77 , where the @xmath78 are the kraus operators acting on the @xmath79th qubit . in subsequent sections , we determine the precisions attainable under our model described by eq . ( [ eq.master ] ) for specific input states and measurements . the model has four kraus operators , which have the form @xmath80 here , the coefficients @xmath81 , @xmath82 are real and depend on the frequency @xmath3 , the noise parameters @xmath29 , @xmath83 , @xmath84 , @xmath85 , and the time @xmath9 ( see appendix [ app.mapcoeffs ] ) . however , to simplify notation we suppress these dependences . because of trace preservation , @xmath86 , the coefficients must satisfy @xmath87 for later calculations , it is useful to compute the evolution of both @xmath88 and @xmath89 under the kraus map . for @xmath88 , we have @xmath90 using eq . ( [ eq.coeffsum ] ) , the evolution under the channel can then be written as ( pauli operators with no explicit time dependence are time independent ) @xmath91 where the coefficients @xmath92 , @xmath93 are again real and encode the full dependence of the evolved operator on time , frequency , and the noise parameters . they are given in appendix [ app.mapcoeffs ] . similarly , one obtains @xmath94 with @xmath95 , @xmath96 . several experiments have demonstrated superclassical sensitivity of magnetometry with atomic ensembles by squeezing the collective atomic spin @xcite . considering the perpendicular model noise , we now show that spin - squeezed states and ramsey - type measurements together with interrogation - time optimisation are sufficient not only to reach precisions unattainable by classical protocols but also to maintain superclassical precision scaling with the particle number . ramsey interferometry performed on a collection of spin-@xmath97 particles ( qubits ) effectively corresponds to collective spin measurements @xcite . here , we consider the components of collective spin along @xmath51 and @xmath49 , @xmath98 which specify the observables measured in scenarios ( b ) and ( a ) of fig . [ fig.magnetosetup ] , respectively . the evolution of @xmath52 under the model of eq . ( [ eq.master ] ) follows directly from eq . ( [ eq.sigmaxevol ] ) , @xmath99 and similarly for @xmath50 using eq . ( [ eq.sigmayevol ] ) . the derivatives with respect to the estimated parameter then read @xmath100 and similarly for @xmath50 after interchanging @xmath101 . we also compute [ note that taking the square and evolving do not commute because the evolution is not unitary , i.e , @xmath102 @xmath103 so that from eqs . ( [ eq.jxevolv ] ) and ( [ eq.jxsqevolv ] ) we obtain the variance @xmath104 with @xmath105 denoting the covariance . the variance for @xmath50 is again obtained by just replacing @xmath106 . for a specific initial state of the atomic ensemble with both its expectation values and variances known at @xmath107 , we can substitute the above expressions into eq . ( [ eq.freqerrprop ] ) , in order to quantify the precision attained in scenarios ( a ) and ( b ) of fig . [ fig.magnetosetup ] for a given interrogation time @xmath9 . there is no unique definition of spin squeezing @xcite , but generally spin - squeezed states are states in which fluctuations of the collective spin component are reduced in a particular direction , when compared to the value they would have in a state with all individual spins aligned , i.e , in a _ coherent spin state _ ( css ) , an eigenstate of the corresponding spin component with maximal eigenvalue . spin - squeezed states are useful for metrology due to their enhanced sensitivity to any change of the collective spin in the squeezed direction , e.g. , caused by precession in a magnetic field . a number of experiments , in particular ref . @xcite , employ the so - called two - axis - twisted spin - squeezed states , which can be generated by quantum nondemolition measurement of the collective atomic spin mediated by light . however , here we focus on _ one - axis - twisted spin - squeezed states _ ( oatsss ) because they are amenable to analytical treatment . as two axis - twisted states allow for stronger suppression of the collective spin variance in a particular direction , i.e , stronger squeezing , we expect them to attain precisions at least as good as those derived below for oatsss . at the same time , quantum advantage with oatsss in magnetometry has been demonstrated in the experiment of ref . @xcite using a bose - einstein condensate , and the generation of oatsss using nitrogen - vacancy centres in diamond has been studied @xcite . oatsss are a particular kind of spin - squeezed states first introduced by kitagawa and ueda @xcite . they can be produced by first preparing atoms in a css along one direction , and then applying an evolution with a hamiltonian quadratic in one of the perpendicular spin components . for example , for spin-@xmath97 particles , one can start from an eigenstate of @xmath108 with eigenvalue @xmath109 ( all spins aligned along @xmath51 ) and apply an evolution with a hamiltonian proportional to @xmath110 . this will generate a state with minimum uncertainty at an angle to both the @xmath49 and @xmath48 axes , which depends on the strength of the evolution . the state can then be rotated to align the direction of minimum uncertainty with one of the axes . for scenarios ( a ) and ( b ) of fig . [ fig.magnetosetup ] , we consider two cases where the initial css is along either @xmath51 or @xmath49 , and the collective spin component with minimum uncertainty is @xmath111 or @xmath108 , respectively . for scenario ( a ) , the mean values of the collective spin are @xcite @xmath112 whereas the variances read @xmath113\\ ( \delta^{2}\hat{j}_{y})_0 & = \frac{n}{4}\left[1+\frac{1}{4}\left(n-1\right)\left[a-\sqrt{a^{2}+b^{2}}\right]\right]\\ ( \delta^{2}\hat{j}_{z})_0 & = \frac{n}{4}\left[1+\frac{1}{4}\left(n-1\right)\left[a+\sqrt{a^{2}+b^{2}}\right]\right ] , \end{split}\ ] ] with @xmath114 being the squeezing parameter , @xmath115 , and @xmath116 . we note that the covariance @xmath117 vanishes for this state . the equivalents of eqs . ( [ eq.oatssexpec ] ) and ( [ eq.oatssvar ] ) for scenario ( b ) are obtained by interchanging @xmath101 . the mean - squared errors of estimation , which are achieved in scenarios ( a ) and ( b ) of fig . [ fig.magnetosetup ] , can be calculated by using eqs . ( [ eq.oatssexpec ] ) , ( [ eq.oatssvar ] ) , ( [ eq.jxexpecevolv ] ) , ( [ eq.jxvarevolv ] ) ( and the equivalents for @xmath50 ) and substituting into eq . ( [ eq.freqerrprop ] ) . the best precision is then obtained by optimising the evolution time @xmath9 and the squeezing @xmath114 for each @xmath1 . the general expressions are rather involved , and we have been able to obtain their minima only numerically . however , any explicit choice of @xmath12 and @xmath118 provides a precision that is guaranteed to be attainable . specifically , for scenario ( b ) a choice that appears to be nearly optimal is @xmath119 and @xmath120 . is not perfectly known in advance , this choice is not a problem . replacing @xmath121 in @xmath114 , @xmath9 has the effect @xmath122 in eq . ( [ eq.squeezenonzeroperpscaling ] ) . similarly , @xmath123 in @xmath114 , @xmath9 leads to @xmath122 in eq . ( [ eq.squeezenonzeroperpscaling ] ) . the scaling remains unchanged . ] for this choice , we expand eq . ( [ eq.freqerrprop ] ) in @xmath0 to find the expression for the asymptotic mean - squared error @xmath124 since the scaling is better than the @xmath0 of the sql , this demonstrates that superclassical precision scaling is indeed possible with spin - squeezed states and ramsey - type measurements in the presence of transversal noise . the possibility for a large quantum enhancement depends strongly on the geometry . we can see this by comparing with scenario ( a ) . there , we find that @xmath125 which coincides with the best achievable precision for the parallel - noise setting @xcite constrained by eq . ( [ eq.cz ] ) . as discussed in appendix [ app.scenarioabound ] , we find an analytical proof of eq . ( [ eq.squeezenonzeroperpscalinga ] ) in the limit @xmath126 , and strong numerical evidence for arbitrary @xmath3 , which indicates very clearly that no better precision can be achieved . the value is attainable by any choice of @xmath127 and @xmath128 with @xmath129 . thus , for this geometry , under transversal noise only sql - like scaling is possible and the quantum enhancement over classical , nonentangled strategies is bounded , while for scenario ( b ) , the quantum enhancement is unbounded with increasing @xmath1 . one can understand intuitively why scenario ( b ) is more robust than ( a ) from the pictures of the collective spin in fig . [ fig.magnetosetup ] . in both cases the dephasing is directed along @xmath51 and so causes random rotations around the @xmath51 axis . the effect of such rotations on the state in ( a ) is to smear the squeezed state into a more circular distribution at the pole . this reduces the sensitivity of a @xmath50 measurement to a small rotation around the z axis , which is the signal we want to detect . in scenario ( b ) , the effect of @xmath51 rotations is to smear the state along the equator . however , this does not affect the sensitivity of the @xmath52 measurement to a @xmath48 rotation as strongly . using the same picture , one can also understand why there is a finite optimal value of the squeezing parameter @xmath114 . when @xmath114 becomes large , the ellipse starts to stretch around the ball . referring to scenario ( b ) , a @xmath48 rotation will then increase the projection of the state onto the @xmath51 axis and hence the @xmath52 measurement loses sensitivity . , @xmath130 , @xmath131 . in ( i ) the interrogation time is fixed at @xmath132 while in ( ii ) the squeezing is fixed at @xmath133 ( indicated by the dotted lines , these values correspond to the numbers discussed in the text ) . the results for scenario ( a ) ( _ yellow line _ ) , scenario ( b ) ( _ blue line _ ) , and for css ( _ dashed gray line _ ) are shown . ] the difference between the two geometries shows up only for quantum strategies , that is , when squeezed states are employed . if the initial states are not squeezed but are simply css states along @xmath51 for scenario ( a ) or @xmath49 for scenario ( b ) , then the precision takes the same form in both cases : @xmath134 } { \left[2\left(1\!-\!{\mathrm{e}}^{-\gamma t\bar{\gamma}}\right)-\gamma t\gamma^{2}\bar{\gamma}\left(1\!+\!{\mathrm{e}}^{-\gamma t\bar{\gamma}}\right)\right]^{2 } } \frac{1}{n},\ ] ] where @xmath135 and @xmath136 . thus , we can benchmark the quantum enhancement in either scenario against this classical value . in particular , we consider the numbers from ref . @xcite . in this experiment , @xmath137 , @xmath138 , @xmath139 , and @xmath140 , which gives @xmath141 and @xmath142 , and the measurement time is @xmath143 . the experiment is not performed with oatsss , but we compute the quantum enhancements that oatsss would provide . we insert the numbers in the full expressions [ from eq . ( [ eq.freqerrprop ] ) ] for @xmath144 and @xmath145 and vary the squeezing , which we quantify in db ; introduced in @xcite and use eqs . ( 112 ) and ( 139 ) of @xcite . ] see fig . [ fig.var_vs_tmu](i ) . the precision in scenario ( a ) saturates with increasing squeezing , and the best quantum enhancement attainable is @xmath146 . in scenario ( b ) , on the other hand , the enhancement can reach @xmath147 , corresponding to a 4500-fold improvement in precision . in this case , the precision does not saturate but is limited by the maximal squeezing attainable by the oatss . this underlines the advantage offered by geometry . however , these maximal enhancements require rather prohibitive squeezings of @xmath148 and @xmath149 , respectively . if we restrict the squeezing to at most @xmath133 as discussed in the outlook of ref . @xcite , then scenario ( b ) provides an enhancement of @xmath150 , corresponding to a factor of 2.5 in precision , while scenario ( a ) for the same numbers gives @xmath151 corresponding to a factor of 1.95 . however , the behaviour in the two scenarios for varying interrogation time is very different , as explicitly shown in fig . [ fig.var_vs_tmu](ii ) . the performance is similar at short times , but when the interrogation time is increased , the quantum advantage in scenario ( a ) is lost , while the advantage in scenario ( b ) is maintained . this is a nice feature of the optimized geometry , as in practice experimental constraints may impose a lower limit on @xmath9 . as seen from fig . [ fig.var_vs_tmu](ii ) , at a squeezing of @xmath133 , the best precision in scenario ( a ) ( optimising @xmath9 ) is not significantly below the best precision attainable with a css , while the advantage of scenario ( b ) importantly remains even when experimental constraints do not allow for arbitrarily small @xmath9 . we note that , as for eq . ( [ eq.prevbound ] ) , the error ( [ eq.squeezenonzeroperpscaling ] ) vanishes as @xmath152 . we refer the reader to appendix [ app.omega_zero ] for a discussion of this limit . in a realistic implementation , in addition to the dominant transversal noise , other sources of decoherence will be present . for example , in some setups different from ref . @xcite , e.g. , spin - exchange relaxation - free magnetometers @xcite , spin relaxation can not be neglected . typically @xcite , this is modelled as uncorrelated depolarising noise [ @xmath153 in eq . ( [ lindblad ] ) ] with a strength dictated by the spin - relaxation time @xmath55 ( see appendix [ app.master ] ) . as opposed to the directional dephasing caused by spin decoherence , such noise is isotropic and always yields a parallel - noise component independently of the geometry . in the case of ref . @xcite , any spatial misalignment between @xmath46 , @xmath45 , and the direction of squeezing , temporal mismatch between the @xmath45 rotations , and the larmor frequency of @xmath46 , or violation of the condition @xmath154 , may be phenomenologically included into such a depolarisation model . we assess the effect of such additional noise sources by considering a deviation from perfect transversality . in particular , we take a small component of dephasing directed along the @xmath48 axis , such that @xmath155 and @xmath156 . as discussed in ref . @xcite , once any such parallel - dephasing contribution is present , the asymptotic scaling must return to its sql - like behaviour ; that is , @xmath157 with @xmath158 lower - bounded by the minimum of eq . ( [ eq.cz ] ) , i.e , @xmath159 . for instance , for the depolarisation - based spin - relaxation model , @xmath160 ( see appendix [ app.master ] ) . we illustrate the resulting crossover behaviour in fig . [ fig.deviations ] . [ ! t ] from eq . ( [ eq.squeezenonzeroperpscalinga ] ) , @xmath161 from the parallel noise component , and @xmath162 from eq . ( [ eq.squeezenonzeroperpscaling ] ) . we also show the performance of a css without squeezing ( _ grey dashed line _ ) as described by eq . ( [ eq.cssscaling ] ) . the locations where the oatss strategies reach 90% of their asymptotic gain over this css strategy are indicated for scenarios ( a ) ( _ open triangle _ ) and ( b ) ( _ closed triangle_).,title="fig : " ] although the asymptotic scaling is now again sql - like , geometry can strongly influence the achievable quantum gain and the effective @xmath1 at which the crossover to sql - like scaling happens . in fig . [ fig.deviations ] we show the mean - squared - error scalings attained in scenarios ( a ) and ( b ) using , respectively , oatsss along @xmath51 and @xmath49 and measurements of @xmath50 and @xmath52 and compare them to a strategy without entanglement , simply using a css along @xmath51 and measurement of @xmath50 corresponding to the nonentangled strategy implemented in ref . we see that while the strategy in ( b ) can saturate the bound @xmath163 , the strategy in ( a ) only reaches @xmath164 as imposed by eq . ( [ eq.squeezenonzeroperpscalinga ] ) . thus , the mean - squared error of geometry ( b ) is a factor @xmath165 lower than ( a ) , which may be significant when the noise is dominantly transversal . furthermore , superclassical scaling persists over a larger range of @xmath1 in geometry ( b ) . in the figure , the locations where the oatss strategies for ( a ) and ( b ) reach 90% of their asymptotic gain over the nonentangled css strategy are indicated . clearly , the crossover happens at much larger @xmath1 in scenario ( b ) . as @xmath166 , the crossover must go to infinity . to get an idea of the behaviour , we can take the @xmath1 at which the asymptotic bound @xmath167 crosses the asymptote ( [ eq.squeezenonzeroperpscaling ] ) for perfectly transversal noise . this intersection scales as @xmath168 . thus , significant gain in precision by squeezing is attained over a larger range of @xmath1 if the geometry is chosen correctly . for quantum metrology to be relevant in practical situations , it is important that good performance can be attained under realistic noise with states and measurements that are amenable to implementation in the laboratory . while recent results have shown that for many noise types precision scaling can only improve over the classical limit by a constant , here we demonstrate that under transversal dephasing , superclassical scaling can be preserved with experimentally accessible states and measurements , and we argue that this noise model is relevant to recent atomic magnetometry experiments . we show that the choice of geometry is important for the attainable quantum improvement both asymptotically and for parameter settings corresponding to recent experiments . furthermore , we assess the robustness of the model to other nontransversal sources of noise and find that quantum enhancement could still be achieved for atomic ensembles of macroscopic size with an adequate choice of geometry . our results give a clear message that quantum - enhanced metrology maintains its relevance even in the presence of noise , and we hope that they will encourage the search for other practically motivated scenarios where quantum strategies provide an advantage . for instance , it has been suggested that the transversal - noise model applies also to nitrogen - vacancy centres in diamonds @xcite , and that one - axis - twisted spin - squeezed states could be prepared in such systems @xcite . very recently , a noise - robust magnetometry scheme employing squid junctions has been proposed @xcite . finally , in the appendices , we speculate about the potential of adaptive techniques that bias the estimated parameter towards the zero value , for which our current precision bounds fail . we expect that the question of what happens in this limit could be consistently resolved by employing bayesian techniques @xcite , which account explicitly for the prior knowledge about the estimated parameter . we are grateful to antonio acn , stefan christensen , rafa demkowicz - dobrzaski , marcin jarzyna and wojciech wasilewski for valuable feedback and discussions . j. b. b. acknowledges support from the swiss national science foundation ( qsit director s reserve ) , sefri ( cost action mp1006 ) , and fp7 ip project siqs . r. c. acknowledges support from the excellence initiative of the german federal and state governments ( grant zuk 43 and 81 ) , the research innovation fund from the university of freiburg , and the u.s . army research office under contracts no . w911nf-14 - 1 - 0098 and no . w911nf-14 - 1 - 0133 ( quantum characterisation , verification , and validation ) . j. k. acknowledges support from the fp7 ip project siqs co - financed by the polish ministry of science and higher education , start scholarship granted by foundation for polish science , the erc consolidator grant qitbox , as well as eu msca individual fellowship q - metapp . following ref . @xcite , we assume correlated effects in the atomic ensemble to play a role only at time scales longer than the interrogation time , i.e , the ensemble relaxation and decoherence times @xmath169 @xcite ( in ref . @xcite , @xmath170 ) . we may then describe the dynamics by a master equation where each atom evolves independently @xmath171\ ] ] where @xmath172 generates the evolution for a single atom ( assumed to be the same for all atoms ) and reads @xmath173 & = & -{\mathrm{i}}\frac{\omega_b}{2}\left[\hat{\sigma}_{x},\rho(t)\right]-{\mathrm{i}}\frac{\omega}{2}\left[\hat{\sigma}_{yz}(t),\rho(t)\right ] \label{eq.gen_sing_part_h}\\ & & + \frac{1}{t_{2}}\left(\hat{\sigma}_{x}\rho(t)\hat{\sigma}_{x}-\rho(t)\right ) \label{eq.gen_sing_part_t2}\\ & & + \frac{1}{t_{1}}\left(\frac{1}{3}\sum_{i=1}^{3}\hat{\sigma}_{i}\,\rho(t)\hat{\sigma}_{i}-\rho(t)\right ) . \label{eq.gen_sing_part_t1}\end{aligned}\ ] ] the terms ( [ eq.gen_sing_part_h ] ) represent the hamiltonian part of the dynamics : @xmath174 is the larmor frequency of the strong field @xmath46 ( see fig . [ fig.magnetosetup ] ) , whereas @xmath175 is the frequency corresponding to the weak field @xmath45 being sensed . having ref . @xcite in mind , we allow @xmath45 to rotate in the @xmath176 plane @xmath177 . the term listed in eq . ( [ eq.gen_sing_part_t2 ] ) represents dephasing noise . it is a consequence of @xmath46-field fluctuations independently affecting each atom ( e.g. , arising from the atomic motion and/or local field inhomogeneities ) . equivalently , it may be derived by considering an average - description of a noiseless evolution but with @xmath178 fluctuating according to a gaussian distribution with variance @xmath179 ( note that the variance diverges in the @xmath18 limit manifesting the markovianity of the noise ) . such fluctuations define the _ spin - decoherence time _ @xmath54 @xcite and constitute the transversal @xcite noise which is the main focus of this paper . the term ( [ eq.gen_sing_part_t1 ] ) represents the spin - relaxation process occurring predominantly due to spin - destruction or spin - exchange atomic collisions @xcite ( the latter may be eliminated in the accordingly called spin - exchange relaxation - free magnetometers @xcite ) . note that the term ( [ eq.gen_sing_part_t1 ] ) effectively yields a depolarising channel [ as defined in eq . ( [ lindblad ] ) ] with strength parametrised by the _ spin - relaxation time _ @xmath55 @xcite . by moving to the rotating frame precessing with the larmor frequency , in which @xmath180 , and by using that @xmath181 , we obtain the single - atom master equation in the rf as @xmath182 \nonumber\\ & & + \frac{1}{t_{2}}\left(\hat{\sigma}_{x}\rho_{\textrm{rf}}(t)\hat{\sigma}_{x}-\rho_{\textrm{rf}}(t)\right)\nonumber \\ & & + \frac{1}{t_{1}}\left(\frac{1}{3}\sum_{i=1}^{3}\hat{\sigma}_{i}\,\rho_{\textrm{rf}}(t)\hat{\sigma}_{i}-\rho_{\textrm{rf}}(t)\right ) . \label{eq.gen_sing_part_rf}\end{aligned}\ ] ] with @xmath183 . as a consequence , if the larmor frequency is chosen to exactly match the rotation of @xmath45 , i.e , @xmath184 , and if we consider the limit @xmath185 in which the spin relaxation is completely ignored ( as in @xcite ) , then we indeed recover eq . ( [ eq.master_magnetometry ] ) in the main text . note that eq . ( [ eq.gen_sing_part_rf ] ) indicates that by _ not _ exactly matching @xmath186 with the larmor frequency we nonetheless preserve the required transversal geometry between the dephasing noise eq . ( [ eq.gen_sing_part_t2 ] ) and the @xmath3-encoding part , so that the perpendicular - noise model of eq . ( [ lindblad ] ) introduced in ref . @xcite , in principle , still applies . however , let us emphasize that for our analysis of squeezed states to be valid , the geometry of squeezing must always be adjusted for a particular choice of @xmath45 direction in the rf , e.g. , depicted in fig . [ fig.magnetosetup ] for @xmath45 chosen to be stationary and pointing along @xmath48 in the rf . in an experiment , we expect there to be some mismatch between the @xmath45-rotation frequency and @xmath178 , which would constitute an extra source of _ global _ ( correlated across particles ) dephasing noise along @xmath51 , i.e , decreasing the ensemble spin - decoherence time @xmath187 neglected in ref . @xcite . in contrast , any such fluctuations felt locally by atoms would just lower @xmath54 in eq . ( [ eq.gen_sing_part_t2 ] ) , which would thus not impair the robustness of the setup . on the other hand , any geometrical misalignment of the fields or any instabilities of @xmath186 can always be modelled in eq . ( [ eq.gen_sing_part_rf ] ) by lowering the effective spin - relaxation time @xmath55 . importantly , the spin - relaxation process may be interpreted as a nontransversal - noise source , which we study in sec . [ sub.non_trans_noise ] . similarly to the parallel dephasing case [ @xmath37 in eq . ( [ lindblad ] ) ] , which is known to asymptotically restrict the precision to @xmath188 @xcite , it has been shown @xcite that depolarising noise [ @xmath153 in eq . ( [ lindblad ] ) ] yields a sql - like asymptotic precision @xmath189 , thus giving just a @xmath190-factor improvement as compared to the pure parallel - noise model . a finite spin - relaxation time must thus impose an asymptotic sql - like behaviour , which we may estimate by taking @xmath191 of @xmath55 in eq . ( [ eq.gen_sing_part_rf ] ) to contribute solely to the parallel - dephasing component . then , after substituting for @xmath192 and @xmath193 , we can directly utilise eq . ( [ eq.xzbound ] ) to obtain an asymptotic bound with @xmath194 dictated by the spin relaxation . crucially , the analysis of sec . [ sub.non_trans_noise ] thus shows that in the regime of @xmath195 , the optimal geometry of scenario ( b ) not only allows the counterbalance of the spin - decoherence effects , but also postpones the inevitable sql - bounding impact of spin relaxation to much higher @xmath1 ( see the triangular marks in fig . [ fig.deviations ] ) . the map corresponding to evolution under the master equation eq . ( [ eq.master ] ) during time @xmath9 can be written as a composite map of the form @xmath196 . following andersson _ et al . _ @xcite , the single - qubit maps are then given by @xmath197 where @xmath198 are the normalised pauli operators @xmath199 and @xmath200 denotes the identity . all elements of the matrix s are zero , except @xmath201 , @xmath202 , @xmath203 , @xmath204 , @xmath205 , @xmath206 . where we have defined @xmath135 , @xmath207 , and @xmath208 , and the coefficients @xmath209 a kraus representation of the map @xmath210 can be obtained by diagonalising the matrix @xmath211 . denoting the eigenvalues and normalised eigenvectors of @xmath211 by @xmath212 and @xmath213 respectively , one can find a valid set of kraus operators for the channel : @xmath214 with @xmath215 , which gives the set in eq . ( [ eq.krausops ] ) . the coefficients in eq . ( [ eq.krausops ] ) are rather involved , and we do not explicitly state them here . instead , we directly give the expressions for @xmath216 , @xmath217 , @xmath218 , and @xmath219 of eqs . ( [ eq.sigmaxevol ] ) and ( [ eq.sigmayevol ] ) . for general noise they read @xmath220 , \nonumber \\ \chi_y & = - \,{\mathrm{e}}^{-\frac{\gamma t}{2}(1+\alpha_{z})}\frac{\gamma}{\tilde{\alpha}}\sinh\!\left(\frac{\gamma t}{2}\tilde{\alpha}\right ) , \end{aligned}\ ] ] and @xmath221 , \nonumber \\ \chi_y & = { \mathrm{e}}^{-\frac{\gamma t}{2}(1+\alpha_{z})}\frac{\gamma}{\tilde{\alpha}}\sinh\!\left(\frac{\gamma t}{2}\tilde{\alpha}\right).\end{aligned}\ ] ] in the case of perfectly transversal noise they further simplify , since @xmath36 implies @xmath222 and @xmath223 . stemming from the error - propagation method [ see eq . ( [ eq.freqerrprop ] ) ] utilised in the main text , we can also confirm the results of ref . @xcite analytically for the _ ghz input states _ : @xmath224 by considering the _ parity operator _ in the @xmath51 direction : @xmath225 as the observable being measured . similarly to the case of collective spin operators and eq . ( [ eq.jxevolv ] ) , we may utilise eq . ( [ eq.sigmaxevol ] ) to write the form of the parity operator at time @xmath9 as @xmath226 in the computational basis @xmath227 , such an operator just flips all of the qubits , and hence only the off - diagonal terms contribute when calculating its expectation value for a ghz state of eq . ( [ eq.ghz ] ) . every @xmath88 contributes a factor of 1 while @xmath89 contributes a factor of @xmath228 . thus , the expectation value of the measurement becomes @xmath229,\ ] ] and , since @xmath230 , it follows that @xmath231 . we compute the mean - squared error of estimation via eq . ( [ eq.freqerrprop ] ) , after setting the interrogation time to @xmath232 , as was found in ref . @xcite from numerical analysis . expanding the corresponding @xmath233 in @xmath0 , we find the asymptotic scaling to read : @xmath234 where @xmath235 represents oscillating terms that are lower - bounded by @xmath236 . the constant prefactor here is larger than the prefactor @xmath237 , which was numerically verified to be optimal optimised over all possible measurements for ghz states @xcite . nevertheless , although this suggests that either parity measurement is suboptimal or the above interrogation time @xmath9 dependence should be improved in the parity - based scenario , eq . ( [ eq : parity_bound ] ) suffices to prove the superclassical precision scaling , @xmath19 , as well as the @xmath238 behaviour of the asymptotic coefficient . for @xmath239 , both the ghz - achievable bound ( [ eq.prevbound ] ) and the oatss - based expression ( [ eq.squeezenonzeroperpscaling ] ) vanish . this does not mean that the precision is unbounded for the two cases , but rather suggests that the results give no information in such a limit . it is therefore not clear , what precision scaling can then be achieved . in general , for the channel described by eq . ( [ map ] ) at @xmath240 , we get @xmath241 , @xmath242 , and @xmath243 , @xmath244 . for a ghz state ( [ eq.ghz ] ) and parity measurement eq . ( [ eq.parity ] ) , one can show utilising eq . ( [ eq.freqerrprop ] ) that @xmath245 for fixed @xmath9 . this is minimised at @xmath246 , where @xmath59 is a numerical constant . similarly , at @xmath247 for an oatss along @xmath49 squeezed in @xmath51 [ as in scenario ( b ) of fig . [ fig.magnetosetup ] of the main text ] with squeezing parameter @xmath248 one finds @xmath249 and again the optimal time is @xmath250 . thus , on the one hand , the local estimation approach which we employ above indicates that an improved scaling , even reaching the heisenberg limit for ghz states , is possible at the special @xmath239 parameter value . even if the value of @xmath3 is _ a priori _ nonzero , one might then think that the precision scaling can be improved by adopting an iterative , adaptive strategy @xcite . by applying a bias ( e.g. , in the case of magnetometry , a magnetic field in opposite direction to the estimated field ) to decrease the parameter after obtaining its first estimate , a better estimate is obtained with a precision which less heavily constrained by bounds eqs . ( [ eq.prevbound ] ) and ( [ eq.squeezenonzeroperpscaling ] ) , due to the lower effective value of @xmath3 . on the other hand , the prior information on @xmath3 required to adjust the bias may scale prohibitively . we can compute the estimated mean - squared error for ghz states and parity measurements ( see above ) and expand it in @xmath251 to obtain @xmath252 , with @xmath13 given by eq . ( [ eq.ghzzeroscaling ] ) . for the heisenberg - scaling term to dominate for a fixed @xmath9 , the higher - order terms in the expansion must be negligible in comparison . however , we find ( for even terms , odd terms vanish ) that the @xmath25th term scales as @xmath253 , which implies that we need @xmath254 to neglect the higher - order terms . for this to hold for all @xmath25 , @xmath255 , which means that the prior information on @xmath3 must already be heisenberg limited , as in the case of a decoherence - free local estimation scenario @xcite . at first sight , this may indicate that such an adaptive scheme may not be successful for any prior distribution of finite width , and that the value of @xmath3 must be perfectly known and set to zero for the above improved scalings to be observed . however , recent results @xcite , based on the bayesian approach to estimation , indicate that in the decoherence - free case the heisenberg scaling is attained irrespectively of the prior knowledge of @xmath3 . hence , we expect the transversal - noise model to behave similarly due to its decoherence - free - like regime at short interrogation times , which would then prove the above adaptive strategy to also be efficient . we give below an analytical proof of eq . ( [ eq.squeezenonzeroperpscalinga ] ) in the limit @xmath126 and strong numerical evidence for arbitrary @xmath3 . following the method outlined at the beginning of sec . [ sub.varscaling ] , we find expressions for @xmath256 and @xmath257 . we are looking for a lower bound on the former . from the expressions it can be seen that , in the case @xmath247 @xmath258 now , since both the prefactor @xmath259 and @xmath257 are positive , any lower bound on this quantity is also a lower bound on @xmath256 . the @xmath9-dependent factor @xmath260 is lower bounded by @xmath261 ( attained when @xmath18 ) , while @xmath262 is lower bounded by 1 . it follows that @xmath263 is lower bounded by @xmath188 as desired . when @xmath267 , the expressions for @xmath256 and @xmath257 become significantly more complicated . however , we can again look at a quantity @xmath268 where @xmath269 since @xmath270 and @xmath257 are positive , a lower bound on @xmath271 is again also a lower bound on @xmath256 . we have not been able to prove an analytical bound for @xmath271 , but for given values of @xmath3 , @xmath29 , @xmath1 we can numerically minimise @xmath271 over @xmath114 and @xmath9 . in fig . [ fig.scenarioabound ] we plot the results of several such minimisations . as can be seen , the numerics give very clear evidence that @xmath272 and hence @xmath256 is lower bounded by @xmath188 as claimed . 61ifxundefined [ 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 10.1126/science.1104149 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.010401 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.133601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.093602 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.253605 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.143001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.160802 [ * * , ( ) ] link:\doibase 10.1063/1.4901588 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.103004 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.86.5870 [ * * , ( ) ] link:\doibase 10.1126/science.1097576 [ * * , ( ) ] link:\doibase 10.1073/pnas.0901550106 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/12/6/065032 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.79.3865 [ * * , ( ) ] link:\doibase 10.1103/physreva.64.052106 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.032111 [ * * , ( ) ] link:\doibase 10.1088/1751 - 8113/41/25/255304 [ * * , ( ) ] link:\doibase 10.1109/tit.2008.929940 [ * * , ( ) ] link:\doibase 10.1007/s00220 - 011 - 1239 - 4 [ * * , ( ) ] link:\doibase 10.1038/ncomms2067 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/15/7/073043 [ * * , ( ) ] _ _ , http://arxiv.org/abs/1409.0535[ph.d . thesis ] , ( ) link:\doibase 10.1103/physrevlett.113.250801 [ * * , ( ) ] link:\doibase 10.1038/nphys1958 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.012103 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.233601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.120401 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.150802 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.080801 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.150801 [ * * , ( ) ] http://stacks.iop.org/1367-2630/16/i=8/a=083010 [ * * , ( ) ] http://arxiv.org/abs/1402.0495 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/17/1/013010 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.93.173002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.073002 [ * * , ( ) ] http://stacks.iop.org/0953-4075/48/i=3/a=035502 [ * * , ( ) ] link:\doibase 10.1103/physreva.46.r6797 [ * * , ( ) ] link:\doibase 10.1103/physreva.50.67 [ * * , ( ) ] link:\doibase 10.1103/physreva.54.r4649 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.82.2207 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.230801 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/16/11/113002 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.123601 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.090801 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.190403 [ * * , ( ) ] @noop link:\doibase 10.1103/physrevlett.72.3439 [ * * , ( ) ] `` , '' ( , ) link:\doibase 10.1016/j.physrep.2011.08.003 [ * * , ( ) ] link:\doibase 10.1103/physrevb.72.121303 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.156402 [ * * , ( ) ] link:\doibase 10.1103/physreva.47.5138 [ * * , ( ) ] link:\doibase 10.1103/physreva.77.033408 [ * * , ( ) ] link:\doibase 10.1364/oe.18.027167 [ * * , ( ) ] http://arxiv.org/abs/1410.7556 [ * * , ( ) ] link:\doibase 10.1080/09500340701352581 [ * * , ( ) ] link:\doibase 10.1109/jstqe.2009.2020810 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.5098 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.063603 [ * * , ( ) ]

for toric calabi - yau 3-folds , open gromov - witten invariants associated to riemann surfaces with one boundary component can be written as the product of a disk factor and a closed invariant . using the brini - cavalieri - ross formalism , these disk factors can often be expressed in terms of gamma classes . when the lagrangian boundary cycle is preserved by the torus action and can be locally described as the fixed locus of an anti - holomorphic involution , we prove a formula that expresses the disk factor in terms of a gamma class and combinatorial data about the image of the lagrangian cycle in the moment polytope . we verify that this formula encodes the expected invariants obtained from localization by comparing with several examples . we then examine a novel application of this formula to disk enumeration on the quintic 3-fold . finally , motivated by large duality , we show that this formula also unexpectedly applies to lagrangian cycles on constructed from torus knots .

1601.00203

gromov - witten theory has a rich history , both in physics and mathematics . physically , gromov - witten invariants appear in type iia topological string theory as instanton counts associated to interactions between particles . mathematically , they are invariants associated to symplectic manifolds that , roughly speaking , count pseudoholomorphic curves in the manifold . the relationship between these two perspectives is conceptually straightforward : as a string moves in time , it sweeps out a compact riemann surface ( its ` worldsheet ' ) . the amplitudes in string theory encode counts of maps from riemann surfaces into a 3-(complex)-dimensional calabi - yau manifold , and gromov - witten theory assigns invariants to spaces of such maps . in general , counting holomorphic maps from riemann surfaces to a given target space is a difficult problem in enumerative geometry . gromov - witten theory has famously benefited from its connections with string dualities , first with mirror symmetry @xcite , and more recently , large @xmath0 duality @xcite . beginning with @xcite , for toric manifolds , gromov - witten invariants associated to maps of closed surfaces have also been systematically computed using localization @xcite . `` closed '' gromov - witten theory is a natural mathematical counterpart to closed topological string theory , and , in contrast to the `` open '' theory ( i.e. , for maps of riemann surfaces with boundary ) , the moduli spaces involved are rigorously defined . open gromov - witten theory is the subject of this paper . by analogy with the closed case , open gromov - witten theory is a mathematical counterpart to open topological string theory : open strings sweep out compact riemannian surfaces with boundary , and the boundary of the strings are constrained to lie on branes . these boundary constraints are expressed mathematically as lagrangian submanifolds @xmath2 , and the string amplitudes are encoded by counts of holomorphic maps @xmath3 , with the image of the boundary constrained to lie on @xmath4 : @xmath5 . however , as observed in @xcite , there are additional subtleties in adapting the methods of the closed theory to the open case . in particular , even for well - behaved lagrangian boundary cycles , open gromov - witten invariants depend on an additional integral parameter ( in localization , this parameter corresponds to the weights of the torus action ) . in spite of this , the same computational tools of mirror symmetry , large @xmath0 duality , and localization can still be used . in fact , through these string dualities , open gromov - witten theory can be connected to both classical and homological knot theory @xcite . motivated by relationships with the crepant resolution conjecture @xcite , open gromov - witten theory can also be generalized to orbifolds @xcite . this setting has driven a more abstract formulation of open gromov - witten invariants , which has led to a deeper understanding of the a - model . in particular , as will be discussed in detail below , the open invariants contain gamma classes coming from disk terms @xcite . in addition , the open gromov - witten generating function can be obtained from a modification of givental s @xmath6 function @xcite . the primary goal of this paper is to describe a concise and consistent framework for computing open gromov - witten invariants directly , via localization . leveraging the formalism of @xcite , the main tool is a formula for open gromov - witten invariants expressed in terms of local combinatorial data and a gamma class . as expected from @xcite , the construction depends only on the local geometry near a vertex of the moment polytope of @xmath7 . in the case where the associated moduli space of open maps is rigorously defined ( @xcite ) , this formula is proven to be correct . most intriguingly , this result is shown to apply in two unexpected contexts : enumerative invariants associated to the quintic 3-fold , and lagrangian cycles obtained from torus knots appearing in large @xmath0 duality . the author hopes that the approach described herein will lead to a more general construction of open gromov - witten invariants . the paper is organized in the following way . section [ sec : preliminaries ] reviews some general facts about open gromov - witten theory , including deformation theory and localization . most importantly , this section describes how to express an open gromov - witten invariant as the product of a `` disk term '' and an invariant of closed maps . section [ sec : main - results ] contains the proof of the main computational tool of this paper : let @xmath7 be a calabi - yau 3-fold and @xmath2 a lagrangian submanifold . let @xmath8 act on @xmath7 such that the @xmath8 action preserves @xmath4 , and @xmath4 intersects a rigid circle - invariant curve @xmath9 . suppose that @xmath4 can be described in a neighborhood of @xmath10 as the fixed locus of an anti - holomorphic involution . let @xmath11 . then , the genus @xmath12 , 1 boundary component , degree @xmath13 , winding @xmath14 open gromov - witten invariant with lagrangian boundary @xmath4 is @xmath15 where @xmath16 is the disk function @xmath17 here , @xmath18 is the homogeneous iritani gamma class , @xmath19 is the weight of the @xmath8 action along a normal direction to @xmath9 , @xmath20 is the equivariant chern class of the induced representation of @xmath8 at the attachment point of the disk , and @xmath21 is the equivariant class of the image @xmath22 of the disk attachment point . this gamma class formula was encountered previously in @xcite , where the authors study a lagrangian locally described as the fixed locus of the antiholomorphic involution @xmath23 . as will be seen ( lemma [ lem : involution - form ] ) , this result implies the formula above after a change of coordinates . section [ sec : comparisons ] describes how to apply the gamma class formula to several examples where the resulting invariant is already known , and demonstrates that this formula reproduces the expected result . the last two sections study two examples where the assumptions on the local geometry of @xmath4 are not satisfied . section [ sec : quintic ] examines disk enumeration on the quintic 3-fold and finds that a slight modification of this formula again applies . finally , section [ sec : torus - knot - lagrangians ] applies this formula to a novel class of lagrangian cycles motivated by large @xmath0 duality . these lagrangian cycles are obtained from the conormal bundles of torus knots in @xmath24 after the conifold transition , and do not have the same local description required in the above theorem . nevertheless , the main result of this paper is still found to apply to these cycles . the examples in sections [ sec : quintic ] and [ sec : torus - knot - lagrangians ] hint that a version of the theorem may hold for a broader class of lagrangian cycles . the author thanks h. gao , h. jockers , c - c . m. liu , and p. zhou for valuable discussions , and a. brini for helpful corrections . the author is especially grateful to r. cavalieri for the suggestion of this project and many related conversations , and to e. zaslow for guidance and suggestions . open gromov - witten invariants are enumerative invariants of maps @xmath3 from riemann surfaces with boundary into a calabi - yau manifold @xmath7 with a chosen lagrangian submanifold @xmath4 , such that @xmath5 . by analogy with the definition for closed stable maps appearing in @xcite , @xcite define the open gromov - witten invariant in the following way . fix integers @xmath12 ( the genus of @xmath25 ) and @xmath26 ( the number of connected components of @xmath27 ) , and a relative homology class @xmath28 with @xmath29 . then , the open gromov - witten invariant @xmath30 is a virtual count of continuous maps @xmath31 satisfying : * @xmath32 is a riemann surface of genus @xmath12 with boundary @xmath33@xmath27 consisting of @xmath26 oriented circles , * @xmath34 is holomorphic in the interior of @xmath25 , * @xmath35=d$ ] , and * @xmath36=\sum w_{i}$ ] . for brevity , @xmath37 will sometimes be denoted by @xmath38 . in order to define such an invariant , @xcite construct a moduli space @xmath39 of stable maps which compactify the maps described above , and give a local description of an orientation and a virtual fundamental class on this moduli space . in particular , the authors generalize the deformation complex in ordinary gromov - witten theory to the open case . recall that for smooth , closed @xmath25 in ordinary gromov - witten theory , there is a normal bundle exact sequence of vector bundles on @xmath25 : @xmath40 & t_{\sigma}\ar[r ] & f^{*}t_{x}\ar[r ] & n_{\sigma / x}\ar[r ] & 0 } .\ ] ] the corresponding long exact sequence in cohomology is \(m ) [ matrix of math nodes , row sep=1em , column sep=2.5em , text height=1.5ex , text depth=0.25ex ] 0 & h^0(,t _ ) & h^0 ( , f^*t_x ) & h^0 ( , n_/x ) & + & h^1(,t _ ) & h^1 ( , f^*t_x ) & h^1 ( , n_/x ) & 0 + ; ( m-1 - 1 ) edge ( m-1 - 2 ) ( m-1 - 2 ) edge ( m-1 - 3 ) ( m-1 - 3 ) edge ( m-1 - 4 ) ( m-1 - 4 ) edge[out=355,in=175 ] ( m-2 - 2 ) ( m-2 - 2 ) edge ( m-2 - 3 ) ( m-2 - 3 ) edge ( m-2 - 4 ) ( m-2 - 4 ) edge ( m-2 - 5 ) ; the terms in this sequence can be interpreted as infinitesimal automorphisms , deformations , and obstructions to deformations for @xmath25 and @xmath34 , so this sequence can be re - written as the deformation complex : @xmath41 \matrix ( m ) [ matrix of math nodes , row sep=1em , column sep=2.5em , text height=1.5ex , text depth=0.25ex ] { 0 & { \mathrm{aut}}(\sigma ) & { \mathrm{def}}(f ) & { \mathrm{def}}(\sigma , f ) & \\ & { \mathrm{def}}(\sigma ) & { \mathrm{obs}}(f ) & { \mathrm{obs}}(\sigma , f ) & 0 . \\ } ; \path[overlay,- > , font=\scriptsize,>=latex ] ( m-1 - 1 ) edge ( m-1 - 2 ) ( m-1 - 2 ) edge ( m-1 - 3 ) ( m-1 - 3 ) edge ( m-1 - 4 ) ( m-1 - 4 ) edge[out=355,in=175 ] ( m-2 - 2 ) ( m-2 - 2 ) edge ( m-2 - 3 ) ( m-2 - 3 ) edge ( m-2 - 4 ) ( m-2 - 4 ) edge ( m-2 - 5 ) ; \end{tikzpicture } \label{eq : deformation - complex}\ ] ] suitably interpreted , the same sequence holds for nodal , open curves in open gromov - witten theory : over a smooth point @xmath42 in @xmath39 , @xmath43 are the cohomology groups associated to sections @xmath44 of @xmath45 satisfying @xmath46 . the expected ( virtual ) dimension of @xmath39 is @xmath47 where @xmath48 denotes the generalized maslov index of the real subbundle @xmath49 @xcite . when @xmath7 is a complex manifold and @xmath4 is the fixed locus of an anti - holomorphic involution , @xmath50 , and if @xmath7 is a calabi - yau threefold , @xmath51 . when @xmath39 has a well - behaved torus action , @xcite give an explicit description for the localization of the virtual fundamental class to the fixed loci of the torus action . in contrast to closed gromov - witten theory , the virtual cycle found in @xcite depends on the torus action . additionally , the invariants defined in @xcite depend on a choice of orientation . this choice is reflected in the overall sign of the invariant , and the invariant formula proposed in this note has an analogous orientation - dependent sign . gromov - witten invariants are , in general , difficult to compute . the primary computational tool is the atiyah - bott fixed - point formula @xcite . when applied to computations of gromov - witten invariants for toric varieties , this `` localizes '' integrals over the entire moduli space of stable maps to integrals over only those maps which are fixed by the torus action @xcite . the atiyah - bott fixed point formula for the integral of a class @xmath52 over a manifold ( or more generally , a deligne - mumford stack ) @xmath53 is @xmath54 where the sum is over the fixed point sets @xmath55 , @xmath56 is the embedding of @xmath55 into @xmath53 , and @xmath57 is the ( equivariant ) euler class of the normal bundle of @xmath55 in @xmath53 . following @xcite , a stable map @xmath42 can be naturally described as a decorated graph . the vertices @xmath58 of the graph correspond to contracted components of the nodal curve @xmath25 , and are labeled by the genus @xmath59 of that component . the edges correspond to @xmath60 s which are not contracted by @xmath34 , and are labeled by the degree @xmath61 of the map @xmath62 associated to the edge . when the fixed stable maps are described as decorated graphs in this way , becomes @xmath63^{vir}}1=\sum_{\gamma}\frac{1}{\left|a_{\gamma}\right|}\int_{m_{\gamma}}\frac{1}{e\left(n_{\gamma}^{vir}\right)}.\label{eq : localization - for - closed - gw - invt}\ ] ] as observed in @xcite , the graph description of stable maps can be extended to the open stable maps defined in @xcite by treating the open disk component as a `` leg '' of the graph . a crucial consequence of this is that open gromov - witten invariants can be expressed as a closed gromov - witten invariant multiplied by a `` disk term . '' for simplicity , restrict attention to surfaces with one boundary component . let @xmath7 be a calabi - yau manifold equipped with an @xmath8 action that fixes a lagrangian submanifold @xmath2 . suppose that @xmath64 is a stable map from a genus @xmath12 riemann surface with one boundary component such that @xmath35=d\in h_{2}\left(x;\mathbb{z}\right)$ ] and @xmath65 has winding @xmath14 as a map between homotopy circles . let @xmath66 denote the the moduli space of such maps ( genus @xmath12 , 1 boundary component , @xmath67 marked points ) . the @xmath8 action on @xmath7 naturally induces an @xmath8 action on @xmath68 . if @xmath69 is fixed by the @xmath8 action , then @xmath25 must take the form @xmath70 where @xmath71 is a closed genus @xmath12 riemann surface , @xmath72 is a disk , and @xmath73 is a simple node on @xmath71 at which @xmath72 is attached . @xmath8 invariance further requires that @xmath74 is fixed by the induced action on @xmath75 . then , a virtual localization formula analogous to for the genus @xmath12 , degree @xmath13 , winding @xmath14 open gromov - witten invariant would take the form @xmath76 where @xmath77 ( @xmath78 is the graph associated to the closed curve @xmath79 ) . note that @xmath75 is equipped with a natural map @xmath80 given by evaluation at the marked point . the conditions on @xmath42 specified above ( in particular , that @xmath81 ) imply that the fixed locus @xmath82 is isomorphic to the fixed subspace @xmath83 . as in the closed case , the equivariant normal bundle @xmath84 and the virtual fundamental cycle are determined , respectively , by the moving and fixed parts of the deformation complex : \(m ) [ matrix of math nodes , row sep=1em , column sep=2.5em , text height=1.5ex , text depth=0.25ex ] 0 & ( ) & ( f ) & ( , f ) & + & ( ) & ( f ) & ( , f ) & 0 + ; ( m-1 - 1 ) edge ( m-1 - 2 ) ( m-1 - 2 ) edge ( m-1 - 3 ) ( m-1 - 3 ) edge ( m-1 - 4 ) ( m-1 - 4 ) edge[out=355,in=175 ] ( m-2 - 2 ) ( m-2 - 2 ) edge ( m-2 - 3 ) ( m-2 - 3 ) edge ( m-2 - 4 ) ( m-2 - 4 ) edge ( m-2 - 5 ) ; this gives the following relationship in the representation ring of @xmath8 : @xmath85 with @xmath86 , @xmath87 , @xmath88 , and @xmath89 . ( here , @xmath90 is the divisor associated to the nodal points of @xmath25 . when @xmath25 is smooth , these spaces are just @xmath91 and @xmath92 , respectively ) . now , relate the terms in this sequence to the terms concerning @xmath71 and @xmath72 : let @xmath93 and @xmath94 . suppose that @xmath72 is parametrized by @xmath95 , with @xmath73 identified with the point @xmath96 . then , there is an exact sequence \(m ) [ matrix of math nodes , row sep=1em , column sep=2.5em , text height=1.5ex , text depth=0.25ex ] 0 & _ & _ _ 0 _ & _ & 0 . + ; ( m-1 - 1 ) edge ( m-1 - 2 ) ( m-1 - 2 ) edge ( m-1 - 3 ) ( m-1 - 3 ) edge ( m-1 - 4 ) ( m-1 - 4 ) edge ( m-1 - 5 ) ; this becomes the exact sequence on cohomology : \(m ) [ matrix of math nodes , row sep=1em , column sep=2.5em , text height=1.5ex , text depth=0.25ex ] 0 & ( f ) & h^0(,t_(,f _ ) ) ( f_0 ) & t_px & + & ( f ) & h^1(,t_(,f _ ) ) ( f_0 ) & 0 & + ; ( m-1 - 1 ) edge ( m-1 - 2 ) ( m-1 - 2 ) edge ( m-1 - 3 ) ( m-1 - 3 ) edge ( m-1 - 4 ) ( m-1 - 4 ) edge[out=355,in=175 ] ( m-2 - 2 ) ( m-2 - 2 ) edge ( m-2 - 3 ) ( m-2 - 3 ) edge ( m-2 - 4 ) ; which yields the following relations in the representation ring : @xmath97 @xmath98 where @xmath99 and the @xmath34 , @xmath100 superscripts denote fixed and moving terms with respect to the @xmath8 action . similarly , @xmath101 and @xmath102 note that @xmath103 consists of the infinitesimal automorphisms of @xmath72 preserving the origin @xmath96 , which are generated by the sections @xmath104 over @xmath105 . therefore , @xmath106 is trivial , and @xmath107 . collecting the above observations , @xmath108 and @xmath109 the first equation implies that the virtual fundamental cycle of the fixed locus is the restriction of the natural virtual cycle of the fixed locus @xmath110^{vir}$ ] to the subspace @xmath111 . the second equation yields the following relationship between the normal bundles @xmath112 and @xmath113 ( where @xmath78 is the graph for the closed curve with the disk `` leg '' removed , i.e. , the stable map associated to @xmath71 with one marked point ) : @xmath114 here @xmath115 is the representation of @xmath8 on @xmath116 induced by the pullback of the @xmath8 action on @xmath117 , and @xmath118 is the tautological cotangent line bundle on @xmath119 associated to the marked point @xmath73 , i.e. , the line bundle whose fiber at the point @xmath120 is @xmath121 . @xmath122 is contribution from the term @xmath123 : @xmath124 is the fiber of @xmath118 and @xmath125 is a constant vector space which carriers the representation @xmath115 by @xmath8 . hence , @xmath126 where @xmath127 and @xmath128 ( so that @xmath129 ) . as before , @xmath130 denotes the @xmath8-equivariant euler class of the specified bundle . denote by @xmath131 the `` disk factor '' @xmath132 then , becomes @xmath133 where @xmath21 is the equivariant thom class of the point @xmath22 , and @xmath134 the pullback to the fixed locus @xmath135 . comparing this formula with shows that the parenthetical quantity is the localization of a closed gromov - witten invariant : @xmath136^{vir}}\frac{ev^{*}\left(\phi_{p}\right)}{\left(\alpha-\psi\right)}.\label{eq : delocalized - separation - of - disk}\ ] ] the proposed formula for open gromov - witten invariants is obtained by composition of a disk function @xmath16 and a descendant invariant . @xmath16 is built from combinatorial data about the moment polytope , and a characteristic class . recall that if @xmath137 are the chern roots of a complex vector bundle @xmath138 , iritani s gamma class @xcite is a characteristic class associated to @xmath138 defined by @xmath139 as observed in @xcite , in some cases the disk term @xmath131 can be expressed using gamma classes . the gamma class also appears in quantum cohomology , and can be regarded as a localization contribution from constant maps in floer theory @xcite . the inputs of @xmath16 are the torus weight @xmath19 of a normal direction to @xmath140 , and the homogenized iritani gamma class @xmath141 , defined by @xmath142 where @xmath137 are the chern roots of the tangent bundle @xmath143 . ( when @xmath144 , @xmath145 ) . with these definitions , the main result is : [ thm : main - result]let @xmath7 be a calabi - yau 3-fold and @xmath2 a lagrangian submanifold . let @xmath8 act on @xmath7 such that the @xmath8 action preserves @xmath4 , and @xmath4 intersects a rigid circle - invariant curve @xmath9 . suppose that @xmath4 can be described in a neighborhood of @xmath10 as the fixed locus of an anti - holomorphic involution . let @xmath11 . then , the genus @xmath12 , 1 boundary component , degree @xmath13 , winding @xmath14 open gromov - witten invariant with lagrangian boundary @xmath4 is @xmath146 where @xmath16 is the disk function @xmath147 here , @xmath18 is the homogeneous iritani gamma class , @xmath19 is the weight of the @xmath8 action along a normal direction to @xmath9 , @xmath20 is the equivariant chern class of the induced representation of @xmath8 at the attachment point of the disk , and @xmath21 is the equivariant class of the image @xmath22 of the disk attachment point . the setup in theorem [ thm : main - result ] is depicted in figure [ fig : local - vertex ] . in @xcite , the authors study the orbifold version of this scenario with a lagrangian obtained by the anti - holomorphic involution @xmath148 , and obtain an analogous result . the disk term only differs from ( * ? ? ? * equation 28 ) in notation : here , @xmath149 , and the winding @xmath14 is included to account for automorphisms of the leg . although the result above is stated in a more general form , it will be shown below ( lemma [ lem : involution - form ] ) that after a change of coordinates , @xmath148 is actually the only @xmath8-equivariant anti - holomorphic involution satisfying these assumptions . in particular , the result proven in this note applies to aganagic - vafa branes @xcite . this author is unsure how to prove the result without any assumptions on the local geometry of @xmath4in order to apply localization , one requires a decription of the torus - fixed disks with boundary on @xmath4 . however , as will be shown in section [ sec : torus - knot - lagrangians ] , this result also applies without modification to a family of lagrangian cycles which are not obtained from the fixed locus of an anti - holomorphic involution . a lagrangian @xmath4 , obtained as the fixed locus of an anti - holomorphic involution , intersects an edge of the toric polytope , labeled by the local coordinate @xmath150 . the only torus - fixed disks are the hemispheres @xmath90 with @xmath151 . such a map is given locally by @xmath152 . the @xmath153-@xmath154 hyperplane is normal to the disk . the normal directions to @xmath9 are spanned by @xmath155 and @xmath156 , so the weight @xmath19 appearing in @xmath16 can be the weight of any @xmath8-invariant line spanned by these vectors ( for example , @xmath157 or @xmath158 ) . in the genus @xmath67 case , the closed gromov - witten invariants in formula also appear as terms in givental s @xmath6 function @xcite . givental s @xmath6 function is the map on quantum cohomology @xmath159 given by @xmath160 where @xmath161 is a basis for the cohomology of @xmath7 , @xmath162 is the dual basis with respect to the poincar pairing , and @xmath163 is a power series of gravitational descendant closed gromov - witten invariants . to obtain the genus-@xmath12 generating function , a higher - genus version of the @xmath6-function is needed . define the genus-@xmath12 modified @xmath6-function @xmath164 to be @xmath165 where @xmath166 and @xmath167 is the complexified khler class of @xmath7 . from and theorem [ thm : main - result ] , it is easy to write a generating function for open gromov - witten invariants . the generating function for the one - boundary - component , winding @xmath14 open gromov - witten invariants @xmath168 is the function @xmath169 where @xmath170 is the string coupling constant , @xmath171 , and the summation is only over combinations of @xmath12 , @xmath172 , and @xmath13 where the summands are defined . [ cor : generating - function]let @xmath7 and @xmath4 be as in theorem [ thm : main - result ] . then , a generating function for the winding-@xmath14 open gromov - witten invariants of @xmath173 is given by the formula @xmath174 two observations about this generating function merit mention . first , in @xcite , the authors use a similar procedure to obtain a generating function for open invariants from a modification of the @xmath6 function . the main distinction here is the presentation of the disk term . second , for @xmath175 , @xmath176 has the expression @xmath177 as will be discussed in section [ sec : torus - knot - lagrangians ] , @xcite have found that , for a certain class of lagrangian cycles originating from torus knots , the expression above encodes the homfly polynomial associated to the original knot . the main content of the proof of theorem [ thm : main - result ] is the comparison of the predicted disk term from with an explicit localization calculation of the open gromov - witten invariant . first , recall the virtual localization technique : as described in section [ sub : separating - the - disc ] , the open gromov - witten invariant @xmath178 can be expressed as a product of a disk term @xmath131 and a descendant invariant . the surface @xmath25 can be written as @xmath179 , with @xmath71 is a closed surface , @xmath72 a disk , and @xmath73 the point of attachment . in terms of the cohomology of sheaves over @xmath72 , @xmath131 was found to have the following expression : @xmath180 where @xmath181 is the restriction of the map @xmath3 to the disk @xmath72 , @xmath182 , and @xmath130 denotes the @xmath8-equivariant euler classes of the specified bundles . to compute the disk contribution to @xmath178 , one must compute each of these cohomology groups . in contrast to the analogous computation of closed invariants , the lagrangian @xmath4 imposes boundary conditions on the sections of @xmath183 . let @xmath184 denote @xmath185 . then , @xmath183 consists of sections of @xmath186 satisfying @xmath187 . to obtain an explicit presentation of the boundary conditions , let @xmath188 be the subbundle of the cotangent bundle @xmath189 which annihilates the tangent bundle @xmath190 . choose a basis of sections @xmath191 of @xmath192 along the boundary @xmath193 of the disk . ( the @xmath194 can be obtained by , for example , linearizing the equations defining @xmath4 ) . @xmath183 consists of the sheaf of germs of holomorphic sections of the bundle @xmath186 satisfying the boundary conditions @xmath195 with this presentation of the boundary conditions , computing @xmath196 becomes an exercise in ech cohomology : let @xmath197 be an open cover of @xmath72 , and let @xmath198 be local coordinates such that @xmath81 is the origin @xmath199 . then , local sections over @xmath200 and @xmath201 are of the form @xmath202 and the coefficients @xmath203 , @xmath204 , and @xmath205 are subject to boundary conditions imposed by . finally , to apply localization , let @xmath206 denote the @xmath8 action determined by @xmath207 the weights @xmath157 , @xmath158 , @xmath208 are required to satisfy @xmath209 so that the holomorphic volume form is preserved by the @xmath8 action . now , suppose that near the intersection @xmath10 , @xmath4 is described as the fixed locus of an anti - holomorphic involution @xmath210 . choose the local coordinates @xmath153 , @xmath154 , and @xmath150 such that @xmath10 is defined by @xmath211 , @xmath212 . [ lem : involution - form]for a calabi - yau @xmath8 action on @xmath213 , the only @xmath8-equivariant anti - holomorphic involution defined on @xmath214 and fixing the circle @xmath211 , @xmath212 is @xmath215 where @xmath216 are two constants satisfying @xmath217 . note that the calabi - yau condition means that the holomorphic 3-form @xmath218 is preserved by the @xmath8 action , i.e. , if @xmath8 acts with weights @xmath219 , then @xmath209 . @xmath8-equivariance is the requirement that @xmath220 , and is a necessary condition to apply localization to @xmath4 . ( by assumption , in a neighborhood of @xmath10 , @xmath221 must satisfy @xmath222 . because @xmath4 is defined locally by @xmath223 , this is equivalent to @xmath220 ) . generically , @xmath210 takes the form @xmath224 where @xmath225 , @xmath226 , and @xmath227 are laurent series in the variables @xmath228 , @xmath229 , @xmath230 : @xmath231 @xmath8-equivariance imposes restrictions on the coefficients @xmath232 , @xmath233 , @xmath234 . for example , @xmath227 must satisfy @xmath235 for all @xmath236 . expanding , this is @xmath237 hence , @xmath238 . recalling that @xmath239 , this forces @xmath240 . similar results hold for @xmath225 and @xmath226 . therefore , anti - holomorphic involutions commuting with the @xmath8 action along @xmath4 must take the form @xmath241 in fact there are further restrictions on @xmath210 . because @xmath210 must be defined along @xmath242 , @xmath243 for @xmath244 , and @xmath245 , @xmath246 for @xmath247 . substituting these relations into @xmath248 , the equation @xmath249 becomes @xmath250 where @xmath12 is a power series in @xmath198 . therefore , @xmath217 , and , for degree reasons , @xmath251 . an analogous relation holds for @xmath226 . similarly , @xmath252 becomes @xmath253 as with @xmath225 , @xmath254 . a careful examination of the coefficients in the above equations reveals that the only solutions to @xmath248 are @xmath246 for @xmath255 and @xmath243 for @xmath256 . this completes the proof of the lemma . applying the above lemma , the defining equations of @xmath4 near @xmath10 are @xmath257 linearizing these equations yields @xmath258 and at @xmath211 , these equations simplify to @xmath259 as @xmath260 along @xmath212 , a basis for @xmath192 along @xmath193 is given by the 1-forms @xmath261 recall that local sections over @xmath262 are of the form @xmath263 parameterizing @xmath264 by @xmath265 , the map @xmath184 takes the form @xmath266 . the boundary conditions @xmath267 impose restrictions on the coefficients @xmath203 , @xmath204 , @xmath205 : @xmath268 these yield the following equations for the coefficients @xmath203 , @xmath204 , @xmath205 : @xmath269 the first two equations are actually equivalent : after complex conjugation , relabeling of indices , and substituting @xmath217 , the second equation becomes the first . so , the lagrangian boundary conditions on sections over @xmath200 are @xmath270 from these boundary conditions , the cohomology groups @xmath196 can be computed explicitly . @xmath271 consists of the global sections , i.e. , holomorphic sections @xmath44 on @xmath72 . these take the form @xmath272 with @xmath203 , @xmath204 and @xmath205 subject to the boundary conditions in , and @xmath273 for @xmath274 . in particular , the equation @xmath275 implies that @xmath276 for all @xmath277 . shifting indices @xmath278 , this equation also implies that @xmath279 for all @xmath277 . finally , from the last boundary equation @xmath280 , @xmath281 for @xmath282 . so , @xmath271 consists of sections of the form @xmath283 where @xmath284 is real . as a vector space , @xmath271 is isomorphic to @xmath285 the map @xmath286 takes @xmath287 , so @xmath8 action the section @xmath288 with weight @xmath289 . @xmath290 is fixed by the @xmath8 action , so @xmath291 is just the complex part of this vector space . hence , @xmath292 @xmath293 consists of the cokernel to the ech differential . sections over @xmath294 can be written as @xmath295 the image of the ech differential consists of sections @xmath296 of the form @xmath297 . in terms of the coefficients , this is @xmath298 where again , @xmath203 , @xmath204 , and @xmath205 are subject to the boundary conditions . solutions always exist for @xmath299 : set @xmath300 for @xmath274 , and set @xmath301 for @xmath302 . similarly , because @xmath303 and @xmath304 are completely free , any @xmath305 and @xmath306 for @xmath302 can be solved for . however , to solve for @xmath305 and @xmath306 for @xmath274 , it must be the case that @xmath307 . the first boundary equation @xmath275 then implies that @xmath203 is fixed for @xmath308 . so , there are no solutions if @xmath309 in @xmath308 . when @xmath310 , @xmath311 , so setting @xmath312 and @xmath313 will solve these equations . therefore , the cokernel of the ech differential is isomorphic to the space of sections @xmath296 of the form @xmath314 the induced @xmath8 action on @xmath315 has weight @xmath316 . as a vector space , @xmath293 is @xmath317 and @xmath318 substituting and in yields @xmath319 the proof will be complete if is equivalent to the claimed expression : @xmath320 first , observe that @xmath321 similarly , @xmath322 recall that , by assumption , @xmath209 , so @xmath323 . therefore , @xmath324 . the above manipulations show that @xmath325 the induced action on @xmath125 carries weight @xmath326 . substitute @xmath327 and apply euler s reflection formula to get @xmath328 generically , there are two @xmath8-invariant normal directions to @xmath10 in @xmath7 , given by the tangent vectors @xmath155 and @xmath156 . in the formula in , @xmath19 may be the weight of either of these directions , i.e. , @xmath329 or @xmath157 . the choice of @xmath19 changes the sign of because @xmath330 . this sign ambiguity reflects an overall choice of orientation of @xmath331 @xcite . this completes the proof of the main result . in this section , is compared to previous virtual localization calculations of open gromov - witten invariants . in the interest of brevity , only the geometric setup and final results are stated below ; the reader interested in further localization calculations is referred to the original sources , or the computation appearing in the proof of theorem [ thm : main - result ] . this situation was first described in @xcite , and the gamma class formula for an orbifold generalization of this case was given in @xcite . for completeness , the scenario will also be described here . let @xmath333 . @xmath7 appears often in gromov - witten theory and mirror symmetry : it is the small resolution of the conifold singularity , the normal bundle to a smooth rational line in a calabi - yau 3-fold , and it can be obtained from a @xmath334 gauge theory with 4 chiral fields with charges @xmath335 . @xmath7 can be described symplectically using symplectic reduction on @xmath336 , and in this setting it is easiest to obtain the moment polytope of @xmath7 . let @xmath8 act on @xmath336 with weights @xmath335 . then , the moment map for this action is @xmath337 and it can be seen ( for example , by choosing appropriate local coordinates and checking transition functions ) that @xmath338 for @xmath339 ( @xmath340 determines the symplectic volume of the base @xmath60 ) . there s a natural anti - holomorphic involution @xmath210 on @xmath336 given by @xmath341 the fixed locus of this involution is a lagrangian submanifold @xmath342 of @xmath336 defined by the equations @xmath343 because @xmath342 is preserved by the @xmath8 action , @xmath344 defines a lagrangian @xmath2 . this lagrangian is easy to visualize in the moment polytope of @xmath7 . the moment polytope is the image of @xmath7 in @xmath345 under the projection @xmath346 . then , @xmath4 is the intersection of the two planes @xmath347 and @xmath348 in the polytope . @xmath4 intersects the zero section @xmath60 along its equator , so that @xmath349 as depicted in figure [ fig : kl - lagrangian ] . this figure depicts the geometry described in section [ sub : kl - lagrangians ] . a lagrangian @xmath4 , obtained as the fixed locus of an anti - holomorphic involution , intersects the zero section of @xmath7 . local coordinates @xmath350 and @xmath351 parametrize the fibers of @xmath7 in a neighborhood of the two vertices of the moment polytope . one possible torus - fixed disk @xmath90 with boundary @xmath352 is depicted . there are two unique disks with boundary on @xmath4 , corresponding to the two hemispheres of the @xmath60 . in local coordinates @xmath353 defined by @xmath354 @xmath4 is defined by the fixed locus of the anti - holomorphic involution @xmath355 it is readily checked that at @xmath356 , and the disk based at the @xmath357 pole of the @xmath60 takes the form @xmath358 . the winding @xmath14 disk map is @xmath359 in this situation , @xcite computed @xmath360 comparing with in section [ sub : pf - comparing - disk - terms ] , this is @xmath361 , so this result agrees with the proposed formula . this situation was described in @xcite . as in the previous example , @xmath363 can be obtained via symplectic reduction . let @xmath8 act on @xmath336 with weights @xmath364 . then , @xmath365 for @xmath339 . @xcite consider the lagrangian submanifold @xmath366 defined by @xmath367 where @xmath368 . because @xmath369 is preserved by the @xmath8 action , it descends to a lagrangian @xmath370 , as depicted in figure [ fig : gz - lagrangian ] . this figure depicts the geometry described in section [ sub : kp2 ] . a lagrangian @xmath4 , obtained as the fixed locus of an anti - holomorphic involution , intersects the zero section of @xmath7 along one edge of the moment polytope . one possible torus - fixed disk @xmath90 with boundary @xmath371 is depicted . note that @xmath372 parametrizes the intersection of @xmath373 with the @xmath60 given by the image of @xmath374 in the quotient space . at @xmath375 , @xmath373 intersects this curve along its equator . for simplicity , restrict attention to @xmath376 ( locally , other values of @xmath372 can be obtained by a coordinate transformation ) . in local coordinates @xmath377 the lagrangian @xmath4 is the fixed locus of the anti - holomorphic involution @xmath355 the disk is @xmath378 , and the winding @xmath14 disk map is @xmath359 so , locally , the situation computed in @xcite is identical to @xcite . as seen in section [ sub : kl - lagrangians ] , this agrees with theorem [ thm : main - result ] . slightly extending the computation in @xcite , theorem [ thm : main - result ] can also be used to compute the invariants associated to a lagrangian cycle intersecting an external leg of the moment polytope . let @xmath379 be the submanifold of @xmath336 defined by @xmath380 again , these equations are preserved by the @xmath8 action , so the image of @xmath379 in the quotient space @xmath7 is a well - defined lagrangian submanifold @xmath4 . @xmath4 can be equivalently described as the fixed locus of the anti - holomorphic involution @xmath381 . in local coordinates @xmath353 , @xmath4 is defined by @xmath382 ( this lagrangian cycle is shown in figure [ fig : kp2-fiber - lagrangian ] ) . a lagrangian cycle intersects an external leg of the moment polytope of @xmath7 . the lagrangian is obtained as the fixed locus of an anti - holomorphic involution . there is only one torus - fixed disk @xmath90 with boundary on @xmath4 , as depicted above . the disk is @xmath383 , and the winding @xmath14 disk map is @xmath384 . so , @xmath385 . applying theorem [ thm : main - result ] , @xmath386 here , the normal direction weight @xmath387 has been chosen in . choosing @xmath388 instead changes the sign , reflecting the overall dependence of these counts on the choice of torus weights . this can be seen from the product identity : @xmath389 in @xcite , the authors investigate open invariants for disks with boundary on a lagrangian submanifold @xmath390 of a quintic 3-fold @xmath391 . @xmath390 is obtained as the restriction to @xmath391 of the fixed locus of an anti - holomorphic involution on @xmath392 ; however , the associated invariants are computed by an integral over the moduli of stable disk maps to @xmath392 , rather than @xmath391 . as will be seen in section [ sub : quintic - and - main - result ] , the result of theorem [ thm : main - result ] still applies in this setting . this section reviews the geometric setup of @xcite . let @xmath393 be a nonsingular quintic hypersurface with symplectic form @xmath167 obtained from the fubini - study metric on @xmath392 . let @xmath210 be the anti - holomorphic involution on @xmath392 , described in homogeneous coordinates as @xmath394\right)=\left[\overline{z_{0}}:\overline{z_{2}}:\overline{z_{1}}:\overline{z_{4}}:\overline{z_{3}}\right].\ ] ] the fixed locus of @xmath210 is a real lagrangian submanifold @xmath395 , and its restriction to @xmath391 is a real lagrangian submanifold @xmath396 . for the diagonal @xmath397 action on @xmath392 , there are five torus fixed points : @xmath398 , & \ldots , & & \zeta_{4 } & = \left[0:0:0:0:1\right].\end{aligned}\ ] ] only @xmath399 is fixed by both the @xmath397 action and @xmath210 , so it is the unique real @xmath397-fixed point . now , consider the rank-2 subtorus @xmath400 acting by @xmath401=\left[z_{0}:e^{i\lambda\theta_{1}}z_{1}:e^{-i\lambda\theta_{1}}z_{2}:e^{i\lambda'\theta_{3}}z_{3}:e^{-i\lambda'\theta_{3}}z_{4}\right].\label{eq : psw - torus - action}\ ] ] there are two rational lines invariant with respect to this @xmath402 action : @xmath4 , connecting @xmath403 to @xmath404 ; and @xmath405 , connecting @xmath406 to @xmath407 . as in the doubling construction in @xcite , a holomorphic degree-@xmath13 map @xmath408 ( or @xmath4 ) can be halved to obtain two winding-@xmath13 disk maps corresponding to the two hemispheres of @xmath4 . the boundaries of these disks are the intersections @xmath409 and @xmath410 . however , when @xmath13 is even , not every stable disk map can be obtained in this way @xcite , so only odd - winding disk maps will be considered in the remainder of this section . to enumerate the disk maps @xmath411 , @xcite perform an analogous computation to the closed quintic genus - zero enumeration computation described in @xcite . let @xmath412 denote the moduli space of stable disk maps @xmath413 of winding @xmath13 satisfying @xmath414 , as in section [ sub : stable - open - maps ] , and let @xmath415 be the real vector bundle over this moduli space with fiber @xmath416}=h^{0}\left(c,\tilde{f}^{*}\mathcal{o}_{\mathbb{p}^{4}}\left(5\right)\right)_{\mathbb{r}},\ ] ] where @xmath417 $ ] is the stable rational map obtained from the stable disk map via reflection , and the @xmath105 subscript denotes real sections . generically , the boundary of @xmath412 consists of maps with two boundary components . because @xmath415 is not trivial near this boundary , the integral @xmath418 over the moduli space is not well - defined . to remedy this , consider a generic stable disk map @xmath419 . @xmath420 has two components , @xmath421 and @xmath422 . modifying @xmath34 by replacing the image of one of the two components by its image under the involution @xmath210 ( e.g. , defining @xmath423 such that @xmath424 ) yields another two - component map @xmath425 let @xmath426 denote the quotient of @xmath412 under the equivalence relation identifying two - component maps of the form @xmath427 and @xmath425 . the vector bundle @xmath415 descends to a bundle @xmath428 over @xmath426 , and here @xcite prove that the virtual disk count for maps @xmath411 is given by @xmath429 ( * ? ? ? * theorem 3 ) . localization can be used to compute the right - hand side of this equation . there are two rational lines @xmath4 , @xmath405 fixed by the @xmath402 action on @xmath392 . consequently , there are four @xmath402-fixed disks , identified by their incident torus - fixed points @xmath430 . these disk maps are completely characterized by the winding @xmath431 and the incident point @xmath430 . in the notation of @xcite , the disk count is related to the contributions from each of these disk maps by the formula @xmath432 where @xmath433 denotes the contribution of the disk map @xmath434 to the disk count . the intersection disk term @xmath435 of @xmath433 is the contribution of the unique @xmath402-fixed map @xmath436 with winding @xmath431 and incident point @xmath430 . in ( * lemma 6 ) , the terms @xmath435 , @xmath437 are computed explicitly . in fact , this expression can also be obtained using the result of theorem [ thm : main - result ] . for simplicity , restrict attention to the disk term associated to @xmath438 . here , using localization , @xcite compute @xmath439 where @xmath19 and @xmath440 are the torus weights from . the vertex of the disk is the point @xmath403 , so the induced torus representation @xmath115 on @xmath125 has weight @xmath441 . now , re - write the terms of using gamma functions ( recalling the identity @xmath442 for @xmath172 odd ) : @xmath443 substituting these expressions into the third fraction of yields @xmath444 @xmath445 after the additional substitution @xmath446 . this formula now closely resembles . by definition , the gamma class is multiplicative . hence , @xmath447 can be obtained from the normal bundle exact sequence of sheaves @xmath40 & t_{q}\ar[r ] & t_{\mathbb{p}^{4}}\ar[r ] & \mathcal{o}_{\mathbb{p}^{4}}\left(5\right)\ar[r ] & 0 } , \ ] ] yielding @xmath448 an additional subtlety is that the above exact sequence is not torus - equivariant for the real quintic . however , from the hard lefshetz theorem , it suffices to write the ( non - equivariant ) gamma class as a ratio , and then choose an equivariant lift for each of the factors . the weights of the tangent space @xmath449 at the fixed point @xmath403 are @xmath450 , @xmath19 , @xmath451 , and @xmath452 , and the bundle @xmath453 has weight @xmath454 at this fixed point . therefore , the second term in the product above is the localized form of the inverse of @xmath447 . finally , because @xmath455 , @xmath456 . combining these observations , can be expressed as @xmath457 which has a similar form to . the disk is contained in the rational line @xmath4 . in local coordinates @xmath458 near the vertex @xmath403 , the disk map is @xmath459 , and @xmath451 is the weight associated to a normal direction to the disk ( namely , @xmath460 ) . the only discrepancy between and is an additional factor of @xmath461 in the latter equation . as these are invariants appearing from the localization of an integral over @xmath426 , rather than @xmath462 , there is no a priori reason to expect a form of theorem [ thm : main - result ] to apply . in addition to lagrangians appearing as the fixed loci of anti - holomorphic involutions , there is another family of lagrangians on @xmath333 motivated by large @xmath0 duality and knot theory . recent work in this area has yielded many connections between knot theory and gromov - witten theory ( @xcite ) ; this section reviews the geometric relationship between knots on @xmath24 and open gromov - witten theory on @xmath7 . recall that @xmath7 can be identified with the resolved conifold@xmath7 is the small resolution of the conifold singularity @xmath463 in @xmath336 . in particular , by blowing up the subspace @xmath464 , @xmath7 can be described by the equations @xmath465 where @xmath466 and @xmath467\in\mathbb{p}^{1}$ ] . the conifold singularity is also the singular limit of the smooth hypersurface threefold @xmath468 defined by @xmath469 where @xmath470 . as described in @xcite , @xmath471 is symplectomorphic to the cotangent bundle @xmath472 . the base @xmath473 is the fixed locus of the anti - holomorphic involution @xmath474 , expressed by the equations @xmath475 . the large @xmath0 duality conjecture states that the large @xmath0 limit of the topological a - model on @xmath471 with @xmath0 lagrangian branes wrapping @xmath476 is equivalent the topological a - model on @xmath7 @xcite . this has been checked in several ways . first , according to @xcite , the topological a - model on @xmath471 with @xmath0 lagrangian branes wrapping @xmath476 is equivalent to the @xmath477 chern - simons theory on @xmath476 . then , in the large @xmath0 expansion , the partition function @xmath478 is equivalent to the topological a - model partition function @xmath479 @xcite . the parameters determining the a - model theory on @xmath7 are the string coupling constant @xmath170 and the symplectic area @xmath480 of the zero section @xmath481 , which are related to the chern - simons parameters @xmath277 and @xmath0 by @xmath482 large @xmath0 duality is extended to incorporate wilson loops in @xcite . following @xcite , wilson loop observables in the chern - simons theory on @xmath24 correspond to colored homfly polynomials of knots @xmath483 . the conormal bundle @xmath484 to a knot @xmath483 is lagrangian submanifold of @xmath472 . the main difficulty in extending large @xmath0 duality in this manner is determining the corresponding a - model on @xmath7 : @xmath484 intersects the zero section @xmath24 in the knot @xmath485 , which becomes contracted after the conifold transition . to remedy this difficulty , the lagrangian cycle @xmath484 must be lifted to a new lagrangian @xmath379 disjoint from the zero section before performing the conifold transition @xcite , as depicted in figure [ fig : conifold - transition ] . this figure depicts the conifold transition . the lagrangian @xmath486 is constructed by shifting the conormal bundle of a knot @xmath483 off of the zero section . this lift introduces a holomorphic cylinder @xmath9 connecting the knot on @xmath24 to its image in @xmath379 . @xmath487 is the conifold singularity @xmath488 in @xmath336 . the map @xmath489 is a symplectomorphism away from the zero section , so @xmath490 is a lagrangian submanifold of @xmath487 . @xmath491 is the small resolution of the conifold singularity , and @xmath492 is the corresponding natural map . in fact , there are a family of such symplectomorphisms , where @xmath493 parametrizes the symplectic form on the zero section @xmath481 . hence , @xmath494 is a lagrangian submanifold of @xmath7 . the holomorphic disk @xmath90 is the image of @xmath9 under the conifold transition . such a lift is easy to construct : define coordinates @xmath495 for @xmath472 by @xmath496 any knot @xmath483 is given by a parametrization @xmath497 . then , the conormal bundle @xmath484 can be expressed as @xmath498 lifts of @xmath484 can be specified by maps @xmath499 such that @xmath500 : for such a @xmath12 , define the lifted conormal bundle @xmath379 to be @xmath501 the image of @xmath379 under the conifold transition will be a lagrangian @xmath2 , and the open a - model on @xmath7 with this lagrangian boundary can be computed . shifting @xmath484 off of the zero section modifies large @xmath0 duality in the following ways : the lift of @xmath484 to @xmath379 introduces corrections to the wilson loop observables in the chern - simons theory proportional to the area of the holomorphic cylinder @xmath9 connecting the lift of the knot to its image in the zero section @xcite . instead of the closed a - model on @xmath7 , the corresponding theory should be an open a - model with lagrangian boundary @xmath4 . this statement of large @xmath0 duality is found to be true for torus knots in @xcite , and their construction provides a novel source of lagrangians . [ [ toric - lagrangian - cycles - and - theoremthmmain - resultsubdsv - comparison ] ] toric lagrangian cycles and theorem [ thm : main - result][sub : dsv - comparison ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ it is important to note that the lagrangians considered in @xcite are not obtained as the fixed loci of anti - holomorphic involutions , so there is no a priori reason to expect that the formula proposed in theorem [ thm : main - result ] should apply in this situation . for the @xmath502 torus knot , the corresponding lagrangian @xmath4 is found to be fixed under the torus action @xmath503\right)=\left(e^{is\theta}x , e^{ir\theta}y , e^{-is\theta}z , e^{-ir\theta}w\right),\left[e^{-i\left(r+s\right)\theta}\lambda:\rho\right].\ ] ] there is only one holomorphic disk in @xmath7 fixed by this @xmath8 action , and it lies entirely in the @xmath153-@xmath154 face of the moment polytope , as depicted in fig . [ fig : dsv - lagrangian ] . this figure depicts the geometry described in section [ sub : dsv - comparison ] . the lagrangian @xmath4 is the image of a shifted conormal bundle to a knot in @xmath24 under the conifold transition . local coordinates @xmath350 and @xmath351 parametrize the fibers of @xmath7 in a neighborhood of the two vertices of the moment polytope . the boundary of the disk @xmath90 is symplectomorphic to the torus knot ( in the depiction above , the trefoil ) . @xmath90 is contained entirely in the @xmath153-@xmath154 face of the polytope , and the disk map can be written in local coordinates as @xmath504 . a neighborhood of the disk can be described by local coordinates @xmath153 , @xmath154 , @xmath505 . in these coordinates , the disk map is @xmath506 where @xmath507 and @xmath508 is a constant obtained from the geometric construction in @xcite . after a lengthy localization calculation , @xcite compute the winding-1 open gromov - witten invariants with lagrangian boundary @xmath4 . this computation readily generalizes to higher winding @xcite , and gives the following expression for @xmath131 : @xmath509 this can be re - written in terms of gamma functions in the following way : @xmath510 locally ( figure [ fig : dsv - lagrangian ] ) , the weights of the torus action are @xmath511 , @xmath512 , @xmath513 . the induced torus action on @xmath72 is @xmath514 , so @xmath515 . replacing @xmath516 and substituting these weights into the above formula yields @xmath517 after euler s reflection identity . as remarked above , the lagrangian @xmath4 is not the fixed locus of an anti - holomorphic involution . however , the result still applies . let @xmath9 be the @xmath8-invariant curve given locally by @xmath518 for @xmath519 . then , @xmath520 and @xmath208 is the weight of an @xmath8-invariant normal direction to @xmath9 as @xmath9 is entirely contained in the @xmath153-@xmath154 plane of the moment polytope . the author finds it curious that the result of theorem [ thm : main - result ] appears to apply in this situation , and hopes that this is evidence that , properly formulated , a more general version of this theorem exists . m. aganagic , a. klemm , c. vafa . `` disk instantons , mirror symmetry , and the duality web . '' z. naturforsch . * a57 * ( 2002 ) 128 mr 1906661 . http://arxiv.org/abs/hep-th/0105045[`arxiv:hep-th/0105045 [ hep - th ] ` ] t .- chiang , a. klemm , s .- yau , e. zaslow . `` local mirror symmetry : calculations and interpretations , '' adv . * 3 * ( 2000 ) , 495565 . http://arxiv.org/abs/hep-th/9903053v4[`arxiv:hep-th/9903053v4 ` ] diaconescu , v. shende , c. vafa . `` large @xmath0 duality , lagrangian cycles , and algebraic knots . '' * 319 * ( 2013 ) , 813863 . http://arxiv.org/abs/1111.6533v1[`arxiv:1111.6533v1 [ hep - th ] ` ] j. gu , h. jockers , a. klemm , m. soroush . `` knot invariants from topological recursion on augmentation varieties . '' * 336 * ( 2015 ) , 9871051 . http://arxiv.org/abs/1401.5095v3[`arxiv:1401.5095v3 [ hep - th ] ` ] t. graber and e. zaslow . `` open - string gromov - witten invariants : calculations and a mirror ` theorem ' '' in _ orbifolds in mathematics and physics _ , 107121 , contemp . math . * 310 * , amer . ( providence ) 2002 . http://arxiv.org/abs/hep-th/0109075[`arxiv:hep-th/0109075 ` ] s. katz and c .- c . `` enumerative geometry of stable maps with lagrangian boundary conditions and multiple covers of the disc , '' adv . phys . * 5 * ( 2001 ) , 149 . http://arxiv.org/abs/math/0103074[`arxiv:math/0103074 [ math.ag ] ` ] m. kontsevich . `` enumeration of rational curves via torus actions , '' in _ the moduli space of curves _ , dijkgraaf et al eds . , progress in mathematics * 129 * , birkhauser ( boston ) 1995 . http://arxiv.org/abs/hep-th/9405035v2[`arxiv:hep-th/9405035v2 ` ] y. ruan . `` cohomology ring of crepant resolutions of orbifolds , '' in _ gromov - witten theory of spin curves and orbifolds _ , contemp . math . * 403 * , amer . ( providence ) 2006 . http://arxiv.org/abs/math/0108195[`arxiv:math/0108195 [ math.ag ] ` ] e. witten . `` chern - simons gauge theory as a string theory . '' in _ the floer memorial volume _ , volume 133 of _ progr . _ , 637678 . birkh@xmath522user ( basel ) 1995 . http://arxiv.org/abs/hep-th/9207094[`arxiv:hep-th/9207094 ` ]

in this paper , we describe a new method to compute the minimum of a real polynomial function and the ideal defining the points which minimize this polynomial function , assuming that the minimizer ideal is zero - dimensional . our method is a generalization of lasserre relaxation method and stops in a finite number of steps . the proposed algorithm combines border basis , moment matrices and semidefinite programming . in the case where the minimum is reached at a finite number of points , it provides a border basis of the minimizer ideal

1301.5298

optimization appears in many areas of scientific computing , since the solution of a problem can often be described as the minimum of an optimization problem . local methods such as gradient descent are often employed to handle global minimization problems . they can be very efficient to compute a local minimum , but the output depends on the initial guess and they give no guarantee of a global solution . in the case where the function @xmath0 to minimize is a polynomial , it is possible to develop methods which ensure the computation of a global solution . reformulating the problem as the computation of a ( minimal ) critical value of the polynomial @xmath0 , different polynomial system solvers can be used to tackle it ( see e.g. @xcite , @xcite ) . but in this case , the complex solutions of the underlying algebraic system come into play and additional computation efforts should be spent to remove these extraneous solutions . semi - algebraic techniques such as cylindrical algebraic decomposition or extensions @xcite may also be considered here but suffer from similar issues . though the global minimization problem is known to be np - hard ( see e.g. @xcite ) , a practical challenge is to device methods which take into account `` only '' the real solutions of the problem or which can approximate them efficiently . * previous works . * about a decade ago , a relaxation approach has been proposed in @xcite ( see also @xcite , @xcite ) to solve this difficult problem . instead of searching points where the polynomial @xmath0 reaches its minimum @xmath1 , a probability measure which minimizes the function @xmath0 is searched . this problem is relaxed into a hierarchy of finite dimensional convex minimization problems , that can be solved by semi - definite programming ( sdp ) techniques , and which converges to the minimum @xmath1 @xcite . this hierarchy of sdp problems can be formulated in terms of linear matrix inequalities on moment matrices associated to the set of monomials of degree @xmath2 for increasing values of @xmath3 . the dual hierarchy can be described as a sequence of maximization problems over the cone of polynomials which are sums of squares ( sos ) . a feasibility condition is needed to prove that this dual hierarchy of maximization problems also converges to the minimum @xmath1 , ie . that there is no duality gap . from a computational point of view , this approach suffers from two drawbacks : 1 . the hierarchy of optimization problems may not be _ exact _ , ie . it may not always convergence in a finite number of steps ; 2 . the size of the sdp problems to be solved _ grows exponentially _ with @xmath3 . to address the first issue , the following strategy has been considered : add polynomial inequalities or equalities satisfied by the points where the function @xmath0 is minimum . inequality constraints can for instance be added to restrict the optimization problem to a compact subset of @xmath4 and to make the hierarchies exact @xcite , @xcite . natural constraints which do not require apriori bounds on the solutions are for instance the vanishing of the partial derivatives of @xmath0 . a result in @xcite shows that if the gradient ideal ( generated by the first differentials of @xmath0 ) is zero - dimensional , then the hierarchy extended with constraints from the gradient ideal is exact . it is also proved that there is no duality gap if `` good '' generators of the gradient ideal are used . in @xcite , it is proved that the extended hierarchy is exact when the gradient ideal is radical and in @xcite the extended hierarchy is proved to be exact for any polynomial @xmath0 when the minimum is reached in @xmath4 ( see also @xcite ) . in @xcite , the relaxation techniques are analyzed for functions for which the minimum is not reached and which satisfies some special properties `` at infinity '' . from an algorithmic point of view , this is not ending the investigations since a criteria is needed to determine at which step the minimum is reached . it is also important to known if all the minimizers can be recovered at that step . methods based on moment matrices have been proposed to compute generators of the ideal characterizing the real solutions of a polynomial system . in @xcite , the computation of generators of the ( real ) radical of an ideal is based on moment matrices which involve all monomials of a given degree and a stopping criterion related to curto - fialkow flat extension condition @xcite is used . this method is improved in @xcite . combining the border basis algorithm of @xcite and a weaker flat extension condition @xcite , a new algorithm which involves sdp problems of significantly smaller size is proposed to compute the ( real ) radical of an ideal , when this ( real ) radical ideal is zero - dimensional . in @xcite , an algorithm is also proposed to compute the global minimum of the polynomial function @xmath0 based on techniques from @xcite . it terminates when the gradient ideal is radical zero - dimensional . an interesting feature of these hierarchies of sdp problems is that , at any step they provide a lower bound of @xmath1 and the sos hierarchy gives certificates for these lower bounds ( see e.g. @xcite and reference therein ) . in @xcite it is also shown how to obtain `` good '' upper bounds by perturbation techniques , which can directly be generalized to the approach we propose in this paper . * contributions . * we show that if the minimum @xmath1 is reached in @xmath4 , a generalized hierarchy of relaxation problems which involve the gradient ideal @xmath5 is exact and yields the generators of the minimizer ideal @xmath6 from a sufficiently high degree . in the case where the minimizer ideal is zero dimensional , we give a criterion for deciding when the minimum is reached , based on the flat extension condition in @xcite . this criterion is used in a new algorithm which computes a border basis of the minimizer ideal of a polynomial function , when this minimizer ideal is zero - dimensional . the algorithm is an extension of the real radical algorithm described in @xcite . the rows and columns of the matrices involved in the semi - definite programming problem are associated with the family of monomials candidates for being a basis of the quotient space , i.e , a subset of monomials of size much smaller than the number of monomials of the same degree . thus , the size of the sdp problems involved in this computation is significantly smaller than the one in @xcite , @xcite or @xcite . we show that by solving this sequence of sdp problems , we obtain in a finite number of steps the minimum @xmath1 of @xmath0 , with no duality gap . when this minimum is reached , the kernel of the hankel matrix associated to the solution of the sdp problem yields generators of the minimizer ideal which are not in the gradient ideal . assuming that the minimizer ideal @xmath6 is zero dimensional , computing the border basis of this kernel yields a representation of the quotient algebra by @xmath6 and thus a way to compute effectively the minimizer points , using eigenvector solvers . an implementation of this method has been developed , which integrates a border basis implementation and a numerical sdp solver . it is demonstrated on typical examples . * content . * the paper is organized as follows . section 2 recalls the concepts of algebraic tools as ideals , varieties , dual space , quotient algebra , the definitions and theorems about border basis , and the hankel operators involved in the computation of ( real ) radical ideals and in the computation of our minimizer ideal . in section 3 , we describe the main results on the exactness of the hierarchy of sdp problems on truncated hankel operators and show that the minimizer ideal can be computed from the kernel of these hankel . in section 4 , we analyze more precisely the case where the minimizer ideal is zero - dimensional . in section 5 , we describe our algorithm and we prove its correctness . finally in section 6 , we illustrate the algorithm on some classical examples . in this section , we set our notation and recall the eigenvalue techniques for solving polynomial equations and the border basis method . let @xmath7 $ ] be the set of the polynomials in the variables @xmath8 , @xmath9 , with coefficients in the field @xmath10 . hereafter , we will choose and @xmath11 can be replaced respectively by any real closed field and any field containing its algebraic closure . ] @xmath12 or @xmath11 . let @xmath13 denotes the algebraic closure of @xmath14 . for @xmath15 , @xmath16 is the monomial with exponent @xmath17 and degree @xmath18 . the set of all monomials in @xmath19 is denoted @xmath20 . we say that @xmath21 if @xmath22 divides @xmath23 , i.e. , if @xmath24 coordinate - wise . for a polynomial @xmath25 , its support is @xmath26 , the set of monomials occurring with a nonzero coefficient in @xmath0 . for @xmath27 and @xmath28 $ ] , we introduce the following sets : * @xmath29 is the set of elements of @xmath30 of degree @xmath31 , * @xmath32 is the linear span of @xmath30 , * @xmath33 , f \in s \big\}$ ] is the ideal in @xmath7 $ ] generated by @xmath30 , * @xmath34_{t-\deg(f)}\big\}$ ] is the vector space spanned by @xmath35 , * @xmath36 is the prolongation of @xmath30 by one degree , * @xmath37 is the border of @xmath30 , * @xmath38}:= s^{\stackrel{t\ \mathrm{times}}{+\cdots+}}$ ] is the result of applying @xmath3 times the prolongation operator ` @xmath39 ' on @xmath30 , with @xmath40}=s^+$ ] and , by convention , @xmath41}=s$ ] . therefore , @xmath42_t$ ] , @xmath38}=\{x^\alpha f\mid f\in s , |\alpha|\le t\}$ ] , @xmath43_t= ( s)_t$ ] , but the inclusion may be strict . if @xmath44 contains @xmath45 then , for any monomial @xmath46 , there exists an integer @xmath47 for which @xmath48}$ ] . @xmath49-index _ of @xmath50 , denoted by @xmath51 , is defined as the smallest integer @xmath47 for which @xmath48}$ ] . a set of monomials @xmath49 is said to be _ connected to @xmath45 _ if @xmath52 and for every monomial @xmath53 in @xmath49 , @xmath54 for some @xmath55 $ ] and @xmath56 . given a vector space @xmath57 , its prolongation @xmath58 is again a vector space . the vector space @xmath59 is said to be _ connected to 1 _ if @xmath60 and any non - constant polynomial @xmath61 can be written as @xmath62 for some polynomials @xmath63 with @xmath64 , @xmath65 for @xmath66 $ ] . obviously , @xmath59 is connected to 1 when @xmath67 for some monomial set @xmath68 which is connected to 1 . moreover , @xmath69 if @xmath67 . given an ideal @xmath70 $ ] and a field @xmath71 , we denote by @xmath72 its associated variety in @xmath73 . by convention @xmath74 . for a set @xmath75 , we define its vanishing ideal @xmath76\mid f(v)=0\ \forall v\in v\}.\ ] ] furthermore , we denote by @xmath77 \,\mid\ , f^m\in i \ \text { for some } m\in { \mathbb{n}}\setminus \{0\}\}\ ] ] the radical of @xmath78 . for @xmath12 , we have @xmath79 , but one may also be interested in the subset of real solutions , namely the real variety @xmath80 the corresponding vanishing ideal is @xmath81 and the _ real radical ideal _ is @xmath82i:=\{p\in { \mathbb{r}}[{\mathbf x}]\mid p^{2 m } + \sum_j q_j^2\in i \ \text { for some } q_j\in { \mathbb{r}}[{\mathbf x } ] , m\in { \mathbb{n}}\setminus \{0\}\}.\ ] ] obviously , @xmath83i\subseteq i(v_{\mathbb{r}}(i)).\ ] ] an ideal @xmath78 is said to be _ radical _ ( resp . , _ real radical _ ) if @xmath84 ( resp . @xmath85i$ ] ) . obviously , @xmath86 . hence , if @xmath87 is real radical , then @xmath78 is radical and moreover , @xmath88 if @xmath89 . the following two famous theorems relate vanishing and radical ideals : * * hilbert s nullstellensatz * ( see , e.g. , @xcite ) @xmath90 for an ideal @xmath91 $ ] . * * real nullstellensatz * ( see , e.g. , @xcite ) @xmath92i = i(v_{\mathbb{r}}(i))$ ] for an ideal @xmath93 $ ] . given an ideal @xmath70 $ ] , the quotient set @xmath94/i}$ ] consists of all cosets @xmath95:=f+i=\{f+q \mid q \,\in\,i\}$ ] for @xmath96 $ ] , i.e. , all equivalent classes of polynomials of @xmath7 $ ] modulo the ideal @xmath78 . the quotient set @xmath94/i}$ ] is an algebra with addition @xmath95+[g]:=[f+g]$ ] , scalar multiplication @xmath97:=[\lambda f]$ ] and with multiplication @xmath95[g]:=[fg]$ ] , for @xmath98 , @xmath99 $ ] . we will say that an ideal @xmath78 is zero - dimensional if @xmath100 . then @xmath94/i}$ ] is a finite - dimensional vector space and its dimension is the number of roots counted with multiciplicity ( see e.g. @xcite , @xcite ) . thus @xmath101/i}$ ] , with equality if and only if @xmath78 is radical . assume that @xmath102 and set @xmath103/i}$ ] so that @xmath104 . consider a set @xmath105 $ ] for which @xmath106,\ldots,[b_n]\}$ ] is a basis of @xmath94/i}$ ] ; by abuse of language we also say that @xmath49 itself is a basis of @xmath94/i}$ ] . then every @xmath96 $ ] can be written in a unique way as @xmath107 where @xmath108 @xmath109 the polynomial @xmath110 is called the remainder of @xmath0 modulo @xmath78 , or its _ normal form _ , with respect to the basis @xmath49 . in other words , @xmath111 and @xmath94/i}$ ] are isomorphic vector spaces . given a polynomial @xmath112 $ ] , we can define the _ multiplication ( by @xmath113 ) operator _ as @xmath114/i } & \longrightarrow & { { \mathbb{k}}[{\mathbf x}]/i}\\ & [ f ] & \longmapsto & { \mathcal{m}}_h([f])\,:=\ , [ hf]\ , , \end{array}\ ] ] assume that @xmath115/i}<\infty$ ] . then the multiplication operator @xmath116 can be represented by its matrix , again denoted @xmath116 for simplicity , with respect to a given basis @xmath117 of @xmath94/i}$ ] . namely , setting @xmath118 for some scalars @xmath119 , the @xmath120th column of @xmath116 is the vector @xmath121 . define the vector @xmath122 , whose coordinates are the evaluations of the polynomials @xmath123 at the point @xmath124 . the following famous result ( see e.g. , ( * ? ? ? * chapter 24 ) , @xcite ) relates the eigenvalues of the multiplication operators in @xmath94/i}$ ] to the algebraic variety @xmath125 . this result underlies the so - called eigenvalue method for solving polynomial equations @xcite and plays a central role in many algorithms , also in the present paper . [ thm::muloperator ] let @xmath78 be a zero - dimensional ideal in @xmath7 $ ] , @xmath49 a basis of @xmath94/i}$ ] , and @xmath112 $ ] . the eigenvalues of the multiplication operator @xmath116 are the evaluations @xmath126 of the polynomial @xmath113 at the points @xmath127 . moreover , @xmath128 and the set of common eigenvectors of @xmath129}$ ] are up to a non - zero scalar multiple the vectors @xmath130 for @xmath131 . throughout the paper we also denote by @xmath132 the matrix of the multiplication operator by the variable @xmath133 . by the above theorem , the eigenvalues of the matrices @xmath134 are the @xmath135th coordinates of the points @xmath131 . thus the task of solving a system of polynomial equations is reduced to a task of numerical linear algebra once a basis of @xmath94/i}$ ] and a normal form algorithm are available , permitting the construction of the multiplication matrices @xmath136 . the eigenvalue method for solving polynomial equations from the above section requires the knowledge of a basis of @xmath94/i}$ ] and an algorithm to compute the normal form of a polynomial with respect to this basis . in this section we will recall a general method for computing such a basis and a method to reduce polynomials to their normal form . throughout @xmath44 is a finite set of monomials . a rewriting family @xmath137 for a ( monomial ) set @xmath49 is a set of polynomials @xmath138 such that * @xmath139 , * @xmath140 has exactly * one * monomial in @xmath141 , denoted as @xmath142 and called the _ leading monomial _ of @xmath140 . ( the polynomial @xmath140 is normalized so that the coefficient of @xmath142 is @xmath45 . ) * if @xmath143 then @xmath144 . we say that the rewriting family @xmath137 is _ graded _ if @xmath145 for all @xmath146 . [ defnormfam ] a rewriting family @xmath137 for @xmath49 is said to be _ complete _ in degree @xmath3 if it is graded and satisfies @xmath147 ; that is , each monomial @xmath148 of degree at most @xmath3 is the leading monomial of some ( necessarily unique ) @xmath146 . [ defpibf ] let @xmath137 be a rewriting family for @xmath49 , complete in degree @xmath3 . let @xmath149 be the projection on @xmath111 along @xmath137 defined recursively on the monomials @xmath150 in the following way : * if @xmath151 , then @xmath152 , * if @xmath153}\setminus { \mathcal{b}}^{[0]})_{t})$ ] , then @xmath154 , where @xmath0 is the ( unique ) polynomial in @xmath137 for which @xmath155 , * if @xmath156}\setminus { \mathcal{b}}^{[k-1]})_{t}$ ] for some integer @xmath157 , write @xmath158 , where @xmath159}$ ] and @xmath55 $ ] is the smallest possible variable index for which such a decomposition exists , then @xmath160 . if @xmath137 is a graded rewriting family , one can easily verify that @xmath161 for @xmath162 . the map @xmath163 extends by linearity to a linear map from @xmath7_t$ ] onto @xmath164 . by construction , @xmath165 and @xmath166 for all @xmath167 . the next theorems show that , under some natural commutativity condition , the map @xmath163 coincides with the linear projection from @xmath7_t$ ] onto @xmath168 along the vector space @xmath169 . it leads to the notion of border basis . [ def : borderbasis ] let @xmath170 be connected to @xmath45 . a family @xmath171 $ ] is a border basis for @xmath49 if it is a rewriting family for @xmath49 , complete in all degrees , and such that @xmath7= { \<{{\mathcal{b}}}\ > } \oplus ( f).$ ] an algorithmic way to check that we have a border basis is based on the following result , that we recall from @xcite : [ thmcom ] assume that @xmath49 is connected to @xmath45 and let @xmath137 be a rewriting family for @xmath49 , complete in degree @xmath27 suppose that , for all @xmath172 , @xmath173.\ ] ] then @xmath149 coincides with the linear projection of @xmath7_{t}$ ] on @xmath168 along the vector space @xmath169 ; that is , @xmath7_t={\<{{\mathcal{b}}}\>}_t\oplus { \<f\,|\ , { t}\>}.$ ] in order to have a simple test and effective way to test the commutation relations , we introduce now the commutation polynomials . let @xmath137 be a rewriting family and @xmath174 . let @xmath175 be the smallest degree monomials for which @xmath176 . then the polynomial @xmath177 is called the _ commutation polynomial _ of @xmath178 . for a rewriting family @xmath137 with respecet to @xmath49 , we denote by @xmath179 the set of polynomials of the form @xmath180 where @xmath174 and + @xmath181 satisfy * either @xmath182 , * or @xmath183 and @xmath184 . therefore , @xmath185 and @xmath186 contains all commutation polynomials @xmath187 for @xmath174 whose monomial multipliers @xmath188 are of degree @xmath189 . the next result can be deduced using theorem [ thmcom ] . [ thmnfdegd ] let @xmath170 be connected to @xmath45 and let @xmath137 be a rewriting family for @xmath49 , complete in degree @xmath3 . if for all @xmath190 of degree @xmath31 , @xmath191 , then @xmath149 is the projection of @xmath192 on @xmath168 along @xmath169 , ie . @xmath193 . if such a property is satisfied we say that @xmath137 is a border basis for @xmath194 in degree @xmath195 . [ thmnfanyt ] let @xmath170 be connected to 1 and let @xmath137 be a rewriting family for @xmath49 , complete in any degree . assume that @xmath191 for all @xmath196.then @xmath49 is a basis of @xmath7/(f)$ ] , @xmath197 , and @xmath198 for all @xmath27 ; the set @xmath137 is a _ border basis _ of the ideal @xmath199 with respect to @xmath49 . this implies the following characterization of border bases using the commutation property . [ thmcomanyt ] let @xmath170 be connected to 1 and let @xmath137 be a rewriting family for @xmath49 , complete in any degree . if for all @xmath200 and all indices @xmath201 $ ] , we have : @xmath202 then @xmath49 is a basis of @xmath203 , @xmath197 , and @xmath198 for all @xmath204 . this section is based on @xcite . for @xmath205^*$ ] , the hankel operator @xmath206 is the operator from @xmath207 $ ] to @xmath207^*$ ] defined by @xmath208 \mapsto p\cdot \lambda \in { \mathbb{r}}[{\mathbf x}]^*\ ] ] we define the kernel of the hankel operator : @xmath209 \mid p \cdot \lambda = 0 , i.e , \ \lambda(pq)=0 \ \forall q \in { \mathbb{r}}[{\mathbf x } ] \}\ ] ] to analyse the properties of @xmath210 , we study the quotient algebra @xmath7/\ker h_\lambda={\mathcal{a}}_{\lambda}$ ] . a first result is the following ( see e.g. @xcite ) : [ lem : basis ] the rank of the operator @xmath206 is finite if and only if @xmath211 is a zero - dimensional ideal , in which case @xmath212/\ker h_\lambda = rank h_{\lambda}$ ] . for a zero - dimensional ideal @xmath213 $ ] with simple zeros @xmath214 , we have @xmath215 and the ideal @xmath78 is radical as a consequence of hilbert s nullstellensatz . when @xmath216 , this yields the following property . [ propradideal ] let @xmath217 and assume that @xmath218 . then , the ideal @xmath211 is radical if and only if @xmath219 in which case @xmath220 is the vanishing ideal of the points @xmath221 . as a corollary , we deduce that when @xmath222 and @xmath223 , the ideal @xmath211 is real radical if and only if the points @xmath224 are in @xmath225 . we say that @xmath205^*$ ] is positive , which we denote @xmath226 , if @xmath227 for all @xmath228 $ ] . we have that @xmath229 iff @xmath230 . if moreover @xmath231 , we say that @xmath210 is a probability measure . this term is justified by a theorem of riesz - haviland @xcite , which states that a lebesgue measure on @xmath4 is uniquely determined by its value on the polynomials @xmath232 $ ] . in particular , if @xmath229 and @xmath231 , there exists a unique probability measure @xmath233 such that @xmath234 , \lambda ( p ) = \int p\ , d\mu$ ] . an important property of positive forms is the following : [ proppositive ] assume @xmath218 . then @xmath226 if and only if @xmath210 has a decomposition ( [ eqlambdarad ] ) with @xmath235 and distinct @xmath236 , in which case @xmath237 . in particular , it shows that if @xmath226 , then @xmath238 is a real radical ideal . in this section , we give the main results which shows how the minimum of @xmath0 and the ideal defining the points where this minimum is reached , can be computed . hereafter , we will assume that the minimum @xmath1 of @xmath0 is reached at a point @xmath239 . we define the gradient ideal of @xmath240 : @xmath241 we define the minimizer ideal of f(x ) : @xmath242 by construction , @xmath243 and @xmath244 if the minimum @xmath1 is reached in @xmath4 . the objective of this section is to describe a method to compute generators of @xmath6 from generators of @xmath5 . for that purpose , first of all we need to restrict our analysis to matrices of finite size . for this reason , we consider here truncated hankel operators , which will play a central role in the construction of the minimizer ideal of @xmath0 . for a vector space @xmath245 $ ] , let @xmath246 . for a linear form @xmath247 , we define the map @xmath248 by @xmath249 for @xmath250 . thus @xmath251 is called a truncated hankel operator , defined on the subspace @xmath59 . we define the kernel of the truncated hankel operator : @xmath252 when @xmath253_t$ ] for @xmath27 , @xmath251 is also denoted @xmath254 . given a subspace @xmath255 , @xmath210 induces a linear map on @xmath256 and we can consider the induced truncated hankel operator @xmath257 as a restriction of @xmath258 . below we will deal with the following sets and minimums , in order to define our primal - dual problems . given a vector space @xmath259 $ ] and @xmath260 , we define @xmath261 if @xmath253_t$ ] and @xmath262 , we also denote @xmath263 by @xmath264 . notice that if @xmath265 and @xmath266 then @xmath267 . when @xmath59 and @xmath268 are vector spaces of finite dimension , @xmath269 is the intersection of the closed convex cone of semi - definite positive quadratic forms on @xmath270 with a linear space , thus it is a convex closed semi - algebraic set . more details on its description will be given in section [ sec-0dim ] . we will need the following result : [ positive ] for any vector space @xmath59 and @xmath271 and @xmath272 , we have : * @xmath273 , @xmath274 implies @xmath275 . * @xmath276 . the first point is a consequence of the positivity of @xmath277 . for the second point , @xmath278 , @xmath279 , @xmath280 . dividing by @xmath3 and letting @xmath3 go to zero yields @xmath281 , thus showing @xmath282 . the inclusion @xmath283 is obvious . conversely , let @xmath284 . in particular , @xmath285 , which implies @xmath286 ( since @xmath287 ) and thus @xmath288 . given a vector space @xmath259 $ ] and @xmath260 , we define @xmath289 \mid p= \sum_{i=1}^s h_i^2 + h , \ h \in g \ } \label{eq2}.\ ] ] if @xmath253_t$ ] and @xmath262 , we also denote @xmath290 by @xmath291 . notice that if @xmath265 and @xmath266 then @xmath292 . when @xmath59 and @xmath268 are vector spaces of finite dimension , @xmath293 is the projection of the sum of a linear space and the convex cone of positive quadratic forms on @xmath294 . [ mintruncated ] let @xmath59 be a subspace of @xmath207 $ ] such that @xmath295 and @xmath296 and let @xmath297 . we assume that @xmath0 attains its minimum at some points @xmath298 . we define the following extrema : * @xmath299 * @xmath300 stands for measure ] s.t . @xmath301 * @xmath302 s.t . @xmath303 if @xmath253_t$ ] ( resp . @xmath253 $ ] ) and @xmath262 we also denote @xmath304 by @xmath305 ( resp . @xmath306 ) and @xmath307 by @xmath308 ( resp . @xmath309 by convention if the sets are empty , @xmath310 and @xmath311 . we now analyse the relations between these different extrema . [ subset ] let @xmath266 be two subspaces of @xmath207 $ ] and @xmath312 , @xmath313 with @xmath265 then we directly deduce that * @xmath314 , * @xmath315 , from the fact that @xmath316 and @xmath317 . if we take @xmath318 , we have the following relations between the extrema . [ minor ] let @xmath319 . then @xmath320 . [ subspaces ] we have @xmath321 because if there exists @xmath322 such that @xmath323 , i.e. @xmath324 with @xmath325 and @xmath326 then @xmath327 , @xmath328 we deduce that @xmath329 and we conclude @xmath321 . for the second inequality , let @xmath330 be a point of @xmath225 such that @xmath331 is the mininum of @xmath0 and let @xmath332^ * : p \longmapsto p({\mathbf x}^*)$ ] be the evaluation at @xmath330 . then we have @xmath333 , @xmath334 and @xmath335 since @xmath336 , so that @xmath337 we deduce the inequality @xmath338 . following the relaxation approach proposed in @xcite , we are now going to consider a hierarchy of convex optimization problems and show that for such hierarchy , the minimum @xmath1 is always reached in a finite number of steps . let us consider the sequence of spaces @xmath339 for @xmath27 , @xmath340 . using remark [ subset ] and the fact that @xmath341 , we check that * the increasing sequence @xmath342 converges to @xmath343 , * the increasing sequence @xmath344 converges to @xmath345 . we are going to show that these limits are attained for some @xmath27 . the next result which is a slight variation of a result in @xcite ( and also used in @xcite ) shows that for a high enough degree , the kernel of some truncated hankel operators allows us to compute generators of the real radical of an ideal . [ radical ] for @xmath346 $ ] with @xmath347 , there exists @xmath348 such that @xmath349 , @xmath350{i_{grad } ( f ) } \subset ( \ker h_{\lambda^*}^t)$ ] . let @xmath351 be generators of @xmath352 and @xmath353 be generators of the ideal @xmath354{i}$ ] . by the real nullstellensatz , for @xmath355 , there exist @xmath356 and polynomials @xmath357 and sums of squares @xmath358 such that @xmath359 . as @xmath360 , there exist @xmath361 such that @xmath362 . set @xmath363 , @xmath364 and let @xmath365 . as @xmath366 , @xmath367 for all @xmath368 . hence @xmath369 , which implies that @xmath370 since @xmath371 . to compute generators of @xmath6 , we use the decomposition of @xmath372 in components where @xmath0 has a constant value . [ pis ] let @xmath373 be the decomposition of the variety in disjoint real subvarieties , such that @xmath374 with @xmath375 . then there exist polynomials @xmath376 $ ] such that @xmath377 , where @xmath378 is the kronecker delta function . as in @xcite , we decompose @xmath379 as an union of complex varieties + @xmath380 such that @xmath381 , @xmath382 have real points and @xmath383 is constant on @xmath381 and @xmath384 has no real point . we number these varieties so that @xmath385 and @xmath386 . by construction , the varieties @xmath387 are disjoint , @xmath0 is constant on @xmath388 and @xmath389 . let us take @xmath390 where @xmath391 are the lagrange interpolation polynomials at the value @xmath392 . they satisfy @xmath377 . [ idempotents ] by definition the polynomials @xmath393 have the following properties : * @xmath394 modulo @xmath395{i_{grad } ( f)}$ ] . * @xmath396 modulo @xmath395{i_{grad } ( f)}$ ] @xmath397 . the next results shows that in the sequence of optimization problems that we consider , the minimum of @xmath0 is reached from a certain degree . [ mayor][pi - kernel ] for @xmath346 $ ] with @xmath398 , there exists @xmath399 such that @xmath400 , @xmath401 and @xmath402 with @xmath403 , we have @xmath404 . by @xcite[theorem 2.3 ] , there exists @xmath405 such that @xmath406 , @xmath407 where @xmath408 . as @xmath409 , there exists @xmath410 $ ] such that @xmath411 . let @xmath412 . then @xmath413 . by remark [ subset ] and proposition [ minor ] , for @xmath414 , @xmath415 . thus for @xmath416 , we have @xmath417 let @xmath418{i_{grad } ( f)}$ ] . by lemma [ pis ] and remark [ idempotents ] , we can write @xmath419 where @xmath420 and @xmath421 . then @xmath422 hence @xmath423 with @xmath424 . by theorem [ radical ] , there exists @xmath425 such that @xmath426 for all @xmath427 . let us take @xmath428 , @xmath429 and @xmath430 such that @xmath403 . from equation , we deduce that @xmath431 this implies that @xmath432 and @xmath433 for @xmath434 , since @xmath435 and @xmath436 on @xmath207_t$ ] . the example 3.4 in @xcite shows that it may not always be possible to write @xmath437 as a sum of squares modulo @xmath5 but , from the previous result we see that @xmath437 is the limit of sums of squares modulo @xmath438_{t}$ ] for a fixed @xmath439 . under the assumption that there exists @xmath239 such that @xmath440 , we can construct @xmath441 such that @xmath442 . this means that @xmath305 is reached for @xmath439 . a direct corollary of the previous theorem is that for any @xmath346 $ ] with @xmath443 , we have @xmath444 as for the construction of generators of the real radical @xmath395{i_{grad } ( f)}$ ] ( proposition [ radical ] ) , we can construct generators of @xmath6 from the kernel of a truncated hankel operator associated to any linear form which minimizes @xmath0 : [ minker ] for @xmath346 $ ] with @xmath398 , there exists @xmath445 such that @xmath446 , for @xmath447 with @xmath403 , we have @xmath448 . let @xmath449 and @xmath418{i}$ ] . we consider again the above decomposition @xmath450 with @xmath451 . we denote by @xmath452 a family of generators of the ideal @xmath453 and @xmath454 . let us fix @xmath455 and show that @xmath456 for @xmath3 sufficiently large . we know that @xmath457 and that for @xmath458 , @xmath459 which implies that @xmath460{i_{grad } ( f)}$ ] . by theorem [ radical ] , there exists @xmath461 such that @xmath462 for all @xmath368 with @xmath463 . by theorem [ pi - kernel ] , @xmath464 for @xmath465 . by remark [ idempotents ] , @xmath394 modulo @xmath466 . by theorem [ radical ] , there exists @xmath467 such that @xmath468 for all @xmath368 with @xmath469 . let us take @xmath470 and @xmath471 . then for @xmath447 with @xmath472 , we have @xmath473 hence , @xmath456 for @xmath474 and @xmath448 . we introduce now the notion of _ generic linear form for @xmath475$]_. such a linear form will allow us to compute @xmath6 as we will see . [ generic ] for @xmath476 and @xmath477 $ ] , the following assertions are equivalent : * @xmath478 . * @xmath327 such that @xmath479 , @xmath480 . * @xmath481 for any subspace @xmath482 . we say that @xmath476 is generic for @xmath483 if it satisfies one of the equivalent conditions ( i)-(iii ) . \(i ) @xmath484 ( ii ) : note that @xmath485 and @xmath486 and @xmath487 ( using lemma [ positive ] ) . hence , @xmath488 and thus equality holds . this implies that @xmath489 is thus contained in @xmath490 . + ( ii ) @xmath484 ( iii ) : given @xmath255 , we show that @xmath491 . by lemma [ positive ] , we have @xmath492 and , by the above , we have @xmath493 . + ( iii ) @xmath484 ( i ) this implication is obvious . the next result , which refines theorem [ minker ] , shows only elements in @xmath6 are involved in the kernel of a truncated hankel operator associated to a generic linear form for @xmath0 . [ kermin ] let @xmath494 be as in definition [ mintruncated ] with @xmath495 . if @xmath476 is generic for @xmath0 and such that @xmath442 , then @xmath496 . let @xmath298 such that @xmath497 is minimum . let @xmath498 denotes the evaluation at @xmath330 restricted to @xmath499 and @xmath500 . our objective is to show that @xmath501 . suppose for contradiction that @xmath502 . we know that @xmath503 since @xmath340 and @xmath504 . we define @xmath505 . by definition @xmath506 and @xmath507 . as @xmath500 , @xmath508 thus @xmath509 and by the maximality of the rank of @xmath510 , @xmath511 . hence there exits @xmath512 . then @xmath513 . as @xmath514 implies @xmath515 . on the other hand,@xmath516 as @xmath517 , then @xmath518 as @xmath515 , since we have supposed that @xmath502 it yields a contradiction . the last result of this section shows that a generic linear form for @xmath0 yields the generators of the minimizer ideal @xmath6 in high enough degree . [ thprin ] for @xmath346 $ ] with @xmath398 , there exists @xmath519 ( defined in theorem [ minker ] ) such that such that @xmath446 , for @xmath520 generic for @xmath0 , we have @xmath521 and @xmath522 . we obtain the result as consequence of theorem [ mayor ] , theorem [ minker ] and theorem [ kermin ] . as a consequence , in a finite number of steps , the sequence of optimization problems that we consider gives the minimum of @xmath0 and the generators of @xmath6 . in this section , we describe a criterion to detect when the kernel of a truncated hankel operator associated to a generic linear form for @xmath0 yields the generators of the minimizer ideal . it is based on a flat extension property @xcite and applies to global polynomial optimization problems where the minimizer ideal @xmath6 is zero - dimensional . [ defflatextension ] given vector subspaces @xmath523 $ ] and @xmath524 , @xmath251 is said to be a _ flat extension _ of its restriction @xmath525 if @xmath526 . we recall here a result from @xcite , which gives a rank condition for the existence of a flat extension of a truncated hankel operator . [ theoflatextension ] consider a monomial set @xmath527 connected to 1 and a linear function @xmath210 defined on @xmath528 . let @xmath529 and @xmath530 . assume that @xmath531 . then there exists a ( unique ) linear form @xmath532^*$ ] which extends @xmath210 , i.e. , @xmath533 for all @xmath534 , satisfying @xmath535 . moreover , we have @xmath536 . in other words , the condition @xmath537 implies that the truncated hankel operator @xmath538 has a ( unique ) flat extension to a ( full ) hankel operator @xmath539 . [ general ] let @xmath59 , @xmath268 be as in definition [ mintruncated ] with @xmath540 . if @xmath476 coincides with a probability measure @xmath233 on @xmath541 and satisfies @xmath542 , then @xmath543 . by proposition [ minor ] , @xmath338 . conversely as @xmath544 for all @xmath545 , we have @xmath546 . [ nogap ] if there exists @xmath547 with @xmath548 , then @xmath549 . if @xmath548 implies that @xmath550 , i.e , @xmath551 . hence by slater s theorem of @xcite we have the strong duality then @xmath549 . [ caszero ] let @xmath194 is a monomial set connected to @xmath45 , @xmath552 and @xmath553 . let @xmath554 such that @xmath555 is generic for @xmath0 and satisfies the flat extension property : @xmath556 . then there is no duality gap , @xmath557 and @xmath558 as @xmath559 , theorem [ theoflatextension ] implies that there exists a ( unique ) linear function @xmath560^*$ ] which extends @xmath555 . as @xmath561 and @xmath562 , any polynomial @xmath228 $ ] can be reduced modulo @xmath563 to a polynomial @xmath564 so that @xmath565 . then @xmath566 since @xmath554 . this implies that @xmath567 . by proposition [ proppositive ] , @xmath568 has a decomposition @xmath569 with @xmath570 and @xmath571 . as @xmath572 , @xmath573 is a probability measure . by proposition [ general ] , @xmath574 . as @xmath569 with @xmath570 and @xmath571 and as @xmath575 and @xmath576 , we deduce that @xmath577 for @xmath578 so that @xmath579 . by proposition [ propradideal ] , theorem [ theoflatextension ] and theorem [ kermin ] , we have @xmath580 we deduce that @xmath581 so that @xmath582 . as we have the flat extension condition , @xmath583 and by proposition [ nogap ] there is not duality gap : @xmath557 . notice that if the hypotheses of this theorem are satisfied , then necessarily @xmath6 is zero - dimensional . in this section we describe the algorithm to compute the global minimum of a polynomial , assuming @xmath1 is reached in @xmath4 and @xmath6 is zero dimensional . it can be seen as a type of border basis algorithm , for we insert an additional step in the main loop . it is closely connected to the real radical border basis algorithm presented in @xcite but instead of `` minimizing zero '' to generate new elements in the real radical , we minimize @xmath0 to compute generators of the minimizer ideal @xmath6 . the convex optimization problems that we consider are the following : + * input : * @xmath475 $ ] , @xmath584 a monomial set containing @xmath45 with @xmath585 , @xmath346 $ ] . + * output : * the minimum @xmath586 of @xmath587 subject to : * @xmath588 , * @xmath589 satisfies the hankel constraints @xmath590 , and @xmath591 + if @xmath592 , * @xmath593 for all @xmath594 . and @xmath595 represented by the vector @xmath596_{\alpha \in a + a}$ ] . this optimization problem is a semi - definite programming problem , corresponding to the optimization of a linear functional on a linear subspace of the cone of positive semi - definite matrices . it is a convex optimization problem , which can be solved efficiently by sdp solvers . if an interior point methods is used , the solution @xmath555 is in the interior of a face on which the minimum @xmath597 is reached so that @xmath555 is generic for @xmath0 . this is the case for tools such as ` csdp ` or ` sdpa ` , that we will use in the experimentations . in the case , where @xmath598 is a monomial set connected to @xmath45 , @xmath137 is a complete rewriting family for @xmath194 in degree @xmath599 , and @xmath600 , we will also denote optimal linear form @xmath601:=optimal linear form @xmath602 . @xmath603 ; @xmath604 ; @xmath194 : = set of monomials of degree @xmath195 ; + @xmath605 ; stop:=false ; + while not stop + 1 . compute the commuting relations for @xmath137 with respect to @xmath194 in degree @xmath606 . 2 . reduce them by the existing relations @xmath137 . 3 . add the non - zero reduced relations to @xmath137 and update @xmath194 . 4 . let @xmath607:= \textsc{optimal linear form } ( f , b_{t},\pi_{f , b_{t}})$ ] 1 . if there is a duality gap then go to 1 with @xmath608 . 2 . if @xmath609 then @xmath610 ; go to 1 with @xmath608 . 3 . compute the border basis @xmath611 of @xmath612 and the basis @xmath613 in degree @xmath614 . if there exists @xmath615 with @xmath616 then go to ( 1 ) with @xmath608 . 5 . let @xmath617 : = \textsc{optimal linear form } ( f , b',\pi_{f',b'})$ ] . 6 . if @xmath618 and @xmath619 stop = true , @xmath620 , @xmath621 , @xmath622 . 7 . else go to 1 with @xmath608 . in this subsection , we analyse the correctness of the algorithm . [ lemextend ] let @xmath27 , @xmath623_{2t}$ ] be a monomial set connected to @xmath45 , @xmath624 $ ] be a border basis for @xmath194 in degree @xmath625 and @xmath626 . let @xmath245_{t}$ ] be a vector space containing @xmath627_{t}$ ] and @xmath628 . for all @xmath629 , there exists a unique @xmath630 which extends @xmath210 . moreover , @xmath631 satisfies @xmath632 and @xmath633 . suppose that @xmath624 $ ] be a border basis for @xmath194 in degree @xmath625 , that is , all boundary polynomials of @xmath634 reduces to @xmath635 by @xmath636 . then by theorem [ thmnfdegd ] , we have @xmath207_{2\,t}={\<{b}\>}_{2\,t } \oplus { \<f\,|\ , { 2\,t}\>}$ ] and @xmath637 , @xmath638 . thus for all @xmath639 , there exists a unique @xmath640 such that @xmath641 and @xmath642 for all @xmath643 . as @xmath644 , we have @xmath645 so that @xmath646 for any element @xmath647 we have @xmath648 . as @xmath649 and @xmath650 , for any element @xmath651 , @xmath652 . this proves that @xmath653 conversely as @xmath654 , any element of @xmath59 can be reduced modulo @xmath655 to an element of @xmath656 , which shows that @xmath657 from the inclusions , and , we deduce that @xmath658 and that @xmath632 . as @xmath226 , by projection along @xmath659 , @xmath273 there exists @xmath660 such that @xmath661 . thus @xmath662 which ends the proof of this lemma . [ algostopbien ] if the algorithm terminates , then @xmath611 is a border basis of @xmath663 for @xmath613 . if the algorithme stops , this can happen only in step ( 4 ) in some degree @xmath664 such that * @xmath665:= \textsc{optimal linear form } ( f , b_{t+1},f)$ ] , * @xmath611 is a border basis of @xmath666 for @xmath613 in degree @xmath667 , * the monomials in @xmath613 are of degree @xmath668 so that @xmath669 , * @xmath617 : = \textsc{optimal linear form } ( f , b',f')$ ] with @xmath670 and @xmath671 . then @xmath672 . by lemma [ lemextend ] with @xmath673 , the linear form @xmath674 can be extended to a linear form @xmath675 with @xmath676 such that @xmath677 . as @xmath678 and @xmath672 , we have @xmath679 and the flat extension theorem ( theorem [ theoflatextension ] ) applies : @xmath680 is the restriction to @xmath681 of a positive linear form @xmath682 with @xmath683 , @xmath684 distinct , @xmath685 and @xmath686 . then @xmath687 . by hypothesis , @xmath688 since @xmath689 . we deduce that @xmath690 , that @xmath691 by theorem [ kermin ] and that @xmath692 by theorem [ caszero ] . therefore @xmath611 is a border basis of @xmath693 for @xmath613 . assume that @xmath693 is zero - dimensional . then the algorithm terminates . it outputs a border basis @xmath137 for @xmath194 connected to 1 , such that @xmath207=\ < b \ > \oplus ( f)$ ] and @xmath694 . first , we are going to prove by contradiction that when @xmath693 is zero - dimensional , the algorithm terminates . suppose that the loop goes for ever . notice that at each step either @xmath137 is extended by adding new linearly independent polynomials or it moves to @xmath664 . since the number of linearly independent polynomials added to @xmath137 in degree @xmath695 is finite , there is a step in the loop from which @xmath137 is not modified any more in degree @xmath696 2 t. in this case , all boundary c - polynomials of elements of @xmath137 of degree @xmath695 are reduced to @xmath635 by @xmath697 and @xmath698 is a border basis for @xmath699 in degree @xmath625 . by lemma [ lemextend ] with @xmath700 , @xmath555 extends to a linear form @xmath701_{2\ , t}^{*}$ ] such that @xmath702 by theorem [ thprin ] , for @xmath471 we have @xmath703 and @xmath704 . as @xmath6 is zero - dimensional , for @xmath3 high enough , a border basis @xmath611 of @xmath705 in degree @xmath695 is a border basis of @xmath703 . let @xmath613 be the corresponding monomial basis . then @xmath706 is the number of minimizers ( ie . the points in @xmath707 ) . by lemma [ lemextend ] with @xmath673 and theorem [ caszero ] , any linear form @xmath708 generic for @xmath0 is the restriction of a positive linear form @xmath709 with @xmath581 and @xmath685 and @xmath686 . in this case , @xmath710 and @xmath711 . we arrive at a contradiction , which shows that the algorithm should stops in step ( 4 ) , for some degree @xmath3 . by lemma [ algostopbien ] , @xmath611 is a border basis of @xmath6 with basis @xmath613 and @xmath712 . this section contains examples which illustrate the behavior of the algorithm . in the first example of motzkin polynomial , the gradient ideal is not zero - dimensional whereas that in the second example of robinson polynomial , the gradient ideal is zero - dimensional . in all the examples the minimizer ideal is zero - dimensional hence when we apply our algorithm we obtain the good result in a finite number of steps . semidefinite positive hankel operators are computed using the semi definite programming routine of sdpa software . for the link with sdpa we use a file interface since sdpa is not distributed as a library . for the computation of border basis , we use as a choice function that is tolerant to numerical instability i.e. a choice function that chooses as leading monomial a monomial whose coefficient is maximal among the choosable monomials . this , according to @xcite , makes the border basis computation stable with respect to numerical perturbations . this property is fundamental as we use results from sdp solvers to get new equations . and as these equations are computed numerically , the computation is subject to numerical errors . * in the first iteration the degree is 3 , the size of the hankel matrix @xmath718 is 10 , @xmath719 , there is a duality gap hence we try with degree 4 . * in the second iteration the degree is 4 , the size of the hankel matrix @xmath720 is 15 , min @xmath721 , there is no duality gap . the minimum differs from the previous minimum , so a new iteration is needed . * in the third iteration the degree is 5 , the size of the hankel matrix @xmath722 is 19 and @xmath723 . the minimums are equal hence we compute the kernel of @xmath720 , which is generated by 5 polynomials . we compute the border basis and obtain the basis @xmath724 . all the elements of @xmath194 have degree @xmath725 . * we compute a generic form for @xmath0 with this border basis , the size of the hankel matrix @xmath720 is 4 , @xmath726 = 0 and @xmath727 . * in the first iteration the degree is 3 , the size of the hankel matrix @xmath718 is 10 , min @xmath734 , there is no duality gap hence we compare the minimum with degree 4 . * in the second iteration the degree is 4 , the size of the hankel matrix @xmath720 is 15 , @xmath723 . as the minimums are different , another iteration is needed . * in the third iteration the degree is 5 , the size of the hankel matrix @xmath722 is 19 and @xmath726 = 0 . the minimums are equal , hence we compute the kernel of @xmath720 , which is generated by 6 polynomials . we compute the border basis and we obtain + @xmath735 . there exists an element of @xmath194 with degree @xmath736 , so we go to the next degree . * we compute a generic form @xmath210 for @xmath0 in degree @xmath737 , the size of the hankel matrix @xmath738 is 22 and @xmath726 = 0 . the minimums are equal , hence we compute the kernel of @xmath722 , which is generated by 11 polynomials . we compute the border basis and obtain the basis @xmath739 . all the elements of @xmath194 have degree @xmath740 . * we compute a generic form for @xmath0 with this border basis , the size of the hankel matrix @xmath722 is 8 , @xmath726 = 0 and @xmath741 . * in the first iteration the degree is 2 , the size of the hankel matrix @xmath748 is 4 , min @xmath749 . the minimum differs from the previous minimum @xmath750 . hence , a new iteration is needed . * in the second iteration the degree is 3 , the size of the hankel matrix @xmath718 is 4 and @xmath751 . the minimums are equal hence we compute the kernel of @xmath748 , which is generated by 3 polynomials . we compute the border basis and we obtain @xmath752 . all the elements of @xmath194 have degree @xmath753 . * we compute a generic form for @xmath0 with this border basis , the size of the hankel matrix @xmath748 is 1 , @xmath751 and @xmath754 . * in the first iteration the degree is 3 , the size of the hankel matrix @xmath718 is 10 , min @xmath761 , there is duality gap we try with degree 4 but we change the basis we try again with degree 3 . * in the second iteration the degree is 3 , the size of the hankel matrix @xmath718 is 10 , min @xmath762 , there is no duality gap . as the minimum differs from the minimum in degree @xmath763 , a new iteration step is performed . * in the third iteration the degree is 4 , the size of the hankel matrix @xmath720 is 11 ( with the reduction ) and min @xmath764 . the minimums are equal , hence we compute the kernel of @xmath718 , which is generated by 9 polynomials . we compute the border basis and obtain the basis @xmath752 . all the elements of @xmath194 have degree @xmath765 . * we compute a generic form for @xmath0 with this border basis , the size of the hankel matrix @xmath718 is 1 , @xmath766 and @xmath767 . the experiments suggest that due to the small size of the matrices @xmath771 most of the resolution time is spent during a classical border basis computation , we all the more emphasize this , as we used a file interface for communicating with sdpa which is rather slow . a. greuet and m. safey el din . deciding reachability of the infimum of a multivariate polynomial . in _ proceedings of the 36th international symposium on symbolic and algebraic computation _ , issac 11 , pages 131138 , new york , ny , usa , 2011 . f. guo , m. safey el din , and l. zhi . global optimization of polynomials using generalized critical values and sums of squares . in _ proceedings of the 2010 international symposium on symbolic and algebraic computation _ , issac 10 , pages 107114 , new york , ny , usa , 2010 . acm . lasserre , m. laurent , and p. rostalski . a unified approach for real and complex zeros of zero - dimensional ideals . in m. putinar and s. sullivant , editors , _ emerging applications of algebraic geometry . _ , volume 149 , pages 125156 . springer , 2009 . b. mourrain and p. trbuchet . generalized normal forms and polynomials system solving . in m. kauers , editor , _ issac : proceedings of the acm sigsam international symposium on symbolic and algebraic computation _ , pages 253260 , 2005 . y. nesterov . squared functional systems and optimization problems . in h. frenk , k. roos , t. terlaky , and s. zhang , editors , _ high performance optimization _ , chapter 17 , pages 405440 . kluwer academic publishers , dordrecht , the netherlands , 2000 . parrilo and b. sturmfels . minimizing polynomial functions . in _ proceedings of the dimacs workshop on algorithmic and quantitative aspects of real algebraic geometry in mathematics and computer science _ , pages 83100 . american mathematical society , 2003 . m. safey el din . computing the global optimum of a multivariate polynomial over the reals . in _ proceedings of the twenty - first international symposium on symbolic and algebraic computation _ , issac 08 , pages 7178 , new york , ny , usa , 2008 .

we present an analysis of the x - ray properties of sources detected in thexmm-_newton _ deep ( 200ks ) field . in order to constrain the absorbed agn population , we use extensive monte carlo simulations to directly compare the x - ray colours of observed sources with those predicted by several model distributions . in particular , we have carried out our comparisons over the entire 0.210 kev energy range of the xmm-_newton _ cameras , making our analysis sensitive to a large range of absorbing column densities . we have tested the simplest form of the unified scheme , whereby the intrinsic luminosity function of absorbed agn is set to be the same as that of their unabsorbed brethren , coupled with various model distributions of absorption . of the tested distributions , the best fitting model has the fraction of agn with absorbing column , proportional to ( log) . we have also tested two extensions to the unified scheme : an evolving absorption scenario , in which the fraction of absorbed sources is larger at higher redshifts , and a luminosity dependent model in which high luminosity agn are less likely to be absorbed . both of these models provide poorer matches to the observed x - ray colour distributions than the best fitting simple unified model . we find that a luminosity dependent density evolution luminosity function reproduces poorly the 0.52 kev source counts seen in thefield . field to field variations could be the cause of this disparity . computing the x - ray colours with a simple absorbed power - law spectral model is found to over - predict , by a factor of two , the fraction of hard sources that are completely absorbed below 0.5 kev , implying that an additional source of soft - band flux must be present in a number of the absorbed sources . the tested synthesis models predict that around 16% of the detections in thefield are due to agn at . however , so far , only a single agn with has been identified in our approximately complete optical spectroscopy follow up program . finally , we use our simulations to demonstrate the efficacy of a hardness ratio selection scheme at selecting absorbed sources for further study . using this selection scheme , we show that around 40% of thesample are expected to be agn with . [ firstpage ] surveys galaxies : active x - rays : galaxies quasars : general

astro-ph0504503

deep x - ray surveys have progressively resolved an increasing fraction of the soft x - ray background ( xrb ) , into faint point - like sources . most recently , the ultra - deep _ chandra _ surveys ( @xcite , @xcite ) , have resolved over @xmath8 of the 0.52 kev xrb . source counts have been measured in these fields to limiting fluxes of a few times @xmath9erg s@xmath10 @xmath11 , with corresponding sky densities of over @xmath12 sources deg@xmath13 , @xcite . optical identification of faint x - ray sources reveals a heterogeneous mixture of objects , with the dominant class being active galactic nuclei ( agn ) , ( @xcite , @xcite ) . these x - ray selected agn have a range of luminosities spanning several orders of magnitude , are found at redshifts up to 5 , and exhibit a wide range of observational properties . the unification scheme of @xcite attributes the wide variety of radio , optical and x - ray characteristics seen between agn classes , to the geometry and relative orientation of a dusty torus surrounding the central black hole . the dusty torus model is used to to explain both the absence of broadened optical lines , and the strong x - ray absorption seen in many agn . the simplest version of the unified scheme is one in which the geometry of the inner regions , and hence the distribution of absorbing column densities ( @xmath1 ) , is independent of all other agn properties . a number of refinements to this model have been suggested , in order to explain recent observations which are at odds with the simple unified scheme . the demographics of the 0.52 kev agn population have been well measured , albeit to relatively bright flux limits , using _ rosat _ /_einstein _ samples , for example by @xcite , @xcite , @xcite and @xcite . each of these studies indicates strong evolution of the agn x - ray luminosity function ( xlf ) , with a peak in agn activity at @xmath14 , although the best description of this evolution is disputed . the soft xlf is characterised by a double power - law with a knee at @xmath15 erg s@xmath10 . the bulk of the xrb is the result of the integrated emission of agn having luminosities in the vicinity of this knee . the spectral slope of the 0.11 kev xrb is @xmath16 @xcite , approximately matching that of a typical soft band detected agn , @xmath17 , ( @xcite , @xcite , @xcite ) . however , at harder x - ray energies the xrb flattens dramatically , and has @xmath18 in the 110 kev band ( e.g. @xcite , @xcite , @xcite ) , and can not be produced by a simple superposition of canonical @xmath19 agn spectra . clearly , some additional sources of hard x - rays must exist , and a large population of heavily absorbed agn is postulated to fill this role . population synthesis models have been formulated , such as those of @xcite , @xcite , and @xcite , and have been successful in reproducing both the broad band spectrum of the xrb , as well as the agn source counts observed below 10 kev . @xcite predict that the majority ( @xmath20 ) , of the agn population is heavily absorbed . a number of deep surveys of the 210 kev sky have been performed with _ chandra _ and xmm-_newton _ ( e.g. @xcite , @xcite , @xcite ) . in the 2ms observations of the _ chandra _ deep field - north ( cdf - north ) , sources have been detected to a limiting flux of @xmath21 erg s@xmath10 @xmath11 in the 210 kev band @xcite . the 1ms observations of the _ chandra _ deep field - south ( cdf - south ) , are estimated to have resolved more than @xmath22 of the 210 kev xrb @xcite . however , due to the relatively large uncertainty ( @xmath23 ) in the total intensity of the 210 kev cosmic xrb , the precise resolved fraction is still somewhat unknown @xcite . it has also been shown that extrapolating the source counts seen in the chandra deep fields to much lower fluxes does not reproduce fully the total level of the cosmic xrb , suggesting the existence of an additional very faint x - ray population ( e.g. @xcite , @xcite ) . the luminosity function of agn that are selected in the 210 kev band , has been measured by several studies . @xcite demonstrated that strong evolution of the hard xlf has occurred between the @xmath24 and @xmath25 epochs . @xcite used a sample of 247 agn , including some from the cdf - north , to show that the xlf is best represented by a complex luminosity dependent density evolution ( ldde ) model . it should be noted that this sample does not reach to the limiting flux of the cdf - n data , but to @xmath26 erg s@xmath10 @xmath11 , where the optical identifications are reasonably complete . the major stumbling block in understanding the nature of the faint , hard x - ray selected agn population , is the difficulty of obtaining optical spectroscopic identifications . the soft x - ray selected samples used for xlf determinations , for example @xcite and @xcite , are primarily , or wholly , composed of bright ( @xmath27 ) , broad - line agn counterparts , which are relatively easy to optically identify . at fainter x - ray fluxes , such as those probed in the cdf - north , agn without broad lines , together with normal galaxies , make up a large fraction of the identified objects @xcite . those non - broad - line agn having spectroscopic identifications are almost all found with @xmath28 , in contrast to the peak of the broad - line sample , which lies at @xmath29 . the large numbers of @xmath30 type - ii quasars ( having log@xmath31 , log@xmath32 ) , predicted by synthesis models have not been detected in these surveys . the obvious conclusion to be drawn is that the absorbed and unabsorbed agn are taken from separate populations , a direct contradiction of the simplest unified scheme . however , the optical follow up programs in these deep _ chandra _ fields are by no means complete to the faint x - ray limit . for example , in the cdf - north , only 55% of the x - ray detections have optical counterparts with @xmath33 @xcite . objects fainter than this limit are practically unidentifiable with current optical spectroscopic techniques . there is a wide range of x - ray to optical flux ratios , @xmath34 in these samples , and so the unidentifiable objects are not necessarily the faintest x - ray sources , and produce a significant fraction of the xrb . the nature of these optically faint , hard x - ray objects is still not well understood . they could be narrow line agn at @xmath35 , with their strongest emission lines shifted out of the optical band . a number of optically - faint _ chandra _ sources have been identified from their near ir properties to be agn located in luminous , evolved host galaxies at high redshifts ( e.g. @xcite ) . alternatively , the unidentifiable sources may be agn embedded in optically thick dusty galaxies at moderate redshifts , the faintness of the hosts precluding identification @xcite . optical studies of nearby seyfert galaxies have found that the ratio , r , of type 2 to type 1 seyferts , is approximately 4 @xcite . the hard x - ray study of these type 2 seyferts by @xcite discovered a wide distribution of absorbing columns , but with @xmath36 of the agn having @xmath37 @xmath7 . however , this study was limited to the very local universe , @xmath38 mpc , and to low nuclear luminosities , @xmath39 ; the behaviour in the rest of the redshift - luminosity plane is less well understood . the distribution of absorption in x - ray selected agn is poorly constrained , the prime difficulty being that the greater an object s @xmath1 , the lower is its chance of being detected , or optically identified . there have been several published cases of agn in which the absorbing columns inferred by optical and x - ray measurements differ significantly ( e.g. @xcite , @xcite , @xcite ) . despite these limitations , the @xmath1 distribution , for hard x - ray selected agn , has been estimated , for relatively bright samples , by @xcite . these authors have primarily determined @xmath1 in their optically identified sample by examination of x - ray hardness ratio between the 0.52 and 210 kev bands . the distribution of absorption within the sample is described with a luminosity dependent @xmath1 model , in which high intrinsic luminosity agn are less likely to be heavily absorbed . this model does require some additional compton thick agn to reproduce fully the xrb when extrapolated to harder energies . so , despite the progress made in resolving , and to some extent optically identifying , the hard x - ray population , it has still not been possible to delimit the distribution of absorption in agn . this problem is particularly acute for the heavily absorbed , high-@xmath40 agn ; few of which have been detected and identified . however , by better constraining the @xmath41 in faint agn , we can hope to answer many questions about the geometry , composition and evolution of the dusty torus . for example , the strength of the luminosity dependence of @xmath41 can tell us about how the radiation from the accretion disk influences the surrounding torus , and/or how the torus geometry scales with black hole mass . if some redshift evolution of @xmath41 is detected , is it related to the overall evolution of the agn luminosity function ? does the dusty torus form coevally with the black hole , and is the amount of absorbing material related to the black hole mass ? in this study we use x - ray hardness ratios as an indicator of absorption in the spectra of the sources in our sample . many authors ( e.g. @xcite , @xcite , and @xcite ) have shown that colour based analyses are effective in deriving the properties of xmm-_newton _ sources which are detected with too few counts to permit full spectral fitting . in these optically identified samples , the agn with and without broad emission lines are seen to occupy separate regions in x - ray colour - colour plots . we present in this paper an analysis of the x - ray properties of sources detected by xmm-_newton _ in the @xmath0deep field . in section [ section_observations ] we describe the xmm-_newton _ observations . in order to escape the possible biases introduced by the incompleteness of optical identification programmes , we have devised a method to probe the @xmath41 of our sample . our technique does not depend on optical identification of the sample , permitting the inclusion of the optically faint x - ray detections . the simulations use a model xlf to describe the intrinsic distribution of all agn in redshift and ( de - absorbed ) luminosity space ; this is coupled to a model @xmath1 function , to generate a synthetic agn population ( described in section [ section_simulation_method ] ) . we simulate how this model population would be seen with xmm-_newton _ , accounting for both the selection function caused by the complex epic detector imaging characteristics , and the nuances of the source detection process ( see section [ section_capabilities ] ) . the output products of the simulation allow direct comparison of each of the model @xmath1 distributions with the @xmath0sample . we then compare the predictions of several simple unified scheme models of the @xmath1 distribution , by using a statistical comparison of the x - ray colour distributions found in the data and models ( section [ section_colour_stats ] ) . furthermore , we test two examples of more complex @xmath1 distribution models taken from the literature , and compare them to the @xmath0sample . in section [ section_source_counts ] we compare the source counts found in the @xmath0field and those predicted by the model simulations . finally , in section [ section_discussion ] we discuss our results and their implications for agn torus models , and for the evolving xlf model of @xcite . throughout the paper we use a lambda - dominated flat cosmology with @xmath42 km s@xmath10mpc@xmath10 , @xmath43 . @xmath44 refers to an object s de - absorbed x - ray luminosity in the observed @xmath45 band . @xmath1 is the equivalent hydrogen column density in units of @xmath7 . we refer to the @xmath1 distribution function as @xmath41 , and define it to be the fraction of all agn , per unit log@xmath1 , which have absorbing column @xmath1 . we define a power - law spectrum to be @xmath46 , where @xmath47 is the flux in units of photons kev@xmath10 s@xmath10 @xmath7 , @xmath48 is the photon energy in kev , and @xmath49 is the photon index . the unified model attributes the x - ray absorption seen in agn to a dusty torus surrounding the central super massive black hole @xcite . for certain orientations , the torus obscures the observer s line of sight to the x - ray / uv emitting accretion disk . there have been a wide range of absorbing columns inferred from the x - ray spectra of various agn , ranging from effectively zero absorption , to column densities over @xmath50@xmath7 . the unified scheme states that all agn are intrinsically similar , and that the observational differences between the various agn types are due to the orientation of the observer . therefore , it is the geometry of the dust torus which determines the amount of obscuring material along the observer s line of sight to the central x - ray emitting regions . if we assume that all agn have the same geometry , then it is only the properties of the torus which determine the observed @xmath41 in the agn population as a whole . a typical zeroth order approach is to speculate that this characteristic geometry is independent of the luminosity of the central engine , and has not evolved over cosmological timescales . however , alternative scenarios are postulated , for example by @xcite and @xcite , which imply more complex forms for @xmath41 . the @xcite model suggests that some evolution of the average torus properties has occurred over cosmological timescales , with more absorbed agn at high redshifts . the @xcite model predicts that the geometry , specifically the opening angle , of the obscuring torus is determined by the luminosity of the central engine . in this study , we compare the predictions of several different forms of @xmath41 . a very simple description of @xmath41 , is a continuous distribution , in which the number of agn per unit log@xmath1 is proportional to @xmath51log@xmath52 , over the range @xmath53 log @xmath54 . a similar parameterisation was adopted in the synthesis model of @xcite , who found that setting @xmath55 or @xmath56 gave acceptable fits to the xrb ( it should be noted that the authors used a separately evolving luminosity function for absorbed agn ) . we have tested three such @xmath41 , and refer to them as the @xmath57 , @xmath58 and @xmath59 models . model a of @xcite combines the @xmath41 observed in the optically selected seyfert-2 galaxies @xcite , with a fixed ratio , @xmath60 , of absorbed to unabsorbed agn . we have tested two similar @xmath41 models here , where @xmath60 is constant and set to 4 , and 8 , and refer to these as the @xmath61 , and @xmath62 models respectively . the measured distribution of @xcite contains a number of agn where only the lower or upper limit on absorption is known . so , for the purposes of our study , all those absorbed agn having log@xmath63 are evenly distributed in the @xmath64 log@xmath63 interval , and those with log@xmath65 are set to have log@xmath66 . we also compare the predictions of model b of @xcite , which is similar to model a , above , but with @xmath61 at @xmath67 increasing to @xmath68 for @xmath69 ; we refer to this as the @xmath70 model . in addition , we test the luminosity dependent @xmath41 function of @xcite , in which high luminosity agn are more likely to have lower absorbing columns . we have converted from our observed frame @xmath71 to rest - frame @xmath72 using the specific spectral slope , and redshift of each simulated agn ; we refer to this as the @xmath73 model . the @xmath73 model distribution does not include any agn having absorbing columns outside @xmath74 log@xmath75 . finally , we employ a zero absorption scenario to provide a base - line to the more realistic models ; we call this the @xmath76 model . a subset of the model @xmath41 are shown in fig . [ nh_histo ] . to reiterate , for all the tested models , we take only the @xmath41 part from the published model , and always use the ldde1 model xlf of @xcite to describe both absorbed and unabsorbed agn . recent x - ray spectral fitting analyses have found that after absorption is considered , the mean agn photon index @xmath49 is @xmath77 ( @xcite , @xcite , @xcite ) . the absorbed and unabsorbed agn show similar mean slopes , and there is no significant evolution seen even up to @xmath78 . however , there is still an intrinsic scatter of slopes about the mean , and this will have some effect on the observed colours and/or detectability of sources . therefore , we have used a gaussian distribution of slopes @xmath79 , to represent the spectra of the simulated agn , with @xmath17 , and @xmath80 . we have not considered sources with slopes outside the range @xmath81 . @xcite found no apparent dependence of de - absorbed @xmath49 on @xmath40 , @xmath1 , or flux , therefore we assume that @xmath79 is independent of the other agn spectral parameters . are shown in the leftmost bin , and those with log @xmath65 in the rightmost bin.,width=302 ] the xmm-_newton _ data consist of three observations of the @xmath0field totalling 200 ks , of which @xmath82ks is unaffected by soft proton flaring . this field was the location of one of the deepest _ rosat _ surveys @xcite , due to its unusually low galactic absorbing column ( @xmath83@xmath7 ) . in addition , the @xmath0field has been the subject of a host of multi - wavelength observations , including a mosaic of four 30ks _ chandra _ pointings covering the xmm-_newton _ field of view @xcite , and extensive , very deep radio mapping with the vla @xcite . the data from the european photon imaging cameras ( epic ) were reduced with version 6.0 of the science analysis software ( sas ) task - set to produce images in four energy bands ( 0.20.5 , 0.52 , 25 , and 510 kev ) . our source detection process uses the standard sas tasks eboxdetect and emldetect together with a custom background fitting task . we perform the source detection routines on the combined data from all three ( mos1 , mos2 and pn ) epic detectors . we used simulations to determine detection likelihood limits such that we expect only 3% of the final sourcelist to be spurious detections . using these likelihood limits we detected 225 sources . the approximate limiting fluxes ( in units of @xmath84 erg s@xmath10 @xmath11 ) , are 0.5 ( 0.20.5 kev ) , 0.5 ( 0.52 kev ) , 1.2 ( 25 kev ) , and 5 ( 510 kev ) . in the 25 kev band our sample reaches a factor @xmath85 fainter than the knee of the source counts , where the contribution to the xrb , per unit log flux , is greatest . a full description of the data reduction , source searching , and detection threshold determination processes is given in @xcite . the purpose of this study is to test the predictions of a number of model agn populations against agn in the @xmath0field . however , we do expect to find a small number of non - agn sources in the full x - ray sample , and these could affect our comparisons . our ongoing optical spectroscopic follow - up program has identified counterparts to over 100 of the xmm-_newton _ sources . in particular , the brightest @xmath86 optical counterparts are @xmath87 identified . four of the sources , including the brightest source in the field , are associated with foreground stars , and therefore are not included in this study . we do not expect many of the remaining optically faint , @xmath88 , counterparts to be identified with stars . emldetect finds four x - ray sources with high likelihood of being extended , and with fwhm @xmath89 . of these four sources , three were identified as clusters by @xcite in an analysis of the _ rosat _ imaging of the @xmath0field . @xcite found two additional clusters in the @xmath0field , however , these are both located at the very edge of the epic field of view , ( where the vignetting is most pronounced ) , and are not detected as extended sources by emldetect . our agn population models do not include clusters of galaxies , so we discount the four extended sources from our analysis . a small number of fainter clusters are expected to remain in the sample , but not be flagged as extended by emldetect . we estimate this number by extrapolating the @xmath90 plot of @xcite to lower fluxes , whilst incorporating the effective area determination of our survey @xcite . assuming that the flux limit for detecting extended sources is twice that for point sources , and using a @xmath90 slope of @xmath91 , we predict that approximately five additional clusters will remain in the sample . after the stars and obvious clusters are removed , the resulting xmm-_newton _ sample contains a total of 217 sources of which the vast majority are likely to be agn . we have devised a monte carlo simulation technique which allows direct comparison of the pattern of x - ray colours produced by agn absorption models , with the pattern seen in the @xmath0sample . for this , we have extended the xmm-_newton _ imaging simulation method of @xcite , to a multi - band approach . we model the epic point spread function , vignetting and diffuse background in the same way as before . this method accounts for observational biases and the complex selection function at work in the sample . each iteration of the simulation method consists of four steps . i ) we generate an input source population , with each member having a set of randomly distributed parameters @xmath92 and @xmath49 , from the xlf , @xmath41 , and @xmath79 models . ii ) multi - band count rates are calculated for each simulated source according to , @xmath93 , and the chosen spectral model . iii ) random source positions are assigned , and the sourcelist is folded through a model of the epic imaging response to create multi - band images for each epic detector . iv ) a source - detection chain is carried out on the resulting multi - band images to create an output sourcelist . we repeat these steps for 100 simulated fields for each @xmath41 model , and for two agn spectral models . simulated images are produced separately for the mos1 , mos2 and pn cameras , then combined to produce a single image in each of the four energy bands for source searching purposes . to generate the catalogue for the @xmath0field , @xcite used model @xmath94 curves to represent the agn population independently in each of four energy bands . while valid for monochromatic studies , this technique is not suitable for colour analyses , since it takes no account of the multi - band properties of individual sources . we assume that there exists a single _ intrinsic _ xlf which describes all agn , which is modified by some distribution of absorption to produce the observed xrb , source counts and source colours . of the various models for the soft xlf ( e.g. @xcite , @xcite , @xcite , @xcite ) , we have chosen to use the luminosity dependent density evolution ( ldde1 ) xlf model of @xcite . this was primarily because it is based on a large sample of agn , and its model parameters have been determined for the currently preferred lambda - dominated cosmology . the sample used to fit this xlf model contains a mixture of agn both with , and without , broad lines , suggesting that it contains a subset of absorbed agn . we adopt the best fitting parameter values presented in table 3 of @xcite , and where appropriate , have corrected for the @xmath42 km s@xmath10 mpc@xmath10 used in this study . we integrate the xlf over the range @xmath95 log@xmath96 , @xmath97 to calculate the total number of agn expected in the field . a 1d cumulative probability distribution is generated by integrating the 2d xlf via an arbitrary path in @xmath40 , @xmath71 space . it is then possible to build a list of agn which are randomly distributed in @xmath40 , @xmath71 according to the model xlf . each of these agn are assigned a random value of @xmath1 according to the @xmath41 model being tested , and a spectral slope taken from @xmath79 . the absolute normalisation of the xlf is iteratively adjusted , so that the simulated fields contain the same source counts as the @xmath0sample at @xmath98 erg s@xmath10 @xmath11 . we determine the x - ray colours of the simulated agn by using a simple absorbed power - law ( apl ) model , which also includes a correction for the small galactic absorbing column ( @xmath99@xmath7 ) found in the @xmath0field . in order to compare simulated images with the observations , we must convert from the simulated agn parameters to multi - band epic count rates . we use the spectral fitting package xspec to generate fake spectra , incorporating both the instrumental response ( for the mos1 , mos2 and pn cameras ) , and the agn parameters @xmath40 , @xmath1 and @xmath49 . these spectra are summed over the appropriate energy bands , to derive the relevant conversion factors . the cost in processing time would be prohibitive if we were to individually recalculate these conversion factors for each of the thousands of simulated agn . hence , we have built lookup tables of conversion ratios , which finely sample @xmath100 parameter space , covering the range @xmath101 , @xmath102 log@xmath103 , @xmath81 . the conversion ratios are calculated for a single luminosity , but then scaled according to the luminosity of each simulated agn . these tables are used to convert rapidly from any set of simulated agn parameters @xmath93 to count rates , for each epic detector , and energy band . we have also examined the effect of including a small reflection component in the spectral model . this has the net effect of hardening the spectrum at higher energies , making simulated agn slightly more detectable above 5 kev . we use the pexrav model of @xcite , with the reflecting material covering @xmath104 steradians , a viewing angle of @xmath105 , and solar abundances . we call this the apl+r spectral model . it is beyond the scope of this study to include more complex agn spectral features , such as @xmath106 lines , or scattered soft x - ray emission . we expect the effect of these features on agn colours to be small relative to the effects of continuum obscuration . however , some of our results suggest that a number of detected sources in the @xmath0field have an additional soft component , as discussed later . we expect to be able to detect very few ( if any ) very heavily absorbed agn having log@xmath65 , and so have not included such objects in the simulated populations . in fact , the simulations show us that we expect agn having log@xmath107 to account for only @xmath108 of the detections in the @xmath0sample . therefore , any additional attenuation due to compton scattering within the dusty torus is ignored , since it has little effect for agn with log@xmath75 . the simulation method incorporates the effects of the epic response function , effective area , point spread function ( psf ) , vignetting , and background to produce multi - band images . we use the energy and off - axis angle dependent `` medium '' accuracy psf model , taken from the xmm-_newton _ calibration library . this psf model has been measured to be accurate to better than @xmath109 at @xmath110kev @xcite . the effective exposure time and vignetting are calculated from the sas generated exposure maps for the @xmath0field . a synthetic background is added to the simulated images to reproduce the level observed in the observations . the correct level of this additional background was determined through an iterative process to account for the contribution from the faint unresolved simulated sources . we use the simultaneous , multi - band source detection process on the combined simulated mos1+mos2+pn images in the same fashion as described for the @xmath0data @xcite . however , only one iteration of the background determination process is carried out , in order to conserve computation time . we have searched for sources over the entire useful field of view of the combined epic detectors , giving a total sky area of 0.185 deg@xmath111 . the inherent capabilities and limitations of the @xmath0survey data can be precisely evaluated using our simulation method . in this section we refer to sources found in the simulated images by emldetect as `` output '' sources . we have employed a simple algorithm , that for each output source , associates an input source , in order that the colours of the output sources can be related to the input parameters ( @xmath40 , @xmath112 , @xmath1 , and @xmath49 ) . for the majority of output sources , there is a single nearby input candidate , which we consider to be the progenitor . however , in some cases , an output source can have several nearby input candidate sources . this problem is exacerbated at high off - axis angles because of the degradation of the psf ( and consequently , the precision of positions reported by emldetect ) . therefore , we have employed a simple algorithm that matches each output source to the brightest input source within a small radius , @xmath113 of the detected position . we make @xmath113 dependent on off - axis angle by setting it to 5 , 8and 10 for off - axis angles @xmath1149 , 9 - 12 , and @xmath11512 respectively . any output sources with no input candidates within @xmath113 are almost certainly caused by poissonian background fluctuations , and so are not considered in this section . however , we expect a small number of the detections in the @xmath0sample to be caused by this phenomenon , and hence unrelated to any real x - ray source . therefore , in section [ section_results ] , we do include those output simulated sources having no input candidates when comparing the simulated agn populations to the sample . to determine the selection function of the simulated agn , we evaluate the fraction of matched output / input sources , as a function of the input parameters @xmath116 , and @xmath49 . fig . [ det_frac_zl ] shows the 50% completeness limit , as a function of redshift and luminosity , for several different levels of absorption . the contours show the loci in @xmath117 parameter space , at which half of the input sources are detected . there is a clear reduction in detection probability for agn having absorbing columns above @xmath118 @xmath7 , and this effect is more marked at low-@xmath40 . the plot shows that the @xmath0survey is able to detect the majority of moderately luminous agn ( log@xmath119 ) , with moderate - absorption ( @xmath120log@xmath121 ) up to @xmath122 . and @xmath71 for the absorbed power - law spectral model . the results for four ranges of absorption are shown : @xmath123 log @xmath124 ( solid ) , @xmath125 log@xmath126 ( long dash ) , @xmath127 log@xmath128 ( short dash ) , and @xmath129 log@xmath130 ( dotted ) . , width=302 ] fig . [ nh_det_frac ] shows the fraction , as a function of absorption , of all input sources that are matched to output sources in the simulated images . this highlights the small differences in detection probability between the two spectral models . the addition of a reflection component in the agn spectra has a rather small effect on the detectability of simulated sources . ) , as a function of @xmath1 , normalised to the output / input fraction for log@xmath131 sources . results are shown for the apl spectral model ( solid line ) , and the apl+r model ( dashed line ) . the plot is compiled from the simulations for the @xmath59 @xmath41 model.,width=302 ] ) , as a function of @xmath49 , normalised to the output / input fraction for @xmath19 sources . results are shown for the apl spectral model ( solid line ) , and the apl+r model ( dashed line ) . the plot is compiled from the simulations for the @xmath59 @xmath41 model . the inset shows the number of output sources , per simulated field , as a function of spectral slope . , width=340 ] the dependence of the selection function on @xmath49 can be seen in fig . [ gamma_det_frac ] , which shows the fraction of simulated input sources with output counterparts , as a function of spectral slope . it can be seen that the spectral slope of an agn has a small but measurable bearing on its probability of detection . a strong increase in detection probability is seen for very hard sources ( @xmath132 ) , however , the inset histogram shows that very few of these objects are predicted by the @xmath79 model . we have used the 0.52 kev de - absorbed flux to normalise the model spectra , and so the hard - slope agn have a relatively high countrate above 2 kev , and are more likely to be detected . this effect is larger for moderate to heavily absorbed sources , since they are primarily detected at these harder energies . the impact on the overall selection function is largest for the @xmath41 models containing the largest fraction of absorbed sources , i.e. the @xmath62 , @xmath70 models . constraints on @xmath41 models can be made from analysis of x - ray colour ( i.e hardness ratio ) distributions . for example , @xcite compared the @xmath1 of xmm-_newton _ sources determined from full spectral fits , with the @xmath1 estimated using a hardness ratio method ( over the 0.510 kev range ) , and showed that they were in good agreement for log@xmath32 . for this study we define the hardness ratios @xmath133 , @xmath134 and @xmath135 , where @xmath136 is the source count rate , corrected for vignetting , in the given energy band . the corresponding measurement errors are denoted by @xmath137 , @xmath138 , @xmath139 respectively . the count rates , hardness ratios , and errors are computed within emldetect using the combined dataset from the mos1 , mos2 and pn cameras . if any hardness ratio measurement is undetermined ( zero countrates in two energy bands ) , we set it to @xmath140 . the dependence of @xmath141 , @xmath142 and @xmath143 on absorption is illustrated in fig . [ colour_nh_ranges ] , which shows the measured colour - colour distributions of `` output '' simulated sources grouped into a number of @xmath1 bins . for each @xmath1 bin , we have over - plotted the `` perfect '' @xmath40-track in colour - colour space for an agn with mid - bin absorption , @xmath19 , and @xmath144 . both the width of the @xmath1 bins , and the range of @xmath49 in the simulated sources , act to distribute sources about this track . the relative density of the distribution along the track is mostly determined by the evolution of the xlf , which peaks above @xmath145 . in addition , a significant amount of scatter is caused by measurement uncertainties within the source detection process , particularly for the faintest sources . we see from the left - most upper panel of fig . [ colour_nh_ranges ] , that the colour distribution of output simulated sources with log@xmath146 is compact , and approximately centered on ( @xmath147 ) = ( 0.2,-0.5 ) . the study of xmm-_newton _ sources in the lockman hole by @xcite , found that the vast majority of agn in this part of hardness ratio space had broad line optical counterparts and , at most , weak absorption ( log@xmath148 ) in their x - ray spectra . in contrast most of the identified agn having hard x - ray colours had narrow line optical counterparts , although only a small fraction of the hard sources had optical identifications . examination of the three upper right panels reveals that the moderately to heavily absorbed sources ( log@xmath149 ) , occupy a measurably different region of @xmath141,@xmath142 space compared to their less absorbed counterparts . in particular , @xmath141 is sensitive to absorption in the range @xmath125log@xmath150 , and @xmath142 to absorption above log@xmath151 . in the study of @xcite , the hardness ratio between the 0.52 and 28 kev bands did not appear to separate the broad and narrow line agn ; however , relatively few of the harder agn in this sample had spectroscopic identifications . @xcite showed that the majority of agn with broad line counterparts fall in the range @xmath152 , consistent with the location of the low absorption agn ( log@xmath148 ) produced by our simulations . as we would expect , the majority of the simulated faint unabsorbed agn do not have good measurements of @xmath143 . these sources have noise dominated countrate measurements above 2 kev , and hence have @xmath143 measurements randomly scattered in the interval @xmath153 $ ] . of the simulated agn having log@xmath154 , it is only the most luminous ( @xmath155 erg s@xmath10 ) , that are detectable in our survey , as shown in figs . [ det_frac_zl ] and [ nh_det_frac ] . the bottom right hand panel illustrates that @xmath143 is sensitive to absorption above log@xmath156 for all but the highest redshift agn . hard band x - ray count rates were well determined for sources in the bright sample of @xcite , and of the four objects with a higher count rate in the 4.57.5 kev band than in the 24.5 kev band , three were associated with narrow line optical counterparts , and one with a seyfert 1.9 galaxy . figure [ colours_sims_models ] shows the @xmath141 _ vs. _ @xmath142 ( upper row ) and @xmath142 _ vs. _ @xmath143 ( lower row ) distributions produced by three of the @xmath41 models . the most immediately noticeable difference between the plots , is the fraction of sources that appear to the right of @xmath157 for the various @xmath41 models . sample . the levels of the contours are set such that they include @xmath5 ( short dash ) , @xmath158 ( long dash ) , and @xmath8 ( solid line ) of the smoothed source distribution , and were generated in the same way as for fig . [ colour_nh_ranges ] . typical sizes of @xmath137 , @xmath138 , and @xmath139 are shown with boxes , for sources having `` two - band '' fluxes of @xmath159 and @xmath160 erg s@xmath10 @xmath11 , where the `` two - band '' flux is the flux measured over the two energy bands used to calculate the hardness ratio.,title="fig:",width=302 ] sample . the levels of the contours are set such that they include @xmath5 ( short dash ) , @xmath158 ( long dash ) , and @xmath8 ( solid line ) of the smoothed source distribution , and were generated in the same way as for fig . [ colour_nh_ranges ] . typical sizes of @xmath137 , @xmath138 , and @xmath139 are shown with boxes , for sources having `` two - band '' fluxes of @xmath159 and @xmath160 erg s@xmath10 @xmath11 , where the `` two - band '' flux is the flux measured over the two energy bands used to calculate the hardness ratio.,title="fig:",width=302 ] the two left - most panels of fig . [ colours_sims_models ] show the colour - colour distributions of the @xmath0sample , with grey - scale and contours generated in the same way as for fig [ colour_nh_ranges ] . [ colours_13hr ] shows the same contours , but with the individual data points overlaid [ colours_sims_models ] and [ colours_13hr ] show that there is a strong concentration of sources in the @xmath161 = ( 0.4,-0.5,-0.5 ) region , slightly harder in @xmath141 than the nominal position of an unabsorbed agn with log@xmath162 , @xmath19 . a large number of sources have much harder values of @xmath141 and @xmath142 than the nominal unabsorbed position , indicating that strong absorption is present in a significant fraction of the population . however , the majority of the sources in the @xmath163 , @xmath164 region are actually faint soft sources having large @xmath143 measurement uncertainties . the bimodality apparent either side of @xmath157 is probably due to the fast increase in @xmath141 over the range @xmath165 log@xmath166 , which limits the number of sources in this region . a similarly sparse region occurs at @xmath167 and again , this is probably related to the fast increase in @xmath142 over the range @xmath168 log@xmath169 . curves measured in the @xmath0sample ( filled circles ) , compared to those produced by the @xmath170 ( solid line ) , @xmath59 ( long dash ) , @xmath61 ( short dash ) , @xmath70 ( dotted ) , and @xmath171 ( dot - dash ) , @xmath41 models . these results are for the absorbed power - law spectral model and are normalised to the euclidean - slope . the equivalent points for the sample of @xcite ( taken from fig . 6 ) , are also shown ( open symbols with errorbars ) , and have been normalised assuming a sky area of 0.185 deg@xmath111 . , width=340 ] we have compared the 0.52 kev band integral source counts , @xmath90 , measured in the @xmath0sample with those produced by the simulated model distributions . we make no correction for sky coverage , since the @xmath0sample and the simulated fields have an identical survey - depth / sky area relation . we find a large disparity between the @xmath0and simulated fields , especially around the knee of the observed @xmath172 at @xmath173 erg s@xmath10 @xmath11 , as can be seen in fig . [ source_counts_soft ] . each of the @xmath41 models produced similar @xmath90 curves , especially at faint fluxes , where the statistical errors are better . thus we deduce that the data - model disparity is primarily caused by differences between the data and the model xlf ( and/or its evolution ) . we discuss this later . the primary purpose of this study is to compare the @xmath41 models , so it is important that we minimise the effect on the statistical analysis caused by the disparity between the data and xlf / evolution model . therefore , we have examined the x - ray colour distribution of sources , rather than the distributions of their absolute fluxes . we expect the colour - colour distributions to be more sensitive to @xmath41 than to the xlf , because a small change of the position of an absorbed agn in the @xmath40-@xmath112 plane , has a strong effect on its overall brightness , but only a small effect on its x - ray colours . for example , if the peak of agn space density is actually at @xmath174 , ( rather than at @xmath175 as predicted by the xlf / evolution model ) , then the resulting change in hardness ratios for an agn , having an absorbed power - law spectrum , log@xmath176 , at this peak redshift , would be @xmath177 = + 0.07(+0.03 ) . however , an increase of 0.5 dex in the absorption of the same agn , would result in @xmath177 = + 0.23(+0.17 ) . therefore , a colour analysis of the @xmath41 models is more strongly dependent on the tested @xmath41 , than on differences between the data and xlf / evolution model . we have used the kolmogorov - smirnov test ( ks ) , to determine how well the simulated data reproduce the x - ray colour distribution measured in the @xmath0sample . the ks test has the advantage that it requires no rebinning of data , utilising the full information content of the data set . however , it does not take into account the relative errors on data points , meaning that low signal to noise measurements can , to some extent , `` wash out '' the signal from the more precise measurements . a three dimensional extension of the ks test ( 3d - ks ) , as devised by @xcite , was used to compare the sample with the simulation results in the full ( hr1,hr2,hr3 ) variable space . in order to examine more closely how the models reproduce the sample distribution , we have carried out one dimensional ks tests separately on @xmath141 , @xmath142 and @xmath143 . the conversion from the 3d - ks test statistic , @xmath178 , to the probability that two samples were taken from the same underlying population , @xmath179 , is strongly dependent on the number of sources in the tested samples , and the degree of correlation between the tested variables . @xcite numerically generated lookup tables to allow this conversion at a number of confidence levels , for a range of sample sizes , and values of the correlation parameter . however , these tables give only a relatively small number of conversion values , at discrete confidence levels , sample sizes , and values of the correlation parameter . therefore , we have run a set of simulations , the results of which permit conversion from @xmath178 directly into @xmath179 conversion using the precise sample sizes and correlations seen in the @xmath0sample . we calculated the three - dimensional probability density map ( 3d - pdm ) , of the @xmath0sample in ( hr1,hr2,hr3 ) space . the contribution from each source to the 3d - pdm is calculated from a 3d - gaussian that has widths equivalent to @xmath137,@xmath138 , and @xmath139 . the normalisation of the 3d - gaussian is set such that the total contribution of each source is unity . this 3d - pdm is used to generate pairs of random populations , having 217 and 25000 members respectively , for which @xmath178 is calculated . the latter step is repeated for 100000 iterations . the equivalent probability for any particular value of the 3d - ks statistic , is equal to the fraction of these iterations having @xmath178 greater than this value . the absolute lower limit at which we can evaluate the probability is given by the reciprocal of the number of simulation iterations , i.e. 0.001% , although the errors are large at this level . this limit is determined by the processing time available . table [ ks_results ] shows @xmath179 , for the eight @xmath41 models , and for both of the tested spectral models . in order to determine where the biggest differences arise between the data and models , we have calculated the ks probabilities ( @xmath180 ) , separately for each of @xmath141 , @xmath142 and @xmath143 , the results of which are also shown in table [ ks_results ] . [ cols= " < , > , > , > , > , > , > , > , > " , ] for the simplest toy model of a torus with uniformly density , and a typical opening angle , @xmath181 , the fraction of agn that are heavily absorbed is approximately @xmath182 . so , if we use the size of the `` hard '' fraction of the @xmath0sample as a measure of the number of absorbed agn , we can infer a rather wide opening angle of @xmath183 . however , this estimate does not take into account the effect of the drop in the selection function toward high @xmath1 , and can only be seen as an upper limit on @xmath181 . we estimate the relative selection function for hard sources by counting the fraction of simulated `` hard '' input sources that have output counterparts relative to that for all input sources . applying this correction to the @xmath0sample , we predict an _ intrinsic _ `` hard '' fraction of @xmath184 , implying an opening angle of @xmath185 . if in our correction for the relative selection function , we exclude those sources with absorbing column above log@xmath186 , where our sample constrains the models only weakly , then we find @xmath187 . we are also able to examine the range of torus parameters that would best match the @xmath41 models . for the best fitting @xmath188 model , where the fraction of input sources with log@xmath32 is @xmath189 , the predicted opening angle is @xmath190 . as fig . [ colour_nh_ranges ] shows , @xmath141 and @xmath142 are sensitive to the shape of the distribution over a wide range of @xmath1 , particularly for intermediately absorbed sources . we have seen that the @xmath188 is strongly favoured over the @xmath61 @xmath41 model ( see table [ ks_results ] ) . these two models are very similar in the range @xmath191 log@xmath75 , contain similar numbers of unabsorbed agn ( log @xmath162 ) , and produce comparable numbers of `` hard '' sources . therefore , the difference must lie primarily in the @xmath64 log@xmath63 range , in which the @xmath188 model contains many more agn . a major problem with the uniformly dense torus model is that it predicts that nearly all agn will be either heavily absorbed or completely unabsorbed . however , more complex models , incorporating a wide distribution of torus densities , predict larger numbers of intermediately absorbed agn . for example , a model in which the density falls off exponentially with angle away from the plane of the torus , predicts a much flatter @xmath41 ( e.g. @xcite ) . it is possible , with some tuning of such a model s parameters , to approximately match the best fitting @xmath188 distribution . since absorption in the @xmath64 log@xmath63 range has only a significant effect on @xmath141 , it would not have been detectable in the colour distributions if the @xmath192 kev band had not been considered . a number of studies of absorption in faint agn have based their estimates of @xmath1 on hardness ratios between the @xmath193 and @xmath194 kev bands , and therefore may have underestimated the number of intermediately absorbed agn ( e.g. @xcite,@xcite ) . a better determination of mean torus properties will be possible when the @xmath0field is covered by _ spitzer _ , and we are able to correlate x - ray colours with mid / far - ir data . each of the simulated @xmath41 models produced similar 0.52.0 kev source count - flux relations , @xmath90 . however , these are seen to reproduce poorly the @xmath90 relation observed in the @xmath0sample ( see fig . [ source_counts_soft ] ) . the models under - produce the @xmath90 above the normalisation flux ( @xmath195 erg s@xmath10 @xmath11 ) , and over - produce the @xmath90 below this flux ( see fig . [ source_counts_soft ] ) . in fact , at @xmath160 erg s@xmath10 @xmath11the models under - produce the source counts seen in the @xmath0sample by a factor of about two . this disparity is seen to a similar degree in each of the @xmath41 models , suggesting that it is related to the difference between the data and xlf / evolution model . the @xmath90 of the @xcite sample , is also shown in fig . [ source_counts_soft ] , plotted assuming our field has a uniform sky area of 0.185 deg@xmath111 . this illustrates that in the flux range @xmath196 erg s@xmath10 @xmath11 , the ldde1 xlf also under - produces the source counts of the sample from which it was derived . the shape of the @xmath90 relation of the @xcite sample is closer to that seen in the @xmath0sample than to the models . the faintest agn in the @xcite sample are from the deepest _ rosat _ observations of the lockman hole field , where the flux limit of the data was @xmath197 erg s@xmath10 @xmath11 . our significantly deeper flux limit means that we are using part of @xmath198 space outside that constrained by the sample of @xcite . a previous comparison of source counts from _ rosat _ observations in the lockman hole and @xmath0fields , revealed a @xmath199 over - abundance near @xmath200 erg s@xmath10 @xmath11 in the @xmath0field with respect to the lockman hole @xcite . in addition , @xcite found that the @xmath0field is slightly over - dense in the 0.52 kev band with respect to both of the chandra deep fields . therefore we conclude that the differences between model and sample are caused by a combination of these factors . in particular , our extrapolation of the ldde1 xlf / evolution model to faint fluxes , suggests that this complex scheme requires some revision . the shape of the xlf at high redshift is poorly known because of the difficulties in obtaining a large spectroscopically identified sample of these objects . we can use the simulated source population to make predictions about the number of high-@xmath40 agn in the @xmath0sample . each of the @xmath41 models predict that around 16% of the total number of x - ray detections are due to agn with @xmath201 . therefore , it can be inferred that the fraction of agn with @xmath3 in the x - ray population is primarily dependent on the shape of the underlying xlf and its evolution , rather than the @xmath1 distribution within the high-@xmath40 population . the model predictions suggest that we should expect around 35 high-@xmath40 agn in the @xmath0field . however , only a single agn has been identified having @xmath3 by our follow up optical spectroscopy program ( which has secure ids for over 100 sources ) . this disparity is maybe due to the over - production of faint sources by the xlf / evolution model ; these are more likely to be at high @xmath40 . the x - ray detection probability of the @xmath3 agn is much less dependent on @xmath1 than for the low-@xmath40 agn , since most absorption is redshifted below 2 kev . therefore , most of the @xmath41 models predict that absorbed agn make up the majority of the _ detected _ high-@xmath40 population , the precise fraction being dependent on the particular @xmath41 model . however , the absorption of optical and uv spectral features does severely affect the probability of identification for these objects . we have recently obtained further deep optical imaging of the @xmath0field in several bands , which will permit us to make photometric redshift estimates for some of the optically faint sources . the forthcoming deep coverage of the @xmath0field in the infrared with _ spitzer _ will further constrain the nature of the high-@xmath40 population . we have demonstrated how a colour - based analysis of deep xmm-_newton _ data can be used to constrain models of absorption in the agn population without requiring complete optical spectroscopic follow up . by using a detailed simulation technique , we have been able to take account of the complex selection function at work in the sample , and demonstrate how this modulates the input population . we have shown that a simple @xmath41 model together with an absorbed power - law spectral model ( including a reflection component ) , reproduces the observed @xmath141/@xmath142/@xmath143 colour distribution with probability 6% . all of the other model @xmath1 distributions that we compared were rejected at greater than 99.5% probablity . in particular , two more complex @xmath41 models are strongly rejected by the 3d - ks test ; the redshift dependent @xmath70 model produces too many hard sources , and the luminosity dependent @xmath171 model produces too few . in general , the addition of a reflection component to the absorbed power - law spectral model improved the match between the colour distributions of the models and the sample . the reflection component serves to harden the spectral slope at higher energies , and its effect was most evident in the @xmath142 and @xmath143 distributions . we have shown that there is a large disparity between the shape of the @xmath90 produced by the models and that found in our sample . we suggest that this is for the most part due to differences between the actual @xmath198 distribution of sources in the @xmath0field , and the xlf / evolution model that we have used . these xlf / evolution differences will have had some effect on the colour distributions produced by the @xmath41 models , and could explain the surfeit of @xmath202 sources produced by all of the @xmath41 models . we have seen some evidence that suggests that the spectra of a significant fraction of absorbed sources in the @xmath0sample have an additional soft x - ray component . this feature was not included in our spectral models , and therefore contributed a large part of the disparity between the @xmath141 distributions of models and sample . considering this factor , together with the xlf / evolution differences , we conclude that the 6% probability for the @xmath59 model shows that it provides a rather good fit to the data . the shape of the @xmath188 distribution can be broadly reproduced using a toy model for the torus in which the density falls away rapidly for viewing angles away from the plane of the torus . we have shown that agn having log@xmath32 can be efficiently selected by choosing sources in the regions @xmath203 and @xmath204 . we intend to extend the methods described here to further xmm-_newton _ deep fields in order to increase the sample size , and to reach to fainter x - ray fluxes . based on observations obtained with xmm - newton , an esa science mission with instruments and contributions directly funded by esa member states and nasa . td acknowledges the support of a pparc studentship . 99 alexander d.m . , bauer f.e . , et al . , 2003 , apj , 126 , 539 antonucci r. , 1993 , annu.rev . a&a , 31 , 473 barger a. , cowie l. , et al . 2002 , apj , 124 , 1839 barger a. , cowie l. , et al . 2003 , apj , 126 , 632 boyle b.j . , griffiths r.e . , et al . 1993 , mnras , 260 , 49 boyle b.j . , shanks t. , et al . 1994 , mnras , 271 , 639 brandt w.n . , alexander d.m . , et al . , 2001 , 122 , 2910 caccianiga a. , severgnini p. , et al . 2004 , a&a , 416 , 901 carrera f.j . , page m.j . , mittaz j.p.d . , 2004 , a&a , 420 , 163 comastri a. , setti g. , et al . , 1995 , a&a , 296 , 1 cowie l.l . , barger a.j . , et al . , 2001 , apj , 551 , l9 cowie l.l . , barger a.j . , et al . , 2003 , apj , 584 , l57 della ceca r. , maccacaro t. , et al . , 2004 , a&a , 428 , 383 de luca a. , molendi s. , 2004 , a&a , 419 , 837 fabian a.c . , barcons x. , et al . , 1998 , mnras , 297 , l11 fasano g. , franceschini a. , 1987 , mnras , 225 , 155 franceschini a. , braito v. , fadda d. , 2002 , mnras , 335 , l51 georgantopoulos i. , georgakakis a. , et al . , 2004 , mnras , 352 , 91 gandhi , p. , fabian , a. , 2003 , mnras , 339 , 1095 giacconi r. , zirm a. , et al . , 2002 , apjs , 139 , 369 gilli r. , salvati m. , hasinger g. , 2001 , a&a , 366 , 407 gondoin p. , xmm-_newton _ ccf release note , xmm - ccf - rel-2 , 2000 hasinger g. , burg r. , et al . , 1998 , a&a , 329 , 482 hasinger g. , altieri b. , et al . , 2001 , a&a , 365 , l45 jones l.r . , mchardy i.m . , et al . , 1997 , mnras , 285 , 547 jones l.r . , mchardy i.m . , et al . , 2002 , mnras , 334 , 219 loaring n.s . , page m.j . , ramsay g. , 2003 , mnras , 345 , 865 loaring n.s . , dwelly t. , et al . , submitted . lumb d.h . , warwick r.s . , et al . , 2002 , a&a , 389 , 93 maccacaro t. , della ceca r. , et al . , 1991 , apj , 374 , 117 madau p. , ghisellini g. , and fabian a.c . , 1994 , mnras , 270 , l17 mchardy i.m . , jones l.r , et al , 1998 , mnras , 295 , 641 mchardy i.m . , gunn , k.f . , et al . , 2003 , mnras , 342 , 802 magdziarz p. , zdziarski a. a. , 1995 , mnras , 273 , 837 mainieri v. , bergeron j. , et al . , 2002 , a&a , 393 , 425 maiolino r. , rieke g.h . , 1995 , apj , 454 , 95 maiolino r. , marconi a. , et al . 2001 , a&a , 365 , 28 mateos s. , barcons x. , et al . , 2005 , a&a 433 , 855 miyaji t. , ishisaki y. , et al . , 1998 , a&a , 334 , l13 miyaji t. , hasinger g. , schmidt m. , 2000 , a&a , 353 , 25 moretti a. , campana s. , et al . , 2003 , apj , 588 , 696 page m.j . , mason k.o . , et al . , 1997 , mnras , 291 , 324 page m.j . , mittaz , p.d . , carrera f.j . , 2001 , mnras , 325 , 575 page m.j . , mchardy i.m . , et al . , 2003 , astron . nachr , 324 , 101 page m.j . , lehmann i. , et al . , 2005 , in prep . perola g.c . , puccetti s. , et al . , 2004 , a&a , 421 , 491 piconcelli e. , cappi m. , et al . , 2003 , a&a , 412 , 689 risaliti g. , maiolino r. , salvati m. , 1999 , apj , 522 , 157 rosati p. , giacconi r. , et al . , 2002 , apj , 556 , 667 severgnini p. , caccianiga , a. , et al . , 2003 , a&a , 406 , 483 seymour n. , mchardy i.m . , gunn k. , 2004 , mnras , 352 , 131 treister e. , urry c.m . , et al . , 2004 , apj 616 , 123 ueda y. , akiyama m. , et al . , 2003 , apj , 598 , 886 wolf c. , meisenheimer k. , et al . , 2004 , a&a , 421 , 913

we study how well magnetic models can be implemented with ultracold bosonic atoms of two different hyperfine states in an optical superlattice . the system is captured by a two - species bose - hubbard model , but realizes in a certain parameter regime actually the physics of a spin-1/2 heisenberg magnet , describing the second order hopping processes . tuning of the superlattice allows for controlling the effect of fast first order processes versus the slower second order ones . using the density - matrix renormalization - group method , we provide the evolution of typical experimentally available observables . the validity of the description via the heisenberg model , depending on the parameters of the hubbard model , is studied numerically and analytically . the analysis is also motivated by recent experiments [ s. flling _ et al . _ , nature * 448 * , 1029 ( 2007 ) ; s. trotzky _ et al . _ , sience * 319 * , 295 ( 2008 ) ] where coherent two - particle dynamics with ultracold bosonic atoms in isolated double wells were realized . we provide theoretical background for the next step , the observation of coherent many - particle dynamics after coupling the double wells . contrary to the case of isolated double wells , relaxation of local observables can be observed . the tunability between the bose - hubbard model and the heisenberg model in this setup could be used to study experimentally the differences in equilibration processes for nonintegrable and bethe ansatz integrable models . we show that the relaxation in the heisenberg model is connected to a phase averaging effect , which is in contrast to the typical scattering driven thermalization in nonintegrable models . we discuss the preparation of magnetic groundstates by adiabatic tuning of the superlattice parameters .

0809.5141

one of the most exciting recent events in physics has been the increasing overlap between two previously disjoint fields , quantum optics and condensed matter physics . this has become possible due to the enormous progress in cooling dilute bosonic and also fermionic gases down to temperatures where respectively bose - einstein condensation and fermi degeneracy ( temperatures well below the fermi energy ) are reached . a very attractive feature of this new class of experiments is that they provide the arguably cleanest realization of the ( bosonic ) hubbard model @xcite , which with nearest - neighbor hopping and onsite interaction is the minimal model of strong correlation physics @xcite . here , we describe and analyze numerically a particular setup with ultracold bosons of two species in an optical superlattice , described by a bose - hubbard model . in the limit of strong onsite interactions , the system can be described by the spin-1/2 heisenberg antiferro- or ferromagnet , depending on the parameters of the superlattice . the motivation is fourfold : \(i ) in the vein of feynman s idea to simulate quantum systems by other quantum systems @xcite , it would be a great achievement to implement models of magnets like the heisenberg model with ultracold atoms in optical lattices . in condensed matter systems , collective magnetism arises from the coulomb interaction and the particle statistics which cause ( super)exchange processes @xcite . in particular , exchange interactions resulting from second order hopping processes in the fermi - hubbard model dominate its behavior in the limit of strong onsite interaction and are captured by an effective spin model , namely the heisenberg antiferromagnet @xcite . while collective magnetism has been widely studied in solids over the decades , several experimental restrictions apply quite generally : it is generally far from clear to what extent the typical simplified models are quantitatively realistic , and how to obtain the interaction parameters . external control of these parameters is very difficult . moreover , quantum magnetism becomes particularly interesting in low dimensions . in real effectively low - dimensional solids it is however hard to control or assert the effect of the weaker interactions in the second and/or third dimension . last but not least , solids give us only access to the linear response regime as sampled e.g. by neutron scattering . questions of out - of - equilibrium many - body dynamics are essentially inaccessible . experiments with ultracold atoms in optical lattices constitute clean and well - tunable manifestations of the bose- or fermi - hubbard model @xcite . to implement magnetic systems , the most straight forward approach would hence be to use a gas of ultracold fermions . however , cooling of fermionic gases to the quantum regime is a considerably harder task due to the lack of s - wave scattering among identical fermions @xcite . hence it is desirable to develop an alternative route via gases of ultracold bosons . our investigations follow this idea @xcite . although we focus here on one dimension , analogous setups in higher dimensions @xcite could be used to investigate a plethora of frustrated spin systems that are hard to access analytically and numerically . \(ii ) the superlattice structure chosen in our setup ( in analogy to the recent experiments @xcite ) allows in contrast to @xcite for the tuning of the effective spin - spin interaction by changing an alternating scalar potential @xmath0 ; fig [ fig : systemparameters ] . with the hopping strength @xmath1 and the onsite interaction @xmath2 of the hubbard model , the coupling in the corresponding effective spin model is then @xmath3 . this allows on the one hand to switch for the effective model between the heisenberg ferro- and antiferromagnet . on the other hand one might hope to increase for a fixed onsite interaction @xmath2 the effective coupling , by choosing @xmath4 . in this case , the relevant physics would become visible at correspondingly higher temperatures . however , the validity of the effective model breaks down in the vicinity of @xmath4 . so one has to balance the validity of the heisenberg description and the temperatures needed to observe the quantum effects . to this purpose , the parameter @xmath0 can be easily varied and used to tune to the heisenberg regime in controlled fashion . \(iii ) in recent experiments @xcite by the bloch group , the same optical superlattice as the one discussed here was used . but its parameters were chosen such that the superlattice decomposed actually into isolated double wells . the experiments analyzed dynamics in these double wells and contrasted in particular first order ( hopping ) processes ( hubbard regime ) versus slower second order processes ( heisenberg regime ) . the next step would be to observe coherent many - particle dynamics after coupling the double wells . we analyze such a situation by the time - dependent density matrix renormalization group method ( dmrg ) @xcite . we focus on the coherent evolution of an initial nel state and the differences between the heisenberg and the hubbard regimes and present the experimentally available observables . \(iv ) besides testing the coherence in the experiments , the setup allows to address questions of non - equilibrium many - particle systems , which is in general difficult for all present analytical and numerical methods . contrary to the setup of isolated double wells , one observes for the many - particle dynamics in our setup a relaxation of local quantities . this is an indicator for convergence of subsystems with finite real - space extent to a steady state . recently , the mechanism of how such a relaxation may occur was clarified for ( free ) integrable systems in @xcite . corresponding examples can also be found in @xcite . for a few nonintegrable systems the question was analyzed numerically in @xcite and analytically e.g. in @xcite . in general one expects that in nonintegrable models , thermalization occurs ( due to scattering processes ) , and that in integrable models , relaxation ( to a nonthermal steady state ) occurs via phase averaging effects @xcite . this is demonstrated here analytically for the heisenberg model . our setup could be used to study experimentally such relaxation processes in particular , the qualitative differences between nonintegrable systems , here the bose - hubbard model , and bethe ansatz integrable models @xcite , here the heisenberg model . we also fill a certain gap of current literature on such topics ( see e.g. @xcite ) by emphasizing that the heisenberg spins of the effective model , obtained by the schrieffer - wolff transformation ( in appendix [ sec : derivation ] ) @xcite , should not be identified directly with the two boson species . a spin up of the effective model corresponds rather to a particle of species 1 dressed by hole - double - occupancy fluctuations . the analogy holds only for small @xmath5 . the consequences for experimentally available observables are surprisingly strong . recently in @xcite a setup of coupled double wells ( @xmath6 , but alternating hopping @xmath7 in fig . [ fig : systemparameters ] ) was analyzed numerically in particular , the possibilities to generate entangled pairs of particles were studied . again , a perfect mapping to a spin model was assumed from the outset . the validity of this mapping is one central topic of our article . the paper is organized as follows . section [ sec : setupandmodel ] describes the experimental setup and how it can be described by a bose - hubbard model . restricting to half filling , section [ sec : effectivemodel ] and appendix [ sec : derivation ] derive by a schrieffer - wolff transformation for the limit of large onsite interactions an effective model which is the heisenberg antiferro- or ferromagnet . in section [ sec : evolution ] , we investigate by time - dependent density - matrix renormalization - group ( dmrg ) the evolution of typical observables like magnetization , momentum - space and real - space correlators , where the first two are also available experimentally . the focus is on contrasting the differences between the full hubbard dynamics and the corresponding effective spin model , and also the differences to the case of isolated double wells @xcite . we observe indications for ( local ) relaxation to steady states . this is discussed in section [ sec : relaxation ] where we also explain how the relaxation for the heisenberg model is connected to a phase averaging effect . section [ sec : validity ] addresses in more detail the question why and under what circumstances the effective model is a valid description for the full hubbard hamiltonian , especially for the dynamics . in section [ sec : groundstates ] we argue that the groundstate of the heisenberg antiferromagnet could be prepared by tuning an alternating hopping parameter of the superlattice adiabatically . section [ sec : conclusion ] gives a short conclusion . in the following , we present a setup of ultracold bosonic atoms in a one - dimensional optical superlattice that reduces in certain parameter regimes , where first order hopping processes are suppressed , to the heisenberg ferro- or antiferromagnet . the use of bosons is motivated by the fact that experimentally , access to the low energy quantum physics is at the moment still much harder for fermionic systems . in analogy to the fermionic case , for which the antiferromagnetic heisenberg model describes the effective low - energy physics of the fermionic hubbard model , we choose to have two species @xmath8 of bosons in the lattice two hyperfine states of a bosonic atom . at half filling ( @xmath9 sites , @xmath10 ) , the effective low energy model is , as we will see later , the ferromagnetic heisenberg model . to allow for tuning the effect of first order processes and to switch between a ferromagnetic and antiferromagnetic regime , we employ an alternating onsite potential @xmath11 and call the two sublattices @xmath12 and @xmath13 . the potential minima differ by a value @xmath14 . such superlattices can be generated by the superposition of two laser frequencies of ratio @xmath15 , see ref . @xcite and fig . [ fig : systemparameters ] . further , the tight - binding approximation with restriction to the first bloch band ( one wannier function per site ) is assumed . then the system is described by the two - species bose - hubbard hamiltonian @xmath16 in particular , we choose @xmath17 the superlattice potential is of the form @xmath18 the amplitude @xmath19 of the second potential can be used to tune @xmath0 ; for our purposes , @xmath20 . in it was assumed that hopping from a site @xmath21 to its neighbors @xmath22 occurs with equal amplitude @xmath1 . in principle , the hopping depends exponentially on the distance between the potential minima @xcite . only if we choose the phase difference @xmath23 to vanish as in fig . [ fig : systemparameters]b , the positions of the potential minima will be equidistant for all @xmath19 , and the wannier wave functions for all sites are reflection symmetric . in this case , the hopping will hence be of equal strength for all bonds , i.e. @xmath24 in fig . [ fig : systemparameters ] . this situation is considered in the following . for the actual analysis of dynamics in section [ sec : evolution ] , we will choose equal inter- and intra - species interaction , @xmath25 . this is at the moment the standard situation for the corresponding experiments . with @xmath25 , the effective heisenberg models , describing the second order physics , will turn out to be isotropic . the onsite potential @xmath0 and interaction @xmath2 can be calculated in harmonic approximation of the lattice potential around its minima . for @xmath26 in , the corresponding oscillator frequencies are @xmath27 , where @xmath28 is the recoil energy of the laser potential with the shorter wave length . for the dependence of the onsite potential on the lattice parameters follows @xmath29 in order to achieve an effectively one - dimensional lattice , @xmath30 is superimposed with two transversal laser beam potentials of a ( higher ) amplitude @xmath31 . within the harmonic approximation , and with the s - wave scattering length @xmath32 , the resulting onsite interactions for two neighboring sites are @xcite @xmath33 they are in principle not identical . irrespective of this , the effective spin model derived in section [ sec : effectivemodel ] would be isotropic and translation invariant . as we will see in the following for a set of realistic experimental parameters , we have for the case @xmath34 ( @xmath35 @xmath6 ) that @xmath36 and @xmath37 . the onsite potentials @xmath0 considered in our numerical simulations of the hubbard model are from the interval @xmath38 $ ] . according to , @xmath19 will for nonzero @xmath0 hence obey @xmath39 and we can thus use @xmath40 in good approximation . for the tight - binding approximation to hold , we need that energies @xmath1 , @xmath2 , and @xmath0 are well below @xmath41 , the energy scale for vibrations of an atom in one minimum of the laser potential . with the hopping @xmath42 @xcite , we have for example with @xmath43 , rubidium atoms ( i.e. @xmath44 ) , @xmath45 and @xmath46 ( cmp . e.g. to @xcite ) that @xmath47 , @xmath48 and @xmath49 . so as long as @xmath0 is also well below @xmath50 , the tight - binding approximation for using the lowest bloch band is valid . this will be the case in the rest of the paper . to go to a regime where the physics of the two - species bose - hubbard model reduces to that of a heisenberg magnet , we choose half filling @xmath51 ( @xmath9 is the total number of lattice sites ) and assume the large-@xmath2 limit @xmath52 in this limit , occupation of a single site by more than one boson is energetically unfavorable and will occur only in short lived intermediate states . this means that single ( first order ) hopping processes are suppressed . besides some hybridization effects , exactly one boson sits on each lattice site and can be identified with an effective spin on that site ( up and down orientations corresponding each to one of the boson species ) . the second order hopping processes as depicted in fig . [ fig : superexchange ] lead then to nearest neighbor spin - spin interactions . we will show below , how the effective hamiltonian can be derived by a schrieffer - wolff transformation . while this is a well - known procedure , there is an interesting twist to the interpretation of the result . for the moment let us work with the naive identification of spins up and down of the effective model with the two boson species of the full model . we want to derive an effective hamiltonian describing the physics of the hubbard hamiltonian in the subspace @xmath53 of singly - occupied sites . @xmath54 the effective hamiltonian can be deduced from the following simple recipe : with exactly one spin per site , the onsite interaction is ineffective . hopping processes occur only in second order , leading to a spin - spin interaction @xmath55 the corresponding coupling strengths @xmath56 and @xmath57 are obtained by dividing for each possible second order process ( fig . [ fig : superexchange ] ) the product of the transition matrix elements @xmath58 for the hopping to a neighboring site and back by the energy difference @xmath59 ( @xmath60 ) to the intermediate state , and adding all such terms that contribute to the same effective spin - spin interaction ( see also @xcite ) . @xmath61 the effective hamiltonian is the xxz model . from now on we specialize to @xmath25 , i.e. @xmath62 and have hence the isotropic heisenberg ferromagnet for @xmath63 ( @xmath64 ) and the isotropic antiferromagnet for @xmath65 ( @xmath66 ) . note that the effective heisenberg hamiltonian would also be isotropic and translation invariant if the onsite interaction @xmath2 would be different for even and odd sites . if the hopping would be alternating ( @xmath67 ) , we would obtain the dimerized heisenberg model . a common mathematical approach for the deduction of such effective models is the schrieffer - wolff transformation @xcite . we are interested in the physics of the subspace @xmath53 with exactly one particle per site . the hubbard hamiltonian couples this subspace in first order of the hopping @xmath1 to the rest of the hilbert space ( states with doubly occupied and empty sites ) . the schrieffer - wolff transformation , @xmath68 is a unitary transformation with generator @xmath69 chosen such that the transformed hamiltonian @xmath70 does not contain terms anymore that couple @xmath53 to the rest of the hilbert space , or at least only in some higher order of @xmath1 . in appendix [ sec : derivation ] , a generator @xmath71 is derived , such that effective hamiltonian is @xmath72 the full effective hamiltonian @xmath70 still contains a term @xmath73 $ ] representing the remaining coupling of the subspace @xmath53 to the rest of the hilbert space which is of order @xmath74 . the method is based on the smallness of @xmath69 . according to , it hence breaks down when @xmath75 . the effective hamiltonian is only valid for @xmath76 . only in this regime , the first order hopping processes leading out of @xmath53 are suppressed . see the discussion in section [ sec : validity ] . often spin up ( down ) states of the effective model are then identified with a boson of species @xmath77 ( species @xmath78 ) on the corresponding sites . however , with respect to the original model , eq . , it is _ not _ @xmath53 itself that is weakly coupled to the rest of the hilbert space and evolves according to the heisenberg hamiltonian but the subspace @xmath79 defined in eq . . a spin up in the effective model corresponds rather to a boson of species @xmath77 with a cloud of hole - double - occupancy fluctuations @xmath80 ; see also fig . [ fig : lattice_doublewell_sw - trafo ] . the experimental consequences are surprisingly strong , as we will in the next section . in the following , numerical results for the evolution of the system where the initial state is the nel state @xmath81 in isolated double wells . to compare the effect of first and second order processes , the evolution was done twice for each set of parameters ( @xmath82 , @xmath83 , various @xmath0 ; see section [ sec : setupandmodel ] ) , once with the corresponding heisenberg model ( in the subspace @xmath53 ) and once with the full hubbard hamiltonian ( in the full hilbert space ) , where the initial state was in fact chosen as the tensor product of alternatingly having one boson of species @xmath77 or @xmath78 on each site . the two different time scales of first and second order processes become clearly visible . the qualitative differences to the isolated double well situation ( as analyzed in @xcite ) and resulting interesting questions for experimental investigations are discussed . we shortly want to discuss how well the dynamics of the magnetic model , the heisenberg model , can be implemented experimentally by those of the two - species bose - hubbard model . in the literature on magnetism via ultracold two - species atom gases in optical lattices @xcite , spins up and down of the magnetic system are usually identified directly with atoms of species @xmath77 and @xmath78 of the ultracold gas . in this vein , evolution of the nel state with the heisenberg hamiltonian would be translated into evolution of the state @xmath84 with the hubbard hamiltonian . this is actually correct only to zeroth order in @xmath69 . we want to implement the evolution of a state @xmath85 , , under the effective hamiltonian @xmath86 by the evolution of a state @xmath87 , , under the bose - hubbard hamiltonian @xmath88 . the state @xmath89 is @xmath90 , superimposed with states where starting from @xmath90 , pairs of doubly occupied sites and empty sites were created ( e.g. eq . and fig . [ fig : lattice_doublewell_sw - trafo ] below ) . this can also be interpreted as constructing the nel state with effective spins , each corresponding to a boson accompanied by a cloud of hole - double - occupancy fluctuations @xmath80 . the decisive point is now that it seems not possible to prepare such states from @xmath79 ( and has to our knowledge never been done ) , but only some specific states from @xmath53 . hence , instead of starting the experiment from the initial state @xmath89 , one is forced to neglect the schrieffer - wolff transformation and start from the state @xmath90 in our example the nel state . for observables that do not change the number of doubly occupied sites , this results in an error of @xmath91 . if we had determined the exact schrieffer - wolff transformation @xmath92 ( i.e. @xmath69 exact to all orders in @xmath1 ) and could actually implement it e.g. by time evolution in the experiment , all measurements would be exact . one could prepare the state @xmath90 , apply the schrieffer - wolff transformation by time evolution to obtain @xmath89 , evolve with the hubbard hamiltonian for some time @xmath93 , apply the inverse schrieffer - wolff transformation and measure our observable @xmath94 . this would yield the exact equality @xmath95 where @xmath96 would now of course be a generalization of the heisenberg model with longer ranged interactions . in the appendix [ sec : swtrafo ] , @xmath69 is determined to first order in @xmath97 , , and correspondingly @xmath96 to first order in the effective coupling @xmath56 , . using this approximation of @xmath69 instead of the exact one , the remaining errors in the observables are of order @xmath98 , i.e. @xmath99 . ( it is not @xmath91 , because the operator @xmath69 , given in eq . , changes the number of double occupancies by one and the typical observables @xmath94 we are interested in do not . ) however , failing to implement the schrieffer - wolff transformation completely , i.e. measuring @xmath100 instead of , leads to errors of order @xmath101 . this will be demonstrated in an example ( section [ sec : magnetization ] , figs . [ fig : twositeevolve ] and [ fig : twositeevolveerr ] ) . in addition to the error from neglecting or truncating the schrieffer - wolff transformation , there is the error from truncating the effective hamiltonian . this accumulates with time and is in principle of order @xmath102 , but may also just result in a sort of rescaling of the time axis . also the local observables considered relax relatively quickly , making this second source of error less important . in the remainder of the article , the initial state , evolved with the heisenberg hamiltonian will be called @xmath103 . if it is evolved with the hubbard hamiltonian , it will be called @xmath104 . @xmath105 to illustrate the considerations above , let us shortly regard the case of an isolated double well ( two sites ) . hamiltonian and @xmath69 read ( cf . appendix [ sec : derivation ] ) @xmath106 with this , the effective hamiltonian and the transformed initial state are @xmath107 so a magnetic state with one particle per site corresponds in the experimentally realized hubbard model to the magnetic state plus an admixture of states with doubly occupied and empty sites , fig . [ fig : lattice_doublewell_sw - trafo ] . the original hamiltonian generates with @xmath108 doubly occupied sites to first order in @xmath1 . conversely , in the ( full ) effective model , such terms are at least of order @xmath74 ( in the two site case here , actually of order @xmath109 ) . the nel state @xmath90 , , and the effective hamiltonian are both real in the @xmath110-eigenbasis @xmath111 ( real coefficients and matrix elements ) . typical observables @xmath94 of interest like @xmath112 for the magnetization or @xmath113 and @xmath114 for correlators are real in that basis and selfadjoint . it follows that the corresponding expectation values @xmath115 are identical for the heisenberg ferromagnet ( @xmath116 ) and antiferromagnet ( @xmath117 ) : let @xmath118 , @xmath119 , and @xmath120 for the time evolution operator of the ( anti)ferromagnetic heisenberg model . then @xmath121 the evolution of the corresponding observable under the full hubbard hamiltonian @xmath122 will obey this symmetry to zeroth order in @xmath69 . typically , the resulting curve will coincide well with the corresponding heisenberg curve . the smaller @xmath123 is chosen , the worse the effective model will capture the actual dynamics and the stronger deviations from the corresponding heisenberg results will be . the specific form of the deviations , however , will depend on the choice of @xmath2 , @xmath0 , and @xmath1 . in particular they show no symmetry when switching between the antiferromagnetic and the ferromagnetic regimes ( @xmath124 ) . to illustrate this further , several plots contain the two curves @xmath6 and @xmath125 which have according to the same effective spin spin interaction strength @xmath56 , except for the opposite sign ( fm , afm , respectively ) . for the numerical simulation , a krylov subspace variant @xcite of the time - dependent dmrg algorithm was used @xcite . for the hubbard model , the site basis was restricted to a maximum of two particles for each species . insensitivity of observables to the chosen maximum number of bosons per site was affirmed . we chose lattice sizes of @xmath126 for the bose - hubbard model and @xmath127 sites for the heisenberg model . odd numbers are useful here to have reflection symmetric states . as a matter of fact , boundary effects are much less problematic here than in groundstate calculations , as the initial state is a product state and correlations between sites are generated inside a causal cone ( the analogon of a light cone ; see section [ sec : correlations ] , in particular figs . [ fig : heisenbergantiferrocorrconelog ] and [ fig : afm_avercorrhubheisenlogu8d16 ] ) . so as long as measurements are done in the middle of the system , outside of the causal cones starting from the boundary sites , results are except for exponentially small contributions identical to those of an infinite system ( thermodynamic limit ) . in the time evolution , the absolute difference per physical time unit between exactly evolved state and the state evolved via dmrg @xmath128 was bounded from above by @xmath129 to @xmath130 and the time step chosen appropriately between @xmath131 and @xmath132 . the errors were determined in a rigorous fashion , by calculating the exact value of @xmath133 , where @xmath134 are the krylov vectors . for all calculated observables , convergence in the error bound and @xmath135 was checked . the resulting number of basis states , used to represent the time - evolved state , was @xmath136 . figures [ fig : abweichung_hubbard_afm_heisenberg_long_t][fig : abweichungvoneinfachbesetzung ] show the evolution of the site magnetization @xmath137 in the heisenberg model and the corresponding quantity @xmath138 ( @xmath139 according to section [ sec : errors ] ) for the full hubbard hamiltonian . for the latter , times were rescaled by the coupling constant @xmath56 , , of the corresponding effective spin model . site @xmath140 was chosen to be in the middle of the system in order to avoid finite size effects ( cf . section [ sec : numerics ] ) . for the heisenberg model ( in the thermodynamic limit ) , the site magnetization obeys for symmetry reasons @xmath141 . analogously , due to invariance under translations by an even number of sites and particle number conservation , one has for the hubbard model ( again in the thermodynamic limit ) @xmath142 and hence @xmath143 , and @xmath144 for all times . as discussed in section [ sec : numerics ] , deviations of our numerical results from the thermodynamic limit are negligible , although the simulations are carried out with finite lattice sizes . the larger @xmath123 is ( for fixed @xmath82 ) , the better the curves for the full hubbard hamiltonian coincide with those of the heisenberg model . this is consistent with sec . [ sec : effectivemodel ] as the perturbative derivation of the effective model becomes exact in this limit . note that the deviations between the measurements stem from two contributions here : ( a ) failure of preparing the correct @xmath79 state , i.e. applying the schrieffer - wolff transformation at @xmath145 , ( b ) failure of measuring @xmath146 instead of @xmath147+{\mathcal}o({\hat{{\mathcal}s}}^2)$ ] . the weight of those errors which are of order @xmath101 vanishes only far away from @xmath148 . for the heisenberg model , we observe relaxation of the site magnetization from @xmath149 to @xmath150 . the oscillations of this observable occur on the time scale @xmath151 . the relaxation is possible due to the continuous spectrum of the heisenberg model ( in the thermodynamic limit ) . here , the convergence to a steady state is connected to a phase averaging effect , as is typical for integrable systems . analytically this can be seen in a time - dependent mean field treatment of the heisenberg model which we present in section [ sec : relaxation : heisenbmeanfield ] . for the staggered magnetization ( @xmath152 ) one obtains ( in this approximation ) a damped oscillation with the amplitude decaying as @xmath153 . this coincides well with the dmrg data , giving support to the mean field approach ; see section [ sec : relaxation : heisenbmeanfield ] , fig . [ fig : heisenb_meanfield_evolveneel ] . for large @xmath123 the hubbard dynamics clearly follow the curves obtained with the heisenberg model ( second order processes ) ; fig . [ fig : abweichung_hubbard_afm_heisenberg_long_t ] and [ fig : abweichung_hubbard_fm_heisenberg_long_t ] . on the shorter time - scale @xmath154 ( @xmath155 in the rescaled plots ) , corresponding to first order processes , small oscillations around the heisenberg curves are visible . their amplitude decreases with increasing @xmath123 . the perturbative treatment of the system , leading to the isotropic heisenberg model , breaks down for @xmath156 . in this case , the two boson species can not be interpreted as spin up or down states anymore and one has an appreciable amount of double occupancies in the system as demonstrated in fig . [ fig : abweichungvoneinfachbesetzung ] . finally , we want to compare those results to the dynamics for isolated double wells as addressed experimentally in @xcite . [ fig : twositeevolve ] shows for this case the dynamics of the site magnetization again for the hubbard model at various @xmath65 and the corresponding antiferromagnetic heisenberg model . the decisive difference is that no equilibration is possible in this case . this is due to the fact that the hamiltonian has only a few discrete eigenvalues here , as opposed to a gapless continuous spectrum for the lattice systems in the thermodynamic limit . in the two - site heisenberg model we have only two states in the basis of the @xmath157 hilbert space . the two eigenstates have energy difference @xmath56 . the magnetization curve for the heisenberg curve is hence just a cosine with frequency @xmath56 and constant amplitude @xmath158 . the dynamics of the corresponding two - site hubbard model is determined by three ( discrete ) incommensurate frequencies . the magnetization is hence not completely periodic , but due to the relation to heisenberg model , a frequency @xmath159 is still dominating . no sign of relaxation is visible . as mentioned above , the differences between hubbard and heisenberg dynamics stem from the fact that the two schrieffer - wolff transformations in have been neglected . fig . [ fig : twositeevolve ] shows in the lower panel the site magnetization for the case where both error sources ( a ) and ( b ) have been corrected . although this should be hard to implement experimentally , it is unproblematic for our numerical analysis . we apply the schrieffer - wolff transformation , correct up to @xmath160 , to the initial state before the hubbard time evolution and its inverse before the measurement . as discussed in section [ sec : errors ] , the remaining deviations from the heisenberg curve are then only of order @xmath98 ; figs . [ fig : twositeevolve ] and [ fig : twositeevolveerr ] . the correlation functions in figs . [ fig : afm_corrconelogu8d16][fig : afm_avercorrhubheisenlogu8d16 ] support on the one hand the results already obtained from the magnetization dynamics in section [ sec : magnetization ] . on the other hand one also sees here explicitly that correlations spread out inside a causal cone ( analogon of a light cone ) defined by the maximum group velocity . the latter coincides for large @xmath123 with the maximum group velocity @xmath161 of the heisenberg model . one also notes here that equilibration to a steady state occurs first for small subsystems . this issue will be discussed in section [ sec : relaxation ] . experimental access to onsite magnetization or , correspondingly , the particle number difference ( `` spin imbalance '' ) has already been demonstrated @xcite . however there is no direct access to the real - space correlators . as it turns out , the standard experimental observable for experiments with ultracold atoms , the momentum distribution @xmath162 , is to zeroth order in @xmath69 constant in time . it measures to this order simply the particle density which is in the limit of heisenberg dynamics @xmath163 very close to one . @xmath164 it follows with @xmath165 hence one needs to go beyond the measurement of the momentum distribution . by analysis of shot - noise of absorption images taken by time - of - flight measurements , one obtains momentum - space particle density correlation functions @xcite . experimentally available are @xmath166 and @xmath167 and hence also @xmath168 . in the following , we will again use the approximation @xmath169 . @xmath170 those give information about long - range spin correlations . @xmath171 for the observable @xmath172 and @xmath173 for the observable @xmath168 . this also reflects the fact that to zeroth order of @xmath69 , there are no double occupancies with respect to the original basis . however , there is an admixture of them , contributing in second order of @xmath69 , @xmath174 . inserting and to yields @xmath175 and @xmath176 where @xmath177 , @xmath178 , @xmath179 , and @xmath180 were used . a numerical comparison of the evolution of the momentum - space spin - spin ( density - density ) correlator for the heisenberg and the hubbard models is given in figs . [ fig : sqsq - heisenbergafm]-[fig : sqsq - hubbardd16-pihalf ] . to achieve such a good agreement , two corrections were necessary that are described in more detail in appendix [ sec : postprocess - nk - nk ] . first of all one needs to correct for finite size effects . secondly , single particle green s functions @xmath181 enter which are trivial when evolving with the heisenberg model ( @xmath182 ) , but have contributions of @xmath91 , when evolving with the hubbard hamiltonian . in the comparison of both evolutions , they can hence be understood as a major carrier of disturbance , reflecting first order processes in the hubbard model . to achieve comparability it would be desirable to remove contributions from @xmath181 completely . this would be possible for our numerical analysis . in a corresponding experiment however , the quantities are not available . hence we confined ourselves to removing only the contributions from nearest neighbor correlators @xmath183 . as figures [ fig : sqsq - heisenbergafm]-[fig : sqsq - hubbardd16-pihalf ] demonstrate that this is already sufficient and fig . [ fig : sqsq - hubbardd16_nocorrection ] that it is necessary . the experimental observation of the nearest - neighbor correlators is within reach @xcite . the specific form of the momentum - space correlation function can be understood with the causal cone behavior of the corresponding real - space correlators discussed in section [ sec : correlations ] . at the beginning of time evolution , correlations for small distances build up ( e.g. due to the spin flip terms @xmath184 in the heisenberg model ) . this corresponds in the momentum space representation to correlations for large @xmath185 . as the correlations spread out in real - space , correlations for smaller momenta @xmath185 build up . contrary to the setup of isolated double wells , one observes for the many - particle dynamics in our setup a relaxation for local quantities . this may be seen as an indicator for convergence of the states of subsystems with finite real - space extent to a steady state . recently , the mechanism of how such a relaxation may occur was clarified for ( free ) integrable systems @xcite . corresponding examples can also be found in @xcite . the setup considered in this paper could be used to study experimentally such relaxation processes in particular , the differences for the nonintegrable bose - hubbard model and the bethe ansatz integrable heisenberg model . experimental investigations would be very useful here , as the fast entanglement growth during time evolution , fig . [ fig : entanglementintime ] , prohibits numerical access to long times and for bethe ansatz integrable systems , analytical results are also relatively limited for such purposes . integrable many - particle systems do not relax to the well known canonical , or `` thermal '' , ensembles ( a fact that was already observed experimentally e.g. in @xcite ) . if they relax the steady state is due to the integrals of motion to a much more constrained ensemble @xcite . this could be detected experimentally by comparing the steady state correlation functions after time evolution to those obtained for the corresponding thermal ensemble . the temperature should be chosen such as to have the same energy in both states . for ( free ) integrable models , the relaxation occurs due to a phase averaging ( `` dephasing '' ) effect @xcite . in section [ sec : relaxation : heisenbmeanfield ] , the relaxation in the bethe ansatz integrable heisenberg model is treated within a mean field approximation . also in this case , relaxation is connected to a phase averaging effect . nonintegrable systems are generally believed to relax to a thermal ensemble due to effective scattering processes . recent numerical analysis of such systems @xcite is not yet fully conclusive due to limitations on maximum observation times ( density - matrix renormalization - group ) or system size ( exact diagonalization ) . analytical approaches are usually restricted to rather exotic models or limiting cases . see e.g. @xcite for investigations by dynamical mean - field theory ( dmft ) . in our setup , the nonintegrable two - species bose - hubbard model could be tuned so close to the heisenberg regime ( large @xmath123 ) that thermalization occurs very slowly . one might hence observe first a relaxation to a nonthermal ( almost ) steady state due to the integrable heisenberg dynamics , which would then be followed by slower thermalization due to the remaining nonintegrable first order processes of the full bose - hubbard hamiltonian . in this section , we investigate analytically the relaxation of the heisenberg magnet with the initial state being the nel state . in particular we will derive that the ( staggered ) magnetization decays as @xmath186 due to a phase averaging effect . the model is bethe ansatz integrable @xcite . however , it is in general not possible to solve the equations of motion for arbitrary initial states . with appreciable numerical effort this has been achieved recently ( only ) for the initial state being the groundstate plus a one - particle excitation @xcite . to investigate the dynamics nevertheless , we hence employ a mean field approximation for the @xmath187 term . @xmath188 where the order parameter @xmath189 is the staggered magnetization @xmath190 after a fourier and a jordan - wigner transformation @xcite with @xmath191 and @xmath192 , the mean field hamiltonian and the staggered magnetization read with @xmath193 @xmath194 the initial state is the nel state and reads in the fermionic operators for @xmath145 with @xmath195 @xmath196 is in the initial state only correlated with mode @xmath197 . as the mean field hamiltonian couples for every @xmath198 also just those two modes , the state remains in the form for all times . with @xmath199 $ ] one obtains the equations of motion ( @xmath200 ) [ eq : mfeqofmotion ] @xmath201 a system of @xmath9 coupled nonlinear differential equations . those can be integrated numerically , yielding for the staggered magnetization @xmath202 a damped oscillation decaying as @xmath186 . [ fig : heisenb_meanfield_evolveneel ] compares @xmath202 from the mean field analysis to the corresponding dmrg result ( fig . [ fig : abweichung_hubbard_afm_heisenberg_long_t ] ) and shows good qualitative agreement . as demonstrated in @xcite , where the same equations of motion were obtained for the evolution of a system of spinless fermions , eq . is equivalent to the equations of motion of the classical hamiltonian @xmath203 with the anderson pseudospin variables @xmath204 , @xmath205 for @xmath206 and @xmath207 , @xmath208 , @xcite . this hamiltonian occurred in the mean field analysis of quenches in fermionic condensates ; see e.g. @xcite . from this , it is known that and hence are integrable @xcite due to the @xmath209 integrals of motion @xmath210 , @xmath211 one can now argue that the @xmath140 and @xmath212 components of the vectors @xmath213 will vanish for large times , as done in @xcite . from this one can determine the ( nonthermal ) steady state , by equating @xmath214 with @xmath215 . the result is @xmath216 . with @xmath205 and @xmath217 @xmath218 , it follows that @xmath219 with the knowledge of the steady state , the @xmath186 decay of the magnetization @xmath189 can now be derived . to this purpose let us first recall the general dephasing scenario for @xmath220-dimensional ( free ) integrable models . in @xcite it was demonstrated that local observables @xmath221 ( i.e. correlators ) lead in general to expressions of the form @xmath222 where the amplitude @xmath223 is determined by the chosen observable , the initial state , and the eigenbasis of the hamiltonian . the phase function @xmath224 is determined by the spectrum of the hamiltonian . now , the quantity @xmath221 relaxes to @xmath225 for large times if the phase function varies quickly enough in regions of the @xmath226 space where the amplitude @xmath223 is nonzero . whether and how quickly an observable relaxes is in particular determined by contributions from points where @xmath224 is stationary or @xmath223 diverges . for the paradigmatic scenario of @xmath227 , @xmath228 near a stationary point @xmath229 , the integral in behaves as @xmath230 hence the time - dependent contribution to @xmath221 , for @xmath231 , does not vanish if @xmath232 , vanishes as @xmath233 if @xmath234 , and at least as @xmath235 if @xmath236 , . now we come back to the staggered magnetization . expressed in the variables @xmath237 and @xmath238 , it reads after going to the thermodynamic limit @xmath239 this is , except for the additional time dependence of the amplitude function @xmath240 , an integral of the form . presuming that @xmath189 vanishes for long times , it follows from the equations of motion that for large @xmath93 , the phases of @xmath241 and @xmath242 are roughly @xmath243 and hence @xmath244 , which is stationary ( with @xmath245 ) at @xmath246 ; cf . [ fig : heisenb_meanfield_evolveneel.t_k_nk_uvarg ] . for finite times , the amplitudes of @xmath241 and @xmath242 are finite as @xmath247 and @xmath248 . hence @xmath249 for @xmath246 which is also confirmed numerically in fig . [ fig : heisenb_meanfield_evolveneel.t_k_nk_uvarg ] . the dephasing of the staggered magnetization is determined by the stationary point @xmath246 of @xmath250 . with @xmath251 , @xmath249 , and @xmath252 we have @xmath253 . the phase averaging accounts hence for a factor of @xmath254 for the decay of the staggered magnetization . linearizing the equations of motion around the steady state we find that @xmath255 ( @xmath240 around @xmath246 ) decays as @xmath256 and further that only a vicinity @xmath257 of @xmath246 is contributing to ( the leading order of ) the integral . taking all this together , we infer the @xmath186 decay of the staggered magnetization . one may be wondering why the restriction to the single - occupancy space @xmath53 , , is justified ( if the initial state of the system is in @xmath53 and we evolve with the effective hamiltonian ) , although the coupling to the rest of the hilbert space has the same strength as the coupling for dynamics inside the subspace @xmath53 and although parts of the rest of the hilbert space overlap energetically with @xmath53 . there could be considerable transition rates out of the subspace with ( predominantly ) one particle per site , rendering a description or comparison with dynamics of the effective model derived for that subspace useless . we will assess here that this is not the case ( for the large-@xmath2 limit ) . first of all , the numerical results of section [ sec : evolution ] showed that for large @xmath123 , the hubbard curves follow quite precisely the heisenberg curves , indicating very little transitions to other subspaces . one can also give a somewhat handwaving but rather suggestive argument . we will show in the following that transition matrix elements leading out of @xmath53 occur predominantly to states with energy difference @xmath258 , and diminish in the large-@xmath2 limit . those yield therefore finite small transition amplitudes . in higher orders of the perturbation theory , there are also transitions to states with energy @xmath259 , which will lead to a small ( controllable ) transition rate out of @xmath53 . in the full effective hamiltonian , we regard the term @xmath260 $ ] that generates or destroys double occupancies , i.e. generates transitions between subspace @xmath261 with @xmath262 doubly occupied sites as a perturbation . @xmath263 the subspaces @xmath264 separate ( energetically ) further into @xmath265 with @xmath266 doubly occupied and @xmath267 empty sites on sublattice @xmath12 ( @xmath268 doubly occupied and @xmath269 empty sites on sublattice @xmath13 ) ; i.e. @xmath270 . [ fig : bh2eff_densityofstates ] shows the many - particle spectrum of the effective hamiltonian @xmath271 for the subspaces @xmath272 and @xmath273 as obtained from exact diagonalization in the @xmath157 sector for @xmath274 sites . the single ( quasi-)particle excitations in these subspaces have energies of order @xmath275 spinwaves and hopping of doubly occupied and empty sites . however , the subspaces overlap energetically ( in the thermodynamic limit ) as , in a qualitative picture , one can have @xmath276 quasi - particle excitations resulting in the width @xmath277 of the spectrum for each subspace . specifically for @xmath272 , the lower and upper bounds on the spectrum are determined by the ground state energies of the ferromagnetic and the antiferromagnetic heisenberg models . those are in the thermodynamic limit @xmath278 and @xmath279 , @xcite . if we act on a state @xmath280 of energy @xmath281 with the operator @xmath260 $ ] ( cf . appendix [ sec : derivation ] ) , firstly , @xmath69 generates a double occupancy and an empty site ; @xmath282 on two neighboring sites @xmath21 and @xmath283 . secondly , a corresponding hopping term from @xmath284 acts on @xmath21 ( or @xmath283 ) and @xmath285 ( or @xmath286 ) such that e.g. @xmath287 or @xmath288 . hence both , the doubly occupied and the empty site are on the same sublattice @xmath12 or @xmath13 and the resulting state @xmath289 has energy @xmath290 ; see fig . [ fig : lattice_doubleoccup_subbands ] . let us consider transitions from @xmath272 to @xmath273 . for any initial eigenstate @xmath291 the transition amplitude , to a state @xmath292 is in the born approximation given by @xmath293 this estimate of a small ( oscillating ) transition amplitude follows from the consideration that nonvanishing transition elements exist only for states with energy differences @xmath294 . we have pointed out that the subspaces @xmath272 and @xmath273 ultimately overlap energetically . however , states @xmath295 and @xmath296 with comparable energy will have a vanishing transition matrix element : as argued above , the operator @xmath297 generates states from @xmath273 and causes a change of @xmath258 in energy . besides this it can create or destroy in a qualitative picture only a small number of quasi - particle excitations as it is a product of only four ladder operators this will change the energy only by a small amount of @xmath298 . so @xmath299 will indeed be of order @xmath300 for all nonvanishing transition amplitudes @xmath301 . to illustrate this , figs . [ fig : bh2eff_veigend16 ] and [ fig : bh2eff_veigend10 ] show the transition matrix elements @xmath301 between eigenstates of the effective hamiltonian @xmath271 for the subspaces @xmath272 and @xmath273 as obtained from exact diagonalization in the @xmath157 sector for @xmath274 sites . small matrix elements to states with energy difference @xmath59 remain . for unfortunate choice of @xmath2 and @xmath0 one may hence encounter nonvanishing transitions to states in @xmath302 with @xmath303 . consider e.g. @xmath304 . in this case , two actions of the operator @xmath297 may end up in a state @xmath295 with comparable energy ( @xmath305 ) and hence to a ( finite but small ) transition rate out of @xmath272 . by appropriate choice of the ratio @xmath306 , one can achieve that the effect occurs only in higher orders @xmath297 , resulting in a small transition rate . further , the transition matrix elements itself can be made small by going to the large-@xmath2 limit . in the sections [ sec : evolution ] and [ sec : validity ] we have given arguments and gathered numerical support for the fact that transition rates from the single - occupancy subspace @xmath79 , , to the rest of the hilbert space can be made small for time evolution with the hubbard hamiltonian . if this can also be realized experimentally for sufficiently long times , it would be possible to prepare for example the ground state of the antiferromagnetic heisenberg model by adiabatically switching on the coupling @xmath307 between initially isolated double wells , fig . [ fig : systemparameters ] , i.e. switching from @xmath308 to @xmath309 , while @xmath1 is kept constant . as demonstrated in @xcite for the initial situation of isolated double wells , the groundstate of the single occupancy subspace can be prepared experimentally . for the adiabatic approximation @xcite to be applicable , the system needs to be gapped on the whole path in the space of system parameters except for the end point , where the gap has to close abruptly enough . as argued in section [ sec : validity ] , transitions to other subspaces with ( quasiparticle ) double occupancies can be neglected for a certain period of time @xmath310 that can be made very large . so we only need to worry about transitions from the @xmath79 groundstates to excited states inside the subspace , i.e. we need to derive conditions on the dependence of the corresponding energy gap on the hopping @xmath307 between initially isolated double wells such that @xmath309 can be reached adiabatically in a finite amount of time @xmath311 . the quantitative condition for adiabaticity is generally stated as @xmath312,n\neq 0},\ ] ] where @xmath313 label the energy eigenstates and @xmath314 is the initial state . recently , substantial problems were pointed out @xcite and two more conditions added @xcite @xmath315 if we have one time - dependent system parameter @xmath316 , namely the dimerization @xmath317 , where @xmath318 and one part of the hamiltonian is linear in that parameter ( this is the case for the effective hamiltonian and @xmath319 ) , the numerators of - are proportional to the sweeping speed @xmath320 . the denominator is the spectral gap @xmath321 . only points @xmath322 in parameter space where the gap vanishes are problematic . if the gap vanishes as @xmath323 @xmath238 should ( for @xmath93 close to @xmath324 ) be reduced as @xmath325 . according to , @xmath326 hence , only for @xmath327 , i.e. for gaps that close abruptly enough , adiabaticity can be reached with @xmath328 . the second condition , , is in this scenario fulfilled automatically , the third , , implies @xmath329 which is also true . for @xmath330 , a situation which was examined intensively in the context of spin - peierls systems , the model was first treated by jordan - wigner transformation and subsequent bosonization @xcite . the precise result for the excitation gap can be obtained by a mapping to the four - state potts model @xcite or conformal field theory @xcite ( see also @xcite ) . the gap is given by @xmath331 that means we have a gap with @xmath332 and hence the gap can be closed in a finite amount of time with exponent @xmath333 . this means that the dimerization @xmath334 has to be varied with speed @xmath335 , hence @xmath336 . one needs thus the time @xmath337 . the smaller @xmath338 is , the farther we are in the adiabatic regime but the longer we need for the preparation . an analysis of how small @xmath338 is to be chosen to achieve a given accuracy of the prepared state could be carried out along the lines of ref . @xcite . note that in @xcite , it was recently discussed within a mean - field approach , how the antiferromagnetic phase of the three - dimensional fermi - hubbard model could be reached by adiabatic tuning of the lattice potential . we have studied a setup of two species of ultracold bosonic atoms in an optical superlattice , which realizes in a certain parameter regime the heisenberg ferro- and antiferromagnet . the focus was in particular on time evolution of nonequilibrium states . our numerical results and analytical considerations showed that the physics of bose - hubbard model implemented in the experiment differs for certain parameter ranges considerably from the physics of the effective heisenberg models . note that this would also be true for alternative suggestions as in @xcite . the spin states up and down can in general not be identified directly with a bosonic particle of one specific species . the regime where the correspondence between the two models is good , implies higher requirements on cooling and coherence ( coherence time ) in an experimental realization . the explicit form of the schrieffer - wolff transformation was used to analyze the transition rates out of the magnetic subspace of the full hilbert space . in contrast to the accomplished experiments @xcite for isolated double - wells ( filled each with two particles ) , the setup of coupled double - wells discussed here allows for relaxation of the many - particle state . in the numerics we observed indications for ( local ) relaxation to steady states . for the heisenberg model in a mean field approximation , we explained how the relaxation is connected to a phase averaging effect . this is typical for integrable models which have nonthermal steady states . nonintegrable models are generally believed to thermalize due to effective scattering effects . our setup can be tuned from the nonintegrable bose - hubbard model to the bethe ansatz integrable heisenberg model and could hence be used to study the differences of the relaxation processes experimentally . finally we argued that the groundstate of the heisenberg antiferromagnet could be prepared by tuning an alternating hopping parameter of the superlattice adiabatically . we thank i. bloch , h. capellmann , a. flesch , s. flling , and a. kolezhuk for discussions . this work was supported by the dfg . t. b. also acknowledges financial support by the studienstiftung des deutschen volkes . here we derive the effective spin hamiltonian describing the physics of the two species bose - hubbard model in the subspace @xmath79 , , where every site is occupied by exactly one quasi - particle . the spin - spin interaction is generated by second order hopping processes of the particles . in the large-@xmath2 limit , transitions from @xmath79 to bands with double - occupancies are energetically hindered . in particular , transitions from @xmath339 to the subspace with double - occupancies ( and holes ) , @xmath340 , can be treated perturbatively . the hamiltonian contains terms , linear in the hopping @xmath1 , which couple @xmath53 to the subspace with double - occupancies . we are looking for a canonical transformation @xmath341 such that the single - occupancy subspace @xmath53 of the resulting quasi - particles @xmath342 couples only in second order to double occupancies ( @xmath340 ) . the calculation can be done in analogy to the derivation of the kondo lattice model @xcite from the periodic anderson model @xcite or the @xmath93@xmath56 model @xcite from the fermionic hubbard model @xcite and is modified only by the asymmetry term @xmath11 and the finite intra - species repulsion ( double occupancies @xmath343 and @xmath344 ) . let us rewrite the hamiltonian , restricted to the subspace @xmath345 , in the form @xmath346 where @xmath347 runs over nearest neighbors ( one index from each sublattice ) , @xmath348 increases / decreases the number of doubly occupied sites and @xmath349 leaves it unchanged ; fig . [ fig : hopping - terms ] . @xmath350 where @xmath334 denotes the kronecker delta and in its argument , @xmath351 denote the particle number operators . for the operators @xmath352 , we further distinguish between those which change the number of ( a ) inter - species and ( b ) intra - species double occupancies , see fig . [ fig : hopping - terms ] . @xmath353 the hamiltonian contains terms @xmath354 , linear in the hopping @xmath1 , which couple @xmath53 to the double - occupancy subspace . we are looking for a canonical transformation @xmath355 such that the single - occupancy subspace @xmath53 of the resulting quasi - particles @xmath342 couples only in second order to double occupancies ( @xmath340 ) . @xmath356+{\frac{i^2}{2}}[{\hat{{\mathcal}s}},[{\hat{{\mathcal}s}},\hat h]]+{{\mathcal}o}({\hat{{\mathcal}s}}^3\hat h)\\ & = \hat h_0 + \hat h_{\mathfrak{t}}^0 + \hat h_{\mathfrak{t}}^++\hat h_{\mathfrak{t}}^- + i[{\hat{{\mathcal}s}},\hat h]+{\frac{i^2}{2}}[{\hat{{\mathcal}s}},[{\hat{{\mathcal}s}},\hat h]]+\dots\\\end{aligned}\ ] ] in the first commutator , the contribution @xmath357 $ ] dominates and we therefore look for a generator @xmath69 such that @xmath358 = -(\hat h_{\mathfrak{t}}^++\hat h_{\mathfrak{t}}^-).\ ] ] from this equation follows with @xmath359 and @xmath360 that @xmath361 and hence @xmath362+{\frac{i^2}{2}}[{\hat{{\mathcal}s}},[{\hat{{\mathcal}s}},\hat h]]+{\mathcal}o({\frac{{\mathfrak{t}}^3}{(u , u_s,\delta)^2}}).\ ] ] this yields for the solution @xmath363 with @xmath11 from , we can now state more precisely @xmath364 so the perturbative treatment will break down near the crossing point @xmath365 and for large hopping @xmath1 . the full effective hamiltonian reads @xmath366 + i[{\hat{{\mathcal}s}},\hat h_{\mathfrak{t}}^0]+{\mathcal}o({\mathfrak{t}}^3).\end{gathered}\ ] ] the commutator terms still couple @xmath53 with the rest of the hilbert space ( subspaces with differing numbers of doubly occupied sites ) . however , this coupling is now not @xmath367 , as in the original hamiltonian , but of @xmath368 . this was achieved by the schrieffer - wolff transformation @xmath69 , which replaces our original particles @xmath369 , by particles with a cloud of hole - double - occupancy fluctuations @xmath80 . in the single - occupancy subspace @xmath53 at half filling , , , @xmath284 , @xmath370 , @xmath371 , @xmath73 $ ] and the terms of third order in the hopping are all ineffective such that we are left with @xmath372_{{\mathcal}h_1}+{\mathcal}o({\mathfrak{t}}^4).\ ] ] the commutator consists of hopping terms via virtual double - occupancy states . they are of the form @xmath373 and can be rephrased as spin - spin interactions . with @xmath374 the effective hamiltonian reads @xmath375 where ( see also @xcite ) @xmath376 this hamiltonian is , except for higher order effects , the xxz model . in the bulk of the article we specialize to @xmath25 , i.e. @xmath62 and have hence the isotropic heisenberg ferromagnet for @xmath63 ( @xmath64 ) and the isotropic antiferromagnet for @xmath65 ( @xmath66 ) . as was already pointed out , the full effective hamiltonian still contains a coupling to the subspace with one double - occupancy ( of quasi - particles ) . the approximation made by neglecting it is discussed in section [ sec : validity ] . a peculiarity of our situation is that , due to half - filling of both particle species , we are restricted to the @xmath157 sector of the heisenberg model . one can go to higher orders in the perturbative treatment of the hopping , by adding higher order terms to the generator @xmath69 of the schrieffer - wolff transformation . the next order term @xmath377 has to be chosen such that it eliminates the term @xmath378 $ ] in . this would result in a further contribution to the effective spin hamiltonian , namely next nearest neighbor and four - spin interactions , generated by sequences of four virtual hopping events . the numerics were done for a finite size system ( cf . section [ sec : numerics ] ) . to correct for the resulting finite - size effect is simple for the calculation of the spin - spin correlators in the heisenberg model in the r.h.s . of and . one can use @xmath379 where @xmath380 is some site in the middle of the system that is odd ( even ) for odd ( even ) i. this corresponds to the invariance of the systems under translations by multiples of two sites in the thermodynamic limit . the momentum - space density - density correlators , , in the hubbard model are determined from the real - space four point correlators . to eliminate finite - size effects of those , we note first that due to the restriction of all correlations to a ( causal ) time - space cone , the four point correlators behave for large distances as ( spin indices suppressed ) @xmath381.\end{aligned}\ ] ] consequently , the quantity @xmath382\end{gathered}\ ] ] is localized ; it has support only for @xmath383 and @xmath262 inside the causal cone centered at site @xmath21 ( cmp . to section [ sec : correlations ] ) . so the value of @xmath384 for the thermodynamic limit is approximated well by @xmath385,\end{gathered}\ ] ] where @xmath386 is the ( approximate ) single particle green s function in the thermodynamic limit @xmath387 and for odd ( even ) @xmath21 , @xmath380 is an odd ( even ) site in the middle of the system . + as disclosed by equation or the four point correlators entering the momentum - space density - density correlator contain contributions from products of single - particle correlators @xmath388 . those are trivial when evolving with the heisenberg model : @xmath182 , as there is exactly one particle per site . but according to , they have contributions of @xmath91 , when evolving with the hubbard hamiltonian . in the comparison of observables evolved with both models , those correlators enter hence as a major carrier of disturbance . to achieve comparability it would be desirable to remove contributions from @xmath181 completely . this would be possible for our numerical analysis . in a corresponding experiment however , the quantities are not available . hence we confined ourselves to removing only the contributions from nearest neighbor correlators @xmath183 . that means @xmath386 in is set to zero for @xmath389 . this was already sufficient to demonstrate the correspondence of the dynamics if we are safely in the large-@xmath2 limit . an experimental procedure for the measurement of the nearest - neighbor correlator was suggested in @xcite . hence , the same manipulations might be carried out for experimentally obtained momentum - space density - density correlators .

we study the asymptotic behaviour of families of gradient flows in a general metric setting , when the metric - dissipation potentials degenerate in the limit to a dissipation with linear growth . we present a general variational definition of bv solutions to metric evolutions , showing the different characterization of the solution in the absolutely continuous regime , on the singular cantor part , and along the jump transitions . by using tools of metric analysis , bv functions and blow - up by time rescaling , we show that this variational notion is stable with respect to a wide class of perturbations involving energies , distances , and dissipation potentials . as a particular application , we show that bv solutions to rate - independent problems arise naturally as a limit of-gradient flows , , when the exponents converge to .

1207.2725

the aim of this paper is to study the asymptotic behaviour of the solutions to a sequence of gradient flows ( in a suitable metric setting ) , when the governing energies and metric - dissipation potentials give raise in the limit to a rate - independent evolution or , more generally , to an evolution driven by a dissipation potential with linear growth . in order to explain the problem , let us start from a simple example in a finite dimensional manifold @xmath3 ( see e.g. the motivating discussion in @xcite ) . we fix a time interval @xmath4 $ ] , we denote by @xmath5 the product space @xmath6\times { \mathscr{x}}$ ] , and we consider a sequence of smooth energies @xmath7 indexed by @xmath8 . we are also given a sequence of smooth dissipation potentials @xmath9 of the form @xmath10 smoothly depending on @xmath11 and @xmath12 typical examples are @xmath13 for given initial data @xmath14 we can consider the solutions @xmath15\to { \mathscr{x}}$ ] of the cauchy problem for the doubly nonlinear differential equations @xmath16 in the parameter @xmath8 affects the limit behaviour of the initial data @xmath17 , of the energies @xmath18 in @xmath5 , of the norms @xmath19 on @xmath20 , and of the dissipation potentials @xmath21 on @xmath22 . assuming that ( in a suitable sense that we will describe later on ) @xmath23 , @xmath24 , @xmath25 , @xmath26 as @xmath27 , it is then natural to investigate if a limit curve @xmath28 ( possibly up to subsequence ) of the solutions @xmath29 still satisfies the corresponding limit equation of . we want to address here the singular situation when the limit dissipation potential @xmath30 loses the superlinear growth ; let us focus here on the @xmath2-homogeneous case when @xmath31 corresponding e.g.to @xmath32 or @xmath33 in ( in that cases @xmath34 ) . the limit problem is formally the differential inclusion @xmath35,\quad u(0)=\bar u,\ ] ] where the presence of the subdifferential @xmath36 is motivated by the lack of differentiability of the norm @xmath37 at @xmath38 . since @xmath39 is @xmath2-positively homogeneous , describes a rate - independent evolution and its solutions exhibit a different behavior with respect to the viscous flows . in particular , jumps can occur even for smooth energies @xmath40 and various kinds of solutions have been proposed in the literature ( we refer to the surveys @xcite , the overall presentation in @xcite and the references therein ) . here we focus on the notion of bv solution , proposed in @xcite : for the sake of simplicity , in this introductory section we consider the simplest case of a piecewise smooth curve @xmath28 with a finite number of jump points @xmath41 $ ] ; @xmath42 will denote the left and the right limit of @xmath28 at each @xmath43 ( see also @xcite for an explicit characterization in a one - dimensional setting ) in this case a bv solution @xmath28 can be characterized by two conditions : 1 . in each interval @xmath44 the velocity vector field @xmath45 satisfies the differential inclusion , which yields in particular the local stability condition @xmath46\setminus { { \mathrm j}}_u\ ] ] and the energy dissipation @xmath47 where @xmath48 and @xmath49 denotes the dual norm of the ( opposite ) differential of the energy , @xmath50 it turns out that in the smooth regime and are equivalent to . 2 . at each jump point @xmath43 it is possible to find an optimal transition path @xmath51\to { \mathscr{x}}$ ] , @xmath52 , such that @xmath53 , @xmath54 , @xmath55 in @xmath56 $ ] , and @xmath57 notice that the choice of the interval @xmath58 $ ] is not essential , since the integrals in are invariant with respect to monotone time rescaling . the minimum problem in characterizes the minimal transition cost at each jump point @xmath43 to connect in @xmath59 with @xmath60 passing through @xmath61 . such a cost is influenced both by the norms @xmath62 and by the slope @xmath49 of the energy : we will denote it by @xmath63 . it is a remarkable fact , highlighted in @xcite , that the refined structure given in ( bv1,bv2 ) can be captured by simply imposing the local stability condition and a single energy - dissipation inequality , namely @xmath64 it turns out that is in fact an identity , since the opposite inequality is always satisfied along _ any _ piecewise smooth curve @xmath28 . if holds , then @xmath28 is forced to satisfy along its smooth evolution , and the optimal transition paths obtained by solving the minimum problem in provide the right energy balance between @xmath42 . the link of with the gradient flow becomes more transparent if , following @xcite , one notices that also can be formulated as a energy - dissipation inequality . in fact , setting as before @xmath65 it is not difficult to check ( see the informal discussion in the next section ) that a @xmath66 curve @xmath67 with @xmath68 satisfies if and only if the @xmath21 energy - dissipation inequality holds @xmath69 where @xmath70 is the legendre transform of @xmath21 . [ [ a - more - general - formulation - in - metric - spaces . ] ] a more general formulation in metric spaces . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + here we want to show that the metric - variational approach to gradient flows and rate - independent problems provides a natural framework to study this singular perturbation problem and suggests a robust and general strategy to pass to the limit in a much more general setting where * @xmath3 is a topological space endowed with a family of complete extended distances @xmath71 , * the terms like @xmath72 are replaced by the _ metric velocity _ induced by @xmath71 , * the functions @xmath73 can be characterized as an _ irreversible couple of upper gradients _ in terms of the behaviour of the energies @xmath18 along arbitrary absolutely continuous curves with values in @xmath74 , and * @xmath30 is a general metric dissipation function with linear growth . postponing to the next two sections a more precise review of motivations and definitions , we just remark that whenever sufficiently strong a priori estimates are available to guarantee the pointwise convergence of @xmath67 to some limit function @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] , then the heart of the problem consists in deriving ( in a suitably extended form allowing countably many jumps and cantor - like terms in the metric velocity ) , starting from the viscous inequality . assuming convergence in energy of the initial data , i.e.@xmath76 , some lower - upper semicontinuity conditions on @xmath77 and @xmath78 along arbitrary sequences @xmath79 with equibounded energy and converging to @xmath80 in a fixed reference topology @xmath81 of @xmath3 are naturally suggested by the structure of the main inequalities and : @xmath82 the most challenging point is provided by the limit behaviour of the integral term @xmath83 which has been typically studied by a clever re - parametrization technique , introduced by @xcite and then extended in various directions by @xcite . this approach leads to the notion of the so - called _ parametrized solutions _ to the rate - independent evolution and the crucial assumption concerns the validity of the @xmath84-@xmath85 space - time estimate for the slopes @xmath86 in the present paper we propose a different technique , which avoids parametrized solutions and thus allows for more general non - homogenous dissipation potentials like @xmath87 our approach involves weak convergence of measures to deal with concentrations of the time derivative and blow - up around jump points of the limit solution to recover the variational structure of the transition . in this way , an easier rescaling is sufficient to construct the optimal transition paths ( see ) from the converging family @xmath29 and to obtain the bv energy - dissipation inequality . [ [ particular - cases . ] ] particular cases . + + + + + + + + + + + + + + + + + let us remark that various particular cases of the present setting are interesting by themselves and have been considered from many different points of view . a. a first important case for applications is when @xmath3 is a hilbert space , @xmath88 , and the norms @xmath89 are independent of @xmath90 and coincide with the norm @xmath91 of @xmath3 . in this case we are dealing with the convergence of gradient flows and a typical situation arises when @xmath92 . it is well known , since the pioneering contributions of @xcite , that convexity ( or @xmath93-convexity for some @xmath94 independent of @xmath90 ) of the energies makes it possible to reduce to the simpler mosco - convergence @xcite of @xmath18 ( see e.g. @xcite or @xcite for the connection with the graph - convergence of the differential operators ) . the link between @xmath84-convergence of the energies and convergence of the gradient flows in a metric setting has been considered in @xcite . b. another relevant situation is when both the energies and the distances depend on @xmath90 : in the quadratic case a convergence result can be deduced by a joint @xmath84-convergence , see e.g. @xcite . the role of the @xmath84-@xmath85 condition on the slopes as in in general non - convex setting has been clarified in @xcite . a very general stability result has been given in @xcite . an interesting example where the limit of gradient flows gives raise to a singular limit in a new geometry is discussed in @xcite . c. the particular case when the @xmath90-dependence affects only the dissipation potential @xmath30 and gives raise to a rate - independent problem in the limit has been studied in @xcite . the @xmath84-limit of rate - independent evolutions , in the framework of energetic solutions , has been studied in @xcite . [ [ plan - of - the - paper . ] ] plan of the paper . + + + + + + + + + + + + + + + + + + in the next section we give more details on the simple finite - dimensional example we introduced before , in order to motivate the abstract metric approach , whose setting is explained in section [ sec : preliminaries ] . section [ sec : main ] contains our main results , concerning compactness ( theorem [ thm : compactness ] ) and convergence ( theorem [ thm : convergence ] ) of gradient flows in a general setting . a few examples are briefly presented at the end of the paper . let @xmath3 be the finite - dimensional differentiable manifold discussed in the introduction . let us first recall that the finsler structure @xmath89 on @xmath20 allows us to define the length of a smooth curve @xmath95;{\mathscr{x}})$ ] by @xmath96:=\int_{r_0}^{r_1 } \|\dot{{\sf u}}(r)\|_{{{\sf u}}(r),h}\,{{\mathrm d}}r\ ] ] and a distance @xmath97 : { { \sf u}}\in { { \mathrm c}}^1([r_0,r_1];{\mathscr{x}}),\ { { \sf u}}(r_i)=u_i\big\},\ ] ] which still retains the information of the norms @xmath89 , since @xmath98;{\mathscr{x}}).\ ] ] the limit in can be extended to the general setting of absolutely continuous curves in metric spaces : it is denoted by @xmath99 and it is called _ metric derivative _ of the curve @xmath100 , see definition [ def : metric - derivative ] . a second crucial quantity is the dual norm of the opposite differential of the energy @xmath101 observe that the quantity in also has a nice characterization in terms of curves , since the function @xmath102 is minimal among the functions @xmath103 satisfying the chain - rule inequality @xmath104 along arbitrary curves @xmath95;{\mathscr{x}})$ ] . if one wants to allow for time variation of the energy , it is natural to introduce the partial time derivative @xmath105 , so that is in fact equivalent to @xmath106 along arbitrary regular curves @xmath107 . if we only consider nondecreasing time parametrizations @xmath108 , and we integrate along arbitrary intervals @xmath109 $ ] , we see that the map @xmath110 is maximal among all the functions @xmath111 satisfying @xmath112 in fact , let us suppose that @xmath73 are continuous functions satisfying along arbitrary regular curves with @xmath113 : it would not be difficult to check that this property is equivalent to @xmath114 if holds , we say that the couple @xmath115 is an _ irreversible upper gradient _ for the energy @xmath18 with respect to the distance @xmath71 , see definition [ def : upper - gradients ] , and @xmath116 is an upper - gradient evolution system . this definition is the natural adaptation to time - dependent functionals of the well - known notion of upper - gradient in the frame of analysis in metric spaces ( see @xcite ) ; the interesting fact is that only involves the notion of absolutely continuous curves in @xmath74 . the distinguished role of gradient flows with respect to can be easily seen by recalling the fenchel duality @xmath117 where @xmath118 is the legendre transform of @xmath30 : @xmath119 a crucial feature of fenchel duality is that for _ every _ couple @xmath120 one has the inequality @xmath121 so that in order to check the identity @xmath122 it is sufficient to prove the opposite inequality , i.e. @xmath123 taking into account these remarks and observing that we have the chain rule @xmath124 we deduce that @xmath67 solves if and only if @xmath125 since , as we already noticed in , the opposite inequality is always true , we immediately see that it is sufficient to impose the integrated version of in @xmath126 : @xmath127 now we can make the last step : instead of looking for curves satisfying , we reinforce it by replacing @xmath128 and @xmath129 with a couple @xmath130 of irreversible upper gradients satisfying . since @xmath30 and @xmath118 are nondecreasing maps and holds , it is immediate to see that if a curve @xmath131;{\mathscr{x}})$ ] satisfies the @xmath30-@xmath118 energy - dissipation inequality @xmath132 ( see also the next definition [ def : ed ] ) , then it also satisfies and by the argument above it satisfies ; moreover , along the curve we find a posteriori @xmath133 for every @xmath134 $ ] . we thus have seen that for a couple @xmath115 of irreversible upper gradients provides a natural metric definition of @xmath30-gradient flow , which can be immediately extended to a metric framework . in order to motivate the even more general definition of evolution system considered in section [ subsec : metric ] , where the power functional @xmath135 _ can also depend on a further variable @xmath136 satisfying the constraint @xmath137 _ , let us consider a non smooth situation , where @xmath40 is a _ marginal functional _ : it means that @xmath138 results from a minimization of the form @xmath139 where @xmath140 is a compact topological space and @xmath141 is a continuous function such that @xmath142 for every @xmath143 with uniformly continuous derivatives . even if each single functional @xmath144 is regular , @xmath40 is not @xmath66 in general . referring to @xcite for a more detailed discussion , we recall here that setting @xmath145 it is natural to replace the smooth differential equation with the differential inclusion in @xmath4 $ ] @xmath146 where , just for the purposes of this section , @xmath147 denotes the so - called _ marginal differential _ of @xmath18 , i.e. @xmath148 and @xmath149 is its projection onto the second component , @xmath150 if we want to differentiate the energy along a regular curve @xmath151 as in we get for a.a . @xmath152 @xmath153 where @xmath154 is an arbitrary selection in @xmath155 . setting @xmath156 it is easy to check that for every @xmath157 we have @xmath158 conversely , if a curve @xmath4\ni t\mapsto u_h$ ] satisfies the @xmath21 energy - dissipation inequality @xmath159 for a.a . @xmath160 and for some @xmath161 , we get by and @xmath162 choosing in particular a couple @xmath163 attaining the maximum in for @xmath164 , we obtain @xmath165 which eventually yields by @xmath166 so that @xmath67 solves . notice that we were able to formulate the non - smooth differential inclusion in a metric variational form by looking for a chain - rule inequality with the more general structure given by : this will be reflected in the definition [ def : upper - gradients ] of irreversible upper gradients . the differential inclusion is then characterized by the @xmath21 energy - dissipation inequality : its metric formulation will be considered in definition [ def : ed ] in the superlinear case and in definition [ def : edbv ] in the case of a metric dissipation @xmath30 with linear growth . it is then natural to investigate the stability of inequality with respect to perturbations of the parameter @xmath90 . one of the most difficult points is to guess how to state when the metric dissipation functional @xmath30 has only a linear growth , and therefore one expects a solution in @xmath167;({\mathscr{x}},{{\sf d}}))}$ ] . we have already discussed in the introduction the case of a piecewise smooth curve , but a robust theory should allow for general bv curves , possibly exhibiting countably many jumps and a metric derivative with a singular cantor part . the correct treatment of this case will be discussed in the next section . let @xmath3 be a given set ; an _ extended distance _ on @xmath3 is a map @xmath168 $ ] satisfying @xmath169 we say that @xmath170 is an extended metric space . most of the definitions concerning metric spaces generalize verbatim to extended metric spaces , in particular it makes perfectly sense to speak about a complete extended metric space . let @xmath170 be an extended metric space . [ def : metric - derivative ] we say that a curve @xmath171\to { \mathscr{x}}$ ] is absolutely continuous ( a.c . for short ) and belongs to @xmath172 if there exists @xmath173 such that @xmath174 if @xmath175 then the limit @xmath176 it satisfies @xmath177 @xmath178-a.e . in @xmath179 , belongs to @xmath180 , and it is called _ metric derivative of @xmath28 ; _ @xmath181 provides the minimal function @xmath182 such that holds . the ( pointwise ) @xmath183-variation of @xmath171\to { \mathscr{x}}$ ] in an interval @xmath184\subset [ a , b]$ ] is defined by @xmath185):=\sup\big\{\sum_{j=1}^n { { \sf d}}(u(t_j),u(t_{j-1})):\alpha = t_0<\cdots < t_n = b\big\}.\ ] ] we say that @xmath186;({\mathscr{x}},{{\sf d}}))}$ ] if @xmath187)<\infty$ ] and @xmath28 takes values in a complete subset of @xmath170 ; in this case , @xmath28 admits left and right limits ( denoted by @xmath188 and @xmath189 ) at every point of @xmath190 $ ] and we adopt the convention to extend @xmath28 to @xmath191 $ ] by setting @xmath192 the _ pointwise jump set _ and the _ essential jump set _ of @xmath28 are defined by @xmath193:u(t)\neq u(t_-)\ \text{or } u(t)\neq u(t_+)\big\ } \\ { \mathop{{\rm ess}\text{-}{\rm j}}\nolimits}_u:=&\big\{t\in [ a , b]:u(t_-)\neq u(t_+)\big\ } , \end{aligned}\ ] ] and satisfy the obvious inclusion @xmath194 . if @xmath186;({\mathscr{x}},{{\sf d}}))}$ ] we denote by @xmath195 the bounded monotone function @xmath196)&\text{if } t\in [ a , b],\\ { \mathrm{var}}_{{\sf d}}(u;[a , b])&\text{if } t > b , \end{cases}\ ] ] and by @xmath197 its distributional derivative : @xmath197 is a finite measure in @xmath198 supported in @xmath190 $ ] , and we can decompose it as the sum of a diffuse part and a jump part @xmath199 where @xmath200 denotes the restriction of a measure to a borel set ; thus @xmath201 is concentrated on the ( at most ) countable jump set @xmath202 and @xmath203 the lebesgue decomposition of the diffuse part @xmath204 can be written as @xmath205 we obtain @xmath206)&=\int_\alpha^\beta { { \mathrm d}}\,{|u'|_{{{\sf d}}}}+{\mathrm{jmp}_{{{\sf d}}}}(u;[\alpha,\beta])\\&= \int_\alpha^\beta |\dot u|_{{\sf d}}(t ) \,{{\mathrm d}}t+ \int_\alpha^\beta { { \mathrm d}}\,{|{{\mathrm c}}u|_{{{\sf d}}}}+{\mathrm{jmp}_{{{\sf d}}}}(u;[\alpha,\beta])\end{aligned}\ ] ] where for every subinterval @xmath184\subset [ a , b]$ ] @xmath207):={{\sf d}}(u(\alpha),u(\alpha_+))+\sum_{t\in { { \mathrm j}}_u\cap ( a , b)}{|{{\mathrm j}}u|_{{{\sf d}}}}(\{t\})+{{\sf d}}(u(\beta_-),u(\beta)).\ ] ] let @xmath170 be a complete extended metric space and @xmath4 $ ] a fixed time interval of @xmath198 . we denote by @xmath5 the product space @xmath4\times { \mathscr{x}}$ ] and we say that an a.c . curve @xmath208\to { \mathscr{q}}$ ] is _ time - ordered _ if @xmath209 is non decreasing . if @xmath210 is some interval of @xmath198 , @xmath211 ( resp . @xmath212 ) will denote the collections of borel ( resp . @xmath178-measurable ) maps defined in @xmath210 with values in @xmath213 $ ] . we say that a map @xmath214 is measurable along time - ordered a.c . curves if for every time - ordered a.c . curve @xmath215 in @xmath5 the composition @xmath216 is lebesgue measurable . we denote by @xmath217 the collection of all such functions . an _ evolution system _ @xmath218 consists of 1 . a complete extended metric space @xmath170 , 2 . an energy functional @xmath219 in @xmath220 , 3 . a slope functional @xmath221 $ ] in @xmath217 , 4 . a power functional @xmath222\to \overline { \mathbb{r}}$ ] such that for every @xmath223 the map @xmath224 belongs to @xmath217 and the map @xmath225 is nondecreasing and upper semicontinuous . notice that if @xmath208\to{\mathscr{q}}$ ] is a time - ordered a.c.curve and @xmath226)$ ] , the composition @xmath227 is measurable . the essential feature of this structure is captured by the following definition : [ def : upper - gradients ] we say that @xmath218 is an ( irreversible ) upper gradient system if for every time - ordered a.c . curve @xmath184\ni s\mapsto { { \sf q}}(s)=({{\sf t}}(s),{{\sf u}}(s))\in { \mathscr{q}}$ ] and every @xmath226)$ ] satisfying @xmath228,\quad \int_\alpha^\beta \big[{{\ensuremath{\mathcal p}}}({{\sf q}}(s),{{\sf f}}(s))\big]_-\dot{{\sf t}}(s)\,{{\mathrm d}}s<\infty\ ] ] there holds @xmath229 a _ metric dissipation function _ is a function @xmath230 we say that @xmath30 has @xmath231-linear growth if @xmath232 and that @xmath30 is _ superlinear _ if @xmath233 . its dual @xmath234 $ ] is defined as @xmath235 and it is a convex and superlinear function with @xmath236 as well , satisfying the fenchel duality @xmath237 where @xmath238 $ ] denotes the convex subdifferential of @xmath30 . notice that at @xmath239 we have @xmath240\ ] ] so that @xmath241 is single valued only when the right derivative of @xmath30 at @xmath242 vanishes . the proper domain @xmath243 is related to @xmath231 by the relation @xmath244 so that @xmath118 is finite in @xmath22 if and only if @xmath30 is superlinear . the typical examples are @xmath245 @xmath246 the @xmath30-gradient flows associated with an evolution system can be characterized by a simple family of dissipation inequalities : [ def : ed ] let @xmath218 be an evolution system and let @xmath30 be a metric dissipation function . + a curve @xmath247;({\mathscr{x}},{{\sf d}}))}$ ] with @xmath248 satisfies the @xmath30-@xmath118 energy - dissipation inequality if there exists a measurable map @xmath249)$ ] satisfying @xmath250$},\qquad \int_0^t \big[{{\ensuremath{\mathcal p}}}(t , u(t),{{\sf f}}(t))\big]_+\,{{\mathrm d}}t<\infty,\ ] ] and for every @xmath134 $ ] @xmath251 it is immediate to see that if @xmath218 is an _ upper gradient _ evolution system , then by and the integral characterization is equivalent to the following properties : [ eq : sub1 ] @xmath252 $ , } \\ & \label{eq:6bis } { { \sf f}}(t)\ge { { \ensuremath{\mathcal f}}}(t , u(t))\quad\text{a.a.\ $ t$ in } ( 0,t ) , \\ & \label{eq:7 } \begin{aligned } -\frac{{\mathrm d}}{{{\mathrm d}}t}{{\ensuremath{\mathcal e}}}(t , u(t))&+{{\ensuremath{\mathcal p}}}(t , u(t),{{\sf f}}(t))=|\dot u|_{{\sf d}}(t)\,{{\sf f}}(t)\\&=\psi\big(|\dot u|_{{\sf d}}(t)\big)+\psi^*\big({{\sf f}}(t)\big ) \end{aligned } \qquad{\text{for a.e.}}\ t\in ( 0,t ) . \end{aligned}\ ] ] notice that and yields the velocity - slope relation @xmath253 and , by , @xmath254 realizes the minimal selection property @xmath255 for a.a . @xmath160 . in particular , yields @xmath256 and holds @xmath178-a.e . when @xmath257 . [ def : gf ] let @xmath218 be an upper gradient evolution system , and let @xmath30 be a metric dissipation function as in . a curve @xmath175 is a @xmath30-gradient flow of the system if it satisfies and at @xmath258 , or , equivalently , ( [ eq : sub1]a , b , c ) . let us now consider the case of a dissipation potential @xmath30 with linear growth , corresponding to @xmath259 in . in this case , absolutely continuous solutions to ( [ eq : sub1]a , b , c ) often do not exist , even in the smooth and finite - dimensional setting of section [ sec : intro ] and therefore we have to extend the previous definitions to the bv setting . as before , we fix the time interval @xmath4 $ ] and we denote by @xmath5 the product space @xmath4\times { \mathscr{x}}$ ] and we consider a function @xmath260 $ ] measurable along absolutely continuous curves . relevant examples will be @xmath261 and @xmath262 we interpret @xmath263 as a conformal factor that induces a modified geometry in @xmath3 : the corresponding length of a curve @xmath264 ( notice that the curve is parametrized by a different variable @xmath152 , and @xmath265 remains fixed ) is @xmath266}:=\int_{r_0}^{r_1 } { { { \frak f}}}(t,\vartheta(r))\,|\dot\vartheta(r)|_{{\sf d}}\,{{\mathrm d}}r\ ] ] and the cost of a transition from @xmath267 to @xmath268 in @xmath3 at the time @xmath134 $ ] is then defined by @xmath269}:\vartheta\in { \mathrm{ac}^{}(r_0,r_1;({\mathscr{x}},{{\sf d}}))},\ \vartheta(r_i)=u_i\big\}.\ ] ] we also set @xmath270}:\vartheta\in { \mathrm{ac}^{}(r_0,r_1;({\mathscr{x}},{{\sf d}}))},\\ & \vartheta(r_i)=u_i,\ \vartheta(r)=u\text { for some } r\in [ r_0,r_1]\big\}. \end{aligned } \end{aligned}\ ] ] we can thus consider a modified jump functional @xmath271)&= { \delta_{{{\sf d}},{{{\frak f}}},\alpha}(u(\alpha),u(\alpha_+))}\\&+\sum_{t\in { { \mathrm j}}_u}{\delta_{{{\sf d}},{{{\frak f}}},t}(u(t_-),u(t),u(t_+))}+ { \delta_{{{\sf d}},{{{\frak f}}},\beta}(u(\beta_-),u(\beta))}.\end{aligned}\ ] ] the previous quantities will be quite useful to extend the chain - rule inequality to the bv setting . notice that we are assuming that @xmath136 is a borel map ( instead of lebesgue measurable as in definition [ def : upper - gradients ] ) , since an integration with respect to the possibly singular measure @xmath204 occurs in . [ prop : bv - chain - rule ] let @xmath218 be an upper gradient evolution system , @xmath272 for some @xmath273 , and let @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] satisfy @xmath274)<\infty,\ ] ] for some borel map @xmath275)$ ] with @xmath276 in @xmath4 $ ] . then for every @xmath134 $ ] @xmath277 ) . \end{aligned}\ ] ] it is not restrictive to assume @xmath258 . let us denote by @xmath278 the jump set @xmath202 of @xmath28 and let us first set @xmath279\setminus i,\quad { { \sf t}}:={{\sf s}}^{-1 } : d \to [ 0,t],\quad { { \sf u}}:=u\circ{{\sf t}}:d\to { \mathscr{x}}. \end{gathered}\ ] ] since @xmath280)<\infty $ ] it is not difficult to check that @xmath281 are lipschitz maps ( if we only know @xmath282)<\infty$ ] , it would not clear how to derive a uniform upper bound on the total variation of the function @xmath100 ) . we easily extend @xmath209 to @xmath283 $ ] by setting @xmath284 in order to extend @xmath100 , we fix @xmath285 and for every interval @xmath286 we consider two curves @xmath287\to { \mathscr{x}}$ ] satisfying @xmath288 , taking the value @xmath289 at some point in @xmath286 , and fulfilling @xmath290 for @xmath291 we define @xmath292\setminus i,\\ \vartheta_n(s)&\text{if } s\in i_n,\ n\le n,\\ \zeta_n(s)&\text{if } s\in i_n,\ n > n , \end{cases } \\ { { \sf f}}_n(s)&:= \begin{cases } { { \sf f}}({{\sf t}}(s))&\text{if } s\in [ 0,{{\sf s}}]\setminus i,\\ { { \ensuremath{\mathcal f}}}(t_n,\vartheta_n(s))&\text{if } s\in i_n,\ n\le n,\\ { { \frak f}}(t_n,\zeta_n(s))&\text{if } s\in i_n,\ n > n . \end{cases } \end{aligned}\ ] ] it is not difficult to check that @xmath293 is absolutely continuous , so that yields ( see ( * ? ? ? * lemma 4.1 ) ) @xmath294 passing first to the limit as @xmath295 ( notice that the last term vanishes as @xmath295 since @xmath296)$ ] is finite ) and then as @xmath297 we obtain . [ def : edbv ] let + @xmath218 be an evolution system in the time interval @xmath4 $ ] , let @xmath30 be a metric dissipation function with @xmath231-linear growth , and let @xmath272 . a curve @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] with @xmath248 satisfies the @xmath30-@xmath118 energy dissipation inequality if there exists a borel map @xmath249)$ ] satisfying , @xmath298,\quad \int_0^t \big[{{\ensuremath{\mathcal p}}}(t , u(t),{{\sf f}}(t))\big]_+\,{{\mathrm d}}t<\infty,\ ] ] the _ stability condition on the cantor part _ @xmath299$}\ ] ] and @xmath300 ) \notag\\&\le { { \ensuremath{\mathcal e}}}(0,u(0))+ \int_{0}^{t } { { \ensuremath{\mathcal p}}}(r , u(r),{{\sf f}}(r))\,{{\mathrm d}}r \quad{\text{for every } } t\in [ 0,t ] . \label{eq:5bv } \tag{ed } \end{aligned}\ ] ] since @xmath301 if @xmath302 , yields in fact a stronger version of the local stability condition @xmath303.$ } \tag{$\mathrm s_{\rm loc}$}\ ] ] in the rate - independent case @xmath304 , and are thus equivalent to and @xmath305 ) \\&\le { { \ensuremath{\mathcal e}}}(0,u(0))+ \int_{0}^{t } { { \ensuremath{\mathcal p}}}(r , u(r),{{\sf f}}(r))\,{{\mathrm d}}r \end{aligned}\ ] ] for every @xmath134 $ ] . [ def : bvsol ] let @xmath218 be an upper gradient system in the time interval @xmath4 $ ] , let @xmath30 be a metric dissipation function with @xmath231-linear growth , and let @xmath272 . + a curve @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] with @xmath248 is a @xmath306 _ solution _ of the corresponding evolution if there exists @xmath275)$ ] satisfying , the _ local stability condition _ and the _ energy balance _ @xmath307)\\&= { { \ensuremath{\mathcal e}}}(t_1,u(t_1))+ \int_{t_1}^{t_2 } { { \ensuremath{\mathcal p}}}(r , u(r),{{\sf f}}(r))\,{{\mathrm d}}r \label{eq:22 } \tag{eb } \end{aligned}\ ] ] for every @xmath308\subset [ 0,t].$ ] notice that holds if and only if the curve @xmath309 ( extended to @xmath198 as in ) is of bounded variation , @xmath310 , and its distributional time derivative @xmath311 satisfies @xmath312 where at each jump point @xmath313 we have @xmath314 and the jump part @xmath315 is @xmath316 as for gradient flows , thanks to proposition [ prop : bv - chain - rule ] it is immediate to see that whenever holds the energy balance is equivalent to the energy - dissipation inequality at the final point @xmath258 . moreover , a bv solution @xmath28 satisfies @xmath317 and the minimal selection principle @xmath318 for @xmath178-a.a . @xmath160 . in the rate - independent case @xmath304 , a bv solution is equivalently characterized by the local stability and the energy balance @xmath319 ) \\&= { { \ensuremath{\mathcal e}}}(t_1,u(t_1))+ \int_{t_1}^{t_2 } { { \ensuremath{\mathcal p}}}(r , u(r),{{\sf f}}(r))\,{{\mathrm d}}r \end{aligned}\ ] ] for every @xmath320 . in this section we will state and prove our main results . for the sake of clarity , we distinguish between the compactness ( theorem [ thm : compactness ] ) and the stability ( theorem [ thm : convergence ] ) issues . we also take care to highlight the role of the energy - dissipation inequality for general metric - evolution systems @xmath116 , even if they do not satisfy the irreversible upper gradient condition [ def : upper - gradients ] . therefore , compactness and stability of the energy - dissipation inequality always hold whenever suitable topological properties ( see the next ( c1,2,3,4 ) assumptions ) are satisfied . we also notice that our theorems can also be extremely useful to prove existence results for solutions to the limit evolution system : in this case one could think that @xmath116 is a family of suitably regularized problems ( e.g. with smooth superlinear dissipations and better energies ) for which existence is already known ( see @xcite ) . let @xmath321 be a topological space , and let @xmath6\times{\mathscr{x}}$ ] with the standard product topology , which we will denote by @xmath322 . we consider a family of evolution systems @xmath116 in the time interval @xmath4 $ ] indexed by the parameter @xmath323 , a sequence @xmath17 of initial points , and metric dissipation functions @xmath324 such that @xmath325 since each function @xmath21 is monotone , is equivalent to @xmath326 and also to the following property , valid for arbitrary sequences @xmath327 : @xmath328 typical examples are given in and . we want to study the limit of absolutely continuous @xmath21-gradient flows @xmath329 of the systems @xmath116 with @xmath68 as @xmath27 , assuming that they `` converge '' ( in a variational sense that we are going to make precise ) to a limit system @xmath218 . the most interesting case is when @xmath30 has @xmath231-linear growth , so that we expect a function of bounded variation in the limit . * there exist constants @xmath335 such that @xmath336 * the energies @xmath337 are equi - coercive in @xmath5 : for every sequence @xmath338 with @xmath339 there exists a subsequence @xmath340 such that @xmath341 in the @xmath322-topology . * if two sequences @xmath342 , @xmath343 , @xmath344 , satisfy @xmath345 and @xmath322-converge to @xmath346 , then we have @xmath347 where @xmath348 is a complete extended distance on @xmath3 . [ thm : compactness ] let us suppose that ( c1 ) holds and let @xmath349 , @xmath350)$ ] be sequences satisfying @xmath68 and the corresponding energy - dissipation inequalities and for every @xmath8 , namely @xmath351$},\qquad \int_0^t \big[{{\ensuremath{\mathcal p}}}_h(t , u_h(t),{{\sf f}}_h(t))\big]_+\,{{\mathrm d}}t<\infty,\ ] ] and @xmath352 for every @xmath134 $ ] . if there exists a constant @xmath353 such that @xmath354,\ ] ] then there exists a constant @xmath355 such that for every @xmath134 $ ] and @xmath8 @xmath356 so that @xmath357 if moreover ( c2,3 ) hold , then for every subsequence @xmath331 there exists a further subsequence @xmath340 such that @xmath358 $ , } \label{eq:68}\ ] ] with @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] , @xmath359\ni t_h\to t\in [ 0,t]\setminus{{\mathrm j}}_u$},\ ] ] and , for every interval @xmath308\subset [ 0,t]$ ] , @xmath360)\ ] ] + first of all we show that @xmath361 is uniformly bounded , we choose @xmath362 , and we observe that , the monotonicity of @xmath21 , and the continuity of @xmath118 in @xmath363 yield @xmath364 so that @xmath365 . it follows that @xmath366 and therefore and yield @xmath367 so that @xmath368,\ h\in { \mathbb{n}}.\ ] ] combining ( c1 ) , and we conclude that there exists a constant @xmath369 such that @xmath370,\ h\in { \mathbb{n}}.\ ] ] therefore and yield @xmath371 we eventually obtain , since by the monotonicity of @xmath70 we get from the second of @xmath372 and @xmath373 when @xmath302 by and . by ( c2 ) and the bound , for every @xmath134 $ ] the sequence @xmath374 admits a @xmath81-converging subsequence ( possibly depending on @xmath265 ) . for every @xmath375 and @xmath376 we recall the bound @xmath377 we consider the nonnegative finite measures on @xmath4 $ ] @xmath378 on @xmath4 $ ] ; up to extracting a suitable subsequence , we can suppose that they weakly@xmath379 converge to a finite measure @xmath380 in the duality with continuous functions on @xmath4 $ ] , so that @xmath381)\le \nu_f([t_1,t_2])\ ] ] for every @xmath382 . denoting by @xmath383:\nu_0(\{t\})>0\}$ ] and considering a countable set @xmath384 dense in @xmath4 $ ] , by a standard diagonal argument we can find a subsequence @xmath340 such that @xmath385 for every @xmath386 as @xmath27 , @xmath387 . by ( c3 ) we have @xmath388)\quad{\text{for every } } t_1,t_2\in i.\ ] ] since @xmath170 is complete , the curve @xmath389 can be uniquely extended to a continuous curve in @xmath4\setminus j$ ] , which we still denote by @xmath28 . in order to prove we argue by contradiction and we find a sequence @xmath390 , points @xmath391\setminus j$ ] and a @xmath81-neighborhood @xmath392 of @xmath393 such that @xmath394 for every @xmath395 . up to extracting a further subsequence ( still denoted by @xmath396 ) we can assume that @xmath397 so that by ( c3 ) @xmath398)=\nu_f(\{t\})=0.\ ] ] this yields in particular that @xmath399 converges pointwise to @xmath393 as @xmath27 , @xmath387 ; then holds for every @xmath400 $ ] and shows that @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] . since for an arbitrary subdivision @xmath401 there holds @xmath402 it is easy to check that @xmath403)+\psi^*(f)(\beta-\alpha),\ ] ] so that @xmath404)\le \nu_0([\alpha,\beta ] ) + \psi^*(f)(\beta-\alpha).\ ] ] therefore , the duality formula @xmath405 yields @xmath406)\big)\le ( \beta-\alpha)^{-1}\nu_0([\alpha,\beta ] ) .\ ] ] from we immediately get @xmath407 since @xmath408 and @xmath201 are concentrated in a @xmath178-negligible set , we deduce @xmath409 when @xmath30 is superlinear we conclude that @xmath28 is absolutely continuous , since in this case @xmath410 and yields @xmath411 . + and the monotonicity of @xmath30 yield , for @xmath412 and @xmath413 @xmath414).\ ] ] integrating this inequality from @xmath415 to @xmath416 with respect to @xmath265 we obtain @xmath417)\,{{\mathrm d}}t\\&\le { \varepsilon}^{-1}({{\mathscr l}^{1}}\times\nu_0)(\{(t , s)\in [ t_0,t_1]^2:t\le s\le t+{\varepsilon}\ } ) \\&\le { \varepsilon}^{-1}\int_{t_0}^{t_1 } { { \mathscr l}^{1}}([s-{\varepsilon},s])\,{{\mathrm d}}\nu_0(s)=\nu_0([t_0,t_1])\end{aligned}\ ] ] so that , passing to the limit as @xmath297 in the above inequality and applying fatou s lemma and lebesgue s differentiation theorem we get @xmath418).\ ] ] since @xmath415 and @xmath419 are arbitrary , we conclude that @xmath420 . since @xmath178 is singular with respect to @xmath408 and @xmath201 we eventually get @xmath421 which in particular yields , since @xmath422 and @xmath178 and @xmath408 are diffuse . notice that in we allow for an @xmath90-dependence of @xmath265 in the @xmath84-@xmath85 inequality for @xmath49 , whereas @xmath265 is independent of @xmath90 in . will not be required for the convergence result in the superlinear case , see theorem [ thm : convergence2 ] . the next statement is the main result of our paper : it states that the energy - dissipation inequality is preserved in the limit for arbitrary evolution systems fulfilling ( c1,2,3,4 ) . when the limit @xmath218 is also an upper gradient evolution system , then we recover a bv solution . [ thm : convergence ] let us assume that @xmath30 has @xmath231-linear growth and for @xmath323 let @xmath116 be a family of evolution systems satisfying ( c1,2,3,4 ) with respect to a sequence @xmath14 @xmath81-converging to @xmath428 . let @xmath329 , @xmath350)$ ] , @xmath343 , be sequences satisfying the @xmath21-energy dissipation inequality , such that @xmath67 is pointwise converging to @xmath28 in @xmath429 , @xmath430 , and @xmath431_+ \text { are equi - integrable in $ ( 0,t)$}. \end{aligned}\ ] ] then @xmath75;({\mathscr{x}},{{\sf d}}))}$ ] satisfies the local stability condition @xmath432\setminus{{\mathrm j}}_u$}.\ ] ] if @xmath433 then @xmath28 satisfies the bv energy - dissipation inequality in the formulation of , for some @xmath275)$ ] . in particular , if @xmath218 is an upper gradient evolution system , then @xmath28 satisfies and therefore is a @xmath306-solution of the corresponding rate - independent evolution and we have @xmath434 . \end{gathered}\ ] ] we consider the function @xmath438\to { \mathbb{r}}\cup\{\infty\}$ ] @xmath439 denoting by @xmath440 the lebesgue - measurable subsets of @xmath441 and by @xmath442 the borel sets of @xmath443 , it is easy to check that @xmath444 is @xmath445-measurable , thanks to the monotonicity and upper semicontinuity of @xmath446 . by , it is the immediate to see that for a.a . @xmath447 @xmath448 applying ( * ? ? ? * lemma iii.39 ) we find a lebesgue measurable map @xmath449 such that @xmath450 and , up to a modification of @xmath136 in a negligible set , it is not restrictive to assume @xmath136 borel and @xmath451 on a borel set containing @xmath202 and where @xmath408 is concentrated . setting @xmath452 , by is immediate to see that @xmath453 since and ( c1 ) also yield @xmath454 fatou s lemma and imply @xmath455 and the inequality @xmath456 shows that @xmath457 . and fatou s lemma then yield @xmath458 recalling , we consider now the measures @xmath459 up to extracting a further subsequence , it is not restrictive to assume that @xmath460 @xmath461 in the duality with continuous functions . fatou s lemma and the previous arguments easily imply @xmath462 since @xmath463)=\mu([0,t])$ ] , combining and inequality for @xmath258 follows if we show that @xmath464)&\ge \int_{0}^{t } \big(\psi\big(|\dot u|_{{\sf d}}\big)+\psi^*({{\sf f } } ) -{{\ensuremath{\mathcal p}}}(\cdot , u,{{\sf f}})\big ) \,\,{{\mathrm d}}r \\&+{{\sf l}}\int_0^t { { \mathrm d}}{|{{\mathrm c}}u|_{{{\sf d}}}}+{\text{\slshape jmp}_{{{\sf d}},{{\frak f}}}}(u;[0,t ] ) . \end{aligned}\ ] ] now , and imply that @xmath465 since the atomic and the diffuse part of a measure are mutually singular , ensues if for every @xmath134 $ ] with @xmath466 we have @xmath467 in order to prove we just take two sequences @xmath468 , @xmath8 , converging monotonically to @xmath265 such that @xmath469 and we set @xmath470 since @xmath471 and @xmath472 by , taking into account the definition of @xmath473 we obtain @xmath474)\le \mu(\{t\}),\ ] ] and up to extracting a suitable subsequence we can assume that @xmath475 as @xmath27 . we denote by @xmath476 the inverse map of @xmath477 : @xmath478 is @xmath2-lipschitz , monotone , and maps @xmath479 $ ] onto @xmath480 $ ] . we also set @xmath481,\\ u_h(r_h^+)&\text{if } s\ge { { \sf s}}_h^+,\\ u_h(r_h^-)&\text{if } s\le { { \sf s}}_h^- , \end{cases}\ ] ] so that , in particular , we have @xmath482 we observe that @xmath483 and that the functions @xmath484 are uniformly @xmath71-lipschitz : to show this fact , we choose @xmath485 in such a way that @xmath486 ; the inequality @xmath487 yields @xmath488 so that @xmath489 whence the uniform @xmath71-lipschitz continuity of @xmath484 . by arguing as in the proof of theorem [ thm : compactness ] , we can find a compact interval @xmath210 containing all the intervals @xmath479 $ ] and a further subsequence such that @xmath490 for every @xmath491 : it follows from that @xmath492 is @xmath493-lipschitz with respect to @xmath183 , @xmath494 , @xmath495 , and @xmath496 moreover , by using the elementary inequality @xmath497 and recalling that @xmath498 , @xmath499 , and , we have @xmath500 ) \\&{\stackrel{\eqref{eq:144}}=}\limsup_{h\in h}\int_{r_h^-}^{r_h^+}\big(\psi_h\big(|\dot u_h|_{{{\sf d}}_h}(\tau)\big)+\psi^*_h\big({{\ensuremath{\mathcal f}}}_h(\tau , u_h(\tau))\big)\big)\,{{\mathrm d}}\tau\\\notag & { \stackrel{\eqref{eq:85}}\ge } \liminf_{h\in h}\int_{r_h^-}^{r_h^+ } \big(({{\ensuremath{\mathcal f}}}_h(\tau , u_h(\tau))\lor a ) |\dot u_h|_{{{\sf d}}_h}(\tau)-\psi_h^*(a)\big)\,{{\mathrm d}}\tau \\&\notag \ge \liminf_{h\in h}\int_{{{\sf s}}_h^-}^{{{\sf s}}_h^+ } \big(({{\ensuremath{\mathcal f}}}_h({{\sf r}}_h(s),\vartheta_h(s))\lor a ) |\dot \vartheta_h|_{{{\sf d}}_h}(s)\big)\,{{\mathrm d}}s -(r_h^+-r_h^-)\psi_h^*(a ) \\\notag & \ge \liminf_{h\in h}\int_{i } \big(({{\ensuremath{\mathcal f}}}_h({{\sf r}}_h(s),\vartheta_h(s))\lor a ) |\dot \vartheta_h|_{{{\sf d}}_h}(s)\big)\,{{\mathrm d}}s \\\label{eq:145 } & \ge \int_{i } \big(({{\ensuremath{\mathcal f}}}(t,\vartheta(s))\lor a ) m(s)\big)\,{{\mathrm d}}s \\\notag & \ge \int_{{{\sf s}}^-}^{{{\sf s}}^+ } \big(({{\ensuremath{\mathcal f}}}(t,\vartheta(s))\lor a ) |\dot\vartheta|_{{\sf d}}(s)\big)\,{{\mathrm d}}s \ge { \delta_{{{\sf d}},{{\frak f}},t}(u(t_-),u(t),u(t_+))},\end{aligned}\ ] ] viz . the desired . for the last @xmath85 inequality in we used a result proved in the next lemma . let us set @xmath504 , @xmath505 . since @xmath506 for every @xmath507 and @xmath508 we have for every @xmath505 @xmath509 on the other hand , for @xmath178-a.a . @xmath491 , @xmath510 is a nondecreasing sequence converging to @xmath511 , so that by monotone convergence @xmath512 in the case of a limit metric dissipation function @xmath30 with superlinear growth we have a completely analogous result , which can be compared with ( * ? ? ? * theorem 4.8 ) . we omit the similar proof : it can be carried out without assuming , which has been used only to characterize the contribution of jump part in the energy dissipation inequality . [ thm : convergence2 ] let us assume that @xmath30 is superlinear and for @xmath323 let @xmath116 be a family of evolution systems satisfying ( c1,2,3 ) and ( c4@xmath513 ) with respect to a sequence @xmath14 @xmath81-converging to @xmath428 and let @xmath218 be an upper gradient evolution system . + let @xmath329 , @xmath350)$ ] , @xmath343 , be sequences of curves satisfying the @xmath21-energy dissipation inequality , pointwise converging to @xmath28 in @xmath429 , with @xmath430 , and satisfying . then @xmath514 is a @xmath30-gradient flow of the limit system and @xmath515 . \end{gathered}\ ] ] let @xmath516 , @xmath517 be a family of banach spaces such that @xmath518 with continuous and dense inclusions . setting @xmath519 if @xmath520 we suppose that @xmath521 . let @xmath522 $ ] be a proper , @xmath93-convex functional , i.e. satisfying for every @xmath523 @xmath524.\ ] ] we consider a time - dependent functional @xmath525;{\mathscr{x}}_0')$ ] , we suppose that @xmath526 has compact sublevels on @xmath527 , and we extend it to @xmath3 by setting @xmath528 if @xmath529 . we set @xmath530 and @xmath531 where @xmath532 is the frecht subdifferential of the restriction of @xmath40 in @xmath533 . notice that @xmath534 and it is not difficult to check ( see @xcite ) that @xmath535 and @xmath218 are upper gradient evolution systems and all the assumptions ( c1,2,3,4 ) are satisfied . here is a concrete example of the above setting . consider a bounded open set @xmath536 , a function @xmath537 with @xmath538 @xmath539 , and @xmath540;l^2(\omega))$ ] . in the space @xmath541 endowed with the strong @xmath542-topology we set @xmath543 we choose a sequence of exponents @xmath544 converging to @xmath2 as @xmath27 , and initial data @xmath545 with @xmath546 strongly converging to @xmath428 in @xmath547 with @xmath548 in @xmath549 . we let @xmath71 be the distance induced by the @xmath550 norm , @xmath183 the @xmath549-distance , @xmath551 , and @xmath552 with @xmath553 if @xmath554 ; @xmath49 has the analogous expression in @xmath555 . applying the results of @xcite we see that @xmath535 and @xmath218 are upper gradient evolution systems and for every @xmath8 there exists a solution @xmath67 of the @xmath21-gradient flow . it is also easy to check that all the assumptions ( c1,2,3,4 ) are satisfied so that , up to subsequences , @xmath556 converge to @xmath557 in @xmath558 at every time @xmath265 with convergence of the energies @xmath559 and @xmath28 is a bv solution of the rate - independent evolution governed by @xmath218 . in the finite dimensional setting described in section [ sec : smooth ] ( here we also assume that the norms @xmath560 are independent of @xmath90 ) , let us consider the marginal functional and the couple @xmath561 given by and . it is not difficult to see that @xmath218 is an upper gradient evolution system . @xmath21-gradient flows in the superlinear case can be obtained by applying the results of @xcite : they in particular solve the differential inclusions . existence of a bv solution and convergence of the @xmath21 gradient flows can thus be obtained by applying theorems [ thm : compactness ] and [ thm : convergence ] . , _ oprateurs maximaux monotones et semi - groupes de contractions dans les espaces de hilbert _ , north - holland publishing co. , amsterdam , 1973 . north - holland mathematics studies , no . 5 . notas de matemtica ( 50 ) . , _ evolution in rate - independent systems ( ch . 6 ) _ , in handbook of differential equations , evolutionary equations , vol . 2 , c. dafermos and e. feireisl , eds . , elsevier b.v . , amsterdam , 2005 , pp . 461559 . height 2pt depth -1.6pt width 23pt , _ differential , energetic and metric formulations for rate - independent processes ( ch . 3 ) _ , in nonlinear pdes and applications . c.i.m.e . summer school , cetraro , italy 2008 , l. ambrosio and g. savar , eds . , springer , heidelberg , 2011 , pp . 87170 . height 2pt depth -1.6pt width 23pt , _ nonsmooth analysis of doubly nonlinear equations _ , partial differential equations , ( 2012 ) . doi : 10.1007/s00526 - 011 - 0482-z . in press . published online since january 2012 . wias preprint 1613 . , _ variational convergence of nonlinear diffusion equations : applications to concentrated capacity problems with change of phase _ , atti accad . lincei cl . fis . mat . ( 9 ) mat . appl . , 8 ( 1997 ) ,

we have obtained _ far ultraviolet spectroscopic explorer _ ( 9051187 ) spectra of the non - magnetic cataclysmic variable z cam during a period of quiescence . the spectrum resembles that of a hot ( 57,000k ) metal - enriched white dwarf . a high effective temperature is consistent with the expectation that dwarf nova systems that show standstills , as z cam does , have higher than normal time - averaged mass accretion rates . it is also consistent with current estimates for the mass and distance to z cam . a white dwarf model in which 29% of the surface has a temperature of 72,000 k and the rest of the surface is at 26,000k also reproduces the spectral shape and the continuum flux level at the nominal distance and wd mass , and is a somewhat better statistical fit to the data . a second component could be due to a rotating accretion belt , a remnant of an outburst that ended eleven days prior to the observation , or ongoing accretion . we favor the uniform temperature model for z cam , largely because the data do not require anything more complicated . we have compared the quiescent spectrum with an archival spectrum of z cam in standstill , obtained with . the standstill spectrum is described well by a disk accreting at = , where the errors depend on the assumed inclination . the quiescent spectra cover a full orbital period and are time - resolved tos . no variability was observed in the continuum during the observation , but the depth of many of the absorption lines increased between orbital phases 0.65 and 0.81 . we attribute this effect to absorption by material associated with the accretion stream , which is easier to understand if the inclination is near the maximum allowed value of 68 .

astro-ph0412501

cataclysmic variables ( cvs ) are interacting binary systems in which a white dwarf ( wd ) accretes matter from a low - mass companion . in non - magnetic cvs , the in - falling matter forms a disk around the wd . dwarf novae ( dne ) are disk - accreting cvs that undergo sharp rises in luminosity , during which the optical brightness typically increases by 2 to 5mag . the outbursts occur when a sudden increase in the gas viscosity ( due to a change in the ionization state of the gas ) allows matter to flow more rapidly through the disk , causing an increase in the accretion luminosity . in most dne , outbursts last for a day to a few weeks . in the subclass of dne known as u gem systems , outbursts end with a steady decline to minimum that has a typical timescale of a day however , in z cam systems , the decline from outburst occasionally stalls about 1mag below the peak . such ` standstills ' can last for days to , in a few cases , years . other non - magnetic cvs ( nova - like variables ) appear to remain persistently in the outburst state . the z cam standstill state is comparable to that of a nova - like variable , i.e. a steady , high state , but somewhat lower in luminosity than the peak of a dwarf nova outburst . the difference in outburst behavior between u gem , z cam and nova - like systems is thought to reflect a difference in the rate of mass flow from the secondary ( ) . in dne , is below the upper stability boundary ( ) for which dn outbursts can occur , whereas in nova - like systems @xmath1 . in z cam - type stars @xmath2 lies very near , but below @xmath3 most of the time . standstill phases are triggered when an increase in @xmath2 @xcite is accompanied by a decrease in @xmath3 as a response to heating of the outer edge of the accretion disk by the increased mass transfer @xcite . thus , @xmath2 becomes greater than @xmath3 and the disk assumes a stable high state ( like a nova - like variable ) . it has been difficult to measure mass transfer rates in cvs observationally , and as a result reliable empirical determinations of and are lacking . on theoretical grounds , is thought to be about when heating of the mass transfer stream is taken into account @xcite . z cam is the brightest ( in outburst ) and most extensively studied object of its class . it has an orbital period @xmath4d , which implies that it , like other standstill systems , is above the period gap . ( a complete list of system parameters is presented in table [ t : param ] . ) the difference in the magnitude at outburst maximum ( @xmath5=10.4 ) and quiescence ( @xmath613 ) is small for dne in general , but typical of standstill systems . the standstill @xmath5 averages 11.5 @xcite . far ultraviolet ( fuv ) spectra of z cam in outburst were obtained with the hopkins ultraviolet telescope ( hut ) @xcite . the outburst spectrum resembles that of a steady - state accretion disk with a mass accretion rate = , modified by resonant scattering by material in a wind @xcite . here , we attempt to complete the picture of z cam by describing fuv spectra of z cam in the quiescent and standstill states . we utilize spectra of z cam in quiescence that we have obtained using _ far ultraviolet spectroscopic explorer _ ( ) , and a spectrum of z cam in standstill obtained with telescope that we retrieved from the mast archive . our discussion of the effort to understand the two datasets is organized as follows : in section [ s : obs ] , we describe the data reduction and the mean spectrum of z cam in quiescence . in section [ s : cont ] , we attempt to fit the spectrum of z cam in quiescence , while in section [ s : var ] we analyze the time variability of the spectral lines . in section [ s : stand ] , we describe the spectrum of z cam in standstill as well as our attempts to model that spectrum . finally , in section [ s : discus ] , we evaluate the available spectral models and compare the properties of z cam in its different luminosity states . provides a resolving power of @xmath720,000 , in the 9051187 range @xcite . z cam was observed with on 2002 february 9 , from 13:19:54 ut to 21:06:12 ut . as indicated in fig . [ f : aavso ] , the observations took place about eleven days after the return to quiescence from a short outburst and five days before the next outburst . the observations were made through the lwrs ( @xmath8 ) aperture . the data were obtained in the histogram mode , because z cam can ( in outburst ) exceed the brightness limit for the high - time - resolution mode . a total of 24,436 s of useful data was obtained . this comprised 43 separate exposures , ranging in length from 534s to 584s . the raw data were extracted and calibrated with version 2.2 of the calfuse pipeline and version 12 of the calfuse calibration files . the spectrum from each channel was examined separately to ensure that the flux calibration was consistent enough for the spectra to be spliced together . some data at the edges of each detector segment where the flux calibration is often uncertain were masked out , as was the ` worm ' feature that is present in the lif1b channel . the spectra from the sic1a channel indicated anomalously higher fluxes in the range 1000 - 1065 and so this region was excluded . the remaining data from the four channels were combined to create spectra at 0.1 binning . our process for combining the data consists of summing all the data points that fall within a given wavelength bin , weighting each according to the detector pixel sensitivity . a time - averaged spectrum was created by summing the individual spectra , weighted according to exposure time . this is shown in fig . [ f : spectrum ] . the interstellar lines have measured widths of @xmath70.1 ( about the expected spectral resolution ) , indicating that the four channels remained well - aligned over the course of the observation . this is supported by the flux levels of the overlap regions between spectral channels . the quiescent spectrum has a fairly flat continuum with peak flux lying near 1000 , at a flux level of @xmath9 . the continuum flux falls by only @xmath10% from its peak level before lyman line blanketing sets in at 912 . table [ t : lines ] lists the spectral lines observed in the spectrum of z cam . the equivalent width ( ew ) measurements are made from the time - averaged spectrum , except for those lines that are seen to be obscured by airglow in fig . [ f : spectrum ] ; in this case the measurements are made from a mean spectrum constructed from exposures that were recorded during spacecraft night ( as indicated in the image header ) . the night - time exposures make up about 30% of the total observing time and are relatively free of distortion by airglow lines . of absorption lines intrinsic to z cam , we see the lyman series up to at least @xmath11 , and a range of metals , including c , n , o , s and si. the absorption lines are smooth and are fit well by gaussian profiles with fwhm of @xmath12 . some of the absorption lines cut into broad emission profiles . these include the lower ionization lines of c@xmath13977 and c@xmath131176 , s@xmath141062,1073 and and the higher ionization o@xmath141032,1038 lines . none of the absorption lines fall to zero flux at the line core : the deepest absorption lines are o and c , which have absorption minimum flux levels of , respectively , 30 and 40% of the continuum . a careful inspection of the mean spectrum indicates that the flux below the lyman limit is not zero , but averages . judging from earlier fuv observations ( knigge et al . 1997 , also section [ s : stand ] ) and given the hydrogen column density ( @xmath15 ) to z cam ( mauche 1995 , private communication as quoted by * ? ? ? * ) , we expect to see no flux from z cam in this region . calibration of spectra involves a background subtraction step and for histogram data a time - averaged background is used . if the average background during the observation were higher , then the extra background will appear as extra flux after the pipeline processing . the background is a combination of scattered light and internal detector background and does not follow the effective area curve of for valid photons . as a result , the effect of poor background subtraction is most severe when the effective area is smallest ( or at the short wavelength end of the detector ) . the scattered - light portion of the background is much higher in the spacecraft daytime . therefore , we checked the mean spectrum from the full observation against one constructed from night - time only exposures . for the night - time spectrum , the flux below the lyman limit averages ( one quarter that of the mean spectrum ) . at wavelengths longer than 915 the discrepancy is significantly smaller , with the night - time continuum being on average @xmath16% less than in the mean spectrum , although in the two spectra the depth of the interstellar lyman lines differs by about the same amount as the flux level below 915 . at the minima of absorption lines intrinsic to z cam ( excluding those filled in by airglow in the mean spectrum ) the day - time flux is , at worst 15% less than the night - time flux level , with the effect being greatest in lines lying below 1000 . given that the flux offsets are small we are confident that the background subtraction problem does not affect the spectrum significantly at most wavelengths and does not prevent an accurate qualitative assessment of the spectrum . however , in applying spectral model fits ( in section [ s : cont ] ) we use only the night - time data . this is primarily because the absence of many of the strong airglow lines that are seen in the mean spectrum allows more data points to be included in the model fits , but also because we are more confident that the flux level of the night - time only data is closer to the true observed flux . a comparison of model flux levels with the observed fluxes calls for a reliable measurement of the distance to z cam . this has been estimated variously at @xmath17pc @xcite and @xmath18pc @xcite , based on the fraction of infrared and optical luminosity that can be attributed to the secondary star . a recent parallax measurement , combined with proper motions and absolute magnitude constraints , puts the distance somewhat closer at @xmath19pc @xcite ; we shall use this last distance in this paper . wd model spectra were generated using tlusty and synspec software @xcite . the model spectra covered effective temperatures from 13,000k to 100,000k and rotation rates of 0 to 1200 . we adopted @xmath20@xmath21 ( mauche 1995 , private communication as quoted by * ? ? ? * ) , e(b - v)@xmath220.0 , and @xmath23 ( strictly @xmath24 is expected for a 1@xmath25 wd ) . the parameters of all the model fits are listed in table [ t : fits ] . we first fit non - rotating da models to the mean night - time spectrum , after masking all of spectral lines other than those associated with hydrogen . the models were fit by @xmath26 minimization . the best - fit da model has @xmath2751,200k . as shown in fig . [ f : dafit ] , the model is a reasonably good match to the continuum slope and the peak of the continuum at @xmath71000 . the best - fitting da model falls below the data from the lyman limit to about 950 ; at the limit the model flux lies at about @xmath28 of the observed flux . varying the interstellar @xmath29 about its modest value has a negligible effect on the appearance of the best - fit model spectrum . increasing the interstellar reddening from zero to e(b - v ) @xmath30 ( the upper limit given by * ? ? ? * ) increases the best - fit temperature marginally , and also lowers the model peak flux level slightly , to better match the observed level . assuming the entire wd is visible , the normalization of a model is a measure of the solid angle of the wd and can be combined with the distance to give the wd radius . for the best - fit da model we arrive at , where the errors correspond to the upper and lower distance limits . adopting the wd mass - radius relation of @xcite and @xmath31 ( * quoted in @xcite ) , the expected for z cam is . the value ( and hence the mass ) determined from the fit is therefore consistent with the value of predicted from shafter s mass estimate and the normal wd mass - radius relation . the da model spectrum does not fit the higher order lyman lines particularly well . they are broader and deeper in the model spectrum than the observed lines . increasing , or decreasing the surface gravity of the model narrows the predicted profiles , but not enough to bring the models and the data into good agreement . changing these parameters also causes the spectral slope to become steeper , worsening the model fit to the continuum . one possibility is that there is an emission component , possibly from the disk , that contributes to the narrowness of the observed lines . the h emission is illustrated most effectively by the difference spectrum in fig . [ f : dafit ] . the emission components around other lines ( as noted in section [ s : obs ] ) can also be made out in the difference spectrum , although they appear to be weaker than the h emission . we contrast this to the result of @xcite , who find that the lack of strong lyman absorption in the uv spectrum of am her in the high state can be explained by irradiation of the wd surface by a hot spot that radiates as a blackbody , _ without _ noticeable emission or absorption features . we next considered wd with metal - enriched photospheres . solar abundance models were created for a range of temperatures from 13,000k to 100,000k and wd rotational velocities from 0 to 1200 . before fitting the spectrum , the o@xmath13@xmath321031,1039 , s@xmath13@xmath32933,944 and n@xmath13924 lines were masked out , as wd surface temperatures are not typically high enough to form these high - ionization lines . the best - fit solar - abundance wd model is shown in fig.[f : wdfit1 ] . it has @xmath33 8 cm , @xmath34k and @xmath35 . the peak in the continuum at 1000 and the slope of the continuum are a good match to the data . however , like the da model the solar abundance model underestimates the flux at the long- and short - wavelength ends of the spectrum . the effect is slightly less in the solar abundance case , because the hotter wd does a better job reproducing the short - wavelength emission . as was the case with the da model fits , the best fit parameters indicate that the wd is at somewhat greater distance than 163 pc and/or has some what greater mass than 0.99 . however , given the error bars , the various estimates are self - consistent . the solar abundance wd model does a good job of reproducing some of the stronger absorption lines , e.g. c@xmath13977 , n@xmath131085 , si@xmath13@xmath321122,1128 and c@xmath131176 . in the model the n@xmath13924 , n@xmath13989 , s@xmath131012 - 1021 and s@xmath13@xmath321062,1073 , s@xmath131077 and si@xmath131108 - 1113 lines are significantly weaker then in the data , whereas the si@xmath131066 line is much stronger . in other quiescent cv , two - temperature ( 2 t ) wd model fits have been known to give a significant improvement over the single - temperature models ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in non - magnetic cvs , the second component is attributed to matter arriving onto the surface of the wd from the accretion disk . the accreted matter is hotter than the wd and travels at the keplerian rotational velocity , and so the high - temperature , rapidly - rotating gas can form an ` accretion belt ' on the wd surface @xcite . to see whether model fits of this type provide a better representation of the spectrum , we fit the data to models consisting of two different temperatures , allowing the normalization and rotational velocity of each to vary independently . the best fit has one component with temperature @xmath36 of 26,300 k covering 71% and a second component with @xmath37 of 71,700 k covering 29% of the surface . statistically , the more complex model with of 9.4 is better than the single - temperature fit with of 10.1 , but qualitatively there is little difference between the fits and neither is `` statistically acceptable '' in the sense that is still considerably greater than one . therefore physical arguments will ultimately guide the interpretation . ) for the 2 t fit , the dominant contribution to the emission is due to the higher temperature component . at 1100 , 78% of the flux is from the high temperature component and shortward of 1000 essentially all of the flux is from the hot component . the 2 t fit does result in a larger radius of for the wd , which is typically the case when 1 t and 2 t models are compared . a value of ( at the nominal distance for z cam ) is close to the value of predicted from the estimated mass . we have performed tests to see how the temperature and contribution of the hot component ( @xmath37 ) varies as a function of the temperature of the low temperature ( @xmath36 ) is adjusted . for @xmath36 between 23,000 and 37,500 k , it is possible to fit a two - component wd model with a radius in the range allowed by the error limits on the wd mass ( table [ t : param ] ) . as the temperature of the low - t component increases , the fraction of the surface covered by the hot component also increases . at @xmath36 = 23,000k , the hot component has @xmath37=70,000k and covers 27% of the surface , whereas at 37,500k the hot component has @xmath37=75,000k and covers 41% of the surface . statistically , there is little difference between the goodness of these 2 t fits . they all have in the range 9.4 to 9.9 . the values of the rotation velocity @xmath38 for our best - fit 2 t model are 170 and 320 @xmath39 for the low and high temperature components , respectively . there is no real ( positive ) evidence that the second component , if it exists , is rotating rapidly , and hence there is no `` smoking gun '' for a rapidly rotating accretion belt . this is not entirely surprising though , given the fraction of the flux contributed by the high temperature component . z cam was extensively observed with the _ international ultraviolet explorer _ ( ) between 1979 and 1991 , although only twice during quiescence . the flux level at the long wavelength end of the spectrum provides an excellent overlap with the short - wavelength end of the z cam quiescent spectrum swp18844 ( as presented in * ? ? ? * ) and swp37155 . we rebinned the spectrum to the 1.7 to match the wavelength spacing of the lowres dataset and merged it with swp18844 . we then created model spectra to match the coarser spectra resolution of . at this resolution , there is little sensitivity to @xmath38 and so all of the models were constructed with @xmath38 of 300 @xmath39 . a solar abundance model fit to the combined and spectrum ( fig.[f : dafuseiue ] ) has @xmath40k and = . at @xmath41 the model flux level is somewhat lower than is observed and at around 1000the model continuum level lies slightly above that of the observed spectrum . the larger value of is a direct result of the lower temperature in these fits . in fig . [ f : dafuseiue ] , we also plot the best 2 t wd fit to the combined spectrum . in this case , @xmath36 is 17,600 k and this component covers 97% of the surface , while @xmath37 is 82,000k and covers only 3% of the surface . in terms of flux , the low t component contributes 27% at 1100 and 46% at 1500 . there is a significant improvement in the statistical quality of the fit from of 8.2 to 4.9 . the 1 t model has difficulties at the short and long wavelength ends of the spectrum ; the 2 t model ( predictably ) addresses these difficulties since the high t component provides flux near the lyman limit while the low t component provides flux at the long wavelength portion of the spectrum . however , is now larger than desired . if one constrains the normalization to place the system at the nominal distance and radius ranges , tests show that larger @xmath36 and lower @xmath37 are required . as was the case for when only the data were fitted , there is a locus of values for @xmath36 and @xmath37 that provide qualitatively similar fits to the data . given that z cam was observed in the quiescent state , we do not expect that steady - state accretion disk models would provide a good description of the spectra . nevertheless , it is important to check that this is not the case . therefore , we constructed model disk spectra from summed , area - weighted , doppler - broadened spectra of stellar atmospheres of the appropriate temperature and gravity for each disk annulus . as for the wd models , we adopted @xmath42 and @xmath20@xmath21 . disk spectra were created for a range of from @xmath43 to , and normalized to a distance of @xmath44pc . since the inclination for z cam is not tightly constrained ( @xmath45 ) , we created model spectra for inclinations of 46 , 57 , and 68 . we then fit the data using the same wavelength intervals as the solar abundance wd fits . unless otherwise indicated , these fits have one variable , since in most cases we did not allow the normalization ( or alternatively the distance ) to be a free parameter . [ f : diskfit1 ] shows the best - fit disk spectrum with = . this model matches the slope and level of the continuum between 1050 and 1185 , but bluewards of the model flux declines steeply to zero . as expected from a keplerian disk , the rotational velocities are too high to fit the narrow absorption lines . since we are not allowing the normalization to vary in these fits , the exact values of that we derive are dependent both on the distance to z cam and on the inclination . the inclination of z cam can not be very much larger than 68 or eclipses would be observed . if we assume this inclination , then we obtain of and a slightly better fit in terms of . this apparent improvement in the fit is due to the fact that higher model has a flatter continuum slope at shorter wavelengths . this said , the fits are considerably worse than both the uniform t and two - t wd models . better fits would be obtained if the normalization of the spectra were not fixed . indeed the best fit to the spectral shape would suggest of , a physically implausible value for a quiescent system , and an equally implausible distance of 710 pc . we also combined an accretion disk component with the wd models . we fit two - component models wherein the first component was a wd with @xmath46 and the second component was an accretion disk of @xmath47 ( the resultant spectrum was essentially unchanged when the disk inclination was varied to 68 ) . in the combined disk and wd model the wd is at 62,000k , with @xmath48 and the disk has = . the resultant model differs negligibly from the wd model fits ( see fig . [ f : wdfit1 ] ) ; is 9.8 , slightly worse than produced in the 2 t wd fits ( 9.4 ) . the wd component dominates the model spectrum , while the disk component acts basically only to flatten the continuum slope , by providing a small flux contribution at long - wavelengths . since in quiescence , one does not necessarily expect the spectrum to mimic that produced by steady state disk models , we also experimented with models in which the disk contribution was assumed to be a power law ( @xmath49 ) . these models produce similar fits statistically ( = 8.7 ) to fits using a wd and a steady state disk built from stellar spectra . the power law contributes about 33% of the flux at 1100 and has a slope @xmath50 of 1.6 . the wd contributes the bulk of the flux , and has a temperature of 44,900 k , of , and of 210@xmath39 . for completeness , we carried out the same disk model analysis for the combined and data . the results are almost identical to that of the data alone . a careful inspection of fig . [ f : fuse_iue_disk ] , where the results are shown , indicates that the main observational disagreement between disk and wd models for z cam is in the wavelength range . we note in passing that for the combined wd and disk fit ( 5.0 for 57 ) is better than for simple 1 t wd or disk models , and almost identical to that obtained for 2 t wd models ( 4.9 ) . the inferred wd radius for the combined wd and disk model fit suggests a very massive wd , in excess of 1.35 and much greater than estimated by @xcite . one way to lower the mass would be to allow the distance to be larger . but much more probably , the result is an artifact of our limited knowledge of the spectrum of a quiescent disk , which we , in the absence of any fully - calculated alternative , have represented as that expected from a steady - state optically - thick disk . it has been suggested that in quiescent cv the accretion disk is truncated at its inner edge , due to irradiation by the white dwarf ( e.g. * ? ? ? if we assume that the emission components seen around some of the absorption lines are formed in a keplerian disk , and hence are doppler broadened by the disk rotation , we can estimate the smallest radius at which the emission is formed . using the starlink software dipso , we fit the s@xmath511063,1072 , si@xmath1311081113 and c@xmath131175 lines after first normalizing the continuum flux . we fit a gaussian absorption and emission line to each component of the multiplet ( except in the case of c@xmath131175 , which is fit as a single line ) . these are shown in fig . [ f : emission ] . after subtracting the multiplet splitting , the emission components all have widths of about @xmath74000 ( fwzi ) . this translates to a rotational velocity of 2400 , or a keplerian disk radius of , which is @xmath74 ( * ? ? ? * for example , predicts that the quiescent disk truncation radius is @xmath52 ) . this suggests that the emission lines could be formed by material rotating with the quiescent accretion disk . at present we have only a poor idea of exactly what observational signatures from a cold optically - thin disk would be detectable in the fuv . however , there is substantive evidence that there is an additional component to the quiescent continuum and some attempts have been made to model the structure of this as a coronal layer above the quiescent disk ( e.g. * ? ? ? * ; * ? ? ? there currently exist no quantitative uv spectral models for such a corona , so in order to get some idea of what these properties might be , we constructed tlusty models of an optically thin plasma veil , at various temperatures and densities , which were then rotated as a solid body , to mimic the disk rotation . we then took a simple optically thin emission prescription for the emitted flux ( @xmath53 ) . the optically thin plasma was able to reproduce the relative strengths of the emission components seen around some absorption lines . however , there was some degeneracy between various h column densities , electron densities and temperatures . hence , we found it impossible to estimate , from these models , a possible physical structure for the corona . the 43 histogram exposures cover just over one orbital period of z cam from 0.94 to 0.06 , with a phase uncertainty of 0.015 ( using the spectroscopic ephemeris of thorstensen & ringwald , 1995 ) . the phase coverage is continuous , apart from gaps at phases 0.570.63 and 0.810.88 . we are thus confident that the mean spectrum is representative of the average over the full orbital period , rather than a reflection of some orbital phase - dependent characteristic . we created a continuum light curve by summing the flux over selected spectral windows . for the light curve we used only data from the lif1 channel . is guided on this channel and so it should be less susceptible than the others to channel misalignment . to measure flux variation across the entire lif1 range , we first excluded regions contaminated by airglow and the worm region ( see section [ s : obs ] ) and summed the flux over the ranges 9961024 , 10291038 , 10461081 , 10951103 and 11701185 . continuum flux varied by up to @xmath54% either side of the mean . however , given the noted discrepancy between the day and the night - time flux levels ( see section [ s : obs ] ) , we found no evidence of variability in the continuum flux level in excess of that which could be associated with the periodic transition of the spacecraft from day to night time . the spectroscopic ephemeris of thorstensen & ringwald ( 1995 ) is accurate enough to radial - velocity correct each exposure onto the wd rest frame , with errors of about 25 . the spectra presented in this section are plotted after this correction has been applied . [ f : variation1 ] shows the 10051045region plotted in phase bins of 0.1@xmath55 . the 11001133 spectral range encompasses the c@xmath131010 , s@xmath1310121021 , and o@xmath13@xmath321032,1037 lines and gives a good representative sample of the line variability seen across the spectrum . in the time - averaged spectrum the flux minima of the absorption lines are all centered at between 100 and 200 redwards of the transition rest velocity . throughout the observation there is almost no shift in the position of this feature , except for during phase 0.5 to 0.7 where the entire line profile shifts bluewards by about 100200 ( i.e. back to closer agreement with the rest velocity , see fig . [ f : variation1 ] ) . that the mean line profiles are centered redwards of the rest - velocity could well be an artifact of the radial velocity correction . we take the zero - phase ephemeris from the inferior conjunction of red star emission lines , rather than using the h@xmath50 emission line ephemeris , as the latter may be formed in an extended outflow ( see * ? ? ? this therefore , does not take into account gravitational redshifting , which is 76if @xmath56 ( or 108if @xmath57 ) . if the absorption lines are formed on the surface of the wd , then this accounts for a portion of the redshift of the mean line profile centroids . there is little variability in the strength and width of the absorption lines , except between phases 0.65 and 0.81 ( fig.[f : addabs ] ) . then there is a sharp increase in the strength of most of the absorption lines and we clearly see some lines that are barely apparent in the mean spectrum , e.g. p@xmath13@xmath321118,1128 and the si@xmath13994997 triplet . at the same time there is no sign of an increase or decrease in continuum flux . the fwhm of the lines increases by no more than a few tens of . the additional absorption mainly affects the depth of the line absorption . there is no shift bluewards or redwards from the mean line profile . the increased absorption is seen across the full range of atomic species and ionization stages , for example s@xmath1310121021 , s@xmath13@xmath321062,1073 and s@xmath13@xmath32933,944 . the higher ionization s and s lines increase in ew by around 30% . on average , the increase in line ew at phase 0.650.81 is also @xmath58% . the s triplet shows increased absorption by up to 300% in the mean ew of the s@xmath131013 component , but only 40% in the s@xmath131021 component . some lines , such as si@xmath1311081113 , show a smaller variation ( @xmath59% ) ; others , such as si@xmath13997 and p@xmath131118 , increase their ew by two or three times ; and others , such as si@xmath13@xmath32994,995 , appear only during the period of enhanced absorption . there are a few lines , most notably n@xmath13924 , o@xmath13@xmath321032,1038 , and c@xmath13977 and 1176 that do not increase in ew at phase 0.650.81 . as the lines that show increased absorption cover a range of excitation energies we can rule out any obvious temperature - dependence . an increase in the absorption line strength could stem either from an increase in the column density of absorbing ions , or from an increase in their covering factor against the continuum background ( or a combination of the two ) . given that , at the time of the increased absorption , we see no evidence of an accompanying alteration of the ionization balance , it seems most likely that we are seeing a shift of material into the line of sight and/or an increase in the continuum covering fraction of absorbing ions . we are unable to say whether the narrow absorption endures beyond phase 0.81 as there is a gap in the data from phase 0.81 to 0.88 . however , when the coverage picks up again at phase 0.88 , the additional absorption has disappeared . z cam entered standstill on 1992 september 23 . z cam was observed with the berkeley extreme and far - uv spectrometer ( befs ) , on board the telescope @xcite , on 1993 september 18 . a single 1477s spectrum was obtained . the system remained in standstill for a further 100 days , before declining to quiescence ( see the lightcurve in * ? ? ? during the standstill the system brightness can fluctuate by about 0.3mag either side of the average magnitude of 11.5 , but the timing of this observation so far into the standstill assures us that captured z cam in a true standstill phase . we retrieved the befs spectrum , pre - calibrated , from the mast archive . the spectrum covers the range 7001175 at a spectral resolution of about 0.2 in the two longer - wavelength channels of the spectrometer . however , no flux was detected below 912 , so we have used only data from the longest - wavelength channel ( 9001175 ) . in fig . [ f : standstill ] we have plotted the befs and the spectra together . despite the order - of - magnitude difference in continuum flux , the two spectral shapes are very similar . the standstill spectrum peaks at @xmath60 , at . this is to the red of the peak in the quiescent spectrum ( at @xmath61 ) . on the longer - wavelength side of the peak , both have similar continuum slopes . both spectra show significant flux down to the lyman limit , although the continuum between 912 and the peak flux is slightly flatter in the quiescent spectrum than in standstill . most of the absorption lines seen in the standstill spectrum as we are also seen in quiescence ( see table [ t : standstill ] ) . in the befs spectrum the absorption equivalent widths are generally 30 - 60% greater in the higher - ionization lines ( e.g. o , sand s ) than in the spectrum , whereas the lower - ionization lines ( e.g. si and s ) are weaker by a factor of at least 2 . note also the appearance of the p@xmath13@xmath321118,1128 lines at standstill ( which are only seen at phase 0.650.81 in quiescence ) . this is indicative of a general move upward in the ionization state between quiescence and standstill . the standstill line widths ( fwhm ) are generally about 100200 narrower than the quiescent lines , with the exception of c , n and o. in standstill , there is no sign of the broad line emission that is seen in quiescence . nor for that matter are there indications of the wind that is observed in outburst in z cam , e.g. p - cygni - like profiles or even blue - shifted absorption features . we repeated the model - fitting exercise that was carried out on the quiescent spectrum , beginning with solar abundance wd models ( as for section [ s : cont ] ) . in order not to bias a comparison to the data , we excluded the same regions of the spectra as were used for the analysis . in general , the best - fit values of were lower for the than the data , a fact that is directly attributable to the statistical quality of the two datasets . the best - fit wd model is shown in fig . [ f : standfit2 ] . the wd model spectrum matches fairly well to the observed continuum , although the model flux peaks at around 1000 , bluewards of the observed flux peak at 1060 . the best fit has a temperature of 59,700k and of 240 @xmath39 , not very different from that in the quiescent spectrum . in common with the quiescent wd model fits , the model lyman lines are broader than the observed . however , the normalization at 163pc gives = , which would imply an unacceptably low ( 0.35 ) wd mass . we then fit model accretion disk spectra , as in section [ ss : diskfits ] . we find that the standstill continuum emission is reproduced well by an accretion disk model with = at an inclination of 57 and = at 68 ( see fig.[f : standfit2 ] ) . the disk models are a good fit to the continuum ( the higher - inclination model is slightly better ) , although they are unable to reproduce the absorption lines . the disk model falls off steeply towards the lyman limit , whereas the data ( and to a certain extent the wd models ) level off before dropping steeply to zero . in quiescence , we find that the fuv continuum of z cam as observed by is qualitatively well - described in terms of emission from a hot metal - enriched wd atmosphere . for log g=8.5 , the best fit single temperature models has a of 57,200 k and of . the mass of the wd in z cam is estimated to be [email protected] , which suggests of . thus the two measurements of can be reconciled if the distance to z cam is @xmath63 pc , where the errors here are calculated from the uncertainty in the mass . this is slightly larger than , but wholly consistent with the astrometric distance of @xmath64 pc . if this interpretation is correct , then the wd in z cam is hot when compared to the wds in other cvs . a recent compilation of reliably determined wd temperatures in cvs by @xcite contains 34 objects ( other than z cam ) ranging in temperature from 9,500 k in the polar ef eri to 50,000 k in the nova - like variable and sw sex star dw uma . a temperature of 57,200 k would make z cam the hottest wd in a cv , excepting only v1500 cyg , which is still cooling from its nova outburst in 1975 and for which there is a crude blackbody temperature estimate of 70,000 to 120,000 k for the wd @xcite . is this reasonable ? the surface temperatures of wds in cvs are thought to be primarily determined by the accretion history of the wd @xcite . this is because although the wds in cvs emerge from the common envelope phase as very hot wds , the wds cool to temperatures of 4,500 - 6000 k in the 3 - 4 gyr required to reach the mass - exchanging stage . accretion reheats the surface layers of the wd on the time scales of outbursts ( weeks ) , as is evidenced from observations of temperature changes in wds in cvs such as u gem @xcite and vw hyi @xcite . more importantly , accretion reheats the entire wd over longer time scales ( ) , as the envelope and core adjust to the increase in mass of the wd , an effect known as compressional heating , and through nuclear burning of the accreted material @xcite . recently , @xcite have determined the relationship between time - averaged accretion rate and wd temperature . for a temperature of 57,000 k , the required rate is @xmath7 ( with uncertainties of perhaps a factor of two ) . ( a somewhat lower value of is obtained for 45,000 k , the temperature derived from the combined + spectrum . ) thus the accretion rate required is high , although it may be consistent with the fact that z cam is commonly in ( or near ) the high state . indeed , of the four systems with wd temperatures greater than 40,000 k , mv lyr , tt ari , ru peg , and dw uma , all but ru peg are nova - like variables and hence , like z cam , are systems where there are other indications of high time - averaged accretion rates . given this , we believe a temperature of 57,000 k is reasonable . an alternative interpretation of the quiescent spectrum is that the surface temperature of the wd in z cam is not uniform . our analysis shows that a somewhat better , but still not statistically acceptable fit to the data can be obtained if 2 t models are considered . our best fit had a low temperature component with @xmath36 of 26,300 k covering 71% of the surface and a high temperature component with @xmath37 of 71,700 k covering 29% . if the low temperature component represents energy loss from the interior of the wd while the high temperature component is due to recent or ongoing accretion , then 94% of the luminosity of the wd is due to recent accretion . is this reasonable ? if the excess emission is due to ongoing accretion , then the observed luminosity is @xmath65 , where @xmath66 is the fraction of gravitational energy available that is radiated . for the best 2 t parameters , the excess luminosity is , and the implied value of @xmath67 is . this is a fairly large value for a disk in quiescence , especially since @xmath68 , but is hard to rule out because of our lack of good models for emission from quiescent disks and the brightness of the wd itself . on the other hand , the accretion belt hypothesis is that some of the kinetic energy of material that reaches the boundary layer between the disk and wd is stored in a rapidly rotating belt and released slowly during the interoutburst period . in standard accretion disk theory , half of the gravitational energy of material accreted by the wd is radiated by the disk and half remains as kinetic energy when it enters the boundary layer . for z cam , with a disk accretion rate of , the disk luminosity is . for the outburst that preceded the observation , z cam remained within 1 mag . of outburst maximum for about 4 days ( see fig.[f : aavso ] ) , and hence the total energy release was about . the observation occurred about 11 days from the return to quiescence and , as noted previously , the belt luminosity in the context of this hypothesis was still . we do not have the coverage to determine the time decay of the second component , but a conservative lower limit is obtained by assuming constant luminosity until the observation and zero from then on . by this argument the energy stored in the rotating layer was at least , or 8% of the energy of the outburst . if one assumes an exponential decay , then the energy release is @xmath69 , where t is the time from outburst , l(t ) the luminosity at the time of our observation , and @xmath70 the decay time constant . if @xmath70 was 11 days , then the total energy released would be @xmath71 greater , or 20% of the total outburst energy . in either case , the energy that must be stored in the belt is large , though not energetically disallowed . if the hot component is produced by short - term effects of accretion , then the low t component reflects the long - term average of accretion . for a temperature of 26,300 , this requires an accretion rate of . this is lower than the number inferred from 1 t fits , but still substantial . so how does one decide between the two models we have explored for z cam in quiescence ? the data are not definitive . the model parameters that result from both models are substantially in agreement with other known parameters of the system . is somewhat better for the 2 t model than the uniform temperature model . but is much greater than 1 per degree of freedom in the 2 t model , showing that it is at best a step toward a correct model of z cam in quiescence , and the systematic errors in the spectrum are difficult to quantify . the differences in the two model spectra are quite small ; there is no characteristic of the spectrum that one can point to that requires a second component . evidence that for the high temperature component is high compared to the low t component is lacking . in u gem @xcite , single temperature models lead to a physically implausible result , namely that the wd grows in radius during quiescence . since we do not have multiple observations of z cam through a quiescent period , we do not know whether single t models lead to a similar problem in z cam . there are other well - observed systems , e.g. vw hyi @xcite and wz sge @xcite , which do not exhibit `` radius growth '' in the context of one temperature models , and so one can not use that as a guide . there is also no consensus theoretically . while @xcite have argued that mass transfer from the disk is higher immediately after outburst in order to explain uv delays in dne , the expected x - ray evidence of a gradual decline in hard x - ray emission following outbursts is lacking . furthermore the x - ray data that exist indicates that the x - ray plasma that exists in quiescence is not rotating rapidly @xcite , and therefore it is hard to understand how it could produce a heated annular region around the surface . while @xcite postulated the existence of accretion belts and described an instability that might release energy in an accretion belt , there are to our knowledge no detailed calculations of expected properties of the emitting region and no attempt to calculate the luminosity of the rotating component with time . given these considerations , our answer to the question of which model is most likely correct is to appeal to occam s razor and assert that the simpler 1 t model provide a good physical description of the data and is therefore is the best - bet physical model to pursue . in outburst , @xcite showed that the shape of the fuv ( 9001800 ) continuum of z cam could be modeled as an accretion disk with @xmath72 at an inclination of 57 and distance of 170 pc ( very close to the best astrometric distance of 163 pc ) . in the high state , the fits are self - consistent . in standstill we find @xmath73 . approximately , then , there appears to be at least an order of magnitude increase in from quiescence to standstill , while is an order of magnitude higher still . our derived gives an upper limit on , as it is a reasonable assumption that is also the rate of mass flow from the secondary at that time , which must be greater than . @xcite calculate @xmath74 , using the formula of @xcite and taking into account heating by the stream disk impact ( or if tidal dissipation is also included ) . our upper limit of @xmath75 is just consistent with this and with the upper limit of @xmath76 derived by @xcite . the comparison of the befs standstill spectrum and the spectrum of z cam in quiescence illustrates the problem of separating a wd spectrum from that of a disk . except for the flux level the qualitative differences between the befs spectrum and the spectrum are relatively small . in both states , the continuum slope is almost the same and , other than a slight increase in ionization temperature , their line spectra are very similar . yet , we find that in quiescence a wd alone can make up the continuum flux , whereas in standstill an accretion disk is the best model for the continuum flux . ultimately , the key model parameter that has enabled us to fit one model and reject others is the normalization required to match the observed and the modeled flux levels , which in turn depends on the wd mass and the distance from the sun . without a reasonably accurate knowledge of these parameters it would be impossible to judge between a wide range of models that reproduce the continuum slope and line spectrum equally well . in the time - resolved spectra of z cam in quiescence , we have seen enhanced line absorption between phase 0.65 and 0.81 , which covers the full range of atomic species and excitation levels . the enhanced absorption is very well centered on the mean line profile , which , in turn , is centered 100200 to the red of the transition rest velocity . the change in absorption width is greatest in those lines that are weak or absent from the mean spectrum , but also encompasses lines which are thought to be optically thick in the mean spectrum , leaving only the n@xmath13924 , o@xmath13@xmath321032,1038 and c@xmath13977 and 1176 lines unchanged . flux variations at around phase 0.50.8 are seen in a selection of disk - accreting systems , in phenomena such as dips in the uv @xcite and x - ray @xcite light curve of low - mass x - ray binaries and cv , and humps in the optical light curves of cv ( e.g. * ? ? ? this is typically attributed the interaction of the accretion stream from the secondary with the outer edge of the accretion disk . this can result in a luminous ` bright spot ' at the outer edge of the disk , which is thought to be responsible for the optical light curve modulation . also , the shock of the in - falling stream can cause the disk to bulge outwards at smaller radii or cause the stream to be deflected upward and arc over the surface of the accretion disk ( e.g. * ? ? ? this can cause phase dependent absorption as colder material in the accretion stream moves in front of the continuum source . the additional absorption seen in z cam is qualitatively compatible with some sort of stream - disk interaction . the enhanced absorption is narrow and centered on the mean line profile , indicative of matter that is moving across , rather than parallel to the line of sight . there is no corresponding change in the ionization state and the increase in absorption affects those transitions that appear to be optically thick in the mean spectrum . this is consistent with a disk bulge or stream temporarily moving to occult a larger area of the continuum source , or to place a greater column density of absorbing ions into the line of sight . also , the 0.650.81 phasing is in the phase range that stream - overflow effects are thought to be seen . in these respects z cam compares well with the spectrum of u gem in outburst @xcite , which was observed to show strong enhanced absorption at phase 0.53 to 0.79 . the line variability in u gem is of a similar strength to that seen in z cam , i.e. residual line core flux reducing to as little as 40% of its mean value in some lines , with the o lines being the least affected , although the n@xmath13924 , c@xmath131176 and other weak lines participate in the variation in u gem . smooth particle hydrodynamical models of the disk - stream interaction are now being produced ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? these models make predictions of the observational qualities which we are now able to set against the evidence for stream - related variability seen in the spectrum of z cam . the models are generally in agreement that stream overflow can occur at low mass - accretion rates , i.e. quiescent disks . @xcite predict that efficient cooling of the hot - spot region can occur in systems with low mass - accretion rates ( @xmath77 , or @xmath78 ) , allowing material to overflow the inner disk in a coherent stream , whereas , in higher- systems the stream impact region tends to push the disk into a bulging shape . using a different method , @xcite concurs with armitage & livio that up to @xmath59% of the matter in the accretion stream from the secondary is able to overflow the disk . if the phase - dependent absorption is a result of stream - overflow , it may have implications for the estimated inclination of the system . logically , for absorption by the stream , the absorption strength should have a strong correlation with inclination angle . at higher inclinations , material above the disk plane presents a higher column density to the bright continuum regions , i.e. the wd and inner disk . @xcite estimate that in outburst , enhanced absorption can be seen in systems of inclinations down to 65 , based on the maximum scale height above the disk attained by material in the accretion stream . however , they predict that , at low , material is not as strongly deflected upward as for high and that an inclination angle of at least 75is needed before the overflowing stream can cause accretion dips . at an inclination of only @xmath79(as taken from * ? ? ? * ) , z cam would fall below this lower limit . furthermore , an inclination above the current upper limit of 68 for z cam would produce observable eclipses and therefore can be ruled out . however , given the similarity of the time - dependent absorption to that seen in u gem ( @xmath80 ) , an inclination angle near 68seems far more plausible that the opposite extreme of 48 . we have obtained observations with the satellite of the dwarf nova , z cam , during a period of quiescence . the spectrum is characterized by a fairly flat continuum , with a peak in the flux at 1000 . the line spectrum covers a broad range of atomic species and ionization stages and is dominated by absorption lines of fwhm @xmath12 . qualitatively , the spectrum can be self - consistently described in terms of emission from the wd with a temperature of 57,000 k. modest improvements in the fits are obtained with 2 t models in which most of the emission comes from a small fraction ( 29% ) of the surface heated to @xmath81k . the remainder is at 26,300k . we favor the uniform temperature model for z cam , largely because the data do not require anything more complicated . there are emission lines ( e.g. c@xmath131176 ) that do not come from the wd . the widths of these lines are consistent with a disk origin . in the phase - resolved quiescent spectrum we have observed transient enhanced line absorption . from the 0.650.81 phasing of this absorption and its effect on a wide range of ionization species , we attribute it to material raised from the disk plane into the los , due to interaction of the accretion stream with the disk , or the stream itself moving over the disk . such effects may be easier to understand if inclination of z cam is close to the upper limit of 68 . the standstill continuum is best described by an optically - thick disk accreting at a rate of , depending on the actual inclination of z cam . the continuum slope and the line absorption spectra differ little between quiescence and standstill , despite that , during standstill , an accretion disk creates ( most of ) the continuum flux . the qualitative similarity of the two spectra highlights the difficulty of separating wd from disk signatures without detailed model fitting and , also , underlines the reliance of the model fits on an secure knowledge of the distance and the wd mass . the high temperature of the wd in z cam and the estimates of mass accretion rate in standstill and outburst are all consistent with a mass transfer rate from the secondary star that is higher than in normal dne . additional fuv observations of z cam in the period following a normal outburst are highly desirable both to better understand the physics of standstill systems and to cleanly distinguish between single and multi - temperature models of the wd in z cam . in this research , we have used , and acknowledge with thanks , data from the aavso international database , based on observations submitted to the aavso by variable star observers worldwide . we also gratefully acknowledge the financial support from nasa through grant nag 5 - 11885 . all of the data presented in this paper was obtained from the multimission archive at the space telescope science institute ( mast ) . stsci is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 . 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 . we thank the anonymous referee for their helpful comments and suggestions . anderson , n. 1988 , , 325 , 266 araujo - betancor , s. a. , g " ansicke , b. t. , long , k. s. , beuermann , k. , de martino , d. , sion , e. m. & szkody , p. 2005 , , submitted armitage , p. j. & livio , m. 1996 , , 470 , 1024 armitage , p. j. & livio , m. 1998 , , 493 , 898 baraffe , i. , kolb , u. 2000 , mnras , 318 , 354 baskill , d. , wheatly , p. j. , osbourne , j. p. 2001 , , 328 , 71 berriman , g. , szkody , p. , and capps , r. w. 1985 , , 217 , 327 buat - m ' enard , v. , hameury , j .- m . , & lasota , j .- 2001 , , 369 , 925 froning , c. s. , long , k. s. , drew , j. e. , knigge , c. , proga , d. 2001 , , 963 g " ansicke , b. t. & beuermann , k. 1996 , , 309 , l47 g " ansicke , b. t. , hoard , d. w. , beuermann , k. , sion , e. m. , & szkody , p. 1998 , , 338 , 933 hameury , j. , menou , k. , dubus , g. , lasota , j. , & hure , j. 1998 , , 298 , 1048 hellier , c. , garlick , m. a. , & mason , k. o. 1993 , , 260 , 299 hessman , f. v. 1999 , , 510 , 867 hubeny , i. 1988 , comput . commun . , 52 , 103 hubeny , i. , & lanz , t. 1995 , , 439 , 875 hubeny , i. , lanz , t. , & jeffery , c. s. 1994 , in newsletter on analysis of astronomical spectra ( st . andrews : st . andrews univ . ) , 20 , 30 hurwitz , m. et al.1998 , , 500 , l1 king , a. r. 1997 , , 288 , l16 king , a. r. & cannizzo , j. k. 1998 , , 499 , 348 kiplinger , a. l. 1980 , , 236 , 839 kippenhahn , r. & thomas , h .- c . 1978 , , 63 , 265 knigge , k. , long , k. s. , blair , w. p. , and wade , r. a. 1997 , , 476 , 291 kunze , s. , speith , r. , & hessman , f. v. 2001 , , 322 , 499 lasota , j .- p . 2001 , new astronomy review , 45 , 449 liu , f. k. , meyer , f. , & meyer - hofmeister , e. 1995 , , 300 , 823 long , k. s. et al . 1991 , , 381 , l25 long , k. s. , blair , w. p. , bowers , c. w. , davidsen , a. f. , kriss , g. a. , sion , e. m. , & hubeny , i.1993 , , 405 , 327 mauche , c. w. , wheatley , p. j. , long , k. s. , raymond , j. c. 2004 , & szkody , p. 2004 , in asp conf . the astrophysics of cataclysmic variables and related objects , ed . j. m. hameury & j. p. lasota , in press meyer , f. & meyer - hofmeister , e. 1994 , , 288 , 175 moos , h. w. et al . 2000 , , 538 , l1 oppenheimer , b. d. , kenyon , s. j. , & mattei , j. a. 1998 , , 115 , 1175 sahnow , d. j. et al.2000 , , 538 , l7 schreiber , m. r. & g " ansicke , b. t. 2002 , , 382 , 124 shafter , a. w.1983 , ph.d . thesis , ucla sion , e. m. 1999 , , 111 , 532 schmidt , g. d. , liebert , j. , & stockman , h. s. 1995 , , 441 , 414 sion , e. m. 1985 , , 297 , 538 sion , e. m. 1995 , , 438 , 876 sion , e. m. & urban , j. 2002 , , 572 , 456 sion , e. m. , szkody , p. , g " ansicke , b. , cheng , f. h. , ladous , c. , & hassall , b. 2001 , , 555 , 834 stehle , r. , king , a. , & rudge , c. 2001 , , 323 , 584 szkody , p. & wade , r. a. 1981 , , 251 , 201 szkody , p. & mateo , m. 1986 , , 301 , 286 szkody , p. , g " ansicke , b. t. , howell , s. b. , & sion , e. m. 2002 , , 575 , l79 lcccc n * + h ... & 922924 ; 923.2 & [email protected] & 1030@xmath6265 + h ... & 930.748 & [email protected] & 570@xmath6275 + s ... & 933.4 & [email protected] & 565@xmath6230 + h ... & 937.8 & [email protected] & 545@xmath6245 + s ... & 944.5 & [email protected] & 470@xmath6220 + h+ p ... & 949.7 ; 950.7 & [email protected] & 900@xmath6260 + h ... & 972.5 & 1.09@xmath93 & 600@xmath6250 + c ... & 977.0 & [email protected] & 590@xmath6240 + he * + n ... & 989.8 ; 992.3 & [email protected] & 1150@xmath6240 + si * ... & 997.4 & @xmath94 & @xmath95 + c * ... & 1010.0 & [email protected] & 420@xmath6290 + s ... & 1012.5 & [email protected] & 240@xmath6260 + s ... & 1015.5 & [email protected] & 410@xmath6230 + s ... & 1021.1 & [email protected] & 500@xmath6230 + ... & 1025.7 & [email protected] & 470@xmath6290 + o ... & 1031.9 & [email protected] & 610@xmath6220 + o ... & 1037.6 & [email protected] & 700@xmath6220 + s ... & 1062.7 & [email protected] & 480@xmath6230 + s ... & 1073.0 & [email protected] & 480@xmath6230 + s * ... & 1077.2 & [email protected] & 280@xmath6261 + he * + n ... & 1084.9 ; 1085.0 & [email protected] & 1020@xmath6280 + si * ... & 1108.4 ; 1110.0 & [email protected] & 890@xmath6220 + si * ... & 1113.2 & [email protected] & 500@xmath6220 + p ... & 1118.0 & @xmath96 & @xmath97 + si * ... & 1122.5 & [email protected] & 730@xmath6250 + p+ si * ... & 1128.0 ; 1128.3 & [email protected] & 850@xmath6280 + c * ... & 1175.3 & [email protected] & 940@xmath6230 + cccccccc + da & 4.3 & 58,100 & - & - & 163 & 6.7 & 672 + wd & 4.4 & 57,200 & 330 & - & 163 & 10.1 & 1146 + 2wd & 6.0 ( 71%/29% ) & 26,300/71,700 & 170/370 & - & 163 & 9.4 & 1146 + disk ( 46 ) & - & - & - & 6.6 & 163 & 32.1 & 1146 + disk ( 57 ) & - & - & - & 9.1 & 163 & 27.1 & 1146 + disk ( 68 ) & - & - & - & 12.9 & 163 & 22.0 & 1146 + disk ( 57 ) & - & - & - & 180.0 & 710 & 11.9 & 1146 + disk ( 46 ) & wd & 4.0 & 62,000 & 306 & 1.2 & 163 & 9.8 & 1146 + disk ( 57 ) & wd & 4.0 & 62,000 & 295 & 1.7 & 163 & 9.8 & 1146 + disk ( 68 ) & wd & 3.9 & 62,000 & 293 & 2.5 & 163 & 9.8 & 1146 + + wd & 6.0 & 44,600 & - & - & 163 & 8.2 & 471 + 2wd & 18.4(97%/3% ) & 17,600/82,400 & - & - & 163 & 4.9 & 471 + disk ( 46 ) & - & - & - & 6.0 & 163 & 11.1 & 471 + disk ( 57 ) & - & - & - & 8.7 & 163 & 9.6 & 471 + disk ( 68 ) & - & - & - & 12.9 & 163 & 8.1 & 471 + disk ( 57 ) & - & - & - & 182.0 & 710 & 5.8 & 471 + disk ( 46 ) & wd & 3.0 & 82,000 & - & 3.0 & 163 & 4.7 & 471 + disk ( 57 ) & wd & 3.0 & 81,000 & - & 4.0 & 163 & 5.0 & 471 + disk ( 68 ) & wd & 3.0 & 81,000 & - & 6.3 & 163 & 5.4 & 471 + + wd & 12.5 & 59,700 & 240 & - & 163 & 2.6 & 1505 + disk ( 46 ) & - & - & - & 43.0 & 163 & 2.7 & 1505 + disk ( 57 ) & - & - & - & 69.0 & 163 & 2.6 & 1505 + disk ( 68 ) & - & - & - & 112.0 & 163 & 2.5 & 1505 + disk ( 57 ) & - & - & - & 195.0 & 270 & 2.4 & 1505 + [ fits ] lcccc h ... & 930.748 & @xmath98 & @xmath99 + s ... & 933.4 & @xmath100 & @xmath101 + h ... & 937.8 & @xmath102 & @xmath103 + s ... & 944.5 & @xmath104 & @xmath105 + h + p ... & 949.7 & @xmath106 & @xmath107 + h ... & 972.5 & @xmath108 & @xmath109 + c ... & 977.0 & @xmath110 & @xmath111 + n + si * ... & 989.8 ; 993.5 & @xmath112 & @xmath113 + si * ... & 994.8 & @xmath114 & @xmath115 + si * ... & 997.4 & @xmath116 & @xmath117 + s ... & 1012.5 & @xmath96 & @xmath115 + s ... & 1015.5 & @xmath96 & @xmath115 + s ... & 1021.1 & @xmath118 & @xmath119 + ... & 1025.7 & @xmath120 & @xmath121 + o ... & 1031.9 & @xmath122 & @xmath123 + o ... & 1037.6 & @xmath124 & @xmath125 + s ... & 1062.7 & @xmath126 & @xmath127 + s ... & 1073.0 & @xmath128 & @xmath129 + he * + n ... & 1084.9 ; 1085.0 & @xmath130 & @xmath131 + si * ... & 1108.4 ; 1110.0 & @xmath132 & @xmath133 + si * ... & 1113.2 & @xmath134 & @xmath135 + p ... & 1118.0 & @xmath136 & @xmath137 + si * ... & 1122.5 & @xmath138 & @xmath139 + p + si * ... & 1128.0 ; 1128.3 & @xmath120 & @xmath140 +

in this work we have estimated 10 collisional ages of 9 families for which for different reasons our previous attempts failed . in general , these are difficult cases that required dedicated effort , such as a new family classifications for asteroids in mean motion resonances , in particular the and with jupiter , as well as a revision of the classification inside the resonance . of the families locked in mean motion resonances , by employing a numerical calibration to estimate the yarkovsky effect in proper eccentricity , we succeeded in determining ages of the families of ( 1911 ) schubart and of the `` super - hilda '' family , assuming this is actually a severely eroded original family of ( 153 ) hilda . in the trojan region we found families with almost no yarkovsky evolution , for which we could compute only physically implausible ages . hence , we interpreted their modest dispersions of proper eccentricities and inclinations as implying that the trojan asteroid families are fossil families , frozen at their proper elements determined by the original ejection velocity field . we have found a new family , among the griquas locked in the 2/1 resonance with jupiter , the family of ( 11097 ) 1994 ud1 . we have estimated the ages of 6 families affected by secular resonances : families of ( 5 ) astraea , ( 25 ) phocaea , ( 283 ) emma , ( 363 ) padua , ( 686 ) gersuind , and ( 945 ) barcelona . by using in all these cases a numerical calibration method , we have shown that the secular resonances do not affect significanly the secular change of proper a. for the family of ( 145 ) adeona we could estimate the age only after removal of a number of assumed interlopers . with the present paper we have concluded the series dedicated to the determination of asteroid ages with a uniform method . we computed the age(s ) for a total of 57 families with members . for the future work there remain families too small at present to provide reliable estimates , as well as some complex families ( 221 , 135 , 298 ) which may have more ages than we could currently estimate . future improvement of some already determined family ages is also possible by increasing family membership , revising the calibrations , and using more reliable physical data . * keywords * : asteroids , dynamics ; impact processes ; resonances , orbital ; trojan asteroids .

1607.01998

in our previous work ( @xcite , hereinafter referred to as paper i , and @xcite , paper ii ) , we have introduced new methods to classify asteroids into families , applicable to an extremely large dataset of proper elements , to update continuously this classification , and to estimate the collisional ages of large families . in later work ( @xcite , paper iii , and in @xcite , paper iv ) , we have systematically applied a uniform method ( an improvement of that proposed in paper i ) to estimate asteroid family collisional ages , and solved a number of problems of collisional models , including cases of complex relationship between dynamical families ( identified by clustering in the proper elements space ) and collisional families ( formed at a single time of collision ) . in this paper we solve several difficult cases of families for which either a collisional model had not been obtained , e.g. , because it was not clear how many separate collisions were needed to form a given dynamical family , or because our method based on v - shapes ( in the plane with coordinates proper semimajor axis @xmath4 and inverse of diameter @xmath5 ) did not appear to work properly . in many cases the difficulty had to do with complex dynamics , such as the effects of resonances ( either in mean motion or secular ) . in other cases the problem was due to the presence of interlopers ( members of the family by the automatic classification but not originated from the same parent body ) which can be identified by using physical data , including albedo , color indexes and absolute magnitudes . thus , to assess the level of success of the research presented in this paper , the reader should take into account that all the families discussed in this paper have been selected by the failure of our previous attempts to find a reasonable collisional model and/or to estimate an age . in this paper we discuss new family classifications for asteroids in mean motion resonances , in particular the @xmath0 and @xmath1 with jupiter , plus a revision and critical discussion of the classification inside the @xmath2 resonance . overall we have estimated @xmath6 additional ages for @xmath7 families . almost all the succesful computations have required some additional effort , on top of using the methods developed in our previous papers . two examples : first , removal of interlopers has played a critical role in several cases , to the point that @xmath8 of the @xmath7 families used for age estimation have seen removal of the namesake asteroid as an interloper , resulting in change of the family name ; second , the interaction of the yarkovsky effect with resonances has very different outcomes : outside resonances , the semimajor axis undergoes a secular drift , while inside a mean motion resonance the semimajor axis is locked and other elements can undergo a secular drift . inside a secular resonance the drift in semimajor axis appears not to be significantly affected , as shown by our numerical tests . the paper is organized as follows : in section [ s : meanmot ] we deal with the specific problems of families formed by asteroids locked inside the strongest mean motion resonances with jupiter , namely @xmath2 ( hildas ) , @xmath0 ( trojans ) , and @xmath1 ( griquas ) . in section [ s : secres ] we discuss the families affected by secular resonances ( involving the perihelia and nodes of the asteroid and the planets jupiter and/or saturn ) . in section [ s : complex ] we present two successful interpretations of families with strange shapes and many interlopers . in section [ s : obstest ] we discuss the families for which we either have not found a consistent collisional model , or we have found a model ( and computed an age ) but there are still problems requiring dedicated observational efforts . in section [ s : conclusions ] we draw some conclusion , not just from this paper but from the entire series , in particular discussing the limitations to the possibility of further investigating the collisional history of the asteroid belt with the current data set . the numerical data in connection with the computation of ages presented in this paper are collected in the tables 18 , analogous to those used in papers iii and iv : all the tables are given in the appendix . tables [ tab : tablefite ] and [ tab : tablefita ] describe the fit region in the @xmath9 and @xmath10 planes , respectively , where @xmath11 is the diameter in km . table [ tab : tablealbedo ] contains the albedo data from various sources , used to estimate @xmath11 . tables [ tab : tableslopese ] and [ tab : tableslopesa ] contain the results of the fit for slopes of the v - shapes , again in the two planes . table [ tab : tableyarkoparam ] contains the data to compute the yarkovsky calibration ( see papers i and iii ) , and finally tables [ tab : tableages_dedt ] and [ tab : tableages_dadt ] contain the calibrations , the estimated ages and their uncertainties . tables [ tab : tablefita ] , [ tab : tablealbedo ] , [ tab : tableslopesa ] , [ tab : tableyarkoparam ] and [ tab : tableages_dadt ] are partitioned by horizontal lines into sections for families of type fragmentation , of type cratering , and with one side only . we have found no additional young ages ( @xmath12 my ) . for the sake of brevity , unless absolutely necessary , we do not specifically quote in the text these tables and data they contain , but the reader is encouraged to consult them whenever either some intermediate result is needed , or some quantitative support of the proposed explanation is required . mean motion resonances with jupiter give rise to the kirkwood gaps , where the asteroid main belt has essentially empty regions corresponding to the most chaotic orbits : the main gaps correspond to the resonances @xmath13 and @xmath14 . for the order 1 resonances @xmath2 and @xmath1 there are gaps created in the regions of strongly chaotic orbits but also islands of relative stability , where families can be found . the @xmath0 resonance contains a large stable region containing the two trojan swarms ( preceding and following jupiter ) . the @xmath2 mean motion resonance with jupiter contains a large region in which the critical argument @xmath15 can librate for a very long time . the computation of proper elements specifically for the @xmath2 resonant orbits is possible @xcite , but not really necessary for the purpose of family classification . the reason is that most information is contained in the proper elements @xmath16 ( see figure [ fig : hilda_region ] ) , while the value of the smoothed and averaged @xmath4 is sharply constrained close to the libration center ( see figure [ fig : hildareg_ae ] ) . thus we are using in this region `` semi - proper '' elements @xmath17 computed with the same algorithm used for the non - resonant case , and some families can be identified by the hcm method all the same . two clusters are visible by eye in figure [ fig : hilda_region ] , at low proper @xmath18 the very well defined family of ( 1911 ) schubart and at higher proper @xmath18 a clan , that is a number of small families which may be interpreted as the remains of an eroded , very ancient family , which could have included the largest asteroid in the region , ( 153 ) hilda . the most prominent family we have found in the hilda region with our methods ( paper i , ii ) is the family of ( 1911 ) schubart now grown to 508 members ( of which 136 multiopposition , added in paper iv ) . the methods of paper iii to compute the age are not strictly applicable to this case , because the range in @xmath4 is very narrow , only @xmath19 au . indeed a v - shape in the plane with coordinates proper @xmath4 and @xmath5 is well defined but does not describe the amount of change in the proper elements due to the yarkovsky effect : it is constrained by the slope of the libration curve in the region where the family is . the hildas remain locked in the @xmath2 resonance ( see figure [ fig : hildareg_ae ] ) , thus the yarkovsky effect pushes the eccentricity either up ( for direct spin ) or down ( for retrograde spin ) @xcite . this results in a v - shape in the plane of proper @xmath20 and @xmath5 ( figure [ fig:1911_vshapee ] ) , which can be used to estimate the age , provided the yarkovsky calibration for @xmath21 is available . to obtain the calibration in @xmath21 we have used numerical experiments with a sample of @xmath22 clones of the orbit of ( 1911 ) with values of non - resonant @xmath23 assigned in the range @xmath24 au / y . the integrations were performed for a time span of @xmath6 my , then proper elements were computed for windows of @xmath8 my , shifting by @xmath25 my , giving a total of @xmath7 data points for each clone . to these we fitted a @xmath21 obtaining the results shown in figure [ fig : schubart_calib ] . then a regression line was fitted in the @xmath26 plane : @xmath27 and the coefficient @xmath28 has to be used to convert the @xmath29 calibration , which we obtained with the same procedure used in paper i and iii , into a @xmath21 calibration . the v - shape fit in ( @xmath30 ) plane ( figure [ 1911vshapee ] ) reveals consistent slopes ( see table [ tab : tableslopese ] ) and we can model this family as a single collisional family . the yarkovsky calibration converted in @xmath20 is @xmath31 in 1/y for the in side and @xmath32 for the out side , hence the age estimate is @xmath33 for in , @xmath34 for out side , in my . note that the uncertainty is completely dominated by the relative calibration uncertainty of @xmath35 , thus more work on calibration is needed . the slight curvature of the libration center line of figure [ fig : hildareg_ae ] results in a slow motion also in proper @xmath4 , but the secular @xmath29 is @xmath36 times what it would be outside the resonance . thus , in principle , the age could be computed by using the v - shape in @xmath10 and the calibration above , with similar results but poorer accuracy with respect to the results in the @xmath9 plane . [ 1911vshapee ] the high @xmath18 portion of the @xmath2 resonance shows some density contrast , see figure [ fig : hilda_region ] : the rectangle with @xmath37 and @xmath38 contains @xmath39 asteroids with proper elements . our hcm run , performed in a single step in this region ( see paper i ) , finds there two small families : 6124 with @xmath40 members , and 3561 with only @xmath41 . however , if we definine first core families among the subpopulation with magnitude @xmath42 , we find a core family of ( 153 ) hilda with only @xmath43 members . this family is not recognized as significant in the single step hcm because the limit number for significance , in this case , would be set at @xmath44 members . since this candidate family was important and only marginally discarded , we have decided to use the two step procedure , at least to assess the situation . then the second step , consisting in the attribution of fainter objects to the existing core families , gives a surprisingly small growth , namely the membership of family 153 increases only to @xmath45 members . the conclusion is that a family of ( 153 ) hilda can be considered to exist , but has a peculiar size distribution , strongly depleted in small size members ; this is not due just to observational selection , since in the @xmath2 resonance @xmath46 of the asteroids with proper elements have @xmath47 . the other families of ( 1911 ) schubart , of ( 6124 ) mecklenburg , not far from 153 , and of ( 3561 ) devine at lower @xmath4 and @xmath20 , are not affected at all , that is their membership is the same both with the one step and the two step procedure . at levels of distance above the one which is prescribed by hcm , families 153 and 6124 merge at @xmath48 m / s , 3561 merges at @xmath49 m / s , thus forming a very large `` super hilda '' family ( paper i , section 4.2 ) . an even bigger extension of the hilda family was proposed by @xcite , with a limiting distance of @xmath50 m / s ; they also claim that it is a very ancient family , with an age estimated at about @xmath51 gy . this is a legitimate method to propose a family , although not a way to prove that it is statistically significant . thus we have decided to investigate the properties of the super - hilda family , even if its significance can not be proven by hcm . the fact which is known about very ancient families is that they tend to erode : secular drifts due to yarkovsky , chaotic escape routes , and collisional comminution must necessarily occur and affect all members , but in a size - dependent way . the non - gravitational perturbations typically are proportional to @xmath5 , most resonances containing escape routes become accessible as a result of yarkovsky secular drift , collisional lifetimes are shorter for smaller objects . anyway , whatever the cause , ancient families dissipate by preferentially loosing the smaller members . thus it is possible that a super - hilda family did exist , but is now so much depleted in small members that it can not appear as a connected density contrast ( which is statistically significant ) , but gets disconnected in three smaller islands . the slope of the size distribution of the @xmath39 asteroids in the box cited above is @xmath52 , that is , extremely shallow . if such a family existed , then it could still show a v - shape , in this case of @xmath2 resonance in the proper @xmath53 plane . indeed , figure [ fig : hilda_vshapee ] does show a pronounced v - shape which can be used to compute an age . in the same time , neither of the small isolated groups does show any meaningful v - shape . the calibration of @xmath21 is obtained with exactly the same procedure as for family 1911 , and with the similar result : @xmath54 note that for the drift in @xmath4 due to the slope of the libration center the coefficient ( w.r . to the drift which would occur outside the resonance ) is also similar to the one for the family 1911 : @xmath55 . the slopes ( see table [ tab : tableslopese ] ) are consistent . the yarkovsky calibration converted in @xmath20 is @xmath56 in 1/y for the in side and @xmath57 for the out side . the age estimate is @xmath58 for the in side , @xmath59 for out , in my ; the uncertainty is again dominated by the @xmath35 relative uncertainty in calibration . of course the nominal value and higher values in the formally acceptable range are not possible , but the lowest part of the range of uncertainty gives legitimate ages , older than any other we have measured , and compatible with @xcite results . such an old age estimate can indeed justify the very shallow size distribution ; however , with the current number of identified members we can not claim this is a strong confirmation of the age . note that , if this is really a very old family strongly depleted in small members , we should have here a rare case in which we can expect that future observations , extending the completeness limit to fainter objects , will _ not _ produce a significant increase of the number of family members , as opposite to the case of the large majority of asteroid families . this shall be futher discussed in section [ s : superhilda ] the jupiter trojans dataset includes now @xmath60 objects surrounding l4 ( the `` greek camp '' ) , and @xmath61 surrounding l5 lagrange point ( the `` trojan camp '' ) . the proper elements have been computed with the synthetic method described in @xcite , and are different in that the proper @xmath4 is replaced by the semi - amplitude @xmath62 of the libration in semimajor axis due to the @xmath0 resonance for l4 , @xmath63 for l5 . ] ; the proper elements @xmath20 and @xmath64 are computed in the same way as in the main belt . we have used , separately for the l4 and the l5 swarm , a single step hcm method , with the same metric of @xcite , that is using for the differences in @xmath62 the same @xmath65 coefficient used for the differences in @xmath4 in the main belt ( anyway , the differences in @xmath62 play a minor role due to the fact that their range of values is narrow : @xmath66 au ) . the stalactite diagram for the l4 objects , obtained by using a value of @xmath67 , is shown in fig . [ fig : stalatl4 ] . since we found for the quasi random level a value of @xmath68 m / s , this means that families must be found at @xmath69 m / s . the stalactite diagram for the l5 trojan asteroids is shown in fig . [ fig : stalatl5 ] . the adopted value of @xmath70 for l5 is @xmath71 and the qrl is found at @xmath72 m / s . l5 families are therefore defined at @xmath73 m / s . by looking at the stalactite diagrams for l4 and l5 , one sees a multitude of thin and in several cases very deep stalactites emerging from the bulk of the two populations . for some of them , mainly in l4 , the unusual thing is just their thin and compact morphology . this behaviour is not unexpected _ a priori_. the dynamical environment in the jupiter trojan swarms is such that the usual mechanisms of dynamical dispersion of families at work in the main belt , including primarily the yarkovsky effect , are much weaker here , even over long time scales . as a consequence , swarms of fragments produced by collisions occurred in the trojan swarms in very ancient epochs can be expected to survive until today with very little alterations of their orbital elements , whereas in the main belt identical groupings would tend to disperse and be dissipated over relatively short time scales . so , the general morphology of the stalactite diagrams shown in figs . [ fig : stalatl4 ] and [ fig : stalatl5 ] can be interpreted in terms of very compact groups which may well be the survivors of events that occurred in very ancient times . the relatively small numbers of objects can also be a consequence of the fact that the smallest objects at this heliocentric distance have not yet been discovered . it is also possible that for the jupiter trojan population our criteria for family acceptance , which have been defined to be applicable primarily to the case of main belt asteroids , could be overly severe . in other words , we can not rule out the possibility that some , relatively deep stalactites that do not satisfy our criteria to be accepted as families , might still be the remnants of ancient collisional events . unfortunately , one of the major tools we have at disposal to test similar hypotheses among main belt astreroids , namely the evidence coming from spectroscopic and/or polarimetric evidence , are much less powerful a priori when applied to the jupiter trojans , which are quite homogeneous in terms of spectral properties and sufficiently faint to be a challenge for polarimetric measurements . somewhat less compact , but by far the most populous group in the l4 trojan swarm is the family of ( 3548 ) eurybates . it includes @xmath74 members , @xmath75 of them being multi - opposition objects . strictly linked to eurybates are 8060 and 222861 , two small groupings , consisting of @xmath76 members , including @xmath77 multi - opposition objects , and @xmath78 members ( @xmath79 multi - opposition ) , respectively . these families are very close to eurybates , and join it already at the level of @xmath68 m / s , just @xmath6 m / s above the level for statistical significance . the family of ( 3548 ) eurybates is located in the tip of a larger stalactite branch well visible in fig . [ fig : stalatl4 ] . this also corresponds to the most evident grouping with proper sine of inclination about @xmath80 and proper eccentricity around @xmath81 in the plot shown in fig . [ fig : l4sinie ] . we have used the same calibration method of section [ s : hilda ] to get the right order of magnitudes of the secular effects in all proper elements . on @xmath62 the secular effect of yarkovsky is @xmath82 of the corresponding effect in @xmath4 outside the @xmath0 resonance , thus dispersion in @xmath62 is negligible even over the age of the solar system . for proper @xmath20 the ratio of @xmath21 with the non - resonant @xmath29 is @xmath83 , for proper @xmath18 it is @xmath84 . thus , although it is possible to plot the family members in either a @xmath9 plane or in a @xmath85 plane and measure slopes of the v - shapes , the values of the ages which would result from the hypothesis that these v - shapes describe a yarkovsky effect are by far too large : the smallest one resulting from the use of @xmath85 is @xmath43 gy , little more than the age of the universe . this implies that the distribution of proper elements of the family members is essentially only due to the original distribution of relative velocities of the fragments , with very minor contribution from any non - gravitational perturbation . thus we can not estimate the ages of the trojan families , but we can assess the distribution of the original velocities , something which is not possible in the asteroid main belt ( apart from the case of recent families ) . as a check of this approach , we have computed the standard deviation of the proper elements of the members of 3548 , and found @xmath86 for proper @xmath20 , @xmath87 for proper @xmath18 and @xmath88 au for @xmath89 , corresponding to relative velocities after escaping from the gravitational well of the parent body of @xmath90 , @xmath91 and @xmath92 m / s , respectively . the parent body can be estimated from the family volume to have had @xmath93 km , thus the escape velocity can be roughly estimated ( we have very poor density data ) at about @xmath94 m / s . thus the values of the relative velocities obtained from the present values of the proper elements are fully compatible with an original velocity field . the families of ( 624 ) hektor and of ( 9799 ) 1996 rj , with @xmath44 objects ( @xmath95 multiopposition ) and @xmath92 ( @xmath95 multi - opposition ) , respectively , correspond to two extremely compact , deep and well isolated stalactite branches which are evident at the left and rigth ends of the stalactite diagram shown in fig . [ fig : stalatl4 ] . both families have relatively high inclination ( @xmath96 for 624 , @xmath97 for 9799 ) . their extremely compact structure is clearly visible in the sine of inclination - eccentricity plot ( fig . [ fig : l4sinie ] ) where they correspond to very compact clusters ressembling big blots in the middle - right part of the figure , at proper eccentricity values slightly above and slightly below @xmath81 , respectively . by the same argument used for 3548 , the distributions of proper elements should reflect the original field of relative velocities after the exit from the sphere of influence of the parent body : both these families are of cratering type . ( 624 ) hector is a very large , oddly shaped body ( suspect contact binary ) with a satellite . the rms of proper elements differences , with respect to those of the parent body ( 624 ) , are @xmath98 au for @xmath62 , @xmath99 for @xmath20 , @xmath100 for @xmath18 , corresponding to velocities after escape of @xmath101 and @xmath102 m / s respectively . these numbers are so much smaller than the escape velocity from ( 624 ) hektor , which is about @xmath103 m / s , to point out to a marginal escape from a cratering event , possibly the same resulting in the formation of a satellite and also of the contact binary . the same argument can be applied to the family of ( 9799 ) 1996 rj , which is also of cratering type , with rms of differences in proper elements corresponding to @xmath104 , and @xmath44 m / s for @xmath105 , and @xmath18 , respectively . all the three examples analysed above appear to confirm that the trojan asteroid families are * fossil * families , frozen at their proper elements which are essentially determined by the original ejection velocity field . on the other hand , these velocity fields are remarkably different in cratering vs. fragmentation cases . looking at the same figure , we also note that few other , less compact groupings are present , but they do not correspond to acceptable families according to our hcm criteria . as mentioned above , this might be a consequence of applying overly conservative criteria of identification in a region in which families tend to last over very long timescales without relevant erosion . some of these groups might `` grow '' in the future , and become acceptable , when the magnitude completeness limit for the trojans will improve . a good example may become a family of ( 2148 ) epeios , proposed by @xcite , which can be found by relaxing the hcm significance criteria . the griquas are a comparatively small ( @xmath106 asteroids ) population of asteroids confined in the 2/1 mean motion resonance with jupiter , in a tight range of `` semi - proper '' semimajor axis : @xmath107 . by the same argument used for hildas , the computation of true resonant proper elements is not necessary . griquas are fairly spread in proper eccentricity and sine of inclination , between @xmath108 and @xmath109 and between @xmath108 and @xmath110 , respectively , see figure [ fig : griqua_ie_fam ] . the classification of paper i did not find any statistically significant family in this region . however , the statistics might have been distorted by merging the sparsely populated @xmath1 region with the very populated zone 4 , with @xmath4 above the @xmath14 gap . thus we have rerun a dedicated classification for the griquas only : of course the quasi - random background became sparse , thus allowing to detect smaller families . in this region , we adopt for @xmath70 a value of @xmath111 . the stalactite diagram is shown in fig . [ fig : stalatgriquas_nlim16 ] . the quasi random level for the griquas is found to be @xmath112 m / s , implying that the distance level to identify families is @xmath50 m / s . only one such group is present , the 11097 family , which has @xmath113 members , including @xmath8 multiopposition objects . it corresponds to a group that is apparent in figure [ fig : griqua_ie_fam ] , where it is located at very low proper inclination . this is a small family , which deserves further study , also to be confirmed by observation of a homogenous composition . there are in our classification five families with enough membership to be suitable for age estimation , but affected in significant way by the secular resonances , in which either all or a good portion of the members are locked ; for a map of the secular resonances in the main belt , see @xcite[fig . 7 - 11 ] . to provide a theory of the long term dynamical evolution of asteroids affected by both secular resonances and yarkovsky effect is beyond the scope of this paper ; some cases have been discussed in the literature @xcite . our goal for this paper is just to estimate ages , thus we are going to use an empirical approach very similar to the one we have used for the hildas : by numerical integration we perform a calibration for the yarkovsky effect inside the resonance versus the yarkovsky effect which would occur to the same asteroid if the resonance was not there . in this way we solve all five cases , three with either all or a good portion of the members locked in the resonance ( families 5 , 363 , 945 ) , and two ( families 283 , 686 ) which have only a minor portion of members affected by a secular resonance , such that the v - shape in @xmath10 can not change by the action of the resonance . to understand the shape in proper elements space of the family of ( 5 ) astraea is a complex problem , both because almost all its members are locked in the @xmath114 nonlinear secular resonance ( with values @xmath115 arcsec / y ) , and because the proper @xmath20 has a large spread , up to @xmath116 , a value such that it must be due to a large number of mean motion resonances affecting the stability of the proper elements over very long time spans . note that many of these mean motion resonances are with inner planets , thus they are not included in the computation of the proper elements for the asteroids of the outer main belt ( @xmath117 au ) @xcite . in the case of the family 5 , with lowest value of proper @xmath118 au and with comparatively high eccentricity , to include the inner planets in the dynamical model , as in @xcite , could improve the results . physical observations available for this family are few , because it originated by a cratering , thus has especially small members : @xmath119 of the members have @xmath120 km . wise albedo data @xcite with nominal @xmath121 are available for only @xmath122 of the family members ; out of these , @xmath123 of the albedo values are @xmath124 , that is they indicate interlopers , since ( 5 ) has iras albedo @xcite @xmath125 . this fraction of interlopers is somewhat high , but not extraordinarily so @xcite : it could have been affected by the `` compression '' resulting from the standard computation of proper elements which does not specifically account for the secular resonance . because this compression applies to both the real family family and the background , this can explain a larger fraction of interlopers . anyway , this comparatively large fraction of interlopers is a fact and we need to take it into account in our age estimation procedure . if we analyse the v - shape in the @xmath10 plane ( figure [ fig:5_vshapea ] ) we see that the computations of the slope are possible , although the in side is affected by a group of four members too large for their position , near the inner edge of the family : e.g. , ( 4700 ) carusi has an estimated @xmath126 km . near the center of the family , ( 1044 ) teutonia has an iras estimated @xmath127 km , which is incompatible with the cratering origin of the family . moreover , the fit for the in side was very poor , to the point that the inverse slope was barely significant . we have therefore decided to discard a priori 4 interlopers , marked with a red circle in figure [ fig:5_vshapea ] ; then also ( 1044 ) teutonia was automatically discarded as outlier in the fit . after these removals , the family contains no member with @xmath128 km and the fit has been much improved , being compatible with a single collision origin and with a similar uncertainty on the two sides . our choice of four members interpreted as interlopers is plausible also because of the large fraction of interlopers in general in the family . before interpreting the inverse slopes of the v - shape as measures of age , we need to perform a calibration of the yarkovsky effect taking specifically into account the dynamical environment of the family . for this purpose we have performed a numerical integration for @xmath6 my , with inner planets included , for 200 clones of ( 5 ) astraea . the right hand side of the equations of motion included yarkovsky accelerations corresponding , for a non - resonant objects , to values of secular @xmath29 in the range @xmath129 au / y . after computing for each clone 9 sets of proper elements ( as in section [ s : schubart ] ) , we computed for each clone the `` resonant '' @xmath130 drift , then plotted the result in figure [ fig:5_calib ] with the fit to a line which has slope @xmath131 . in other words : @xmath132 meaning that the amount of secular change in proper @xmath4 due to yarkovsky is not significantly affected by the secular resonance . we can therefore conclude on the age as reported in table [ tab : tableages_dadt ] : @xmath133 for the in side , and @xmath134 for out , in my ; the age of family 5 is old , but far from ancient , and represents another very good example of cratering , significantly younger than vesta . our classification includes a dynamical family of ( 110 ) lydia , whith almost all members locked in the @xmath135 nonlinear secular resonance ( the same as for family 5 ) . however , ( 110 ) has wise albedo @xmath136 while @xmath137 of the members having wise data ( with @xmath121 ) have albedo @xmath124 . thus ( 110 ) lydia is a very likely interloper and the family namesake should be ( 363 ) padua . we applied this change in the tables , starting from table [ tab : tableslopesa ] . the v - shape of figure [ fig:363_vshapea ] , besides confirming the need to remove ( 110 ) from the family list , gives good fits for the inverse slopes with compatible values , suggesting a single age on the in side and @xmath138 on the out side , thus the value obtained for family 363 implies that the two families are not from the same collision , thus confirming the choice made in paper iv of not merging these two . ] . in the same way as we did for family 5 , we have computed , we have performed the 10 my numerical integration with a model including this time only the outer planets . ] the calibration for the secular @xmath130 inside the resonance as a function of the one for @xmath23 outside the resonance , and we got a figure very much like figure [ fig:5_calib ] , with a fit slope of @xmath139 and some noise for positive values of @xmath23 , presumably due to a number of mean motion resonances , including 3j-1s-1a . we do not think that the difference between the calibration @xmath140 of the previous case and this one is significant , anyway it gives a negligible contribution to the relative uncertainty of the calibration . after calibration , the ages reported in table [ tab : tableages_dadt ] indicate a family at the low end of the `` old '' range : @xmath141 for the in , and @xmath142 for the out side , in my , compatible with a single collisional origin of fragmentation type , although with a largest remnant ( 363 ) containing as much as @xmath143 of the family volume . the family of ( 945 ) barcelona has about @xmath144 of the members ( the portion of the family with proper @xmath145 ) strongly affected by the resonance between the perihelion precession of the asteroid @xmath146 and the one of jupiter @xmath147 , with @xmath148 arcsec / y . another unidentified nonlinear secular resonance is also relevant for the dynamics of most members . the family is bound on the out side by the @xmath149 mean resonance with jupiter , thus its v - shape ( figure [ fig:945_vshapea ] ) has only the in side . the in side fit is good , with some outliers , allowing to estimate the inverse slope with good accuracy . we confirm our approach of a purely numerical calibration of @xmath130 , not only because a theory of the evolution under yarkovsky effect plus two overlapping secular resonances is out of the scope of the present paper , but as it also appears exceedingly difficult . the same calibration as in the case of family 363 gives a slope of @xmath130 versus @xmath23 of @xmath150 . the plot shows much more noise than in the previous two cases , some of which is associated with the 11/4 resonance : this noise does not affect the family which has been depleted by the resonance . still our conclusion is the same : the effect of the secular resonance on the rate of yarkovsky is small , to the point of not being significant with respect to the relative calibration uncertainty . the age of @xmath151 my is again at the low end of the `` old '' range , the family is of fragmentation type but with a comparatively large remnant ( 945 ) with @xmath152 of the family volume . our classification contains a dynamical family 194 , with @xmath153 members . the family namesake ( 194 ) prokne has albedo @xmath154 ( wise ) thus it does not belong to the family , which , after removal of members with albedo @xmath155 and @xmath156 , has a mean albedo @xmath157 . the low albedo of ( 194 ) has been confirmed with recent polarimetric measurements by @xcite . after the removal of such a large interloper , ( 686 ) gersuind remains the lowest numbered member , with albedo @xmath158 , thus the family has to be called family 686 ; we applied this change in the tables , starting from table [ tab : tableslopesa ] . a family of ( 686 ) gersuind , without ( 194 ) , had already been proposed by @xcite . as afore mentioned , this family is only partially affected by the @xmath159 nonlinear secular resonance , with only some @xmath160 of members in the out side having the corresponding frequency @xmath161 arcsec / y . the v - shape of the family 686 ( figure [ fig:686_vshapea ] ) , with ( 194 ) and other members with incompatible albedo removed , still results in a poor fit for the in side , with @xmath162 , but in a better one for the out side @xmath163 . after calibration , which yielded regression line slope of 1.003 , the estimated ages are @xmath164 and @xmath165 my , with the out value much more significant . note that the members ( 1609 ) , ( 27496 ) and ( 31402 ) are removed as outliers from the slope fit : since there are no albedo data their status as interlopers remains to be confirmed . the family of ( 283 ) emma contains dark asteroids , with few known higher albedo interlopers : the mean wise albedo is @xmath166 , ( 283 ) itself has an iras albedo @xmath167 . a small fraction ( about @xmath168 ) of its members are affected by the nonlinear secular resonance @xmath169 , which opens a kind of gap between two parts of the family , well visible in the projection on the proper @xmath170 plane , for low values of @xmath4 . the v - shape ( figure [ fig:283_vshapea ] ) results in a good fit on the in side with @xmath171 , thus it does not appear to be much affected by the secular resonance . to the contrary , on the out side the fit is poor : @xmath172 . after calibration performed on the second largest family member ( 32931 ) ferioli ( regression line slope 1.010 ) , the age estimates are @xmath173 on in side , and @xmath174 my on the out side . the family is of cratering type : assuming same albedo for all , the volume of the family members without ( 283 ) is about @xmath175 of the total . thus the occurence of two separate cratering events is a reasonable hypothesis , which is consistent with a double jet shape as shown in the proper @xmath170 projection . the family of ( 25 ) phocaea has a one - sided v - shape in the @xmath10 plane ( figure [ fig:25_vshapea ] ) , with ( 25 ) at the high @xmath4 end . the missing part of the v - shape is due to the resonances bordering the stable region , including the @xmath13 j. the region contains about @xmath176 asteroids , while the dynamical family 25 contains only @xmath177 of these , thus it represents a substructure whith distinctive number density inside the stability region . the fit gives @xmath178 . the only problem with the shape of the family as shown in figure [ fig:25_vshapea ] is that there is obviously a lack of family members near the value of proper @xmath4 of ( 25 ) phocea , which occurs before the high @xmath4 end of the stability region . this might be interpreted as a yorp induced central gap @xcite . @xmath179 of the family members are affected by a nonlinear secular resonance : they have @xmath180 arcsec / y . this fraction is relevant for the age computation because it may affect the side of the v - shape , thus we need calibration , which has been done in the same way by using clones of ( 323 ) brucia ( which is locked in the resonance ) , but only with negative yarkovsky drift in proper @xmath4 . the slope of the @xmath130 with respect to the one outside the resonance was found to be @xmath181 , thus once more the calibration does not need to take into account the resonance . thus the calibrated age is @xmath182 my , an ancient family according to the terminology introduced in paper iii . by irregular families we mean dynamical families with internal structures that can not be interpreted as a simple , single collisional family ; this either prevents the computation of an age , or results in an age which refers to only a portion of the dynamical family . in some cases , an interpretation as two or more collisional families ( partially overlapping ) is possible , e.g. , vesta , eunomia , agnia . alternatively , this apparent complexity could be due to a number of comparatively large interlopers . in other cases the shape of the family can have different , even more peculiar interpretations . in this section we shall propose a solution , including an age estimation , for one of the previously unsolved cases . as explained above , family of ( 283 ) emma obviously fits to this class , but being also crossed by the secular resonance , we preferred to deal with it in section [ s : secres ] . the family of ( 145 ) adeona has a peculiar shape in the proper @xmath4 , @xmath5 plane which appears different from the standard v - shape . there appears to be a common v - shape for a narrow range of values , roughly @xmath183 , while the portion of the family for @xmath184 is much less dense and does not appear to be delimited by the main v - shape . the histogram of number density as a function of proper @xmath4 ( see figure [ fig:145_hista ] ) shows a sharp decline of dynamical family membership in the region with @xmath4 below the three 3-body resonances , 4j-3s-1a at @xmath185 , 2j+2s-1a at @xmath186 ( @xcite ) , and another not identified one at @xmath187 au . this complex structure can have different interpretations : the portion of the dynamical family for @xmath188 could be another , smaller family with lower number density . the same portion could also be a continuation of the family of ( 145 ) adeona with decreased number density , since a large fraction of members could have been removed by the resonances : then the largest members in the low @xmath4 region should be interlopers . there is , however , a certain indication from physical observations : the dynamical family contains mostly dark asteroids - @xmath189 with wise albedo @xmath124 - with a few interlopers . the unexplained group of comparatively large members for @xmath190 au contains a fair fraction of interlopers , characterized either by albedo @xmath191 or by the color index @xmath192 ( see the circled crosses in figure [ fig:145_vshapea_in ] ) . if we just fit the data to a v - shape with the in side only , we get an inverse slope @xmath193 . as shown in figure [ fig:145_vshapea_in ] , the number of outliers removed to obtain this fit is substantial , on top of the interlopers removed by physical observations . the lack of the out side can be due to the strong @xmath194j resonance at @xmath195 au where the family abruptly ends up . thus the age can be estimated as @xmath196 my . as discussed in section [ s : obstest ] , this result needs to be confirmed by showing that to discard that many data points is justified . as described in paper i , our approach to asteroid family analysis is based on using the proper elements data first to identify the families , then adding the absolute magnitude data ( with complementary information from the statistics of the albedo data in each family ) to estimate ages , then use other available physical data to refine the results , e.g. , identifying complex familes with more than one collision and interlopers which can deteriorate some age estimates . however , in many cases we do not have enough physical data to solve the problem . thus we discuss in the following several open problems for which dedicated observations of asteroid physical properties could decisively contribute to the solution of some interesting problems . the single dynamical family of ( 135 ) hertha has been known for a long time to include multiple collisional families , overlapping in the proper elements space @xcite . however , we have been able to compute only one age ( paper iv ) , for the subfamily of ( 650 ) amalasuntha , which is partly separate in the @xmath197 plane and also distinguished by low albedo and other physical properties typical of the c taxonomic complex . a possible splitting of the low - albedo component of the hertha complex in more than one group , in particular the presence of an independent eulalia family @xcite has not been supported by recent spectroscopic observations by @xcite . the complementary `` bright '' subfamily does not have a recognizable v - shape . the recently proposed hertha 1 and 2 families @xcite were found by using color index separation based on sdss parameters . the presence of three density contrast features is confirmed by our last classification , in which family 135 has a total of @xmath198 members projection of family 135 on our family visualizer at http://hamilton.dm.unipi.it/astdys2/plot/index.html ; note the bimodality of the bright component shown by the number density histogram on the right border of the plot . ] ; however , the location of the two `` bright '' subfamilies does not correspond precisely to what was proposed by @xcite . moreover , the role of ( 135 ) hertha as ( core of ) parent body has not been established . targeted physical observations of several members of family 135 , by selecting asteroids in such a way that they belong to the separate density features , could provide a robust collisional interpretation . for the complex of three dynamical families of ( 298 ) baptistina , ( 883 ) matterania , and ( 2076 ) levin ( see figure [ fig : baptistina_ae ] ) we do not have a collisional interpretation . the three families have few intersections in the current classification , but we do not see any way to model them as the result of a single collision . even an interpretation in terms of two collisions does not solve all the problems . the physical data ( see figure [ fig : baptistina_sdss ] for the sdss data ) only make the problem more difficult , because all three families appear heterogeneous in composition ; the same result is obtained by using wise albedo data . therefore we have abstained from merging these three families , in spite of the intersections , and we have computed only one age from the one - sided v - shape of 2076 ( paper iii ) . in the literature there are several `` baptistina families '' with rather different membership , conflicting age estimates and conjectures on composition , all based on the assumption that there is just one collisional family . to solve this problem more physical observations are needed ; they should be of high quality and for a large enough set of asteroids in each of the three families . the family of ( 221 ) eos is not a problem , in that most of its membership is well known and a single age has been computed ( paper iv ) . however , interlopers with low albedo could belong to another family formed by cratering of ( 423 ) diotima , which is much larger than ( 221 ) . targeted physical observations in the region near ( 423 ) in proper elements space are required to confirm the existence of this separate but completely overlapping family . another problem is about the family of ( 179 ) klytaemnestra , which is very close in proper elements space to the family 221 but has no intersections with it . this family v - shape in @xmath199 has a consistent collisional interpretation neither with 1 nor with 2 collisions . the only explanation we may propose is that family 179 could be contaminated by a large number of interlopers ( in the low @xmath4 portion ) transported from family 221 , possibly through the secular resonance @xmath200 , well known to affect eos family @xcite . this could be confirmed by physical observations of two sets of asteroids , belonging to 221 and 179 , respectively , but close in proper elements space . to be able to compute an age for the family of ( 145 ) adeona , we had to discard a comparatively large number of interlopers , based on the existing physical data . additionally , there are many outliers for the slope fit for which we have no physical data . a precisely targeted campaign of physical observations of members of the dynamical family 145 with @xmath201 and @xmath202 km could either confirm our results , including the age , or suggest another interpretation , like the existence of a subfamily . the family of ( 686 ) gersuind , emerged from the dynamical family 194 , may be interesting in connection with the problem of the barbarians , asteroids with peculiar polarimetric properties @xcite ) . several barbarians were found in the neighboring family of ( 729 ) watsonia ( @xcite ) . after the removal of the dark and too bright interlopers from 194 , the family 686 has albedo properties very similar to 729 . thus the members of 686 should be subjected to investigation with polarimetry . the extended hilda family we discussed in section [ s : hilda ] is only a proposal , for which we do not have a statistically significant confirmation . on the other hand , if it was really there , it would be especially interesting as one of the oldest families . given the difficulty of getting confirmation from physical observations ( essentially all the asteroids in the hilda region for which data are available are p types , with low albedo ) , the only way to test the `` super - hilda hypothesis '' observationally is to discover more hildas in the medium inclination region . it should be possible to exclude that the very shallow size distribution is due to observational selection by pushing the complenesss of the catalogs down to @xmath203 km . then this would confirm that the alleged super - family is indeed strongly eroded , thus very ancient . in this paper we have attempted to give a collisional model to a number of families for which the same attempt had previously failed . most of these families were either locked in resonances or anyway significantly affected by resonances , both mean motion and secular . to estimate an age for the family required in each of these resonant cases to apply a specific calibration for the yarkovsky effect , which in principle could be different in each case . for the largest families found in the hilda region , consisting of asteroids locked in the 3/2 resonance with jupiter , the yarkovsky effect results in a secular change in eccentricity , thus the v - shape technique had to be applied in the @xmath9 plane . there we found family 1911 with a good age determination and family 153 of the _ eroded _ type , that is which can be seen visually in the plots in the proper @xmath204 plane but can not be confirmed by the statistical tests of the hcm method . if such a family exists , then its age must be @xmath205 gy , extremely ancient or even primordial , which is consistent with the hypothesis that this family is depleted to the point of not having a significant density contrast with the background . for the trojans , that is the asteroids locked in 1/1 resonance with jupiter , we are presenting in this paper a new classification which identifies a number of families by using synthetic proper elements and a full hcm method . hcm is well established and has been succesfully applied to the main asteroid belt , but the results in the trojan swarms indicate that families there have a very different structure . numerical calibrations have shown that the yarkovsky perturbations are ineffective in determining secular changes in all proper elements . this implies that all trojan families are _ fossil _ families , frozen with the original field of relative velocities , which are small for cratering families , for the fragmentation case somewhat larger , but still limited to the order of the escape velocity from the parent body . thus we find no way to estimate the ages of the families , while they can be a reliable source of information on the original velocity field immediately after the collision . we have found a new family among the griquas , locked in the 2/1 resonance with jupiter . we have analysed 6 large families affected by secular resonances , mostly the nonlinear ones . we have used a numerical calibration method , which has shown in all cases that the secular resonances do not affect significanly the secular change of proper @xmath4 , thus the v - shape method in the @xmath199 plane can be used to compute the age in the standard way ( as in paper iii ) . the solution of some of the cases has been possible only by removal of a number of assumed interlopers . this applies especially to family 145 which otherwise would appear to have a double v - shape . in conclusion , with the 10 ages computed in this paper , the situation is the following : of the 25 families with @xmath206 members , we have computed at least one age for all but 490 , which has a too recent age , already known , unsuitable for our method . of the 19 families with @xmath207 members , excluding 778 which has a too recent age , already known , there is only one case left without age , namely 179 ( see section [ s : eos ] ) . of the 24 families with @xmath208 we have computed 6 ages , the others we believe could only give low reliability results . in this range only for families with a small range in proper @xmath4 we can compute a reliable slope . thus we can compute only young to medium old ages ( @xmath209 my ) . as an example , the family of ( 1222 ) tina , with only @xmath210 members , has a short @xmath4 range and appears to have a well defined v - shape in @xmath10 , but the inverse slopes on the two sides are incompatible ( @xmath211 in , @xmath212 out ) , thus we should conclude it is the result of two separate collisions . with so few data points , this does not appear a mature result , but something to be reanalysed when the number of members is at least doubled . among the problems with family ages we have left open , there are three complex families which we believe have more ages than we have estimated : 221 , 135 , and 298 . two apparently complex cases we believe have been solved : 145 , 283 , although some confirmation would be useful for 145 . overall , we believe we have completed a useful work , which is based on the dataset of proper elements we have produced , and by using physical data only as check ( apart from absolute magnitudes used for ages ) . we think it is anyway a good start towards the goal of an absolute chronology of the main collisions in the asteroid main belt , see in figure [ fig : famage_plotnf ] all the family ages we have been able to compute so far ( in papers iii , iv , and the present one ) . we have used for the error bars , for the cases where two values in and out are available and compatible , standard deviations computed by the formulae from @xcite[section 7.2 ] , applicable to all the cases in which two least square fits can be merged under the assumption that the parameters to be determined are the same . this needs to be applied to the term in the error budget due to the fit of the two slopes , under the assumption that they are two solutions for the same physical quantity . thus the two fits for the slopes reinforce each other ( if compatible ) , while the calibrations are affected by the same errors ( e.g. , in the density ) . this more complex formula is an improvement with respect to what we did in paper iii : @xmath213 given our complete open data policy anyone can try by himself to compute other ages with our proper elements and family classification data . however , we recommend caution : ages computed with insufficient data could be unreliable . it is also possible to improve ages ( and decrease uncertainty ) by using our computed slopes but revising the yarkovsky calibration with a specific effort for each individual family . the authors have been supported for this research by : itn marie curie `` stardust the asteroid and space debris network '' ( fp7-people-2012-itn , project number 317185 ) ( a.m. , z.k , b.n . , and g.t . ) , the department of mathematics of the university of pisa ( a.m. ) , and the ministry of education , science and technological development of serbia , under the project 176011 ( z.k . , b.n . ) . f.s . is a fellow of the cnes postdoctoral program - project called `` asteroid observations in the gaia era '' . 99 bottke , w.f . , d. vokrouhlick , m. bro , d. nesvorn , & a. morbidelli : 2001 , dynamical spreading of asteroid families by the yarkovsky effect , _ science _ * 294 * 16931696 bottke w.f . , d. vokrouhlick , d.f . rubincam & m. bro : 2002 . the effect of yarkovsky thermal forces on the dynamical evolution of asteroids and meteoroids , in _ asteroids iii _ , ( w.f . bottke jr . eds ) , university of arizona press tucson , pp . 395408 . bro , m. , d. vokrouhlick , a. morbidelli , d. nesvorn & w.f . bottke : 2011 , did the hilda collisional family form during the late heavy bombardment ? , _ mnras _ * 414 * , 27162727 . cellino , a. , v. zappal , a. doressoundiram , m. di martino , ph . bendjoya , e. dotto , & f. migliorini : 2001 , the puzzling case of the nysa - polana family , _ icarus _ * 152 * , 225237 . cellino a. , s. j. bus , a. doressoundiram , & d. lazzaro : 2002 , spectroscopic properties of asteroid families , in _ asteroids iii _ , ( w. f. bottke jr . , a. cellino , p. paolicchi , & r. p. binzel , eds ) , university of arizona press , tucson , pp . 633643 . cellino a. , i.n . belskaya , ph . bendjoya , m. di martino , r. gil - hutton , k. muinonen , & e.f . tedesco : 2006 , the strange polarimetric behavior of asteroid ( 234 ) barbara , _ icarus _ * 180 * , 565567 . cellino a. , s. bagnulo , p. tanga , b. novakovi , & m. delb : 2014 , a successful search for hidden barbarians in the watsonia asteroid family , _ mnras letters _ * 439 * , l75l79 . de len j. , n. pinilla - alonso , m. delbo , h. campins , a. cabrera - lavers , p. tanga , a. cellino , ph . bendjoya , j. gayon - markt , j. licandro , v. lorenzi , d. morate , k.j . walsh , f. demeo , z. landsman , & v. ali - lagoa : 2016 , visible spectroscopy of the polana - eulalia family complex : spectral homogeneity , _ icarus _ * 266 * , 5775 . devogele m. , a. cellino , j.p . rivet , s. bagnulo , c. pernechele , d. vernet , g. massone , l. abe , & p. tanga : 2016 , the calern asteroid polarimetric survey using the torino polarimeter : assessment of instrument performances and first scientific results , submitted to _ mnras_. dykhuis m.j . & r. greenberg : 2015 , collisional family structure within the nysa - polana complex , _ icarus _ * 252 * , 199211 . grav , t. , a. k. mainzer , j. bauer , j. masiero , t. spahr , r. s. mcmillan , r. walker , r. cutri , e. wright , p. r. eisenhardt , e. blauvelt , e. debaun , d. elsbury , t. gautier , s. gomillion , e. hand , & a. wilkins : 2012 , wise / neowise observations of the hilda population : preliminary results . _ astrophys . j. _ * 744 * , issue 2 , article i d . 197 , 15 pp . kneevi z. & a. milani : 2000 , synthetic proper elements for outer main belt asteroids . _ astron . _ * 78 * , 1746 . kneevi z. & a. milani : 2003 , proper element catalogs and asteroid families . astrophys . _ * 403 * , 11651173 . kneevi , z. , a. milani , a. cellino , b. novakovi , f. spoto , & p. paolicchi : 2014 . automated classification of asteroids into families at work . _ proceedings iau symposium 310 : complex planetary systems _ ( z. kneevi and a. lemaitre , eds . ) cambridge univ . press . , pp 130133 . masiero , j.r . , a. k. mainzer , t. grav , j. m. bauer , r. m. cutri , j. dailey , p. r. m. eisenhardt , r. s. mcmillan , t. b. spahr , m. f. skrutskie , d. tholen , r. g. walker , e. l. wright , e. debaun , d. elsbury , t. gautier iv , s. gomillion , & a. wilkins : 2011 , main belt asteroids with wise / neowise . i. preliminary albedos and diameters . _ astrophys . j. _ * 743 * , issue 2 , article i d . 156 , 17 pp . migliorini , f. , v. zappal , r. vio , & a. cellino : 1995 , interlopers within asteroid families . _ icarus _ * 118 * , 271291 . milani a. : 1992 , proper elements for trojan asteroids , _ b.a.a.s . _ , * 24 * , 963 . milani a. : 1993 , the trojan asteroid belt : proper elements , stability , chaos and families , _ cel . _ , * 57 * , 5994 . milani , a. & g. f. gronchi : 2010 , _ theory of orbit determination _ , cambridge univ . milani , a. , & z. knev zevi : 1992 , asteroid proper elements and secular resonances , _ icarus _ * 98 * , 211232 . milani , a. , & z. knev zevi : 1994 , asteroid proper elements and the dynamical structure of the asteroid main belt , _ icarus _ * 107 * , 219254 . milani a. , a. cellino , z. kneevi , b. novakovi , f. spoto , & p. paolicchi : 2014 , asteroid families classification : exploiting very large datasets , _ icarus _ * 239 * , 4673 . milani , a. , f. spoto , z. knev zevi , b. novakovi , & g. tsirvoulis : 2016 , families classification including multiopposition asteroids , _ proceedings iau symposium 318 : asteroids : new observations new models_,(s . r. chesley , a. morbidelli , r. jedicke & d. farnocchia , eds . ) , cambridge univ . press , pp . 28 - 45 . novakovi , b. , a. cellino , & z. kneevi : 2011 , families among high - inclination asteroids , _ icarus _ * 216 * , 6981 . paolicchi , p. & z. kneevi : 2016 , footprints of the yorp effect in asteroid families . _ icarus _ , * 274 * , 314326 . radovi , v. , & novakovi , b. : 2014 , excluding interlopers from asteroid families , _ proceedings iau symposium 310 : complex planetary systems _ ( z. kneevi and a. lemaitre , eds . ) cambridge univ . press . , pp 174175 . schubart , j. : 1991 , additional results on orbits of hilda - type asteroids , _ astron . astrophys . _ * 241 * , 297302 . smirnov & shevchenko , i.i . : 2013 , massive identification of asteroids in three - body resonances , _ icarus _ * 222 * , 220228 . spoto f. , a. milani , & z. kneevi : 2015 , asteroid family ages , _ icarus _ * 257 * , 275289 . tedesco , e.f . noah , m. noah , & s.d . price : 2002 , the supplemental iras minor planet survey . j. _ * 123 * , 10561085 . vinogradova , t.a . : 2015 , identification of asteroid families in trojans and hildas , _ mnras _ , * 454 * , 24362440 . vokrouhlick d. , bro m. , morbidelli a. , bottke w.f . , nesvorn d. , lazzaro d. & a.s . rivkin : 2006 , yarkovsky footprints in the eos family , _ icarus _ * 182 * , 92117 . walsh k.j . , m. delbo , w.f . bottke , d. vokrouhlick , & d.s . lauretta : 2013 , introducing the eulalia and new polana asteroid families : re - assessing primitive asteroid families in the inner main belt , _ icarus _ * 225 * , 283297 . zappal , v. , a. cellino , p. farinella , & z. kneevi : 1990 , asteroid families . i - identification by hierarchical clustering and reliability assessment , _ astron . , * 100 * , 20302046 . zappal v. , ph . bendjoya , a. cellino , p. farinella , & c. froeschl : 1995 , asteroid families : search of a 12,487-asteroid sample using two different clustering techniques , _ icarus _ , * 116 * , 291314 . .fit region : family number and name , explanation of the choice , minimum and maximum value of proper @xmath20 , minimum and maximum value of the diameter selected for the inner and the outer side . [ cols="<,^,>,^ , > " , ]

the formation of the beam halo in charged particle accelerators is studied in the framework of a stochastic - hydrodynamic model for the collective motion of the particle beam . in such a stochastic - hydrodynamic theory the density and the phase of the charged beam obey a set of coupled nonlinear hydrodynamic equations with explicit time - reversal invariance . this leads to a linearized theory that describes the collective dynamics of the beam in terms of a classical schrdinger equation . taking into account space - charge effects , we derive a set of coupled nonlinear hydrodynamic equations . these equations define a collective dynamics of self - interacting systems much in the same spirit as in the gross - pitaevskii and landau - ginzburg theories of the collective dynamics for interacting quantum many - body systems . self - consistent solutions of the dynamical equations lead to quasi - stationary beam configurations with enhanced transverse dispersion and transverse emittance growth . in the limit of a frozen space - charge core it is then possible to determine and study the properties of stationary , stable core - plus - halo beam distributions . in this scheme the possible reproduction of the halo after its elimination is a consequence of the stationarity of the transverse distribution which plays the role of an attractor for every other distribution .

cond-mat0302382

high intensity beams of charged particles , in particular in linacs , have been proposed in recent years for a wide variety of accelerator related applications : drivers for sources of neutron spallation ; production of tritium ; transmutation of radioactive wastes to species with shorter lifetimes ; heavy ion drivers for fusion - based production of thermonuclear energy ; production of radioactive isotopes for medical use , and so on . in all these cases it is very important to keep at a low level the beam losses to the wall of the beam pipe , since even small fractional losses in a high current machine can cause exceedingly high levels of radioactivation . it is now widely believed that one of the relevant mechanisms for these losses is the formation of a low intensity halo relatively far from the core of the beam . these halos have been either directly observed @xcite or inferred from experiments @xcite , and have also been predicted from extensive numerical simulations @xcite . often in these studies the particle core model has been used to understand the halo formation and its extent . that notwithstanding , it is widely believed that for the next generation of high intensity machines it is necessary to obtain a more quantitative understanding not only of the physics of the halo , but also of the beam transverse distribution in general @xcite . in fact `` because there is not a consensus about its definition , halo remains an imprecise term '' @xcite so that several proposals have been put forward for its description . charged particle beams are usually described in terms of classical deterministic dynamical systems . the standard model is that of a collisionless plasma where the corresponding dynamics is embodied in a suitable phase space ( see for example @xcite ) . generally speaking , the beams are described in a comoving reference system so that we can confine ourselves to a non relativistic setting . in this framework , the formation of the halo has been studied mainly by means of the particle - core model initially proposed by gluckstern , and the simulations show that the dynamical instabilities due to a parametric resonance can allow ions to escape from the core @xcite . in the present paper we propose and develop a different approach . we introduce a model for the formation of the halo based on the idea that the trajectories of the charged particles are sample paths of a stochastic process , rather than the usual deterministic ( differentiable ) trajectories . a random collisional plasma described in terms of configurational stochastic processes seems a realistic model for the dynamics taking place in particle accelerators . moreover , it can explain the rare escape of particles from a quasi - stable beam core , since it takes into account , statistically , the complicated inter - particle interactions that can not be described in detail . of course , the idea of a stochastic approach is hardly new @xcite , but there are several different ways to implement it . the system we want to describe is the particle beam endowed with some measure of time - reversal invariance . for such a physical situation one is faced with a many - body dynamical system which is at the same time stochastic ( due to the random collisions and the fast , irregular configurational dynamics ) and conservative ( time - reversal invariant ) . one then needs a generalization of classical mechanics for conservative deterministic systems to a classical mechanics of conservative stochastic systems , i.e. a stochastic mechanics . despite the rather widespread misconception that stochastic processes in dynamics are always associated to the description of dissipative and irreversible behaviors , it is in fact possible for specific stochastic dynamical systems to exhibit time - reversal invariance . in general , these systems are such that a deterministic dynamics in terms of known external potentials is ascribed to a system whose kinematics , due to some intrinsic irregularities , is instead stochastic @xcite . this is the case we conside for the collisional beam . in previous papers @xcite we analysed some basic properties of stochastic mechanics ( sm ) and we also introduced some basic criteria of mechanical stability in order to provide phenomenological support to the scheme of sm for classical stochastic dynamical systems . in particular , in ref . @xcite we applied sm to the description of the dynamics of charged particle beams , and we showed how the criteria of mechanical stability allow to connect the ( transverse ) emittance to the characteristic microscopic scales of the system and to the total number of particles in a given beam . moreover , this method allows to implement techniques of active control for the dynamics of the beam . in ref . @xcite these techniques have been proposed to improve the beam focusing and to independently change the frequency of the betatron oscillations . in the present paper , the sm of particle beams is used to investigate the nature , the size and the dynamical characteristics of beam halos . in particular , we determine the effects of the space charge interaction on the formation of the halo by calculating the growth of the beam emittance , and by estimating the probability to find particles far away from the beam core . we then consider the reverse problem i.e. , by imposing reasonable forms of the stationary halo distributions we derive the effective spatial potentials that are associated to such distributions . the paper is organized as follows : in sections [ nelsonism ] and [ control ] we summarize the results of previous papers on the model of sm for particle beams with special emphasis on the techniques of active control needed to engineer a desired beam dynamics . we then apply sm to study the formation of the beam halo . in section [ selfconsistent ] we show how to treat with sm the effects of space charge due to the electromagnetic interactions among the particles . we introduce and solve self - consistently a set of coupled , nonlinear hydrodynamic equations for the beam density and phase , the space - charge potentials , and the external electromagnetic field . solution of the coupled equations leads to the first main result of our paper , i.e. that the final transverse particle distribution is greatly broadened due to the presence of the space - charge potential . this broadening gives rise to a penetration of the distribution well into spatial regions very far from the core of the beam , thus realizing a type of continuous halo , i.e. without intermediate voids of particles ( nodes of the transverse beam distribution ) . this effect is reflected in a corresponding growth of the transverse emittance that we estimate from the solution of the self - consistent equations . in section [ halodistr ] we study the reverse problem : given different stationary transverse distributions of the halo around the core of the beam , such as a continuous broadened distribution or a distribution with a gap ( void ) between the halo and the core , we determine the dynamics that leads to such configurations . this is important since it enables us to deduce the form of the effective potentials which control the stationarity of the halo distribution . finally , in section [ conclusions ] we draw our conclusions and discuss future directions of research . the usual way to introduce a stochastic dynamical system is to modify the dynamics in phase space by adding a wiener noise only in the equation for the momentum , in order to preserve the usual relations between position and velocity : @xmath0 where @xmath1 is the mass of the system , @xmath2 is the position variable , @xmath3 is the momentum variable , @xmath4 is the external force , and @xmath5 is a fluctuating force modelled by a wiener process ; in general , this procedure yields a differentiable , but non markovian process @xmath6 for the position ( here @xmath7 denotes the autocorrelation of the fluctuating force ) . the standard example of this approach is that of a brownian motion in a force field described by the ornstein uhlenbeck system of stochastic differential equations ( sde ) @xcite : @xmath8 where @xmath9 is the spatial gradient of the external potential , and @xmath10 is the damping constant . this system can be reduced to yield a non differentiable , markovian process @xmath6 acted on by a wiener noise , with diffusion coefficient @xmath11 , to the equation for the position . in this case we are forced to drop the equation for the momentum since @xmath12 is no longer differentiable , and we can reduce the stochastic system to the single sde @xmath13 where @xmath14 is the so called forward velocity , and @xmath15 is the increment of a standard wiener noise @xmath16 ; as it is well known , this increment is a gaussian process . the standard example of this reduction is the einstein - smoluchowski approximation of the ornstein uhlenbeck process in the overdamped case . it is however possible to introduce eq . ( [ ito ] ) in a completely different context , as the defining equation for a theory of stochastic dynamical systems in configuration space , rather than in phase space . this amounts to consider a system whose kinematics , rather than its dynamics , is taken _ ab initio _ to be random . by doing this , no external sources of dissipation and irreversibility are introduced on the forces acting on the system . instead , a source of randomness is assumed that perturbs only the configurations of the system . for such systems with random kinematics , the dynamics can then be introduced either by generalizing the newton equation @xcite , or by introducing a stochastic variational principle @xcite . the crucial aspect is that such a dynamics turns out to be both stochastic _ and _ conservative . in this scheme , since @xmath2 is not differentiable , there is no velocity as a standard derivative . instead it is possible to define in a suitable way @xcite an average velocity @xmath17 in the forward time direction , and an average velocity in the backward time direction @xmath18 : these functions of @xmath19 and @xmath20 are different , and they both coincide with the usual velocity only if the process is differentiable , i.e. if the kinematics is deterministic . the relations between @xmath17 and @xmath18 will be introduced in the subsequent discussion . it is important to remark that @xmath21 is no more an a priori given field : it plays now the role of a dynamical variable . the stochastic dynamical scheme sketched above is known as stochastic mechanics ( sm ) , and most of its applications have concerned the problem of developing a classical stochastic model for the simulation of quantum mechanics . nonetheless , it is a very general model which can be applied to very different stochastic dynamical systems endowed with time - reversal invariance @xcite . we will show below how one can derive from the stochastic variational principle two coupled , nonlinear hydrodynamic equations for the density and the phase of a dynamical system in configuration space . these two real , coupled equations can be recast into a single complex equation whose form is completely equivalent to that of a schrdinger differential equation . in this sense , some authors denote classical dynamical systems described by sm as quantum - like systems , in analogy with other recent studies on the collective dynamics of charged particl beams @xcite . in general , sm can be used to describe every conservative , stochastic dynamical system satisfying fairly general conditions : for instance , it is known that for any given conservative diffusion there is a correspondence between the associated diffusion process and a solution of the schrdinger equation with hamiltonians associated to suitable vector and scalar potentials @xcite . under some regularity conditions this correspondence is one - to - one . the usual schrdinger equation , and hence quantum mechanics , are recovered when the diffusion coefficient @xmath11 is independent from the values taken by the conservative diffusion process and coincides with @xmath22 , where @xmath23 denotes the planck constant . here we are not interested in the stochastic modelling of quantum mechanics , but rather in the classical stochastic dynamics of particle beams . in this instance the unit of action appearing in the effective schrdinger equation is linked to the beam emittance as will be clarified below . we will now describe the stochastic process performed by a representative particle of the beam that oscillates , in a comoving reference frame , around the closed ideal orbit in a particle accelerator . we consider the three - dimensional ( 3d ) diffusion process @xmath24 , taking the values @xmath19 , which describes the motion of the representative particle and whose probability density coincides with the particle density of the beam . the evolution of this process is ruled by the it sde ( [ ito ] ) where the diffusion coefficient @xmath11 is supposed to be constant . the quantity @xmath25 , which has the dimensions of an action , will be connected later to the characteristic transverse emittance of the beam ( [ ito ] ) defines the random kinematics performed by the collective degree of freedom , and replaces the simple deterministic kinematics @xmath26 for the differentiable trajectory @xmath27 . we are in a situation in which we have both a random , diffusive kinematics and a time reversal invariance . therefore , to counteract the dissipation , one must impose a conservative dynamics on the stochastic kinematics , at variance with the purely dissipative fokker - planck or langevin dynamics . a conservative dynamics imposed on a random kinematics can be introduced by a suitable stochastic generalization of the least action principle of classical mechanics @xcite . this is achieved by replacing the classical deterministic kinematics ( [ classicalkin ] ) with the random diffusive kinematics ( [ ito ] ) as the independent configurational variables of the lagrangian density entering the action functional . the equations of motion thus obtained by minimizing the action functional take the form of two coupled hydrodynamic equations that describe the dynamical evolution of the beam profile and of the associated velocity field . in the following we briefly sketch the derivation of the hydrodynamic equations , referring for details to references @xcite . given the sde ( [ ito ] ) , we consider the probability density function ( pdf ) @xmath28 associated to the diffusion @xmath12 so that , besides the forward velocity @xmath29 , we can also define the backward velocity @xmath30 it is also useful to introduce the current and the osmotic velocity fields , defined as : @xmath31 the velocities in eq . ( [ osmotic ] ) have a transparent physical meaning : the current velocity @xmath32 represents the global velocity of the density profile , being the stochastic generalization of the velocity field of a classical perfect fluid . on the other hand the osmotic velocity @xmath33 is clearly of intrinsic stochastic nature , for it is a measure of the non differentiability of the stochastic trajectories , and it is related to the spatial variations of the density . in the limit of a deterministic process , i.e. of a diffusion that tends to a deterministic , differentiable trajectory @xmath34 , the current velocity @xmath35 tends to the classical velocity field @xmath36 , and the osmotic velocity @xmath37 tends to zero ( the trajectory becomes differentiable , so that the forward and backward velocities , i.e. the left and right derivatives coincide ) . in order to establish the stochastic generalization of the least action principle one introduces a mean classical action averaged over the probability density function of the diffusion process , in strict analogy to the classical deterministic action . in fact , the main difficulty in the stochastic case is due to the non differentiable character of the sample paths of a diffusion process which does not allow to define the time derivative of the process @xmath12 . hence the definition of a lagrangian density and of an action functional is possible only in an average sense through a suitable limit on expectations . the stochastic action is then defined as @xcite @xmath38 dt \ , , \ ] ] where @xmath39 denotes the expectation of a function of the diffusion process @xmath12 , @xmath40 is an external potential , and @xmath41 is the finite increment of the process . it can be shown that the mean action ( [ lagrangianaction ] ) associated to the diffusive kinematics ( [ ito ] ) can be recast in the following particularly appealing eulerian hydrodynamic form @xcite : @xmath42 \rho({\bf r } , t ) \ , , \ ] ] where @xmath32 and @xmath43 are defined in eq . ( [ osmotic ] ) . this eulerian form of the action immediately shows that when @xmath44 ( and hence the trajectories are differentiable ) it coincides with the action for the conservative dynamics of a classical liouville fluid . the stochastic variational principle now follows by imposing the stationarity of the stochastic action ( @xmath45 ) under smooth and independent variations @xmath46 of the density , and @xmath47 of the current velocity , with vanishing boundary conditions at the initial and final times . as a first consequence we get that the current velocity has the following gradient form : @xmath48 which can be also taken as the definition of the phase @xmath49 . the non linearly coupled lagrange equations of motion for the density @xmath50 and for the current velocity @xmath32 , of the form ( [ gradient ] ) , are a continuity equation typically associated to every diffusion process @xmath51 and an evolution equation for the conservative dynamics @xmath52 the above equations give a complete characterization of time - reversal invariant diffusion processes . the last equation is formally analogous to the hamilton madelung ( hjm ) equation originally introduced in the hydrodynamic formulation of quantum mechanics by madelung @xcite . however , the physical origins of the two equations are profoundly different , as in the quantum mechanical case the diffusion coefficient is related to the fundamental planck constant by the relation @xmath53 . notice that when @xmath54 ( namely when the kinematics is not diffusive ) the equation ( [ hjm ] ) coincides with the usual hamilton jacobi equation for a liouville fluid . since ( [ gradient ] ) holds , the two equations ( [ continuity ] ) and ( [ hjm ] ) can be recast in the following form @xmath55 which now constitutes a coupled , non linear system of partial differential equations for the couple @xmath56 which completely determines the state of our beam . on the other hand , because of ( [ gradient ] ) , this state is equivalently given by the couple @xmath57 . ( [ continuity1 ] ) and ( [ hjm2 ] ) describe the collective behavior of the beam at each instant of time through the evolution of both its particle density and its velocity field . it is important to notice that , introducing the representation @xcite @xmath58 with @xmath59 the two coupled real equations ( [ continuity1 ] ) and ( [ hjm2 ] ) are equivalent to a single complex schrdinger equation for the function @xmath60 , with the planck action constant @xmath23 replaced by the unit of action @xmath61 : @xmath62 in this formulation the phenomenological `` wave function '' @xmath60 carries the information on the dynamics of both the beam density and the beam velocity field , since the velocity field is determined via eq . ( [ gradient ] ) by the phase function @xmath63 . this shows , as previously claimed , that our procedure , starting from a different point of view , leads to a description formally analogous to that of the so - called quantum - like approaches to beam dynamics @xcite . we remark again that obviously eq . ( [ schroed ] ) has not the same meaning as in quantum mechanics . the parameter @xmath61 defined in eq . ( [ alfa ] ) has the dimensions of an action , but in general @xmath64 . in fact , @xmath61 is not a universal constant and its value can not coincide with @xmath23 , as its value depends on the physical system under study . however , @xmath61 plays in some sense a role similar to that of @xmath23 in quantum mechanics since , as we will see in the following , it identifies a lower bound for the beam emittance in phase space , and gives a measure of the position - momentum uncertainty arising from the stochastic dynamics of the beam . thus the schrdinger equation ( [ schroed ] ) for the stochastic mechanics of particle beams presents some features reminiscent of quantum mechanics , but at the same time is a deeply different theory with a different physical meaning . in particular , while in quantum mechanics a system of @xmath65 particles is described by a wave function in a @xmath66dimensional configuration space , in the scheme of sm the normalized probability density distribution @xmath67 is a function of only the three space coordinates in physical space . if there are @xmath65 particles in the beam , then the function @xmath68 is simply the particle density in physical space . in the previous section we have introduced two coupled equations that describe the dynamical behavior of the beam : the first is the it equation ( [ ito ] ) , or equivalently the continuity equation ( either ( [ continuity ] ) or ( [ continuity1 ] ) ) ; the second is the hamilton - jacobi - madelung ( hjm ) equation ( either ( [ hjm ] ) or ( [ hjm2 ] ) ) . here we briefly summarize a general procedure , already exploited in refs . @xcite , to engineer a controlled dynamics of stochastic systems . we will then apply this method to the control of the beam halo in section [ halodistr ] . to this end it is useful to show , by simple substitution from ( [ osmotic ] ) , that eq . ( [ continuity ] ) is equivalent to the fokker planck ( fp ) equation @xmath69 + d \ , \nabla^2 \rho , \ ] ] formally associated to the it equation ( [ ito ] ) . the hjm equation ( [ hjm ] ) can also be cast in a form based on @xmath70 rather than on @xmath71 , namely : @xmath72-v\nonumber\\ & = & -\frac{m}{2}\mathbf{v}_{(+)}^2 + m d\,\mathbf{v}_{(+)}\nabla\ln f\nonumber\\ & & \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; + md^2\nabla^2\ln f - v \ , , \end{aligned}\ ] ] where we have defined the function @xmath73 , and @xmath74 is a constant such that @xmath75 is dimensionless . on the other hand , we know from eqns . ( [ vback ] ) and ( [ osmotic ] ) that also the forward velocity @xmath70 is irrotational : @xmath76 where the scalar functions @xmath77 and @xmath49 are connected by the relation @xmath78 and @xmath79 is an arbitrary time - dependent function . through eq . ( [ gradient1 ] ) , eqns . ( [ fp ] ) and ( [ hjm1 ] ) can be cast in a form analogous to the system of eqns . ( [ continuity1 ] ) and ( [ hjm2 ] ) . in this case , the couple of functions which determine the state of the beam is @xmath80 ( or equivalently @xmath81 ) . it is worth noticing that the possibility of a time reversal invariance @xcite is assured by the fact that the forward velocity @xmath82 is not an _ a priori _ given field , as is the case for dissipative diffusion processes of the langevin type . rather , it is dynamically determined at any instant of time , for any assigned initial condition , through the hjm evolution equation ( [ hjm ] ) . let us suppose now that the beam density @xmath83 is given at some time @xmath20 . we may think for instance of an engineered evolution from some initial density toward a final , required state with suitable properties . in particular , we could imagine to have a halo - ridden particle distribution in the beam that should be steered toward a final halo - free distribution . we know that @xmath50 must be a solution of the fp equation ( [ fp ] ) for some forward velocity field @xmath84 , which we consider here as not given a priori . since also eq . ( [ gradient1 ] ) must be satisfied , this means that for a given @xmath50 we should find an irrotational @xmath70 in such a way that the fp equation ( [ fp ] ) is satisfied . in other words , we are required to solve the partial differential equation @xmath85 for the irrotational vector field @xmath70 , or equivalently , by taking eq . ( [ gradient1 ] ) into account , the second order partial differential equation for the scalar field @xmath77 : @xmath86 this is not , in general , an easy task , but we will give here the solutions for a few simple cases that will be useful in the following . in the one - dimensional ( 1d ) case ( @xmath87 ) , the fp equation defined for @xmath88 ( here we could possibly have @xmath89 , and @xmath90 ) is : @xmath91+d\partial_x^2\rho = \partial_xj , \ ] ] where @xmath92 is the probability current . it is easy to see now that the solution is @xmath93 , \ ] ] where @xmath94 is an arbitrary function . moreover , since the conservation of probability imposes , in particular , that @xmath95 , it is easy to see that we must choose @xmath96 so that finally we have @xmath97 of course , in this case we also have @xmath98 this 1d solution will be useful later for a simplified study of beam distributions when we take @xmath99 as one of the transverse coordinates of the beam . another case of practical interest is that of a 3d system with cylindrical symmetry around the @xmath100-axis ( which will be supposed to be the beam longitudinal axis ) . if we denote with @xmath101 the cylindrical coordinates , where @xmath102 , we will suppose that @xmath103 depends only on @xmath104 and @xmath20 , and that @xmath105 is radially directed , with modulus depending only on @xmath104 and @xmath20 . in this case , it is straightforward to see that the equation for @xmath106 is @xmath107 whose solution is @xmath108 in this case @xmath109 finally , in the 3d stationary case the beam density @xmath50 is independent of @xmath20 . this greatly simplifies the treatment , as it is easy to check by simple substitution , and we simply find @xmath110 with @xmath111 this result corresponds to the well known fact that for stationary forward velocities the fp equation ( [ fp ] ) always admits stationary solutions of the form @xmath112 remark however that also for particular choices of a non stationary @xmath83 it is sometimes possible to find a stationary velocity field @xmath113 such that the fp equation ( [ fp ] ) is satisfied ( think for instance to the ornstein uhlenbeck non stationary solutions ) : this fact will be of use in the following sections . now that @xmath50 and @xmath77 ( or equivalently @xmath70 ) are given and satisfy eq . ( [ fp ] ) , we should also remember that they will qualify as a good description of our system only if they also are a solution of the dynamical problem , namely if they satisfy the hjm equation ( [ hjm1 ] ) . since @xmath77 , @xmath70 and @xmath50 are now fixed , this last equation must be considered as a constraint relation defining an external controlling potential @xmath40 which , after straightforward calculations and in terms of the dimensionless distribution @xmath75 , turns out to be of the general form : @xmath114 this procedure can in principle be applied also to more complicated instances , for example to engineer a beam dynamics that keeps the beam coherent even in the presence of aberrations , or to engineer some desired evolution . however the solutions will not always be available in closed form , and will in general require some approximate treatment . here we present the simple analytic solutions for the three particular cases discussed above . in the 1d case @xmath115 , with @xmath106 given by eq . ( [ 1dim ] ) , the controlling potential is : @xmath116 in the 3d , cylindrically symmetric case @xmath117 , with @xmath106 given by ( [ 3dim ] ) , we have instead : @xmath118 finally , in the 3d stationary case @xmath119 with @xmath70 given by eq . ( [ 3dimstaz ] ) , the potential is still time - dependent because of the presence of the arbitrary function @xmath120 , and it reads : @xmath121 in order to make it stationary it will be enough to require that @xmath122 be a constant @xmath123 , so that @xmath124 in this stationary case the constant @xmath123 fixes the zero of the potential energy , and the phenomenological wave function ( [ schroedpsi ] ) takes the form @xmath125 typical of stationary states . one of the possible mechanisms for the formation of the halo in particle beams is that due to the unavoidable presence of space charge effects . in this section we will investigate this possibility in the framework of our hydrodynamic - stochastic model of beam dynamics . to this end , we take into account the space charge effects by coupling the hydrodynamic equations of stochastic mechanics with the maxwell equations which describe the mutual electromagnetic interactions between the particles of the beam . we thus obtain a self - consistent , stochastic magnetohydrodynamic system of coupled nonlinear differential equations that can be numerically solved to show the effect of the space charge . in the following , the reference physical system will be an ensemble of @xmath65 identical copies of a single charged particle embedded in a particle beam and subject to both an external and a space - charge potential . in a reference frame comoving with the beam , our system is then described by the schrdinger equation ( [ schroedinger ] ) , where @xmath61 is the unit of action ( emittance ) and @xmath126 the hamiltonian operator which will be explicitly determined in the following . since in general @xmath60 is not normalized , we introduce the following notation for its constant norm @xmath127 so that , if @xmath65 is the number of particles with individual charge @xmath128 , the space charge density of the beam will be @xmath129 on the other hand its electrical current density will be @xmath130 which vanishes when the wave function is stationary , namely when @xmath131 for a charged particle in the beam the electromagnetic field is the superposition of the space - charge potential @xmath132 due to the presence of @xmath133 and @xmath134 , plus the external potential @xmath135 , and hence the schrdinger equation takes the form @xmath136 ^ 2 \psi\nonumber\\ & & \;\;\;\;\;\;\;\;\;\;+q_0 \left ( \phi_{sc}+\phi_{ext } \right ) \psi \ , .\end{aligned}\ ] ] eq . ( [ schroedinger ] ) has then to be coupled with the maxwell equations for the vector and scalar potentials @xmath132 . since we are in a reference frame comoving with the beam , we can always assume that the wave function is of the stationary form ( [ stationarywf ] ) . in this case we have @xmath137 , so that @xmath138 . finally , by taking also @xmath139 , our system is reduced to two coupled , nonlinear equations for the pair @xmath140 . for details see appendix [ coupled ] . if the beam with space - charge interactions stays cylindrically symmetric the function @xmath141 will depend only on the cylindrical radius @xmath104 and our system becomes ( see equations ( [ selfconstat1 ] ) and ( [ selfconstat2 ] ) in appendix [ coupled ] ) : @xmath142 where @xmath143 with @xmath144 the length of the beam , and @xmath145 eq . ( [ cylindricalsc1 ] ) is the stationary schrdinger equation in the external scalar potential , and eq . ( [ cylindricalsc2 ] ) is the poisson equation for the space - charge scalar potential . we proceed to solve numerically this nonlinear system and compare its solution with that for a purely external potential @xmath146 which is a cylindrically symmetric , harmonic potential with a proper frequency @xmath147 in absence of space charge ( see appendix [ cylho ] ) : @xmath148 , with @xmath102 . let us first introduce the dimensionless quantities @xmath149 where @xmath150 is the variance of the ground state of the cylindrical harmonic oscillator without space charge . ( [ cylindricalsc1 ] ) and ( [ cylindricalsc2 ] ) can then be recast in the dimensionless form @xmath151\,s w(s)=0 \ , , \label{adimsc1}\\ & & \nonumber\\ & & \!\!\!\!\!\!\!\!s v''(s)+v'(s)+b\,s w^2(s)=0 \ , , \label{adimsc2}\end{aligned}\ ] ] where @xmath152 the effect of the space charge will be finally accounted for by comparing the normalized solution @xmath153 with the unperturbed ground state of the cylindrical harmonic oscillator ( see appendix [ cylho ] ) : @xmath154 we have solved numerically the system ( [ adimsc1 ] ) , ( [ adimsc2 ] ) by tentatively fixing one of the two free parameters @xmath155 and @xmath7 , and then searching by an iterative trial and error method a value of the other such that the solution shows , in a given interval of values of @xmath156 , the correct infinitesimal asymptotic behavior for large values of @xmath156 . we have then normalized the solutions @xmath153 by calculating numerically the value of @xmath157 . it is clear from the definition of the dimensionless parameters @xmath157 , @xmath155 and @xmath7 ( [ parametri ] ) that the value of @xmath7 is a sort of reduced energy eigenvalue of the system , while the product @xmath158 which is by definition a non negative number , will play the role of the interaction strength , since it depends on the space - charge density along the linear extension of the beam . reverting to dimensional quantities , since @xmath159 has the dimensions of an energy , the two relevant parameters are @xmath160 which are respectively the energy of the individual particle embedded in the beam , and the strength of the space - charge interaction . in the following we will limit ourselves to discuss solutions of eqns . ( [ adimsc1 ] ) , and ( [ adimsc2 ] ) without nodes ( a sort of ground state for the system ) . it is possible to see that no solution without nodes can be found for values of the space - charge strength @xmath10 beyond about @xmath161 and that for values of @xmath10 ranging from @xmath162 to @xmath161 , the energy @xmath7 decreases monotonically from @xmath163 to @xmath164 . if the unit of action @xmath61 is fixed at a given value , it is apparent that the value of @xmath10 is directly proportional to the charge per unit length @xmath165 of the beam : a small value of @xmath10 means a rarefied beam ; a large value of @xmath10 indicates that the beam is intense . a halo is supposed to be present in intense beams , while in rarefied beams the behavior of every single particle tends to be affected only by the external harmonic potential . as a function of the space - charge strength @xmath10 . actually these quantities are dimensionless , measured in units of the energy @xmath166.,width=302 ] it is useful to provide a numerical estimate ( in @xmath167 units ) of the relevant parameters . we are assuming a beam made of protons , so that @xmath1 and @xmath128 are the proton mass and charge , while @xmath168 is the vacuum permittivity . the parameter @xmath169 is determined by the strength of the external potential , and it is a measure of the transverse size of the beam when no space charge is taken into account , so that the ground state has the unperturbed form ( [ unpertground ] ) . from empirical data , a reasonable estimate yields @xmath170 . on the other hand the value of @xmath171 clearly depends on the particular beam we are considering : usually a value of @xmath172 particles per meter is considered realistic . as for the parameter @xmath61 ( [ alfa ] ) we have already suggested in section [ introduction ] that its value can be connected to the beam emittance . on the ground of accepted experimental values of the beam emittance ( usually measured in units of length ) we can assume @xmath173 of the order of @xmath174 . in fact , taking @xmath1 , @xmath128 , @xmath169 , and @xmath171 fixed at the above - mentioned values , and assuming a space - charge strength @xmath10 moderately large , i.e. @xmath175 , we can determine @xmath61 from and we get @xmath176 , a number which is in good agreement with the experimentally measured values of the beam emittance . from now on we will fix @xmath61 at its approximate central value : @xmath177 since @xmath1 and @xmath128 have the proton values , and @xmath178 , then the assigned values of the dimensionless parameters @xmath10 and @xmath7 will fix respectively , the values of @xmath171 and @xmath123 by eqns . ( [ parameters ] ) . this implies that by changing the beam intensity ( namely @xmath171 and @xmath10 ) one correspondingly changes the energy ( namely @xmath123 and @xmath7 ) of the individual particle embedded in the beam . in fig . [ bbbeta ] we show the behavior of @xmath7 as a function of @xmath10 . the quantities @xmath7 and @xmath10 are dimensionless : the true physical quantities ( energies ) are obtained by multiplicating them by the unit of energy @xmath179 compared with the distribution in absence of space charge ( dashed line ) . the space - charge strength is @xmath180 . all quantities are dimensionless.,width=302 ] for a space - charge strength of @xmath180.,width=302 ] the overall effect of the space charge in this model is a conspicuous spreading of the transverse distribution of the particles in the beam with respect to the unperturbed ground state distribution ( [ unpertground ] ) . when @xmath181 the potential due to space charge vanishes , and the solution exactly coincides with the ground state of the cylindrical harmonic oscillator with variance @xmath182 . when @xmath183 the transverse distribution begins to spread , as we show in figrs . [ w11 ] and [ w113d ] for a value of the space - charge strength @xmath180 , and in figrs . [ w22 ] and [ w223d ] for a value of the space - charge strength @xmath184 . notice that when comparing distributions , the true radial ( dimensionless ) density is @xmath185 and not just @xmath186 . recall that @xmath153 is a normalized solution of eqns . ( [ adimsc1 ] ) and ( [ adimsc2 ] ) . in figrs . [ scpotential11 ] and [ scpotential22 ] we show the radial transverse form of the total potential , i.e. external plus space - charge given by the solution @xmath187 of eqs . ( [ adimsc1 ] ) and ( [ adimsc2 ] ) , for the values @xmath188 and @xmath189 of the space - charge strength . compared with the distribution in absence of space charge ( dashed line ) . the space - charge strength is @xmath184 . all quantities are dimensionless.,width=302 ] to give a more quantitative measure of the flattening and broadening of the transverse distribution we compare numerically the probabilities of finding a particle at a relatively large distance from the beam longitudinal axis with and without space charge . the quantity @xmath190 is the probability of finding a particle at a distance greater than @xmath191 for systems with a given strength @xmath10 of the space charge coupling . for instance , considering the two different situations @xmath181 ( no space charge ) and @xmath189 ( strong space charge ) we have : @xmath192 we see that the probability of finding particles at a distance larger than @xmath193 from the core of the beam is enhanced by space charge by many orders of magnitude . this means , for example , that if in the beam there are @xmath172 particles per meter , while practically no one is found beyond @xmath193 in absence of space charge , for very strong space - charge intensity we can find up to @xmath194 particles per meter at that distance from the core . .,width=302 ] .,width=302 ] the above analysis shows that in the hydrodynamic - stochastic theory of charged beams , the space - charge potential induces a strong broadening of the unperturbed transverse density distribution of the beam , thus yielding a small , but finite probability of having particles at a distance well away from the core of the beam . the hydrodynamic - stochastic model allows also an estimate on the growth of the emittance due to the presence of the space charge . the emittance can be calculated by exploiting a structure of uncertainty products that is inherent to the sm . in particular , the transverse emittance can be calculated as @xmath195 , where we have : @xmath196 more explicitly : if @xmath153 is a solution of the coupled dynamical equations normalized in the sense that @xmath197 the corresponding 3d probability density is @xmath198 where @xmath199 is the heaviside function and @xmath200 this probability density function is normalized since @xmath201 it is easy to show that the first two moments are : @xmath202 so that the root mean square deviation for the position is : @xmath203 as for the momentum @xmath204 , we recall that in sm one can not introduce a probability distribution directly in phase space . the momentum can nevertheless be recovered from the velocity field which is well defined in sm . since we are considering a stationary state , only the osmotic part of the velocity field is non zero . then the momentum field reads @xmath205 as a consequence , the @xmath99-component of the momentum is @xmath206 and the first and second moments read @xmath207 hence the root mean square deviation for the momentum is : @xmath208 vs. the space - charge strength @xmath10 . the true dimensional emittance , calculated as position momentum uncertainty product , is just @xmath209.,width=302 ] we can then define the emittance @xmath210 as the position - momentum uncertainty product in phase space in the following way : @xmath211 the emittance ( [ emittance ] ) can be thus estimated from knowledge of the numerical solutions @xmath153 and their first derivatives @xmath212 . it can be seen that the emittance is exactly @xmath61 for @xmath181 ( namely in absence of space charge ) , and grows with @xmath10 to a value @xmath213 for @xmath184 . this result is consistent with the expected growth of emittance produced by space - charge effects . the dimensionless ratio @xmath214 is plotted as a function of @xmath10 in fig . [ epsilon ] . this figure and the above discussion provide evidence that in the model of sm of charged beams the constant @xmath61 plays the role of a lower bound for the phase - space emittance . the coupled nonlinear equations introduced in the previous section allow to introduce and explain a possible mechanism of halo formation due to space - charge effects . this mechanism , when considering solutions without nodes , amounts to the broadening of the transverse density distribution of the beam away from the beam core . this is a type of `` continuous '' halo . if we want to investigate the formation and the properties of other possible forms of halos for example ring - like halos separated from the core by an almost void space region we should try to analyze the solutions with nodes of our radial self - consistent equations . being this a time consuming task , we resort for the time being to a different , simplified approach to study the structure and properties of beams of the ringed type . moreover , the scheme introduced in the previous section , although in principle very general , has the drawback that it does not allow for analytical expressions of velocities and potentials , and therefore can not suggest the engineering of controlling external potentials to remove the halo . to this end , in this section we will develop a different approach , still based on sm , by considering some preassigned density distributions for a beam with halo , and then looking for the kind of dynamics that produces them . firs of all we will determine the analytical forms @xmath215 of the dimensionless radial distribution that can best approximate the numerical , self consistent solutions @xmath153 found in the previous section . we will do that by choosing a family of trial functions @xmath215 and by subsequently minimizing the mean square error ( m.s.e . ) with respect to a given solution @xmath153 . since the radial component of the cylindrically symmetric ( @xmath216 ) solutions of the harmonic oscillator ( [ cylindrho ] ) contain only even powers of the radial coordinate ( see appendix b ) , we take as normalized trial functions @xmath217 with @xmath169 , @xmath218 and @xmath155 as free parameters , and @xmath219 as normalization constant . we then take one numerical solution @xmath153 of ( [ adimsc1 ] ) and ( [ adimsc2 ] ) for a given value of @xmath10 and consider the mean square error @xmath220 and numerically minimize its value by varying the parameters @xmath169 , @xmath218 and @xmath155 . as an example , for @xmath221 , we get @xmath222 the corresponding density distribution ( solid line ) is compared in fig . [ fit01 ] with the density distribution obtained by numerically solving the system ( [ adimsc1 ] ) and ( [ adimsc2 ] ) ( dashed line ) . ) with @xmath221.,width=302 ] from this explicit form of the approximating @xmath215 we can now also calculate the expression of the stationary potential which controls the state . from eq . ( [ potential3dimstaz ] ) in the case of a cylindrically symmetric approximating state with @xmath223 we have the controlling potential @xmath224\,.\ ] ] introducing the usual dimensionless quantities @xmath225 we may define the dimensionless controlling potential @xmath226 for an approximating amplitude of the form ( [ trialw ] ) , and with the same values ( [ values ] ) of the parameters , the controlling potential reads @xmath227 this potential is shown ( solid line ) in fig . [ fit03 ] , where it is compared with the potential @xmath228 ( dashed line ) solution of eqns . ( [ adimsc1 ] ) and ( [ adimsc2 ] ) . to a numerical solution for the self consistent equations ( [ adimsc1 ] ) and ( [ adimsc2 ] ) , for trial functions of the form ( [ trialw ] ) with @xmath229 . it is compared with the potential @xmath228 ( dashed line ) obtained as numerical solution of ( [ adimsc1 ] ) and ( [ adimsc2]).,width=302 ] the analytical approximations show a small , but relevant , difference with respect to the numerical solutions of eqns . ( [ adimsc1 ] ) and ( [ adimsc2 ] ) : @xmath215 has a node at about @xmath230 , and correspondingly the potential @xmath231 has a singularity at that point ( out of the @xmath156 range of fig . [ fit03 ] ) . instead , the numerical solutions @xmath153 and @xmath187 of eqns . ( [ adimsc1 ] ) and ( [ adimsc2 ] ) show no such behavior . for values of @xmath156 beyond @xmath230 , @xmath215 shows a small bump , not visible in fig . [ fit01 ] because of the large scale . we show a zoom - up of it in fig . [ fit02 ] , where the node is now clearly visible . that shows the existence of a node in the position density distribution at @xmath232.,width=302 ] this is a feature that we will find as well in other different approaches to the description of the beam distribution , and that could be tentatively connected with the presence of a halo spatially separated from the core of the beam . in our quest for an analytic description of the halo distribution we now depart from the scheme of the self - consistent nonlinear equations ( [ adimsc1 ] ) and ( [ adimsc2 ] ) . instead , while still assuming the space charge as the dominant factor , we introduce a simplified analytical model for the halo production . as a first simplification , valid for small values of the space charge strength @xmath10 , we assume a frozen core for the beam . a frozen core is such that it produces a space - charge effect , induces a spreading of the density distribution , and is left unchanged by this spreading . in this case the potential produced by the charge distribution is a priori given by the distribution of the frozen core and it simply enters linearly in the phenomenological schrdinger equation ( [ schroed ] ) . as a consequence , the latter is now decoupled from the maxwell equations for the electromagnetic field . the frozen charge distribution will be assumed cylindrically symmetric and transversally gaussian : @xmath233 where @xmath65 is the number of particles with elementary charge @xmath128 , @xmath144 is the longitudinal extension of the beam , and @xmath234 denotes the heaviside function . the cylindrical symmetry allows to determine via gauss theorem the potential energy @xmath235 of a test particle of charge @xmath128 embedded in this charge distribution . it is a simple exercise to show that @xmath236 \ , , \ ] ] where @xmath237 is the euler constant , and @xmath238 is the exponential integral function . if we suppose that the test particle is acted upon by this space - charge potential , and by the external , cylindrical harmonic oscillator potential ( [ cylindrho ] ) which , without space charge , would keep the particle in the gaussian state with variance @xmath182 the total potential to be inserted in the schrdinger equation ( [ schroed ] ) reads @xmath239 .\end{aligned}\ ] ] in terms of the usual dimensionless quantities we can also write @xmath240\end{aligned}\ ] ] a plot of the dimensionless potential @xmath187 for @xmath241 is shown in fig . [ frozenpotential ] . produced by a frozen , cylindrically symmetric beam distribution with @xmath241.,width=302 ] for stationary solutions of the form @xmath242 eq . ( [ schroed ] ) becomes @xmath243 and due to the cylindrical symmetry of @xmath40 it reduces to @xmath244+v(r)\,u(r ) \ , .\ ] ] considering the dimensionless quantities @xmath245 we finally obtain the reduced equation @xmath246 despite its simple graphical behavior displayed in fig . [ frozenpotential ] the potential @xmath187 is complicated enough to bar the hope of exactly solving eq . ( [ reduced ] ) in a simple way . reasonable approximate solutions can be obtained by means of the variational method of rayleigh and ritz . we thus look for the minimum of the average energy @xmath247 whose reduced and dimensionless form is @xmath248ds}{\int_0^{\infty}w^2(s)sds } \ , .\ ] ] the minimization can be carried out for different values of the coupling parameter @xmath10 , but the frozen core model is realistic only for small values of @xmath10 : a strong coupling with a frozen core would have the paradoxical effect of totally expelling the distribution function of the test particle from the center of the beam , while keeping the gaussian core always concentrated in the center . to elaborate an example we have chosen @xmath241 and a test function of the form @xmath249 the values of the parameters that minimize the energy functional turn out to be @xmath250 the corresponding optimal radial density distribution @xmath251 is shown in fig . [ frozen01 ] where it is also compared with the gaussian solution ( dashed line ) corresponding to the case of a purely harmonic potential ( @xmath181 ) . ( solid line ) . it is compared with the gaussian solution ( dashed line ) corresponding to the case of a purely harmonic potential ( @xmath181).,width=302 ] notice that this density distribution has a form similar to that of the self consistent solutions elaborated in section [ selfconsistent ] for values of @xmath10 of about @xmath252 ( see fig . [ w11 ] ) . it has a node which is not visible in fig . [ frozen01 ] . we zoom up its plot in fig . [ frozen02 ] . that shows a node in the position density distribution at @xmath253.,width=302 ] this behavior suggests that the halo distribution could be described , in first approximation , as a gaussian core distribution plus a small ring of particles surrounding it and constituting the halo . thus , in the following we will simply assume a specific form of a beam with a halo without deriving it as an effect of space charge interactions . since it is not clearly established that a halo can be due only to space - charge effects , starting with a realistic ring distribution and trying to understand the dynamics that can produce it could be very useful in this respect . the techniques introduced in section [ control ] will be instrumental to this end . for a three - dimensional , cylindrically symmetric beam we introduce the normalized radial density distribution @xmath254 which is composed of a gaussian core with variance @xmath182 ( the simple harmonic oscillator ground state ) , plus a ring - like distribution whose size is fixed through the two parameters @xmath255 and @xmath256 . the parameter @xmath257 is the relative weight of the two parts . this density distribution has the required form for suitable values of the parameters , but , at variance with the two previous examples of approximate distributions ( see figs . [ fit02 ] and [ frozen02 ] ) , it has no nodes . this is convenient for two main reasons : first because it is a rather general requirement for a ground state to have no nodes ( this fact is a rigorous theorem for one - dimensional systems ) . moreover , it has been shown in refs . @xcite that stationary distributions without nodes are also attractors for every other possible ( non extremal ) initial distribution : a property that will be useful in a future discussion of the possible relaxation of the system toward a stable beam halo . for this cylindrically symmetric case the expressions ( [ 3dimstaz ] ) and ( [ potential3dimstaz ] ) for the velocity field and the potential become @xmath258 moving to dimensionless quantities we have @xmath259 a three - dimensional plot of the dimensionless density distribution is given in fig . [ ring01 ] where we have chosen @xmath260 , @xmath261 and @xmath262 . in this example the importance of the halo ring has been exaggerated with respect to the real case in order to make the effect clearly visible . the explicit analytic expression of @xmath263 and @xmath187 is rather lengthy and not particularly illuminating : we simply show the behavior of these functions , for the same values of the parameter for the previous example , respectively in fig . [ ring03 ] and fig . [ ring02 ] , where they are also compared with the corresponding dimensionless quantities ( @xmath264 and @xmath265 ) for the ground state of the harmonic oscillator . the function @xmath187 is the potential that in the stochastic model keeps the beam in the stationary state with a halo ring . given by eq . ( [ adimensionale ] ) , which shows a halo ring surrounding the beam core.,width=302 ] ( solid line ) , and for the ground state of a harmonic oscillator ( dashed line).,width=302 ] ( solid line ) , and for a harmonic oscillator ( dashed line).,width=302 ] in order to illustrate the main effects involved , let us give here the simpler formulae relative to the 1d case . from now on we will consider only 1d processes denoting by @xmath99 one of the transverse space coordinates . we assume that the longitudinal and the transverse beam dynamics can be deemed independent , with the further simplification of considering decoupled evolutions along the transverse directions @xmath99 and @xmath266 . under these conditions the density distribution with a halo ring now reads @xmath267 so that the corresponding velocities and potential are @xmath268 in terms of dimensionless quantities we then have for the distribution @xmath269 while the forward velocity and the potential are @xmath270 in figrs . [ ring1dim01 ] , [ ring1dim03 ] , and [ ring1dim02 ] we respectively show the density distribution , the velocity field , and the potential for @xmath260 , @xmath261 and @xmath271 . as in the three - dimensional case , the importance of the halo ring has been exaggerated with respect to the real case in order to make the effect clearly visible . ) with a halo ring surrounding the beam core.,width=302 ] ( solid line ) , and for a harmonic oscillator ( dashed line).,width=302 ] ( solid line ) , and for a harmonic oscillator ( dashed line).,width=302 ] in this paper we have presented a dynamical , stochastic approach to the description of the beam transverse distribution in the particle accelerators . in the first part we have described the collective beam dynamics in terms of time reversal invariant diffusion processes ( nelson processes ) which are obtained by a stochastic extension of the least action principle of classical mechanics . the diffusion coefficient of the process is identified with a lower bound for the emittance of the beam as discussed in the section [ selfconsistent ] . the collective dynamics of beams is then described by two nonlinearly coupled hydrodynamic equations . this set of equations shows some formal analogies with the effective ginzburg - landau and gross - pitaevskii descriptions of the collective dynamics of self - interacting quantum many body systems in terms of classical , nonlinear schrdinger equation . we have shown that , in the framework of sm , it is possible to have transverse distribution which show a broadening and an emittance growth , typical of the halo formation . the interest of this approach lies in the fact that its dynamical equations allow us , at least in principle , to determine the possible control potentials that are responsible for the stationarity of the halo . a few introductory examples of these potentials and distributions have been discussed in the section [ halodistr ] . the controlling potentials can be engineered by suitable tuning of the external electromagnetic fields . in a previous paper @xcite we have considered evolutions that drive the beam from a less collimated to a better collimated state , and we have furthermore shown that this goal can also be achieved without increasing the frequency of the betatron oscillations which can in fact be independently controlled during the evolution . in forthcoming papers we plan to elaborate on the extension of these techniques to the problem of engineering suitable time - dependent potentials for the control and the elimination of the beam halo . in particular , it will be shown that the transition functions of the nelson processes can be exploited to control these evolutions . we also plan to extend the analysis of the present paper to different , non gaussian beam transverse distributions which could better describe the existence of halo losses in terms of large ratios of the maximum displacement to the rms size of the beam . starting from the maxwell equations for the space - charge field felt by our particle in the beam @xmath272 and introducing the corresponding electro magnetic potentials @xmath132 from @xmath273 with the gauge condition @xmath274 through the usual procedure we get the wave equations @xmath275 it is also well known ( see for example @xcite , chapter xv ) that the schrdinger equation for spinless , charged particles in an electro magnetic field @xmath276 is @xmath277 \psi\,.\ ] ] for a charged particle in the beam the electro magnetic field is the superposition of the space - charge potential @xmath132 plus an external , control potential @xmath135 , and hence we obtain the schrdinger equation ( [ schroedinger ] ) that we rewrite here for convenience : @xmath278 ^ 2\psi\nonumber\\ & & \nonumber\\ & & \;\;\;\;\;\;\;\;+q_0(\phi_{sc}+\phi_{ext})\psi\,.\end{aligned}\ ] ] it is apparent now that ( [ gauge ] ) , ( [ waveqa ] ) , ( [ waveqfi ] ) and ( [ schroedappendix ] ) constitute a self consistent system of coupled , non linear differential equations for the fields @xmath60 , @xmath279 and @xmath280 . since we are in a reference frame comoving with the beam , in the following we will suppose to deal only with stationary wave functions of the form ( [ stationarywf ] ) . in this case , as already remarked , we get @xmath137 , so that we can take the trivial solution @xmath138 of ( [ waveqa ] ) . the gauge condition ( [ gauge ] ) then implies that @xmath281 and the wave equation ( [ waveqfi ] ) reduces itself to the poisson equation for the electrostatic potential . finally , by supposing also @xmath139 , our system is reduced to only two coupled , non linear equations for the couple @xmath140 , namely @xmath282 where we have defined @xmath283 then if we pass to the potential energies by putting @xmath284 our equations are reduced to @xmath285 remark that here the wave functions @xmath60 and @xmath141 , while certainly normalizable , are not supposed in general to be already normalized . however the equations ( [ selfconstat1 ] ) and ( [ selfconstat2 ] ) , albeit non linear , are apparently invariant for the multiplication of @xmath141 by an arbitrary constant . the equations ( [ selfconstat1 ] ) and ( [ selfconstat2 ] ) will be used in the section [ selfconsistent ] . in the model discussed in this paper we suppose that the external potential @xmath286 is a cylindrically symmetric , harmonic potential with a proper frequency @xmath147 , so that the ground state without space - charge interaction will have a variance @xmath287 this means that , in the usual cylindrical coordinates @xmath288 with @xmath289 , we have @xmath290 and the corresponding schrdinger equation @xmath291\psi+v_e\psi\end{aligned}\ ] ] with periodic conditions at @xmath292 ( for a bunch of length @xmath144 ) has the eigenvalues @xmath293 and the normalized eigenfunctions @xmath294 with @xmath295 where @xmath296 are the generalized laguerre polynomials and @xmath297 we suppose that , by neglecting the space - charge interaction and in a comoving frame of reference , the system will be correctly described by the ground state @xmath298 associated to the eigenvalue @xmath299 . n. cufaro petroni , s. de martino , s. de siena , and f. illuminati , phys . e * 63 * , 016501 ( 2001 ) ; n.cufaro petroni , s. de martino , s. de siena , and f. illuminati , in _ quantum aspects of beam physics 2k _ , p. chen ed . ( world scientific , singapore , 2002 ) p.507 ; n. cufaro petroni , s. de martino , s. de siena , and f. illuminati , in _ quantum aspects of beam physics _ , p. chen ed . ( world scientific , singapore , 1999 ) p.710 ; n. cufaro petroni , s. de martino , s. de siena , r. fedele , f. illuminati , and s. i. tzenov , in proceedings of the _ european particle accelerator conference epac98 _ , s. myers et al . eds . ( iop publishing , bristol , 1998 ) p.1259 .

we analyze the quantum process in which a cosmic string breaks in an anti - de sitter ( ads ) background , and a pair of charged or neutral black holes is produced at the ends of the strings . the energy to materialize and accelerate the pair comes from the strings tension . in an ads background this is the only study done in the process of production of a pair of correlated black holes with spherical topology . the acceleration of the produced black holes is necessarily greater than , where is the cosmological constant . only in this case the virtual pair of black holes can overcome the attractive background ads potential well and become real . the instantons that describe this process are constructed through the analytical continuation of the ads c - metric . then , we explicitly compute the pair creation rate of the process , and we verify that ( as occurs with pair creation in other backgrounds ) the pair production of nonextreme black holes is enhanced relative to the pair creation of extreme black holes by a factor of , where is the black hole horizon area . we also conclude that the general behavior of the pair creation rate with the mass and acceleration of the black holes is similar in the ads , flat and de sitter cases , and our ads results reduce to the ones of the flat case when .

hep-th0401069

the quantum schwinger - like process of black hole pair creation in an external field is by now a well studied subject . it supplied a way to produce planck scale black holes , and it gave useful clues to the discussion of issues like the statistical properties of black holes , the black hole information paradox , and the analysis of topology changing processes . in order to turn the pair of virtual black holes real one needs a background field that provides the energy needed to materialize the pair , and that furnishes the force necessary to accelerate away the black holes once they are created . this background field can be : ( i ) an external electromagnetic field with its lorentz force ( see @xcite-@xcite ) , ( ii ) the positive cosmological constant @xmath6 , or inflation ( see @xcite-@xcite ) , ( iii ) a cosmic string with its tension ( see @xcite-@xcite ) , ( iv ) a domain wall with its gravitational repulsive energy ( see @xcite-@xcite ) . one can also have a combination of the above fields , for example , a process involving cosmic string breaking in a background magnetic field @xcite , or a scenario in which a cosmic string breaks in a positive cosmological background @xcite . the evaluation of the black hole pair creation rate has been done at the semiclassical level using the instanton method . an instanton is an euclidean solution that interpolates between the initial and final states of a classically forbidden transition , and is a saddle point for the euclidean path integral that describes the pair creation rate . the instantons that mediate the above processes can be obtained by analytically continuing ( i ) the ernst solution @xcite ( which is available only for @xmath7 ) , ( ii ) the de sitter black hole solutions , ( iii ) the c - metric with a generalized cosmological constant @xmath6 @xcite , or ( iv ) the domain wall solution @xcite . all these exact solutions describe appropriately the evolution of the black hole pair that has been created , since they represent a pair of black holes accelerated by an electromagnetic field , by a cosmological constant , by a string , or by a domain wall , respectively . an important process that accompanies the production of the black hole pair is the emission of electromagnetic and gravitational radiation . in a flat background , an estimate for the amount of gravitational radiation released during the pair creation period has been given in @xcite . after the pair creation , the black hole pair accelerates away and , consequently , the black holes continue to release gravitational and electromagnetic energy . in a @xmath7 background , the gravitational radiation emitted by uniformly accelerated black holes has been computed in @xcite , while in a ds background this analysis has been carried in @xcite , and in an ads background the programme has been completed in @xcite . we ask the reader to see , e.g. , the introduction of @xcite for a recent and detailed description of the works done on black hole pair creation processes . in this paper we want to analyze the process in which a cosmic string breaks and a pair of black holes is produced at the ends of the string , in an anti - de sitter ( ads ) background ( @xmath2 ) . therefore , the energy to materialize and accelerate the pair comes from the strings tension . in an ads background this is the only study done in the process of production of a pair of correlated black holes with spherical topology . the analysis of this process in a flat background ( @xmath7 ) has been done in @xcite , while in a de sitter background ( @xmath8 ) it has been carried in @xcite ) . the instantons for this process can be constructed by analytically continuing the ads c - metric found in @xcite and analyzed in detail in @xcite . contrary to the @xmath7 @xcite and @xmath8 @xcite cases , the ads describes a pair of accelerated black holes only when the acceleration supplied by the strings is greater than @xmath1 @xcite . hence we expect that pair creation of black holes in an ads background is possible only when @xmath9 . we will confirm this expectation . the quantum production of uncorrelated ads black holes has been studied in @xcite , and the pair creation process of topological ads black holes ( with hyperbolic topology ) has been analyzed in @xcite in a domain wall background . the plan of this paper is as follows . in sec . [ sec : ads c - inst ] we construct , from the ads c - metric , the instantons that describe the pair creation process . in sec . [ sec : calc - i ads ] , we explicitly evaluate the pair creation rate of black holes in an ads background when a string breaks . in sec . [ sec : conc ads ] concluding remarks are presented . in the appendix [ sec : a1 ] , we discuss , for the benefit of comparison with the @xmath7 case , the pair creation of black holes in a @xmath7 background . we explicitly compute the numerical value of the pair creation rate , which has not been done yet . in the appendix [ sec : a2 ] a heuristic derivation of the pair creation rates is given . throughout this paper we use units in which @xmath10 . the ads c - metric has been found in @xcite . the physical properties and interpretation of this solution have been analyzed in @xcite . when @xmath9 , and only in this case @xcite , the ads c - metric describes a pair of uniformly accelerated black holes in an ads background , with the acceleration being provided by two strings , from each one of the black holes towards infinity , that pulls them away . since we are interested in black hole pair creation , hereafter we will deal only with the @xmath9 case . for a detailed discussion on the properties of the ads c - metric we ask the reader to see @xcite . here we will only mention those which are really essential . the gravitational field of the lorentzian ads c - metric is given by ( see , e.g. , @xcite ) @xmath11^{-2 } ( -{\cal f}dt^2 + { \cal f}^{-1}dy^2+{\cal g}^{-1}dx^2 + { \cal g}d\phi^2)\ : , \label{c - metric}\ ] ] with @xmath12 where @xmath2 is the cosmological constant , @xmath13 is the acceleration of the black holes , and @xmath14 and @xmath15 are the adm mass and electromagnetic charge of the non - accelerated black hole , respectively . the maxwell field in the magnetic case is given by @xmath16 while in the electric case it is given by @xmath17 the general shape of @xmath18 and @xmath19 is represented in fig . [ g3_pc_ads ] . [ t ] shape of @xmath18 and @xmath20 for a general nonextreme charged massive ads c - metric . the allowed range of @xmath21 is between @xmath22 and @xmath23 where @xmath18 is positive and compact . in the lorentzian sector , the permitted range of @xmath24 is @xmath25 . the presence of an accelerated horizon is indicated by @xmath26 and @xmath27 and the inner and outer charged horizons by @xmath28 and @xmath29 . in the extreme case @xmath28 and @xmath29 coincide , and @xmath18 has only two real roots represented by @xmath22 and @xmath23 . when @xmath30 and @xmath31 , both @xmath18 and @xmath20 are represented by a parabola with the only zeros of @xmath18 being @xmath22 and @xmath23 , and the only zeros of @xmath32 being @xmath27 and @xmath26 . , title="fig:",height=211 ] the solution has a curvature singularity at @xmath33 where the matter source is . the point @xmath34 corresponds to a point that is infinitely far away from the curvature singularity , thus as @xmath24 increases we approach the curvature singularity and @xmath35 is the inverse of a radial coordinate . at most , @xmath20 can have four real zeros which we label in ascending order by @xmath36 . the roots @xmath28 and @xmath29 are respectively the inner and outer charged black hole horizons , and @xmath26 is an acceleration horizon . the negative root @xmath37 satisfies @xmath38 and thus has no physical significance . the angular coordinate @xmath21 belongs to the range @xmath39 $ ] for which @xmath40 ( when we set @xmath41 we have @xmath42 and @xmath43 $ ] ) . in order to avoid a conical singularity in the north pole , the period of @xmath44 must be given by @xmath45 and this leaves a conical singularity in the south pole with deficit angle @xmath46 that signals the presence of strings from each one of the black holes towards infinity with mass density @xmath47 , and with pressure @xmath48 . when we set the acceleration parameter @xmath0 equal to zero , the ads c - metric reduces to the usual ads@xmath49reissner - nordstrm or ads - schwarzschild solutions without conical singularities . later on it will be crucial to note that the number and nature of the horizons crossed by an observer that travels into the black hole singularity depends on the angular direction @xmath21 that he is following ( see fig . [ g3_pc_ads ] ) . this peculiar feature is due to the lower restriction on the value of @xmath24 ( @xmath50 ) , and a detailed discussion and explanation of it can be found in @xcite ( such an angular dependence occurs also in the flat c - metric @xcite , although not so sharply ) . for example , when the observer is travelling towards the black hole singularity following an angular direction in the vicinity of the south pole ( @xmath51 ) , he will cross only the outer @xmath29 and inner @xmath28 black hole horizons ( see fig . [ g3_pc_ads ] ) , while when he does this trip following an angular direction in the vicinity of the equator ( @xmath52 ) , he crosses the acceleration horizon @xmath26 before passing through the black hole horizons @xmath29 and @xmath28 . in order to evaluate the black hole pair creation rate we need to find the instantons of the theory , i.e. , we must look into the euclidean section of the ads c - metric and choose those euclidean solutions which are regular in a way that will be explained soon . the euclidean section of the ads c - metric is obtained from the lorentzian ads c - metric by introducing an imaginary time coordinate @xmath53 in ( [ c - metric])-([f - el ] ) . to have a positive definite euclidean metric we must require that @xmath24 belongs to @xmath54 . in general , when @xmath55 , one then has conical singularities at the horizons @xmath56 and @xmath57 . in order to obtain a regular solution we have to eliminate the conical singularities at both horizons , ensuring in this way that the system is in thermal equilibrium . this is achieved by imposing that the period of @xmath58 is the same for the two horizons , and is equivalent to requiring that the hawking temperature of the two horizons be equal . to eliminate the conical singularity at @xmath56 the period of @xmath58 must be @xmath59 ( where @xmath60 is the surface gravity of the acceleration horizon ) , @xmath61 this choice for the period of @xmath58 also eliminates simultaneously the conical singularity at the outer black hole horizon , @xmath29 , if and only if the parameters of the solution are such that the surface gravities of the black hole and acceleration horizons are equal ( @xmath62 ) , i.e. @xmath63 this condition is satisfied by a regular euclidean solution with @xmath64 that will be referred to as nonextreme ads instanton . we now turn our attention to the case @xmath65 ( and @xmath66 ) . when this happens the allowed range of @xmath24 in the euclidean sector is simply @xmath67 . this occurs because when @xmath68 the proper distance along spatial directions between @xmath26 and @xmath29 goes to infinity . the point @xmath29 disappears from the @xmath69 section which is no longer compact but becomes topologically @xmath70 . thus , in this case we have a conical singularity only at @xmath26 , and so we obtain a regular euclidean solution by simply requiring that the period of @xmath58 be equal to ( [ period tau - ya ] ) . we will label this solution by extreme ads instanton . in a de sitter ( ds ) background there is another @xmath55 instanton which satisfies @xmath71 . it is called nariai instanton . moreover , in the ds background , there is also a special solution that satisfies @xmath72 . it is called ultracold instanton . in the ads c - metric case , the counterparts of these ds instantons are of no interest for the pair creation process because they are out of the allowed range of the angular direction @xmath21 . in secs . [ sec : lukewarm - inst ads ] and [ sec : cold - inst ads ] we will describe in detail the nonextreme ads instanton with @xmath73 and the extreme ads instanton with @xmath74 , respectively . these instantons are the natural ads c - metric counterparts of the lukewarm ds c and cold ds c instantons constructed in @xcite . thus , these instantons could also be labelled as lukewarm ads c and cold ads c instantons . moreover , in sec . [ sec : submaximal - inst ads ] , we will also describe saddle point solutions that have conical singularities . these solutions represent nonextreme instantons with @xmath75 . in sec . [ sec : background string ads ] we will study the background reference spacetime that describes the initial system before the pair creation . these results will allow us to calculate the pair creation rate of accelerated ads black holes in sec . [ sec : calc - i ads ] . as we said above , for the nonextreme ads instanton the gravitational field is given by ( [ c - metric ] ) with the requirement that @xmath20 satisfies @xmath77 and @xmath78 . in this case we can then write @xmath79 with @xmath80 the parameters @xmath0 , @xmath6 , @xmath14 and @xmath15 , written as a function of @xmath26 and @xmath29 , are @xmath81 thus , the mass and the charge of the nonextreme ads instanton are necessarily equal , @xmath73 , as occurs with its flat @xcite and ds counterparts @xcite . the demand that @xmath82 is real requires that @xmath83 and that @xmath84 therefore , as already anticipated , the nonextreme ads instanton is available only when ( [ amin ] ) is satisfied . as we said , the allowed range of @xmath24 in the euclidean sector is @xmath54 . then , the period of @xmath58 , ( [ period tau - ya ] ) , that avoids the conical singularity at both horizons is @xmath85 and @xmath86 is the common temperature of the two horizons . using the fact that @xmath87 [ see ( [ fg pcads ] ) ] we can write @xmath88 and the roots of @xmath18 we are interested in are the south and north pole ( represented , respectively , as @xmath22 and @xmath23 in fig . [ g3_pc_ads ] ) , @xmath89 when @xmath14 and @xmath15 go to zero we have @xmath90 and @xmath91 . this is the reason why we decided to work in between the roots @xmath22 and @xmath23 , instead of working in between the roots @xmath92 and @xmath93 also represented in fig . [ g3_pc_ads ] . indeed , when @xmath94 and @xmath95 these two last roots disappear , and our instanton has no vacuum counterpart . now , the requirement that @xmath96 is real demands that @xmath97 [ note that @xmath98 , see ( [ mq - luk - v0 ] ) ] . if this requirement is not fulfilled then @xmath18 has only two real roots that are represented as @xmath92 and @xmath23 in fig . [ g3_pc_ads ] and , as we have just said , in this case the solution has no counterpart in the @xmath30 and @xmath31 case . therefore we discard the solutions that satisfy @xmath99 , and hereafter when we refer to the mass of the nonextreme ads instanton we will be working in the range @xmath100 the period of @xmath44 , ( [ period phi ] ) , that avoids the conical singularity at the north pole ( and leaves one at the south pole responsible for the presence of the string ) is @xmath101 when @xmath14 and @xmath15 go to zero we have @xmath102 and the conical singularity disappears . the topology of the nonextreme ads instanton is @xmath103 , where @xmath104 represents @xmath105 , @xmath54 , @xmath106 , and @xmath107 , but we have to remove the region , @xmath108 . the lorentzian sector of this nonextreme instanton describes two charged ads black holes being accelerated by the strings , so this instanton describes pair creation of nonextreme black holes with @xmath73 . the gravitational field of the extreme ads instanton is given by ( [ c - metric ] ) with the requirement that the size of the outer charged black hole horizon @xmath29 is equal to the size of the inner charged horizon @xmath28 . let us label this degenerated horizon by @xmath110 : @xmath111 and @xmath112 . in this case , the function @xmath20 can be written as @xmath113 with @xmath114 note that , as occurred with the nonextreme ads instanton , the extreme ads instanton is also available only when @xmath9 . the roots @xmath110 , @xmath27 and @xmath26 are given by @xmath115 the mass and the charge of the solution are written as a function of @xmath110 as @xmath116 and , for a fixed @xmath0 and @xmath6 , the ratio @xmath117 is higher than @xmath118 . the conditions @xmath112 and @xmath119 require that , for the extreme ads instanton , the allowed range of @xmath110 is @xmath120 the value of @xmath26 decreases monotonically with @xmath110 and we have @xmath121 . the mass and the charge of the extreme ads instanton are also monotonically decreasing functions of @xmath110 , and as we come from @xmath122 into @xmath123 we have @xmath124 so , for a fixed @xmath6 , as the acceleration parameter @xmath0 grows the maximum value of the mass and of the charge decreases monotonically . for a fixed @xmath6 and for a fixed mass below @xmath125 , the maximum value of the acceleration is @xmath126 . as we have already said , the allowed range of @xmath24 in the euclidean sector is @xmath67 and does not include @xmath57 . then , the period of @xmath58 , ( [ period tau - ya ] ) , that avoids the conical singularity at the only horizon of the extreme ads instanton is @xmath127 and @xmath86 is the temperature of the acceleration horizon . in what concerns the angular sector of the extreme ads instanton , @xmath18 is given by ( [ fg pcads ] ) , and its only real zeros are the south and north pole ( represented , respectively , as @xmath22 and @xmath23 in fig . [ g3_pc_ads ] ; @xmath92 and @xmath93 also represented in this figure become complex roots in the extreme case ) , @xmath128 with @xmath129 where @xmath14 and @xmath15 are fixed by ( [ mq ] ) , for a given @xmath0 , @xmath6 and @xmath110 . when @xmath94 and @xmath130 we have @xmath90 and @xmath131 . the period of @xmath44 that avoids the conical singularity at the north pole ( and leaves one at the south pole responsible for the presence of the strings ) is given by ( [ period phi ] ) with @xmath23 defined in ( [ polos - cold ] ) . the topology of the extreme ads instanton is @xmath132 , where @xmath133 represents @xmath134 , @xmath67 , @xmath106 , and @xmath107 , but we have to remove the region , @xmath108 . since @xmath135 is at an infinite proper distance , the surface @xmath135 is an internal infinity boundary . the lorentzian sector of this extreme case describes two extreme ( @xmath74 ) charged ads black holes being accelerated by the strings , and the extreme ads instanton describes the pair creation of these extreme black holes . the ads c - metric instantons studied in the two last subsections , namely the nonextreme instantons with @xmath73 and the extreme instantons with @xmath74 , are saddle point solutions free of conical singularities both in the @xmath29 and @xmath26 horizons . the corresponding black holes may then nucleate in the ads background when a cosmic string breaks , and we will compute their pair creation rates in sec . [ sec : calc - i ads ] . however , these particular black holes are not the only ones that can be pair created . indeed , it has been shown in @xcite that euclidean solutions with conical singularities may also be used as saddle points for the pair creation process . these nonextreme instantons have @xmath75 and describe pair creation of nonextreme black holes with @xmath137 . in what follows we will find the range of parameters for which one has nonextreme black holes with conical singularities , i.e. , with @xmath75 . first , when @xmath138 and @xmath139 , we require that @xmath21 belongs to the interval @xmath39 $ ] ( sketched in fig . [ g3_pc_ads ] ) for which the charged solutions are in the same sector of the @xmath30 and @xmath31 solutions . defining @xmath140 the above requirement is fulfilled by the parameter range @xmath141 for which @xmath142 , with @xmath143 being the less negative @xmath21 where the derivative of @xmath18 vanishes . second , in order to insure that one has a nonextreme solution we must require that @xmath144 for which @xmath145 , with @xmath146 being the point in between @xmath29 and @xmath28 where the derivative of @xmath20 vanishes . we have @xmath147 except at @xmath148 where these two functions are equal ; @xmath149 is always positive ; and @xmath150 for @xmath151 and @xmath152 for @xmath153 . the nonextreme black holes with conical singularities are those that satisfy ( [ rangeg<0 ] ) , ( [ rangef>0 ] ) and @xmath9 . the range of parameters of these nonextreme black holes with @xmath75 are sketched in fig . [ fig - range pc ads ] . allowed ranges of the parameters @xmath154 for which one has a solution representing a pair of accelerated black holes . the planar surface whose frontier is the triangle @xmath155 represents the nonextreme ads instanton with @xmath73 . the curved surface delimitated by the closed line @xmath156 represents the extreme ads instanton with @xmath74 and satisfies @xmath157 . the curved surface whose frontier is @xmath158 satisfies @xmath159 [ see ( [ rangeg<0 ] ) ] . the plane surface with boundary given by @xmath160 satisfies @xmath161 . neutral ads instantons ( @xmath31 ) are those that belong to the planar surface with the triangle boundary @xmath162 . the nonextreme ads instantons with @xmath75 are those whose parameters are in the volume with boundary defined by @xmath160 , @xmath156 , @xmath163 , @xmath162 and @xmath158 . , height=226 ] in order to compute the pair creation rate of the nonextreme black holes with @xmath75 , we will need the relation between the parameters @xmath0 , @xmath6 , @xmath14 , @xmath15 , and the horizons @xmath26 , @xmath29 and @xmath28 . in general , for a nonextreme solution with horizons @xmath164 , one has @xmath165 with @xmath166 the parameters @xmath0 , @xmath6 , @xmath14 and @xmath15 can be expressed as a function of @xmath26 , @xmath29 and @xmath28 by @xmath167 the allowed values of parameters @xmath14 and @xmath15 are those contained in the interior region sketched in fig . [ fig - range pc ads ] . the topology of these nonextreme ads instantons with @xmath75 is @xmath103 , where @xmath104 represents @xmath105 , @xmath54 , @xmath106 , and @xmath107 , but we have to remove the region , @xmath108 . these instantons describe the pair creation of nonextreme black holes with @xmath168 . so far , we have described the solution that represents a pair of black holes accelerated by the strings tension . this solution describes the evolution of the black hole pair after its creation . now , we want to find a solution that represents a string in a ads background . this solution will describe the initial system , before the breaking of the cosmic string that leads to the formation of the black hole pair . in order to achieve our aim we note that at spatial infinity the gravitational field of the euclidean ads c - metric reduces to @xmath169 ^ 2}{\biggl [ } -\left ( \frac{|\lambda|}{3a_0^{\,2}}-1+y^2 \right ) dt^2 \nonumber\\ & & + \frac{dy^2}{\frac{|\lambda|}{3a_0^{\,2}}-1+y^2 } + \frac{dx^2}{1-x^2}+ ( 1-x^2)d\phi_0^{\,2 } { \biggr ] } , \label{backg metric}\end{aligned}\ ] ] and the maxwell field goes to zero . @xmath170 is a constant that represents a freedom in the choice of coordinates , and @xmath171 . we want that this metric also describes the solution before the creation of the black hole pair , i.e. , we demand that it describes a string with its conical deficit in an ads background . now , if we want to maintain the intrinsic properties of the string during the process we must impose that its mass density and thus its conical deficit remains constant . after the pair creation we already know that the conical deficit is given by ( [ conic - sing ] ) . hence , the requirement that the background solution describes an ads spacetime with a conical deficit angle given exactly by ( [ conic - sing ] ) leads us to impose that in ( [ backg metric ] ) one has @xmath172 the arbitrary parameter @xmath170 can be fixed as a function of @xmath0 by imposing a matching between ( [ c - metric ] ) and ( [ backg metric ] ) at large spatial distances . we will do this matching in the following section . the black hole pair creation rate is given by the path integral @xmath173d[a_{\mu } ] e^{-\left [ i(g_{\mu\nu},a_{\mu})-i_0(g^0_{\mu\nu},a^0_{\mu})\right ] } \ : , \label{path integral } \end{aligned}\ ] ] where @xmath174 and @xmath175 are the induced metric and electromagnetic potential on the boundary @xmath176 of a compact manifold @xmath177 , @xmath178 $ ] is a measure on the space of the metrics @xmath179 and @xmath180 $ ] is a measure on the space of maxwell field @xmath181 , and @xmath182 is the euclidean action of the instanton that mediates the process . in our case this action is the einstein - maxwell action with a negative cosmological constant @xmath6 . the path integral is over all compact metrics and potentials on manifolds @xmath177 with boundary @xmath176 , which agree with the boundary data on @xmath176 . @xmath183 is the action for the background reference spacetime , the ads background with the string , specified by @xmath184 and @xmath185 ( see sec . [ sec : background string ads ] ) . its presence is required because it describes the initial system . moreover , the geometry of the final system with the black hole pair is noncompact , that is @xmath182 diverges . however , the physical action @xmath186 is finite for fields @xmath179 , @xmath181 that approach asymptotically @xmath184 , @xmath185 in an appropriate way @xcite . specifically , one fixes a boundary near infinity , @xmath187 , and demands that @xmath179 , @xmath181 and @xmath184 , @xmath185 induce fields on this boundary that agree to sufficient order , so that their difference does not contribute to the action in the limit that @xmath187 goes to infinity @xcite . in the semiclassical instanton approximation , the dominant contribution to the path integral ( [ path integral ] ) comes from metrics and maxwell fields which are near the solutions ( instantons ) that extremalize the euclidean action and satisfy the boundary conditions . in this approximation , the pair creation rate of ads black holes is then given by @xmath188 where @xmath189 includes already the contribution from the background reference spacetime . @xmath190 is the one - loop prefactor that accounts for the fluctuations in the gravitational and matter fields , and its evaluation will not be considered in this paper ( see @xcite for a treatment of this factor in some backgrounds ) . hawking and horowitz @xcite ( see also brown and york @xcite ) have shown that the euclidean action of the instanton that mediates pair creation of nonextreme black holes can be written in the form @xmath191 where @xmath192 is the difference in area of the acceleration horizon between the ads c - metric and the background , and @xmath4 is the area of the black hole horizon present in the instanton . the euclidean action of the instanton that describes pair creation of extreme black holes can be written as @xmath193 in these relations , @xmath194 is the period of the euclidean time , and @xmath195 represents the hamiltonian of the system which , for static solutions , is given by @xcite @xmath196 the boundary @xmath197 consists of an initial and final spacelike surface , @xmath198 , of constant @xmath199 with unit normal @xmath200 ( @xmath201 ) and with intrinsic metric @xmath202 plus a timelike surface near infinity , @xmath187 , with unit normal @xmath203 ( @xmath204 , and @xmath205 ) and with intrinsic metric @xmath206 . this surface @xmath187 needs not to be at infinity . it can also be at a black hole horizon or at an internal infinity , but we will generally label it by @xmath187 . this surface @xmath187 is foliated by a family of 2-surfaces @xmath207 that result from the intersection between @xmath187 and @xmath198 . in ( [ hamiltonian ] ) , @xmath208 and @xmath209 are the lapse functions of the system with the pair and of the background , respectively , and @xmath210 in the boundary near infinity , @xmath187 . @xmath211 is the hamiltonian constraint that contains contributions from the gravitational and maxwell fields , and vanishes for solutions of the equations of motion . then , the hamiltonian is simply given by the boundary surface term . in this surface term , @xmath212 represents the trace of the extrinsic curvature of the surface imbedded in the ads c - metric , and @xmath213 is the extrinsic curvature of the surface imbedded in the background spacetime . we will now verify that the boundary surface term in the hamiltonian ( [ hamiltonian ] ) is also zero , and thus the hamiltonian makes no contribution to ( [ i - non - ext ] ) and ( [ i - ext ] ) . we follow the technical procedure applied by hawking , horowitz and ross @xcite in the ernst solution and by hawking and ross @xcite in the flat c - metric . as we said above , one will require that the intrinsic metric @xmath214 on the boundary @xmath207 as embedded in the ads c - metric agrees ( to sufficient order ) with the intrinsic metric on the boundary @xmath207 as embedded in the background spacetime , in order to be sure that one is taking the same near infinity boundary in the evaluation of the quantities in the two spacetimes . in the ads c - metric one takes this boundary to be at @xmath215 , where @xmath216 . the background reference spacetime is described by ( [ backg metric ] ) , subjected to @xmath217 , @xmath218 and ( [ delta phi0 ] ) . in this background spacetime we take the boundary @xmath207 to be at @xmath219 , where @xmath220 . we are now interested in writing the intrinsic metric in the boundary @xmath207 ( the 2-surface @xmath221 const and @xmath222 ) . in the ads c - metric ( [ c - metric ] ) one performs the coordinate transformation @xmath223 in order that @xmath224 , and the analytic continuation of @xmath225 has period @xmath226 , i.e. , @xmath227 . furthermore , one takes @xmath228 where @xmath229 . by making the evaluations up to second order in @xmath230 ( since higher order terms will not contribute to the hamiltonian in the final limit @xmath231 ) , the intrinsic metric on the boundary @xmath207 is then @xmath232}. \label{metric boundary c } \end{aligned}\ ] ] analogously , for the background spacetime ( [ backg metric ] ) one sets @xmath233 and the intrinsic metric on the boundary @xmath207 yields @xmath234 \label{metric boundary bg } \end{aligned}\ ] ] these two intrinsic metrics on the boundary will agree ( up to second order in @xmath235 ) as long as we take the period of @xmath236 to be given by ( [ delta phi0 ] ) and the following matching conditions are satisfied , @xmath237 @xmath238 ^ 2}{2{\cal g}''(x_\mathrm{s})}\,a^2 \ , . \label{match condition a } \end{aligned}\ ] ] note that the maxwell fields of the two solutions agree trivially at the near infinity boundary @xmath207 . in what concerns the lapse function of the ads c - metric , we evaluate it with respect to the time coordinate @xmath225 defined in ( [ redef phi t ] ) and , using @xmath239^{-2}{\cal f}dt^2=n^2 d\tilde{t}^{\,2}$ ] , we find @xmath240 analogously , an evaluation with respect to the time coordinate @xmath199 defined in ( [ backg metric ] ) yields @xmath241 note that these two lapse functions are also matched by the conditions ( [ match condition epsilon ] ) and ( [ match condition a ] ) . the extrinsic curvature to @xmath207 as embedded in @xmath198 is @xmath242 ( where @xmath243 represents the covariant derivative with respect to @xmath179 ) , and the trace of the extrinsic curvature is @xmath244 . the extrinsic curvature of the boundary embedded in the ads c - metric is then @xmath245 while the extrinsic curvature of the boundary embedded in the background reference spacetime is @xmath246 we are now in position to compute the contribution from the surface boundary term in ( [ hamiltonian ] ) . the evaluation in the ads c - metric yields @xmath247 while for the background reference spacetime we have @xmath248 ^ 2}{|{\cal g}'(x_\mathrm{n})|{\cal f}'(y_a ) } \frac{|\lambda|}{3 } \frac{1}{(2a_0^{\,2}\varepsilon_0)^2 } \nonumber\\ & & \times \left ( 1-\frac{3}{|\lambda| } a_0^{\,2}\varepsilon_0 \right ) . \label{int nk bg } \end{aligned}\ ] ] from the matching conditions ( [ match condition epsilon ] ) and ( [ match condition a ] ) we conclude that these two boundary terms are equal . hence , the surface term and the hamiltonian ( [ hamiltonian ] ) vanish . the euclidean action ( [ i - non - ext ] ) of the nonextreme ads instanton that mediates pair creation of nonextreme black holes is then simply @xmath249 while the euclidean action ( [ i - non - ext ] ) of the extreme ads instanton that mediates pair creation of nonextreme black holes is just given by @xmath250 thus as occurs in the ernst case , in the flat c - metric case , in the de sitter case and in the ds c - metric case , the pair creation of nonextreme black holes is enhanced relative to the pair creation of extreme black holes by a factor of @xmath251 . in the next two subsections we will explicitly compute ( [ i - non - ext2 ] ) using the results of sec . [ sec : lukewarm - inst ads ] , and ( [ i - ext2 ] ) using the results of sec . [ sec : cold - inst ads ] . in sec . [ sec : submaximal - rate ads ] we analyze the pair creation rate of black holes discussed in sec . [ sec : submaximal - inst ads ] . remark that the only horizons that contribute with their areas to ( [ i - non - ext2 ] ) and ( [ i - ext2 ] ) are those that belong to the instanton responsible for the pair creation , i.e , only those horizons that are in the euclidean sector of the instanton will make a contribution . the domain of validity of our results is the particle limit , @xmath252 , for which the radius of the black hole , @xmath253 , is much smaller than the typical distance between the black holes at the creation moment , @xmath254 ( this value follows from the rindler motion @xmath255 that describes the uniformly accelerated motion of the black holes ) . in the nonextreme case , the instanton has two horizons in its euclidean section , namely the acceleration horizon at @xmath56 and the black hole horizon at @xmath57 ( the horizons @xmath27 and @xmath28 do not belong to the instanton , see fig [ g3_pc_ads ] ) . the black hole horizon covers the whole range of the angular coordinate @xmath21 , @xmath256 , and its area is @xmath257 where @xmath29 is given in ( [ ya - luk pcads ] ) , @xmath22 and @xmath23 are defined by ( [ polos - luk pcads ] ) , and @xmath258 is given by ( [ period phi - luk pcads ] ) . the acceleration horizon , @xmath26 defined in ( [ ya - luk pcads ] ) , of the nonextreme ads instanton covers the angular range @xmath259 ( i.e. , it is not present in the vicinity of the south pole ) , and is noncompact , i.e. , its area is infinite . we have to deal appropriately with this infinity . in order to do so , we first introduce a boundary at @xmath260 ( @xmath216 ) , and compute the area inside of this boundary , which yields @xmath261 when we let @xmath262 , the term @xmath263 diverges , and the acceleration horizon has an infinite area . this area is renormalized with respect to the area of the acceleration horizon of the background reference spacetime ( [ backg metric ] ) . this background reference spacetime has an acceleration horizon at @xmath264 which is the direct counterpart of the above acceleration horizon of the nonextreme ads instanton , and it covers the angular range @xmath265 . the area inside a boundary at @xmath266 is @xmath267 where in the last step we have replaced @xmath268 by ( [ delta phi0 ] ) . when @xmath269 , the term @xmath270 diverges , and thus the area of this background acceleration horizon is also infinite . now , we have found that the intrinsic metrics at the near infinity boundary match together if ( [ match condition epsilon ] ) and ( [ match condition a ] ) are satisfied . our next task is to verify that these matching conditions between ( @xmath271 ) and ( @xmath272 ) are such that the divergent terms in ( [ area ac - luk pcads ] ) and ( [ area ac - luk - bg ] ) cancel each other . it is straightforward to show that @xmath273 yields a finite value if @xmath274 thus , the matching conditions ( [ match condition epsilon ] ) and ( [ match condition a ] ) satisfy condition ( [ match condition ] ) , i.e. , they indeed eliminate the divergencies in @xmath192 . it is worthy to remark that with the choices ( [ match condition epsilon ] ) and ( [ match condition a ] ) the proper lengths of the boundaries @xmath260 and @xmath275 , ( given , respectively , by @xmath276 and @xmath277 ) do not match . this is in contrast with the flat case @xcite , in which the choice of the matching parameters that avoids the infinities in @xmath278 , also leads to @xmath279 . in the ads case , our main goal was to remove the infinities in @xmath280 . we have achieved this aim by comparing the appropriate acceleration horizons in the instanton and in the reference background . the fact that the matching relations then lead to @xmath281 is not a problem at all . replacing ( [ match condition epsilon ] ) and ( [ match condition a ] ) in ( [ area ac - luk - bg ] ) yields for @xmath282 the result @xmath283 ^ 2 } } } \frac { { \cal g}''(x_\mathrm{s } ) } { { \cal g}'(x_\mathrm{s } ) } { \biggr ) } , \label{diferenca area - luk pcads } \end{aligned}\ ] ] where @xmath284 ^ 2<1 $ ] . adding ( [ area bh - luk pcads ] ) and ( [ diferenca area - luk pcads ] ) , and using the results of sec . [ sec : lukewarm - inst ads ] yields finally the total area of the nonextreme ads instanton with @xmath73 @xmath285 where @xmath286 and @xmath96 are defined in ( [ polos - luk pcads ] ) , @xmath82 is given by ( [ ya - luk pcads ] ) , and condition ( [ mq - luk pcads ] ) must be satisfied . the pair creation rate of nonextreme ads black holes with @xmath73 is then @xmath287 fixing @xmath0 and @xmath6 one concludes that the pair creation rate decreases as the mass of the black holes increases . moreover , fixing @xmath14 and @xmath6 , in the domain of validity of our results , @xmath252 and @xmath9 , as @xmath0 increases the pair creation rate increases . hence , the general behavior of the pair creation rate of nonextreme black holes with @xmath73 in the ads case is analogous to the corresponding behavior in the flat case @xcite ( see also appendix [ sec : a1 ] ) . in the extreme ads case , the instanton has a single horizon , the acceleration horizon at @xmath56 , in its euclidean section , since @xmath57 is an internal infinity . the pair creation rate of extreme ads black holes with @xmath74 is then @xmath288 where @xmath289 is given by ( [ diferenca area - luk pcads ] ) , with @xmath26 defined in ( [ zerosy3-cold ] ) , @xmath22 and @xmath23 given by ( [ polos - cold ] ) , and condition ( [ mq - cold ] ) must be satisfied . the pair creation rate decreases as the mass of the black holes increases , and the pair creation rate increases when @xmath0 increases . the general behavior of the pair creation rate of extreme black holes as a function of @xmath14 and @xmath0 in the ads case is also analogous to the behavior of the flat case , discussed in @xcite ( see also appendix [ sec : a1 ] ) . the nonextreme instantons , that mediate the pair creation of nonextreme black holes with @xmath75 ( including the case @xmath31 ) , have two horizons in their euclidean section , namely the acceleration horizon at @xmath56 and the black hole horizon at @xmath57 . the pair creation rate of nonextreme ads black holes with @xmath75 is then @xmath290 , with @xmath291 given by ( [ area bh - luk pcads ] ) and @xmath292 given by ( [ diferenca area - luk pcads ] ) , subjected to the results found in sec . [ sec : submaximal - inst ads ] . the pair creation rate decreases as the mass of the black holes increases , and the pair creation rate increases when @xmath0 increases . we have studied in detail the quantum process in which a cosmic string breaks in an anti - de sitter ( ads ) background and a pair of black holes is created at the ends of the string . the energy to materialize and accelerate the black holes comes from the strings tension . the analysis of this process in a flat background ( @xmath7 ) has been carried in @xcite , while in a de sitter background ( @xmath8 ) it has been done in @xcite . in an ads background this is the only study done in the process of production of a pair of correlated black holes with spherical topology . we remark that in principle our explicit values for the pair creation rates also apply to the process of pair creation in an external electromagnetic field , with the acceleration being provided in this case by the lorentz force instead of being furnished by the string tension . indeed , there is no ads ernst solution , and thus we can not discuss analytically the process . however , physically we could in principle consider an external electromagnetic field that supplies the same energy and acceleration as our strings and , from the results of the @xmath7 case ( where the pair creation rates in the string and electromagnetic cases agree ) , we expect that the pair creation rates found in this paper do not depend on whether the energy is being provided by an external electromagnetic field or by a string . it is well known that the ads background is attractive , i.e. , an analysis of the geodesic equations indicates that particles in this background are subjected to a potential well that attracts them ( see also appendix [ sec : a2 ] ) . therefore , if we have a virtual pair of black holes and we want to turn them real , we will have to furnish a sufficient force that overcomes this cosmological background attraction . we then expect that pair creation is possible only if the strings tension and the associated acceleration @xmath0 is higher than a critical value . we have confirmed that this is indeed the case : in the ads background , black holes can be pair produced only with an acceleration higher than @xmath1 . we have constructed the saddle point solutions that mediate the pair creation process through the analytic continuation of the ads c - metric , and we have explicitly computed the pair creation rate of the process . the ads pair creation rate reduces to the corresponding one of the flat case @xcite when we set @xmath7 . we have concluded that , for a pair of black holes that is subjected to a fixed @xmath6 and @xmath0 backgrounds , the pair creation probability decreases when the mass or charge of the black holes increases . moreover , when we fix the mass and the charge of the black holes , the probability they have to be pair created increases when the acceleration provided by the string increases . these results are physically expected , as an heuristic derivation done in appendix [ sec : a2 ] confirms . this process has also a clear analogy with a thermodynamical system , with the mass density of the string ( that is proportional to @xmath0 ) being the analogue of the temperature @xmath293 . indeed , from the boltzmann factor , @xmath294 ( where @xmath295 is the boltzmann constant ) , one knows that a higher background temperature makes the nucleation of a particle with energy @xmath296 more probable . equivalently , a higher acceleration provided by the string makes the creation of the black hole pair more probable . [ h ] general behavior of the black hole pair creation rate @xmath297 as a function of the acceleration @xmath0 provided by the strings , when a cosmic string breaks : ( a ) in a flat background ( @xmath7 ) @xcite , ( b ) in a ds background ( @xmath8 ) @xcite , and ( c ) in an ads background ( @xmath2 ) . , title="fig:",height=480 ] for the benefit of comparison , in fig . [ pc rates_ads ] we schematically represent the general behavior of the black hole pair creation rate @xmath297 as a function of the acceleration @xmath0 provided by the strings , when a cosmic string breaks in the three cosmological constant backgrounds . in a flat background [ see fig . [ pc rates_ads].(a ) ] , the pair creation rate is zero when @xmath41 @xcite . in this case , the flat c - metric reduces to a single black hole , and since we are studying the probability of pair creation , the corresponding rate is naturally zero . this does not mean that a single black hole can not be materialized from the quantum vacuum , it only means that this latter process is not described by the c - metric . the creation probability of a single black hole in a hot bath has been considered in @xcite . in a ds background [ see fig . [ pc rates_ads].(b ) ] , the pair creation rate is not zero when @xmath41 @xcite . this means that even in the absence of the string , the positive cosmological constant is enough to provide the energy to materialize the black hole pair @xcite . if in addition one has an extra energy provided by the string , the process becomes more favorable @xcite . in the ads case [ see fig . [ pc rates_ads].(c ) ] , the negative cosmological constant makes a negative contribution to the process , and black hole pair creation is possible only when the acceleration provided by the strings overcomes the ads background attraction . the branch @xmath298 represents the creation probability of a single black hole when the acceleration provided by the broken string is not enough to overcome the ads attraction , and was not studied in this paper . we have also verified that ( as occurs with pair creation in other backgrounds ) the pair production of nonextreme black holes is enhanced relative to the pair creation of extreme black holes by a factor of @xmath299 , where @xmath300 is the gravitational entropy of the black hole . it is a pleasure to acknowledge conversations with jos p. s. lemos , vitor cardoso , and alfredo b. henriques . this work was partially funded by fundao para a cincia e tecnologia ( fct ) through project cern / fis/43797/2001 and peso / pro/2000/4014 . i also acknowledge finantial support from the fct through praxis xxi programme . in a flat background ( @xmath7 ) , the analysis of the process of pair creation of black holes when a cosmic string breaks has been analyzed by hawking and ross @xcite . in this case , there is a direct @xmath7 counterpart of the nonextreme ads instanton and of the extreme ads instanton discussed in our ads case . these instantons describe pair creation of nonextreme ( with @xmath73 ) and extreme black holes with @xmath74 , respectively @xcite . the total area of these @xmath7 instantons can be found in @xcite , and can be obtained by taking the direct @xmath7 limit of ( [ area bh - luk pcads ] ) and ( [ diferenca area - luk pcads ] ) , together with the replacement @xmath301 . this procedure yields that ( [ area bh - luk pcads ] ) also holds in the @xmath7 case , while @xmath192 becomes @xmath302 in @xcite the explicit numerical value of @xmath303 and @xmath192 has not been computed . we will do it here . for the @xmath7 nonextreme case , the discussion of sec . [ sec : lukewarm - inst ads ] applies generically , as long as we set @xmath7 in the corresponding equations . with this data we find the explicit value of the total area of the nonextreme flat instanton @xmath304 in the particle limit , @xmath252 , the above relation reduces to @xmath305 and the mass density of the string is given by @xmath306 . the pair creation rate is then @xmath307 . thus , as occurs with the ads case , the pair creation rate decreases when @xmath14 increases and the rate increases when @xmath0 or @xmath308 increase [ see fig . [ pc rates_ads].(a ) ] . we remark that for @xmath309 , and as occurs in the corresponding ads case , the pair creation rate associated to ( [ area total - flat ] ) starts decreasing when @xmath0 increases . this is a physically unexpected result since a higher acceleration provided by the string background should favor the nucleation of a fixed black hole mass . the sector @xmath309 must then be discarded and the reason is perfectly identified : the domain of validity of the rates is @xmath252 , for which the radius of the black hole , @xmath310 , is much smaller than the typical distance between the black holes at the creation moment , @xmath254 ( this value follows from the rindler motion @xmath255 that describes the uniformly accelerated motion of the black holes ) . so , for @xmath311 one has @xmath312 and the black holes start interacting with each other . in what concerns the pair creation rate of extreme and nonextreme with @xmath76 @xmath7 black holes , an explicit computation shows that its general behavior with @xmath0 and @xmath14 is also similar to the one of the ads case , i.e. , the rate decreases when @xmath14 or @xmath15 increase , and the rate increases when @xmath0 increases . in order to clarify the physical interpretation of the results , in this appendix we heuristically derive some results discussed in the main body of the paper . in particular , we find heuristically the pair creation rates . in an ads background , pair creation of black holes is possible only when the acceleration provided by the strings satisfies @xmath9 . to understand this result one can argue as follows . in general , the time - time component of the gravitational field is given by @xmath313 , where @xmath314 is the newtonian potential . in the ads spacetime , one has @xmath315 and its derivative yields the force per unit mass or acceleration of the ads spacetime , @xmath316 , where we have replaced @xmath317 by the characteristic ads radius @xmath318 . the minus sign indicates that the ads background is attractive and thus , if one wants to have a pair of accelerated black holes driving away from each other , the cosmic string will have to provide a sufficient acceleration @xmath0 that overcomes the ads background attraction , i.e. , @xmath319 . an estimate for the black hole pair creation probability can be given by the boltzmann factor , @xmath320 , where @xmath296 is the energy of the system that nucleates and @xmath321 is the work done by the external force @xmath322 , that provides the energy for the nucleation , through the typical distance @xmath323 separating the created pair . first we ask what is the probability that a black hole pair is created in a @xmath7 background when a string breaks . this process has been discussed in @xcite ( see also appendix [ sec : a1 ] ) where it was found that the pair creation rate is @xmath324 . in this case , @xmath325 , where @xmath14 is the rest energy of the black hole , and @xmath326 is the work provided by the strings . to derive @xmath326 one can argue as follows . the acceleration provided by the string is @xmath0 , the characteristic distance that separates the pair at the creation moment is @xmath327 ( see appendix [ sec : a1 ] ) , and the characteristic mass of the system is a by the compton relation . thus , the characteristic work is @xmath328 . so , from the boltzmann factor we indeed expect that the creation rate of a black hole pair when a string breaks in a @xmath7 background is given by @xmath329 . now we ask what is the probability that a string breaks in an ads background and a pair of black holes is produced . as we saw just above , the presence of the ads background leads in practice to a problem in which we have a net acceleration that satisfies @xmath330 , this is , @xmath6 makes a negative contribution to the process . heuristically , we may then apply the same arguments that have been used in the last paragraph , with the replacement @xmath331 . at the end , the boltzmann factor tells us that the creation rate for the process is @xmath332 . so , given @xmath14 and @xmath6 , when the acceleration provided by the string grows the pair creation rate increases , as the explicit calculations done in the main body of the paper show . g. w. gibbons , in _ fields and geometry _ , proceedings of the 22nd karpacz winter school of theoretical physics , edited by a. jadczyk ( world scientific , singapore , 1986 ) . d. garfinkle , s. b. giddings , phys . lett . b*256 * , 146 ( 1991 ) . d. garfinkle , s. b. giddings and a. strominger , phys . d * 49 * , 958 ( 1994 ) h. f. dowker , j. p. gauntlett , d. a. kastor and j. traschen , phys . rev . d * 49 * , 2909 ( 1994 ) ; h. f. dowker , j. p. gauntlett , s. b. giddings and g. t. horowitz , phys . d * 50 * , 2662 ( 1994 ) . s. ross , phys . 2813 ( 1995 ) . j. d. brown , phys . rev . d * 51 * , 5725 ( 1995 ) ; phys . 5725 ( 1995 ) . s. w. hawking , g. t. horowitz and s. f. ross , phys . d * 51 * , 4302 ( 1995 ) . s. w. hawking and s. f. ross , phys . d * 52 * , 5865 ( 1995 ) . r. emparan , phys . d * 52 * , 6976 ( 1995 ) . f. mellor and i. moss , phys . b * 222 * , 361 ( 1989 ) ; class . * 6 * , 1379 ( 1989 ) . r. b. mann and s. f. ross , phys . d * 52 * , 2254 ( 1995 ) . r. bousso and s. w. hawking , phys . d * 52 * , 5659 ( 1995 ) ; phys . d * 54 * , 6312 ( 1996 ) ; s. w. hawking , phys . d * 53 * , 3099 ( 1996 ) . r. garattini , nucl . b ( proc . suppl . ) * 57 * , 316 ( 1997 ) ; nuovo cimento b*113 * , 963 ( 1998 ) ; mod . phys . lett . a * 13 * , 159 ( 1998 ) . grav . * 18 * , 571 ( 2001 ) . m. volkov and a. wipf , nucl . b*582 * , 313 ( 2000 ) . i. s. booth and r. b. mann , phys . * 81 * , 5052 ( 1998 ) ; nucl . b * 539 * , 267 ( 1999 ) . r. bousso , phys . d * 55 * , 3614 ( 1997 ) . s. w. hawking and s. f. ross , phys . lett . * 75 * , 3382 ( 1995 ) . d. m. eardley , g. t. horowitz , d. a. kastor and j. traschen , phys . * 75 * , 3390 ( 1995 ) . a. achcarro , r. gregory and k. kuijken , phys . d * 52 * , 5729 ( 1995 ) . r. gregory and m. hindmarsh , phys . d * 52 * , 5598 ( 1995 ) . j. preskill and a. vilenkin , phys . d * 47 * , 2324 ( 1993 ) . f. j. ernst , j. math . phys . * 17 * , 515 ( 1976 ) . w. kinnersley and m. walker , phys . d * 2 * , 1359 ( 1970 ) . j. f. plebaski and m. demiaski , annals of phys . ( n.y . ) * 98 * , 98 ( 1976 ) . a. vilenkin , phys . b*133 * , 177 ( 1983 ) ; j. ipser and p. sikivie , phys . rev . d*30 * , 712 ( 1984 ) . v. cardoso , o. j. c. dias and j. p. s. lemos , phys . d * 67 * , 064026 ( 2003 ) ; v. cardoso , s. yoshida , o. j. c. dias and j. p. s. lemos , phys . d * 68 * , r061503 ( 2003 ) . j. bik , proc . a * 302 * , 201 ( 1968 ) ; v. pravda and a. pravdova , czech . * 50 * , 333 ( 2000 ) ; _ on the spinning c - metric _ , in _ gravitation : following the prague inspiration _ , edited by o. semerk , j. podolsk and m. zofka ( world scientific , singapore , 2002 ) , gr - qc/0201025 . j. bik and p. krtou , phys . d * 64 * , 124020 ( 2001 ) ; phys . lett . * 88 * , 211101 ( 2002 ) ; p. krtou and j. podolsk ' y , phys . d * 68 * , 024005 ( 2003 ) ; j. podolsk ' y , m. ortaggio and p. krtou , phys . * 91 * , 061101 ( 2003 ) . j. podolsk ' y , m. ortaggio and p. krtou , phys . d * 68 * , 124004 ( 2003 ) . g. w. gibbons , s. w. hawking and m. j. perry , nucl . b * 138 * , 141 ( 1978 ) ; g. w. gibbons and m. j. perry , nucl . b * 146 * , 90 ( 1978 ) ; s. m. christensen and m. j. duff , nucl . phys . b * 146 * , 90 ( 1978 ) ; p. ginsparg and m. j. perry , nucl . b * 222 * , 245 ( 1983 ) ; r. e. young , phys . d * 28 * , 2420 ( 1983 ) ; phys . d * 28 * , 2436 ( 1983 ) ; p. yi , phys . d * 52 * , 7089 ( 1995 ) . d. j. gross , m. j. perry and l. g. yaffe , phys . d * 25 * , 330 ( 1982 ) .

we explore the feasibility and astrophysical consequences of a new long - range gauge field ( `` dark electromagnetism '' ) that couples only to dark matter , not to the standard model . the dark matter consists of an equal number of positive and negative charges under the new force , but annihilations are suppressed if the dark matter mass is sufficiently high and the dark fine - structure constant is sufficiently small . the correct relic abundance can be obtained if the dark matter also couples to the conventional weak interactions , and we verify that this is consistent with particle - physics constraints . the primary limit on comes from the demand that the dark matter be effectively collisionless in galactic dynamics , which implies for tev - scale dark matter . these values are easily compatible with constraints from structure formation and primordial nucleosynthesis . we raise the prospect of interesting new plasma effects in dark matter dynamics , which remain to be explored .

0810.5126

a wide variety of cosmological observations seem to point to a two - component dark sector , in which approximately 73% of the energy density of the universe is in dark energy and 23% is in non - baryonic dark matter ( dm ) . ordinary matter constitutes the remaining 4% @xcite . the physics of the dark matter sector is plausibly quite minimal : an excellent fit to the data is obtained by assuming that dark matter is a cold , collisionless relic , with only the relic abundance as a free parameter . the well - known `` wimp miracle '' @xcite is the fact that a stable , neutral particle with weak - scale mass and coupling naturally provides a reasonable energy density in dm . particles of this type arise in models of low - scale supersymmetry @xcite or large extra dimensions @xcite , and provide compelling dm candidates . in the contemporary universe , they would be collisionless as far as any conceivable dynamical effects are concerned . nevertheless , it is also possible to imagine a rich phenomenology within the dark sector . the dark matter could be coupled to a relatively strong short - range force that could have interesting consequences for structure on small scales @xcite . alternatively , dm could also be weakly coupled to long - range forces , which might be related to dark energy @xcite . one difficulty with the latter is that such forces are typically mediated by scalar fields , and it is very hard to construct natural models in which the scalar field remains massless ( to provide a long - range force ) while interacting with the dm at an interesting strength . in this paper , we explore the possibility of a long - range _ gauge _ force coupled to dm , in the form of a new unbroken abelian field , dubbed the @xmath3 `` dark photon . '' we imagine that this new gauge boson @xmath4 couples to a dm fermion @xmath5 , but not directly to any standard model ( sm ) fields . our model is effectively parameterized by only two numbers : @xmath6 , the mass of the dm , and @xmath1 , the dark fine - structure constant . if @xmath6 is sufficiently large and @xmath1 is sufficiently small , annihilations of dm particles through the new force freeze out in the early universe and are negligible today , despite there being equal numbers of positively- and negatively - charged particles . the dark matter in our model is therefore a plasma , which could conceivably lead to interesting collective effects in the dm dynamics . remarkably , the allowed values of @xmath6 and @xmath1 seem quite reasonable . we find that the most relevant constraint comes from demanding that accumulated soft scatterings do not appreciably perturb the motion of dm particles in a galaxy over the lifetime of the universe , which can be satisfied by @xmath7 and @xmath8 tev . for values near these bounds , the alterations in dm halo shapes may in fact lead to closer agreement with observation @xcite . however , for such regions of parameter space , if @xmath3 were the only interaction felt by the @xmath5 particles , the resulting relic abundances would be slightly too large , so we need to invoke an additional annihilation channel . we show that @xmath5 can in fact be a wimp , possessing @xmath9 quantum numbers in addition to @xmath3 charge . such a model provides the correct relic abundance , and is consistent with particle - physics constraints so long as the mixing between ordinary photons and dark photons is sufficiently small . we consider a number of other possible observational limits on dark electromagnetism , and show that they do not appreciably constrain the parameter space . since the dm halo is overall neutral under @xmath3 , there is no net long - range force that violates the equivalence principle . although there are new light degrees of freedom , their temperature is naturally lower than that of the sm plasma , thereby avoiding constraints from big - bang nucleosynthesis ( bbn ) . energy loss through dark bremsstrahlung radiation is less important than the soft - scattering effects already mentioned . the coupling of dm to the dark radiation background can in principle suppress the growth of structure on small scales , but we show that the dm decouples from the dark radiation at an extremely high redshift . on the other hand , we find that there are plasma instabilities ( _ e.g. _ the weibel instability ) that can potentially play an important role in the assembly of galactic halos ; however , a detailed analysis of these effects is beyond the scope of this work . the idea of an unbroken @xmath0 coupled to dark matter is not new . forces have , of course , also been considered , see _ e.g. _ ref . @xcite ] de rujula et al . @xcite explored the possibility that dark matter was charged under conventional electromagnetism ( see also @xcite ) . gubser and peebles @xcite considered structure formation in the presence of both scalar and gauge long - range forces , but concentrated on a region of parameter space in which the gauge fields were subdominant . refs . @xcite considered several models for a hidden dark sector , including one manifestation in which the dark matter consists of heavy hidden - sector staus interacting via a copy of electromagnetism . the effect of dimension-6 operators containing a new @xmath0 gauge boson and sm fields was considered in ref . @xcite , for models where the only fields in a hidden sector are charged under the new force . additional models which contain unbroken abelian gauge groups may be found , for example in refs . @xcite . in this paper , we construct minimal models of dark matter coupled to a new unbroken @xmath3 , leaving the dark fine - structure constant and dark - matter mass as free parameters , and explore what regions of parameter space are consistent with astrophysical observations and what new phenomena might arise via the long - range gauge interaction . in section [ sec : earlyuniverse ] , we introduce our notation for a minimal dark - matter sector including a new abelian symmetry @xmath3 . we then consider the bounds on the new dark parameters from successful thermal production of sufficient quantities of dark matter as well as requiring that bbn and cosmic microwave background ( cmb ) predictions remain unchanged . the restrictions of parameter space are closely related to those resulting from standard short - range wimp dark matter . in section [ sec : galacticdynamics ] , we consider the effect of long range interactions on dm particle interactions in the halos of galaxies . by requiring that our model not deviate too greatly from the predictions of collisionless dm , we find that the allowed regions of @xmath10 parameter space from section [ sec : earlyuniverse ] are essentially excluded . in order to evade these constraints , section [ sec : su2 ] describes an extended model , where the dark - matter candidate is charged under both @xmath9 and the new @xmath3 . additional effects of dark radiation are presented in section [ sec : newlimits ] , and we conclude in section [ sec : conclusion ] . we note that our model does not address the hierarchy problem , nor provide a high - energy completion to the sm . however , new gauge groups and hidden sectors may be generic results of many such high - energy theories ( _ e.g. _ string and grand unified theories ) , and a wimp coupled to an unbroken @xmath0 is certainly a plausible low - energy manifestation of such theories . the most important lesson of our model is that interesting physics might be lurking in the dark sector , and it is worthwhile to consider a variety of possible models and explore their consequences for astrophysics and particle physics . we postulate a new `` dark '' abelian gauge group @xmath3 with gauge coupling constant @xmath11 and dark fine - structure constant @xmath12 . in the simplest case , the dark matter sector consists of a single particle @xmath5 with @xmath3 charge of @xmath13 and mass @xmath6 along with its antiparticle @xmath14 . for definiteness , we take @xmath5 to be a fermion , though our results are applicable to scalars as well . as the limits on new long range forces on sm fields are very stringent , we assume that all the sm fields are neutral under @xmath3 . for the moment we take the @xmath5 field to be a singlet under @xmath15 , a restriction that will be relaxed in section [ sec : su2 ] . as a result , this extension of the sm is anomaly free . in this section , we will derive constraints on the mass @xmath6 and coupling @xmath16 from the evolution of dark matter in the early universe . two considerations drive these constraints : the dark matter must provide the right relic abundance at thermal freeze - out , and the dark radiation from the @xmath3 can not contribute too greatly to relativistic degrees of freedom at bbn ( a similar bound coming from the cmb also applies but is weaker ) . the degrees of freedom in the dark sector are thus the heavy dm fermions @xmath5 and massless dark photons @xmath17 . the lagrangian for the dark sector is @xmath18 here @xmath19 and @xmath20 is the field - strength tensor for the dark photons . we assume that the mixing term @xmath21 is set to zero at some high scale ( say the gut scale ) . this is a self - consistent choice , since if there is no mixing between the dark and visible sectors , @xmath22 is preserved by the renormalization group evolution . ( in section [ sec : su2 ] we argue that mixing is not generated by radiative corrections even when @xmath5 carries @xmath9 quantum numbers . ) this choice allows us to bypass constraints on a new @xmath0 coming from mixing between the photon and dark photon , that is , ` paraphotons ' @xcite . we have no _ a priori _ assumptions on the parameters @xmath6 and @xmath16 , though as we shall see , it suffices to think of the former as @xmath23 and the latter @xmath24 . we now follow the thermal history of the dark sector . our analysis follows that of ref . @xcite ; we rehearse it in a slightly simpler context here to illustrate how the results depend on our various assumptions . if the visible sector and the dark sector are decoupled from each other , they may have different temperatures , @xmath25 and @xmath26 , respectively ; rapid interactions between them would equilibrate these two values . after inflation , the two sectors could conceivably reheat to different temperatures , depending on the coupling of the inflaton to the various fields . even if the temperatures are initially equal , once the two sectors decouple as the universe expands and cools , entropy deposited from frozen - out degrees of freedom in one sector will generally prevent the dark temperature @xmath27 from tracking the visible sector temperature @xmath25 . the ratio @xmath28 will depend on the spectrum of both sectors , and is itself a function of @xmath25 . as the temperature drops below a particle s mass , the associated degrees of freedom freeze out and dump entropy into their respective sectors ( dark or visible ) . this causes the temperature of that sector to decline more slowly than @xmath29 , where @xmath30 is the scale factor of the universe . as the entropy density @xmath31 of the visible sector and @xmath32 of the dark sector are individually conserved after decoupling , we must separately count the degrees of freedom in these two sectors . there are two definitions of degrees of freedom of interest to us : @xmath33 and @xmath34 . the former is defined as @xmath35 and is used in calculation of the total relativistic energy density , @xmath36 . here @xmath37 is the number of degree of freedom for particle species @xmath38 , @xmath39 is the temperature of the thermal bath of species @xmath38 , and @xmath25 is the temperature of the photon bath . the sums run over all active degrees of freedom at temperature @xmath25 . separating out the visible fields , @xmath33 can be written as @xmath40 where the sums now run over the dark particles . if we restrict the visible sector to the sm , then the term @xmath41 is @xmath42 above the top mass , dropping gradually to @xmath43 at @xmath44 . between @xmath45 , @xmath46 , and drops again to @xmath47 in the present day . ( see _ e.g. _ ref . @xcite for more detail . ) similarly , the total entropy density @xmath48 ( a conserved quantity ) at a photon temperature @xmath25 is proportional to @xmath49 , where @xmath50 prior to neutrino decoupling , all the relativistic standard model degrees of freedom are in thermal equilibrium at a common temperature . thus , before @xmath51 mev when neutrinos decouple , we have @xmath52 . furthermore , we may split the dark @xmath34 into heavy and light degrees of freedom : @xmath53 and @xmath54 , where the heavy degrees of freedom are non - relativistic at bbn . we are interested in the number of degrees of freedom at bbn ( @xmath51 mev ) because formation of the experimentally observed ratios of nuclei are very sensitive to the expansion of the universe at that time , which is related to the energy density of radiation through the friedmann equation . from this , a bound on the number of relativistic degrees of freedom can be derived @xcite . using the separate conservation of the visible and dark sector entropy and the previous definitions , we see that , at bbn @xmath55 here we have set @xmath56 ( recall that @xmath57 ) . the bbn bound on relativistic degrees of freedom is usually stated in terms of number of light neutrino species in thermal equilibrium at the time : @xmath58 @xcite . here the error bars correspond to @xmath59 ( 95% confidence ) . therefore , assuming three light neutrino species in the visible sector , if the dark sector is not to violate this bound , we must require @xmath60 combining eqs . ( [ eq : gstarbbn ] ) and ( [ eq : bbnbound ] ) , we find that @xmath61^{4/3 } \xi(t_{\rm rh})^4 \leq 2.52 ~~(95\%~\mbox{confidence } ) . \label{eq : bbnbound2}\ ] ] since the high energy completion of the visible sector must at minimum include the sm fields , @xmath62 ; a bound on the dark sector @xmath54 and @xmath53 can be derived for a fixed value of @xmath63 ( see fig . [ fig : gstar ] ) . increasing the number of visible sector degrees of freedom at high temperatures ( for example to that of the mssm ) relaxes this bound . ( those degrees of freedom relativistic at @xmath64 ) and @xmath53 ( the remaining dark degrees of freedom ) arising from bbn constraints eqs . ( [ eq : bbnbound ] ) and ( [ eq : bbnbound2 ] ) . the allowed regions correspond to 95% confidence levels for @xmath65 and a visible sector @xmath66 ( red ) , @xmath65 and @xmath67 ( corresponding to mssm particle content , in blue ) , and @xmath68 and @xmath69 ( in yellow ) . the minimal dark sector model of this paper is noted by a black star at @xmath70 and @xmath71 . [ fig : gstar ] ] in the case of @xmath72 , we see that the minimal model of the dark sector ( only heavy @xmath73 and light @xmath17 ) is safely included . due to the fourth power of @xmath74 entering into eq . ( [ eq : bbnbound2 ] ) , if the minimal dark sector is not to be ruled out , we find @xmath75 for the sm(mssm ) particle content . a similar bound on relativistic degrees of freedom can be derived from the cosmic microwave background , but provides a weaker @xmath59 exclusion limit @xcite . we now turn to bounds on the coupling @xmath16 and dark matter mass @xmath6 coming from the dark matter abundance . at temperatures @xmath27 much above @xmath6 , the @xmath5 particles are kept in thermal equilibrium with the dark photons @xmath17 ( and possibly other particles in the dark sector ) via pair annihilation / creation as in the feynman diagrams of fig . [ fig : feynman ] . since the annihilation can proceed via @xmath31-wave processes , the thermally averaged cross section @xmath76 is , to leading order , independent of @xmath77 : @xcite . this will slightly change the relic abundance @xcite , but we leave the detailed analysis for future work . ] @xmath78 using this , the relic density of the @xmath5 particles may be easily calculated ( see , for example ref . @xcite ) . as a rule of thumb , the dark matter drops out of thermal equilibrium when the rate @xmath79 of annihilation @xmath80 ( and the reverse process ) is outpaced by the expansion of the universe @xmath81 . using the boltzmann equation , the contribution of @xmath5 to the energy density of the universe can be more precisely calculated as @xmath82 here @xmath83 is the ratio @xmath84 where @xmath85 is the dark temperature at time of freeze - out and @xmath86 for @xmath31-wave processes . the quantity @xmath83 is given by @xmath87-\left(n+\frac{1}{2}\right)\ln\ln\left [ 0.038(n+1)\left(\frac{g}{\sqrt{g _ * } } \right ) m_{\rm pl } m_\chi \sigma_0 \right],\label{eq : xf}\ ] ] where @xmath88 is the number of degrees of freedom in the @xmath5 system ( namely 4 ) . as @xmath33 enters into the formula for @xmath83 only logarithmically , we may make the approximation that @xmath89 if @xmath5 freezes out while @xmath25 is above @xmath90 . we make the additional assumptions that the only degrees of freedom in addition to the sm are the @xmath17 and @xmath5 in the dark sector and that @xmath65 . we shall consider how these assumptions may be relaxed later . under these assumptions , the contribution of the dark sector to @xmath91 and @xmath34 is @xmath92 . as no dark degrees of freedom have frozen out yet , @xmath93 . with the measured value @xmath94 @xcite , we may solve for the allowed values of @xmath16 as a function of @xmath6 in eq . ( [ eq : omegadm ] ) . the resulting band is shown in fig . [ fig : alphamass ] . vs. @xmath6 parameter space . the relic abundance allowed region applies to models in which @xmath3 is the only force coupled to the dark matter ; in models where the dm is also weakly interacting , this provides only an upper limit on @xmath1 . the thin yellow line is the allowed region from correct relic abundance assuming @xmath95 , @xmath96 , @xmath97 , and @xmath98 while the surrounding blue region is @xmath99 , @xmath100 , and @xmath101 at the lower(upper ) edge . the diagonal green line is the upper limit on @xmath1 from effects of hard scattering on galactic dynamics ; in the red region , even soft scatterings do not appreciably affect the dm dynamics . we consider this to be the allowed region of parameter space . [ fig : alphamass ] ] in this discussion we have assumed that the process which sets the relic abundance of @xmath5 is annihilation into @xmath4s , as shown in figure [ fig : feynman ] . as we will argue in the next section ( and as is already shown in figure [ fig : alphamass ] ) , the values we obtain for @xmath1 from this calculation are incompatible with bounds from galactic dynamics unless @xmath102 gev ( at which point @xmath16 is non - perturbative ) . however , we can get the correct relic abundance even with much lower values of @xmath1 by adding other annihilation channels , such as the weak interactions , as explored in section [ sec : su2 ] . in that case , the `` relic abundance allowed region '' discussed here really becomes an upper limit ; if the dark fine - structure constant is larger than that value , annihilations are too efficient , and the correct abundance can not be obtained . we now consider how changing our assumptions on @xmath91 and @xmath74 can change our conclusions on the allowed parameter space . the parameter @xmath103 does not enter explicitly into the calculation for @xmath104 , however it does affect the number of active degrees of freedom at freeze - out directly , through eqs . ( [ eq : gstarxi ] ) and ( [ eq : gstarsxi ] ) , and indirectly by allowing the temperature @xmath25 to differ from @xmath27 . if @xmath105 , @xmath106 and there could be many more heavy visible degrees of freedom still active when @xmath5 freezes out . @xmath107 would reduce the visible degrees of freedom . however , as we have seen in eq . ( [ eq : bbnbound2 ] ) , it is difficult to construct a scenario with large @xmath74 , short of a massive increase in @xmath108 and small values of @xmath109 . we include in fig . [ fig : alphamass ] the bounds from both a large and small value of @xmath91 . the large limit is @xmath110 , ( _ i.e. _ equivalent to the mssm degrees of freedom ) , @xmath96 , and @xmath111 , while the small value is given by @xmath112 , ( _ i.e. _ equivalent to the sm degrees of freedom at @xmath90 ) , @xmath113 , and @xmath114 . although freezeout in our scenario is similar to that in the standard wimp scenario , the long - range dm - dm interactions implied by the unbroken @xmath3 may lead to considerably different dm phenomenology in the current universe , and in particular in galactic halos . in this scenario , dark - matter halos are composed of an equal mixture of @xmath5 and @xmath14 . the overall halo will be @xmath3 neutral , eliminating long - range forces that are incompatible with experiment . however , nearest - neighbor interactions between @xmath5 particles remain , and these interactions can be constrained by observations that suggest that dark matter is effectively collisionless . constraints to dark - matter self - interactions arise from evidence for nonspherical cores for some dark - matter halos ( collisions tend to make the cores of halos round ) @xcite and from evidence for dark - matter halos with large phase - space densities ( collisions would reduce phase - space densities ) @xcite . roughly speaking , a bound to dm - dm interactions can be derived by demanding that scattering induces no more than a small fractional change in the energy of a typical dm particle in a galactic halo during the history of the universe @xcite . this translates to an upper bound of @xmath115 on the more familiar quantity @xmath116 . ) , using the age of the universe for @xmath117 , and galactic parameters @xmath118 , @xmath119 . ] a separate bound of @xmath120 can be derived from the bullet cluster @xcite , but as this is less restrictive we ignore it here . to illustrate , we first consider hard scattering of a @xmath5 off another @xmath5 or @xmath14 , where energy on the order of @xmath121 is exchanged . the mean free time @xmath117 for a @xmath5 to undergo a hard scattering with another @xmath122 is given by @xmath123 where @xmath124 is the number density of dark matter , @xmath125 is the hard - scattering cross section , and @xmath77 is the velocity of the dark - matter particles . the number @xmath126 of dark - matter particles in the galaxy is @xmath127 and @xmath128 , where @xmath129 is the radius of the galaxy . the velocity @xmath77 is @xmath130 the dynamical time @xmath131 in the galaxy is @xmath132 taking @xmath133 years for the milky way , the average time for a hard scatter for a dark - matter particle is greater than the age of the universe if @xmath134 a hard scatter occurs when two particles pass close enough so that their kinetic energy is comparable to their potential energy . the impact parameter that defines a hard scatter is thus @xmath135 taking the cross section for hard scatters to be @xmath136 , and using eq . ( [ eq : v ] ) for @xmath77 , we find @xmath137 using @xmath138 we find the hard scattering limit on the @xmath3 coupling constant to be @xmath139 the allowed region arising from this bound is shown in fig . [ fig : alphamass ] . we now turn to the effect of soft scattering on the allowed values of @xmath16 and @xmath6 . here we consider the approach of one @xmath5 particle towards another @xmath122 at impact parameter @xmath140 . by definition , for soft scattering @xmath141 . the velocity change induced by the encounter is @xmath142 as one dark - matter particle orbits the galaxy , it sees a surface density @xmath143 of dark matter . the number of interactions that occur between an impact parameter @xmath140 and @xmath144 is @xmath145 . while the change in @xmath146 over these interactions should average to zero , this is not true for @xmath147 : @xmath148 integrating @xmath147 from @xmath149 to the maximum possible impact parameter in the galaxy , @xmath129 , gives the total change in @xmath150 as the particle orbits once through the halo : @xmath151 the number @xmath152 of orbits it will take for the dark - matter particle to have @xmath153 is @xmath154 the logarithmic suppression in eq . ( [ eq : tausoft ] ) relative to eq . ( [ eq : tauhard ] ) is due to the long - range coulomb force generated by the @xmath3 . as can be seen in fig . [ fig : alphamass ] , the allowed region from these considerations of galactic dynamics completely exclude the @xmath10 band that gives the correct relic abundance up to @xmath155 tev . for @xmath156 tev a dark matter candidate which freezes out due to @xmath3 interactions is ruled out from such considerations . in particular , models such as that in ref . @xcite with @xmath157 and a hidden copy of electromagnetism ( _ i.e. _ @xmath158 ) are ruled out , even though the freeze - out proceeds through hidden - sector weak interactions rather than a @xmath3 . interestingly , @xmath159 is allowed for @xmath160 tev . before considering whether such a model may be valid if our assumptions are loosened , we should ask why galactic dynamics do not similarly exclude wimp dark matter . after all , both models have similar cross sections for annihilations in the early universe ( eq . ( [ eq : sigma ] ) ) as is required for the correct relic density . though the soft scattering limit clearly will not apply due to the short range nature of the broken @xmath9 , naively it would seem that the hard scattering limit eq . ( [ eq : tauhard ] ) should apply to wimps equally well . however , notice that the threshold for hard scattering with a @xmath3 is dependent on energy . as the temperature drops , the cross section rises , as the @xmath5 particles no longer have to approach as close in order for to potential energy @xmath161 to be of the order of the kinetic energy . contrast this to hard scattering from wimps , where the cross section is always proportional to @xmath162 , regardless of the velocity . entering this cross section into eq . ( [ eq : tauconstraint ] ) , results in the uninteresting bound that @xmath163 tev for wimp dark matter from galactic dynamics constraints . it is difficult to see any way of avoiding the bounds from galactic dynamics , so we look to loosen the limits derived in section [ sec : earlyuniverse ] . clearly if the interaction responsible for freezing out the relic density is not the @xmath3 constrained by soft scattering , then @xmath164 is not ruled out . we consider such examples in the next section . however , we first consider the possibility that our assumptions in deriving the relic density are too conservative . from eq . ( [ eq : omegadm ] ) , if we reduce @xmath16 ( and therefore @xmath165 ) in order to satisfy the scattering bounds , we must either decrease @xmath83 or increase @xmath166 . in lowering @xmath16 by a minimum of two orders of magnitude , @xmath167 must likewise increase . as @xmath83 depends only logarithmically on @xmath16 and the number of degrees of freedom , so it is hard to see how it it could be increased sufficiently to counterbalance @xmath16 of order @xmath168 ( rather than @xmath169 ) . we conclude that the number of effective degrees of freedom must be increased . from eqs . ( [ eq : gstarxi ] ) and ( [ eq : gstarsxi ] ) , we see that if @xmath170 , then at freeze - out we must have @xmath171 from eq . ( [ eq : bbnbound2 ] ) , these @xmath172 degrees of freedom must exist in the visible sector at @xmath173 , rather than the dark sector . alternatively , we could imagine that there are no ( or few ) new particles beyond the minimum @xmath5 and @xmath17 at freeze - out , but instead @xmath174 . in this limit @xmath175 this limit is more troublesome ; from eq . ( [ eq : bbnbound2 ] ) we saw that large values of @xmath74 at the reheating scale ( and subsequently @xmath173 ) very quickly violate the bounds on relativistic degrees of freedom at bbn . clearly , by increasing the number of degrees of freedom in the visible sector , this bound could be avoided as well . however , we are left with the conclusion that either @xmath176 or there exist @xmath172 new particles at a few hundred gev to a tev . we leave it to the reader to decide how palatable these alternatives are . a separate , but conceptually similar , bound on scattering can be placed by considering the interaction of galactic dark matter with the hotter dm of the surrounding cluster . scattering will cause heating in galactic dm , and eventually evaporate the halo . from ref . @xcite the characteristic time for this evaporation is given by @xmath177 we may estimate the cross - section for soft - scattering by calling the path length @xmath178 over which a single particle loses of order its initial kinetic energy @xmath179 , where @xmath129 is the radius of the galaxy , and @xmath180 from eq . ( [ eq : totaldeltav2 ] ) is the fractional energy loss as the particle travels once through the halo . this can be expressed as an effective scattering cross section by setting @xmath181 , where @xmath182 is the number density of dm in the halo , we find @xmath183 letting the cluster velocity and density take on the canonical values ( @xmath184 and @xmath185 , where @xmath186 is the solar mass ) , we can place limits on @xmath1 and @xmath6 by requiring that @xmath187 is longer than the age of the universe . numerically , we find this bound less stringent than that from soft - scattering of particles within the galactic halo , eq . [ eq : tausoft ] . it is interesting to note that , aside from logarithmic enhancements , the bound placed on @xmath1 vs. @xmath6 parameter space from soft scattering is essentially a line of constant @xmath116 ( that is , they are , up to log corrections , lines of slope @xmath188 on the log - log plot ) . as mentioned , limiting dm to one hard scattering in the lifetime of the universe is equivalent to bounding @xmath116 in the galaxy to be @xmath189 . it has been suggested in the literature that values of @xmath116 in the range @xmath190 @xcite may provide better agreement between simulation and observation . therefore , our limit from soft - scattering should be considered as the general region at which interaction effects may become relevant . additionally , from eq . ( [ eq : sigmam ] ) as @xmath191 , it should be expected that the soft - scattering bound will vary greatly in dm systems with a range of virial velocities @xmath77 . in particular , we surmise that a bound even stronger than that estimated here can be obtained from the dwarf galaxies that exhibit the highest observed dark - matter phase - space densities @xcite . in this section , we examine an expanded version of our minimal model : one in which the @xmath5 dark matter particles possess @xmath9 quantum numbers in addition to a @xmath3 charge . for such @xmath192 particles , the cross section for freeze - out in the early universe is dominated by the weak interaction @xmath193 , and the @xmath3 contribution is negligible for the small values of @xmath16 under consideration . at late times the situation is reversed . the weak cross section remains small , as it is the result of a short - range force . however the long range cross section for soft scattering increases as the dark matter cools and slows , as exemplified in eq . ( [ eq : totaldeltav2 ] ) . this allows the strength of @xmath16 to be @xmath194 as required by galactic dynamics without running afoul of the relic density conditions , which would require @xmath169 ( when @xmath195 tev ) . we therefore take our dirac fermion @xmath5 to be a @xmath196 multiplet of @xmath197 , where we shall take the @xmath3 coupling to be in the region of fig . [ fig : alphamass ] allowed by soft scattering . thus @xmath164 . the behavior of this model in the early universe is very similar to the ` minimal dark model ' of ref . @xcite , from which we take many of our constraints . in outlining our original model in section [ sec : earlyuniverse ] , we set the coefficient of the mixing term @xmath198 to zero at the high scale . clearly loops involving @xmath5 would generate a non - zero mixing if the @xmath5 field possesses non - zero hypercharge @xmath199 . in order to avoid this complication , we set @xmath200 . our @xmath5 particle must be neutral under @xmath201 . with the assumption of @xmath200 , this requires @xmath5 to sit in an @xmath124-plet of @xmath9 where @xmath124 is odd ( _ i.e. _ @xmath202 ) . in the spirit of simplicity we take @xmath203 , so the @xmath5 triplet contains the neutral @xmath204 and ( electromagnetically ) charged @xmath205 , all with @xmath3 charges of @xmath13 . due to @xmath9 loops , the @xmath205 are @xmath206 mev heavier than the @xmath204 , and decay before bbn . if the dark matter mass is @xmath207 tev , the correct dark matter abundance ( including production and then decay of @xmath205 ) results from thermal freeze out ( see ref . we note that our model does have the nice feature of automatically suppressing unwanted decays of @xmath5 into sm particles , as by assumption @xmath5 is the lightest particle charged under @xmath3 . this minimal model is anomaly free . triangle diagrams with one or three @xmath9 vertexes vanish by the tracelessness of the @xmath9 generators . the diagrams consisting of an odd number of @xmath3 vertexes also vanish as the dark sector contains only two weyl fermions , one with @xmath13 under @xmath3 , and the other with @xmath208 . this model does not run afoul of bbn ( or cmb ) bounds . as in the pure @xmath3 theory , the only new relativistic degrees of freedom at bbn are the two from the @xmath17 . due to the interactions between @xmath5 and the weakly charged sm fields , we expect the temperatures @xmath25 and @xmath27 to track , so @xmath170 until the @xmath5 freeze - out . with small values of @xmath16 , the dark photons may freeze - out earlier , and would thus be colder . however , if we take the worse - case scenario that the dark photons do not decouple until after the @xmath5 undergo freeze - out we find ( from eq . ( [ eq : bbnbound2 ] ) ) that bbn bounds are satisfied as long as freeze - out occurs when @xmath209 this is easily satisfied for any model that freezes out before the qcd phase transition . next we must check that our @xmath5 does not have too large of a coupling to sm particles . we first demonstrate that no mixing occurs between the photon and the dark photon @xmath17 . as indicated previously , we assume that there is no @xmath198 term at high energies . with purely @xmath210 coupling , we find that the diagram fig . [ fig : ghatg]a vanishes . this is because any such vertex can be rewritten as the @xmath17 coupling to a @xmath5 or @xmath14 which then couples to the @xmath211 through some vertex involving sm fermions and @xmath9 couplings ( fig . [ fig : ghatg]b ) . however , since the mass and @xmath9 couplings of @xmath5 are the same as those of @xmath14 yet the @xmath3 charge is opposite , the sum of the two diagrams is zero . mixing . the vertex in a ) can be expanded into that shown in b ) , as the only particle to which the @xmath17 couples is @xmath73 . since the mass and @xmath9 charge of these two particles are the same , yet they possess opposite @xmath3 charge , the sum of the @xmath5 and @xmath14 diagrams in b ) is zero , and the overall mixing vanishes . [ fig : ghatg],title="fig : " ] mixing . the vertex in a ) can be expanded into that shown in b ) , as the only particle to which the @xmath17 couples is @xmath73 . since the mass and @xmath9 charge of these two particles are the same , yet they possess opposite @xmath3 charge , the sum of the @xmath5 and @xmath14 diagrams in b ) is zero , and the overall mixing vanishes . [ fig : ghatg],title="fig : " ] similarly , the coupling between @xmath17 and a standard model fermion @xmath212 is also zero . the relevant diagrams are shown in fig . [ fig : ffhatg ] . again , the vertex between @xmath212 and @xmath17 ( fig . [ fig : ffhatg]a ) can be divided into the @xmath73 vertex connecting with @xmath17 and a vertex between @xmath73 vertex connecting with @xmath212 ( fig . [ fig : ffhatg]b ) . as the latter vertex is identical for @xmath5 and @xmath14 but the former has opposite signs , the overall diagram vanishes . interactions with sm fermions @xmath212 . the vertex in a ) can be expanded into that shown in b ) , as the only particle with an interaction with @xmath17 is the @xmath73 . since the mass and @xmath9 charge of these two particles are the same , yet the @xmath3 charges are opposite , the sum of the @xmath5 and @xmath14 diagrams in b ) is zero , and the overall coupling of @xmath212 to @xmath17 is therefore zero as well . [ fig : ffhatg],title="fig : " ] interactions with sm fermions @xmath212 . the vertex in a ) can be expanded into that shown in b ) , as the only particle with an interaction with @xmath17 is the @xmath73 . since the mass and @xmath9 charge of these two particles are the same , yet the @xmath3 charges are opposite , the sum of the @xmath5 and @xmath14 diagrams in b ) is zero , and the overall coupling of @xmath212 to @xmath17 is therefore zero as well . [ fig : ffhatg],title="fig : " ] the lowest order coupling of sm fermions to @xmath17 occurs at @xmath213 . this is due to a two loop effect , as shown in fig . [ fig : ffgg ] , and unlikely to be accessible in direct detection . we can represent this interaction by an effective lagrangian whose lowest order term is given by @xmath214 where @xmath215 and @xmath216 is the yukawa coupling of the fermion that is involved . let us estimate the order of magnitude of this interaction . to be conservative we use the yukawa coupling of a @xmath217 quark and take @xmath218 ; which by galactic dynamics is the maximum allowed value for @xmath219 tev . with these values we find @xmath220 and @xmath221 . we estimate that the interaction length for dark photons inside the cores of stars would be on the order of @xmath222 km , and thus this interaction would not introduce a potentially dangerous new source of stellar cooling . couple to a loop of @xmath5 particles , which couple through two @xmath9 gauge bosons to sm fermions . coupling through a single @xmath9 boson is zero due to the tracelessness of @xmath223 . [ fig : ffgg ] ] due to the high - order interaction between @xmath17 and sm particles , we can not expect to directly observe the dark radiation . in addition , while the @xmath5 fields would have a direct detection cross section of @xmath224 @xcite and so could be seen in supercdms , any such detection would be indistinguishable from a scenario without the dark photons . therefore , the presence of a new unbroken @xmath3 in the dark sector could only be probed via its effect on galactic dynamics . clearly in the limit that @xmath225 , the galactic structure would remain unchanged . values of @xmath16 near the maximum allowed from soft - scattering ( _ i.e. _ @xmath169 for the @xmath9 triplet candidate with @xmath226 tev ) should have a measurable effect on the halo structure , as in this regime the dark matter is no longer completely collisionless . a full study of this effect requires simulations beyond the scope of this paper , though some additional considerations are discussed in the following section . the existence of a dark matter ` plasma ' may have additional effects that could significantly affect structure formation . we mention three possibilities here : bremsstrahlung , early universe structure formation , and the weibel instability in galactic halos . the first two result in much weaker bounds than those already derived , and are mentioned here only for completeness . the weibel instability may have significant and visible effects in the halo , but requires simulation beyond the scope of this paper . emission of a soft @xmath17 during a @xmath73 collision could conceivably serve as another energy loss mechanism in the halo on par with soft and hard scattering as outlined in section [ sec : galacticdynamics ] . to derive a bound on @xmath16 as related to @xmath6 , we make the same assumption as in the case of soft scatter : over the lifetime of the universe , a dark - matter particle can not lose on order of its initial kinetic energy through bremsstrahlung of dark radiation . by assuming dipole radiation during a soft collision , we find that @xmath227 however this bound is weaker than that from both hard and soft scattering over the parameter space of interest . in the early universe , structure can not grow until after matter / radiation equality . until the matter ( which can clump ) decouples from the dark radiation ( which can not ) , density perturbations remain fixed . we can estimate the scale factor at which this occurs by finding the redshift @xmath228 at which the dissipation time ( the time over which the velocity of a dark matter particle is significantly perturbed by the radiation ) becomes longer than the hubble time @xmath229 . the argument follows that in ref . @xcite for the decoupling of baryons from the photon bath . the dissipation time is the logarithmic derivative of the velocity : @xmath230 here @xmath231 is the force due to radiation pressure , @xmath232 where @xmath233 is the thomson cross - section for dark matter interacting with dark photons and ( as before ) @xmath27 is the temperature of the dark photons . as we shall see , the decoupling occurs when the universe is radiation dominated , so the hubble time is given by @xmath234 here @xmath25 is the photon temperature . the conservation of entropy relates the photon temperature @xmath25 at redshift @xmath228 with the photon temperature today , @xmath235 , @xmath236 combining eqs . ( [ eq : tdiss ] ) and ( [ eq : thubble ] ) , we find the decoupling redshift @xmath228 to be @xmath237 as before @xmath74 is the ratio of dark photon temperature to photon temperature at redshift @xmath228 ( recall that it is difficult to construct models where @xmath74 is much larger than unity ) . the number of degrees of freedom that contribute to the entropy density today , @xmath238 , is of order unity . the decoupling occurs extremely early , before even dark matter freeze - out . as a result , it seems that this effect will be cosmologically irrelevant . in section [ sec : galacticdynamics ] , we constrained @xmath1 by demanding that dark matter be effectively collisionless in galactic halos , under two - body interactions . however , there may be collective plasma effects that affect dm dynamics on timescales much shorter than those due to two - body interactions . unfortunately , it is difficult to state with confidence what the observational consequences of those effects will actually be , even if they are relevant . given theoretical uncertainties about the nonlinear evolution of such instabilities , we leave the detailed implications to future work . as a simple example we consider the weibel instability @xcite , an exponential magnetic - field amplification that arises if the plasma particles have an anisotropic velocity distribution . such anisotropies could arise , for example , during hierarchical structure formation as subhalos merge to form more massive halos . similar instabilities in the baryonic gas have been postulated to account for the magnetic fields in galaxy clusters @xcite . the growth rate @xmath79 of the magnetic field is @xmath239 where @xmath240 is the plasma frequency , @xmath241 is the dark - matter density , and @xmath77 is the velocity of the dark matter within the colliding halos . assuming @xmath242 , we find @xmath243 to be relevant for galactic - halo formation , the timescale @xmath244 for magnetic - field amplification should be shorter than the dynamical timescale @xmath117 of the merging subhalos . the instability will be therefore be of interest when @xmath245 this range of @xmath16 and @xmath6 encompasses the entire parameter space of interest for any reasonable value of @xmath117 . therefore , we suspect that galactic structure will be affected by plasma effects in the dark matter due to the @xmath3 even when @xmath16 is not near the boundary of allowed values from soft scattering . one possibility is that nonlinear evolution would result in a strongly magnetized plasma , and if so , dark matter would be effectively collisional and thus probably inconsistent with data . however , theory and simulations that study the nonlinear evolution of the weibel instability for relativistic pair plasmas and nonrelativistic electron - proton plasmas do not yet agree whether the magnetic fields survive , and simulations for the equal - mass nonrelativistic plasma we are considering have not been performed . it is therefore premature to conclude that these instabilities will result in effectively collisional dark matter ; a more detailed study will be required to assess these effects . given how little direct information we have about the nature of dark matter , it is of crucial importance to explore models in which the dm sector has an interesting phenomenology of its own . in many ways , an unbroken @xmath0 gauge field coupled to dark matter is a natural way to obtain a long - range interaction between dm particles . in contrast to the case of hypothetical long - range scalar fields , the masslessness of the gauge field is protected by a symmetry , and the absence of long - range violations of the equivalence principle is naturally explained by the overall charge neutrality of the dark plasma . new unbroken @xmath0 s can appear naturally in unified models . while a dark @xmath0 may be realized as a broken symmetry with massive vector bosons , it has been pointed out that there are few constraints on the massless , unbroken case from the early universe . we have verified that the minimal model , with just a single massive dirac fermion for the dark matter and a massless dark photon , is consistent with limits obtained from the number of relativistic degrees of freedom at bbn , with relatively mild assumptions on the reheating temperature of the dark sector . more complicated models are also allowed , depending on the details of spectrum and reheating . we found that one can not build a dark matter model charged under a hidden unbroken @xmath3 in which this new gauge group is responsible for thermal freeze out . as can be seen in fig . [ fig : alphamass ] , the required values of @xmath16 and @xmath6 required for the @xmath5 particles to form a thermal relic would violate bounds coming from limits on hard and soft scattering of dark matter in the galactic halo . as an important consequence of this argument , models in which dark matter couples to an exact copy of ordinary electromagnetism ( in particular , with @xmath159 ) are ruled out unless @xmath246 a few tev . this constrains the parameter space of models with hidden copies of the sm or the mssm in which the dark matter is electrically charged , such as the model in ref . @xcite where the stau was suggested as a dark matter candidate . by adding additional interactions to increase the annihilation cross - section , it is possible to build a scenario with an unbroken dark @xmath0 and the correct relic abundance . introducing another short - range force coupling to the @xmath5 , for example the familiar @xmath9 , can provide an appropriately large cross section for @xmath247 annihilation . the new coupling @xmath16 must then be relatively small ( compared to the @xmath9 @xmath248 ) in order to evade galactic dynamics bounds . the simplest model which realizes this situation is a dirac fermion in a triplet of @xmath9 ( in order to avoid @xmath249 mixing ) . bounds from the early universe then force @xmath6 to be on the order of a few tev , which implies @xmath250 . since all couplings between the dark radiation and the sm enter at two loops ( and require two dark photons in the process ) , it would be very difficult to observe the presence of the new gauge group through direct detection . instead , the best search strategy would be an indirect one : looking for the effects on galactic dynamics arising from a soft scattering mediated by a long - range force . clearly , as @xmath16 goes to zero , the model becomes indistinguishable from minimal weakly coupled dark matter . however , if the coupling is near the limit from soft scattering , one would expect detectable deviations from the assumptions of collisionless dark matter currently used in simulations . additionally , since the @xmath3 effectively makes the dark halo a plasma ( albeit a very cold , tenuous one ) , there may be other effects on structure formation that constrain this model @xcite . we have estimated that the timescale for the weibel instability in our model is short compared to relevant timescales for galactic dynamics . if this instability has a dramatic effect when subhalos collide during the assembly of a galactic halo , our @xmath3 could be excluded for the entire range of interesting parameters . further work is required to before we reliably understand the quantitative effects of such instabilities on galactic dynamics . this work opens a window to new phenomenological possibilities within the dark sector . one avenue for further investigation would be the possibility of `` dark atoms , '' which would arise if there were two different stable species with dark charge , each with an asymmetry in the number density of positive and negative charges ( with one balancing the other to maintain overall charge neutrality ) . from there , one is free to contemplate dark chemistry and beyond . dark matter constitutes a large majority of the matter density of the universe , and there is no reason to assume _ a priori _ that physics there is any less rich and interesting than that of ordinary matter . we are very grateful to sonny mantry for collaboration in the early stages of this work , and to jonathan feng , george field , josh frieman , manoj kaplinghat , keith lee , arvind rajaraman , and mark wise for helpful comments . this work was supported by doe de - fg03 - 92-er40701 and the gordon and betty moore foundation . c. amsler _ et al . _ [ particle data group ] , phys . lett . b * 667 * , 1 ( 2008 ) . g. jungman , m. kamionkowski and k. griest , phys . rept . * 267 * , 195 ( 1996 ) [ arxiv : hep - ph/9506380 ] . l. bergstrom , rept . prog . phys . * 63 * , 793 ( 2000 ) [ arxiv : hep - ph/0002126 ] . g. bertone , d. hooper and j. silk , phys . * 405 * , 279 ( 2005 ) [ arxiv : hep - ph/0404175 ] . 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 . * 89 * , 211301 ( 2002 ) [ arxiv : hep - ph/0207125 ] . d. hooper and s. profumo , phys . * 453 * , 29 ( 2007 ) [ arxiv : hep - ph/0701197 ] . d. n. spergel and p. j. steinhardt , phys . * 84 * , 3760 ( 2000 ) [ arxiv : astro - ph/9909386 ] ; m. kaplinghat , l. knox and m. s. turner , phys . lett . * 85 * , 3335 ( 2000 ) [ arxiv : astro - ph/0005210 ] ; a. tasitsiomi , int . j. mod . d * 12 * , 1157 ( 2003 ) [ arxiv : astro - ph/0205464 ] . b. d. wandelt , r. dave , g. r. farrar , p. c. mcguire , d. n. spergel and p. j. steinhardt , arxiv : astro - ph/0006344 . j. a. frieman and b. a. gradwohl , phys . lett . * 67 * , 2926 ( 1991 ) . b. a. gradwohl and j. a. frieman , astrophys . j. * 398 * , 407 ( 1992 ) ; g. w. anderson and s. m. carroll , arxiv : astro - ph/9711288 ; s. m. carroll , phys . lett . * 81 * , 3067 ( 1998 ) [ arxiv : astro - ph/9806099 ] ; g. r. farrar and p. j. e. peebles , astrophys . j. * 604 * , 1 ( 2004 ) [ arxiv : astro - ph/0307316 ] ; s. s. gubser and p. j. e. peebles , phys . d * 70 * , 123511 ( 2004 ) [ arxiv : hep - th/0407097 ] ; s. s. gubser and p. j. e. peebles , phys . d * 70 * , 123510 ( 2004 ) [ arxiv : hep - th/0402225 ] ; o. bertolami and j. paramos , phys . d * 71 * , 023521 ( 2005 ) [ arxiv : astro - ph/0408216 ] ; a. nusser , s. s. gubser and p. j. e. peebles , phys . d * 71 * , 083505 ( 2005 ) [ arxiv : astro - ph/0412586 ] ; r. bean , e. e. flanagan and m. trodden , phys . d * 78 * , 023009 ( 2008 ) [ arxiv:0709.1128 [ astro - ph ] ] ; m. kesden and m. kamionkowski , phys . d * 74 * , 083007 ( 2006 ) [ arxiv : astro - ph/0608095 ] ; m. kesden and m. kamionkowski , phys . lett . * 97 * , 131303 ( 2006 ) [ arxiv : astro - ph/0606566 ] ; j. bovy and g. r. farrar , arxiv:0807.3060 [ hep - ph ] ; s. m. carroll , s. mantry , m. j. ramsey - musolf and c. w. stubbs , arxiv:0807.4363 [ hep - ph ] . d. hooper and k. m. zurek , phys . d * 77 * , 087302 ( 2008 ) [ arxiv:0801.3686 [ hep - ph ] ] . a. de rujula , s. l. glashow and u. sarid , nucl . b * 333 * , 173 ( 1990 ) . b. holdom , phys . b * 166 * , 196 ( 1986 ) . s. davidson , s. hannestad and g. raffelt , jhep * 0005 * , 003 ( 2000 ) [ arxiv : hep - ph/0001179 ] . s. dimopoulos , d. eichler , r. esmailzadeh and g. d. starkman , phys . d * 41 * , 2388 ( 1990 ) . l. chuzhoy and e. w. kolb , arxiv:0809.0436 [ astro - ph ] . s. s. gubser and p. j. e. peebles , phys . d * 70 * , 123510 ( 2004 ) [ arxiv : hep - th/0402225 ] . j. l. feng and j. kumar , arxiv:0803.4196 [ hep - ph ] . j. l. feng , h. tu and h. b. yu , arxiv:0808.2318 [ hep - ph ] . b. a. dobrescu , phys . * 94 * , 151802 ( 2005 ) [ arxiv : hep - ph/0411004 ] . m. pospelov , a. ritz and m. b. voloshin , phys . b * 662 * , 53 ( 2008 ) [ arxiv:0711.4866 [ hep - ph ] ] . d. v. ahluwalia , c. y. lee , d. schritt and t. f. watson , arxiv:0712.4190 [ hep - ph ] . e. kolb and m.s . turner , `` the early universe , '' addison - wesley publishing company , redwood city , ( 1990 ) l. b. okun , sov . jetp * 56 * , 502 ( 1982 ) [ zh . teor . fiz . * 83 * , 892 ( 1982 ) ] . r. h. cyburt , b. d. fields , k. a. olive and e. skillman , astropart . * 23 * , 313 ( 2005 ) [ arxiv : astro - ph/0408033 ] . t. l. smith , e. pierpaoli and m. kamionkowski , phys . lett . * 97 * , 021301 ( 2006 ) [ arxiv : astro - ph/0603144 ] . j. hisano , s. matsumoto , m. m. nojiri and o. saito , phys . d * 71 * , 063528 ( 2005 ) [ arxiv : hep - ph/0412403 ] ; m. cirelli , a. strumia and m. tamburini , nucl . b * 787 * , 152 ( 2007 ) [ arxiv:0706.4071 [ hep - ph ] ] . m. cirelli , m. kadastik , m. raidal and a. strumia , arxiv:0809.2409 [ hep - ph ] . n. arkani - hamed , d. p. finkbeiner , t. slatyer and n. weiner , arxiv:0810.0713 [ hep - ph ] . m. pospelov and a. ritz , arxiv:0810.1502 [ hep - ph ] ; m. kamionkowski and s. profumo , arxiv:0810.3233 [ astro - ph ] . j. miralda - escude , arxiv : astro - ph/0002050 . r. dave , d. n. spergel , p. j. steinhardt and b. d. wandelt , astrophys . j. * 547 * , 574 ( 2001 ) [ arxiv : astro - ph/0006218 ] . n. yoshida , v. springel , s. d. m. white and g. tormen , arxiv : astro - ph/0006134 . d. clowe , m. bradac , a. h. gonzalez , m. markevitch , s. w. randall , c. jones and d. zaritsky , astrophys . j. * 648 * , l109 ( 2006 ) [ arxiv : astro - ph/0608407 ] . s. w. randall , m. markevitch , d. clowe , a. h. gonzalez and m. bradac , arxiv:0704.0261 [ astro - ph ] . o. y. gnedin and j. p. ostriker , arxiv : astro - ph/0010436 . j. j. dalcanton and c. j. hogan , astrophys . j. * 561 * , 35 ( 2001 ) [ arxiv : astro - ph/0004381 ] ; c. j. hogan and j. j. dalcanton , phys . rev . d * 62 * , 063511 ( 2000 ) [ arxiv : astro - ph/0002330 ] . m. cirelli , n. fornengo and a. strumia , nucl . b * 753 * , 178 ( 2006 ) [ arxiv : hep - ph/0512090 ] . j. e. peebles , `` principles of physical cosmology , '' _ princeton , usa : univ . pr . ( 1993 ) 718 p _ m. v. medvedev , l. o. silva and m. kamionkowski , astrophys . j. * 642 * , l1 ( 2006 ) [ arxiv : astro - ph/0512079 ] . j. t. frederiksen , c. b. hededal , t. haugboelle and a. nordlund , astrophys . j. * 608 * , l13 ( 2004 ) [ arxiv : astro - ph/0308104 ] ; k. i. nishikawa , p. hardee , g. richardson , r. preece , h. sol and g. j. fishman , astrophys . j. * 595 * , 555 ( 2003 ) [ arxiv : astro - ph/0305091 ] ; l. o. silva , r. a. fonseca , j. tonge , j. m. dawson , w. b. mori and m. v. medvedev , astrophys . j. * 596 * , l121 ( 2003 ) [ arxiv : astro - ph/0307500 ] ; m. milosavljevic , e. nakar and a. spitkovsky , astrophys . j. * 637 * , 765 ( 2006 ) [ arxiv : astro - ph/0507553 ] . j. march - russell , s. m. west , d. cumberbatch and d. hooper , jhep * 0807 * , 058 ( 2008 ) [ arxiv:0801.3440 [ hep - ph ] ] .

we explore a nonlinear field model to describe the interplay between the ability of excitons to be bose - condensed and their interaction with other modes of a crystal . we apply our consideration to the long - living para - excitons in cuo . taking into account the exciton - phonon interaction and introducing a coherent phonon part of the moving condensate , we derive the dynamic equations for the exciton - phonon condensate . these equations can support localized solutions , and we discuss the conditions for the moving inhomogeneous condensate to appear in the crystal . we calculate the condensate wave function and energy , and a collective excitation spectrum in the semiclassical approximation ; the inside - excitations were found to follow the asymptotic behavior of the macroscopic wave function exactly . the stability conditions of the moving condensate are analyzed by use of landau arguments , and landau critical parameters appear in the theory . finally , we apply our model to describe the recently observed interference and strong nonlinear interaction between two coherent exciton - phonon packets in cuo . -10 mm -0.5 cm * bosons in a lattice : * * exciton - phonon condensate in cuo * d. roubtsov and y. lpine _ groupe de recherche en physique et technologie des couches minces , _ _ dpartement de physique , universit de montral , _ c.p . 6128 , succ . centre - ville , montreal , pq , h3c3j7 , canada e - mail : [email protected] pacs numbers : 71.35.+z , 71.35.lk

cond-mat9910079

excitons in semiconductor crystals @xcite and nanostructures @xcite,@xcite are a very interesting and challenging object to search for the process of bose - einstein condensation ( bec ) . nowadays there is a lot of experimental evidence that the optically inactive para - excitons in cu@xmath0o can form a highly correlated state , or the excitonic bose einstein condensate @xcite,@xcite,@xcite . a moving condensate of para - excitons in a 3d cu@xmath0o crystal turns out to be spatially inhomogeneous in the direction of motion , and the registered velocities of coherent exciton packets turn out to be always less , but approximately equal to the longitudinal sound speed of the crystal @xcite . analyzing recent experimental @xcite,@xcite,@xcite and theoretical @xcite-@xcite studies of bec of excitons in cu@xmath1o , we can conclude that there are essentially two different stages of this process . the first stage is the kinetic one , with the characteristic time scale of @xmath2ns . at this stage , a condensate of long - living para - excitons begins to be formed from a quasi - equilibrium degenerate state of excitons ( @xmath3 , @xmath4 ) when the concentration and the effective temperature of excitons in a cloud meet the conditions of bose - einstein condensation @xcite . note that we do not discuss here the behavior of ortho - excitons ( with the lifetime @xmath5ns ) and their influence on the para - exciton condensation process . for more details about the ortho - excitons in cu@xmath1o , ortho - para - exciton conversion , etc . see @xcite,@xcite , @xcite , @xcite . the most intriguing feature of the kinetic stage is that formation of the para - exciton condensate and the process of momentum transfer to the para - exciton cloud are happening simultaneously . if the diameter of an excitation spot on the crystal surface is large enough , @xmath6 , and the energy of a laser beam satisfies @xmath7 , nonequilibrium acoustic phonons may play the key role in the process of momentum transfer . as a result , the mode with macroscopical occupancy of the excitons appears to be with @xmath8 where @xmath9 and @xmath10 is the packet velocity . indeed , the theoretical results obtained in the framework of the `` phonon wind '' model @xcite,@xcite and the experimental observations @xcite,@xcite,@xcite are the strong arguments in favor of this idea . to the authors knowledge , there are no realistic theoretical models of the kinetic stage of para - exciton condensate formation where quantum degeneracy of the appearing exciton state and possible coherence of nonequilibrium phonons pushing the excitons would be taken into account . indeed , the condensate formation and many other processes involving it are essentially nonliner ones . therefore , the condensate , or , better , the _ macroscopically occupied mode _ , can be different from @xmath11 , and the language of the states in @xmath12-space and their occupation numbers @xmath13 may be not relevant to the problem , see @xcite . in this study , we will not explore the stage of condensate formation . instead , we investigate the second , quasi - equilibrium stage , in which the condensate has already been formed and it moves through a crystal with some constant velocity and characteristic shape of the density profile . in theory , the time scale of this `` transport '' stage , @xmath14 , could be determined by the para - exciton lifetime ( @xmath15s @xcite ) . in practice , it is determined by the characteristic size @xmath16 of a high - quality single crystal available for experiments : @xmath17 where @xmath18 is the longitudinal sound velocity . we assume that at the `` transport '' stage , the temperature of the moving packet ( condensed@xmath19noncondensed particles ) is approximately equal to the lattice temperature , @xmath20 then we can consider the simplest case of @xmath21 and disregard the influence of all sorts of _ nonequilibrium _ phonons ( which appear at the stages of exciton formation , thermalization @xcite ) on the formed moving condensate . any theory of the exciton bec in cu@xmath0o has to point out some physical mechanism(s ) by means of which the key experimental facts can be explained . ( for example , the condensate moves without friction within a narrow interval of velocities localized near @xmath18 , and the shape of the stable macroscopic wave function of excitons resembles soliton profiles @xcite . ) here we explore a simple model of the ballistic exciton - phonon condensate . in this case , the general structure of the hamiltonian of the moving exciton packet and the lattice phonons is the following : @xmath22 here @xmath23 is the bose - field operator describing the excitons , @xmath24 is the field operator of lattice displacements , @xmath25 is the momentum density operator canonically conjugate to @xmath24 , and @xmath26 is the momentum operator . note that the hamiltonian ( [ ham1 ] ) is written in the reference frame moving with the exciton packet , i.e. @xmath27 and @xmath28 is the ballistic velocity of the packet . to derive the equations of motion of the field operators ( and generalize these equations to the case of @xmath29 ) , it is more convenient to start from the lagrangian . in the proposed model , the lagrangian density has the following form in the co - moving frame @xmath30 @xmath31 - _ g^ - ^2 ^ - _ 0 ( ^(*x*,t))^2 ( ( * x*,t))^2 - _ 1(^ ( * x*,t))^3((*x*,t))^3 + @xmath32 + ( _ t)^2 -c_l^2 ( _ j_s)^2 -c_l^2_3 ( _ j_s)^3 -(_t_x + _ x _ t ) + v^2(_x)^2- @xmath31 @xmath33 where @xmath34 is the exciton `` bare''mass ( @xmath35 for @xmath36s excitons in cu@xmath0o ) , @xmath37 is the exciton - exciton interaction constant ( @xmath38 corresponds to the repulsive interaction between para - excitons in cu@xmath0o @xcite ) , @xmath39 is the crystal density , @xmath40 is the exciton - longitudinal phonon coupling constant , and @xmath41 . the energy of a free exciton is @xmath42 . although the validity of the condition @xmath43 ( @xmath44 is the exciton bohr radius ) makes it possible to disregard all the multiple - particle interactions with more than two participating particles in @xmath45 @xcite , we include the hard - core repulsion term originated from the three - particle interaction in @xmath46 , i.e. @xmath47 and @xmath48 ( for 3d case , one has to take @xmath49 because @xmath50 and @xmath51 ; see the discussion in @xcite . ) moreover , in the lagrangian of the displacement field , we include the first nonlinear term @xmath52 . ( the dimensionless parameter @xmath53 originates from taylor s expansion of an interparticle potential @xmath54 of the medium atoms . ) assuming that a dilute excitonic packet moves in a weekly nonlinear medium , we will not take into account more higher nonlinear terms in ( [ lagrangian ] ) . for simplicity s sake , we take all the interaction terms in @xmath55 in the _ local _ form and disregard the interaction between the excitons and transverse phonons of the crystal . note that the ballistic velocity @xmath10 is one of the parameters of the theory , and we will not take into account the excitonic normal component and velocity , i.e. @xmath56 , @xmath57 . this means that we choose the spatial part of the coherent phase of the packet , @xmath58 , to be in the simplest form , @xmath59 the equations of motion can be easily derived by the standard variational method from the following condition : @xmath60 indeed , after transforming the bose - fields @xmath61 and @xmath23 by @xmath62 we can write these equations as follows : @xmath63 @xmath64 @xmath65 we assume that the condensate of excitons _ exists_. this means that the following representation of the exciton bose - field holds : @xmath66 . here @xmath67 is the classical part of the field operator @xmath23 or , in other words , the condensate wave function , and @xmath68 is the fluctuational part of @xmath23 , which describes out - of - condensate particles . one of the important objects in the theory of bec is the correlation functions of bose - fields . the standard way to calculate them in this model ( the excitonic function @xmath69 , for example , ) can be based on the effective action or the effective hamiltonian approaches @xcite . indeed , one can , first , integrate over the phonon variables @xmath70 , get the expression for @xmath71 and , second , use @xmath72 ( or @xmath73 ) to derive the equations of motion for @xmath74 , @xmath68 , correlation functions , etc .. in this work we do not follow that way ; instead , we treat excitons and phonons equally @xcite,@xcite,@xcite . this means that the displacement field @xmath24 can have a nontrivial coherent part too , i.e. @xmath75 and @xmath76 , and the actual moving condensate can be an exciton - phonon one , i.e. @xmath77 . then the equation of motion for the classical parts of the fields @xmath23 and @xmath24 can be derived by use of the variational method from @xmath78 , in which all the fields can be considered as the classical ones . eventually we have @xmath79 @xmath80 @xmath81 notice that deriving these equations we disregarded the interaction between the classical ( condensate ) and the fluctuational ( noncondensate ) parts of the fields . that is certainly a good approximation for @xmath21 and @xmath82 cases @xcite . in this article , a steady - state of the condensate is the object of the main interest . in the co - moving frame of reference , the condensate steady - state is just the stationary solution of eqs . ( [ eq11]),([eq22 ] ) and it can be taken in the form @xmath83 where @xmath84 and @xmath85 are the real number functions , and @xmath86 is the coherent phase of the condensate wave function in the co - moving frame , see eq . ( [ superf ] ) . ( this phase can be taken equal zero if only a single condensate is the subject of interest . ) then , the following equations have to be solved , ( @xmath87 ) : @xmath88 @xmath89 note that in order to simplify eq . ( [ eq22 ] ) to eq . ( [ eq222 ] ) , we assumed only @xmath90 . in this model , it is enough to obtain localized solutions for the displacement field . solving eqs . ( [ eq111]),([eq222 ] ) in 3d space seems to be a difficult problem . however , these equations can be essentially simplified if we assume that the condensate is inhomogeneous along the @xmath91-axis only , that is @xmath92 note that the cross - section area of an excitation spot , @xmath93 , has to be basically constant across the sample cross - section . in this case , the problem can be considered as an effectively one - dimensional one . such an effective reduction of dimensionality transforms difficult ( nonlocal differential ) equations for the condensate wave function into a rather simple differential ones , and obtained in this way the effective 1d model for the condensate wave function @xmath94 conserves all the important properties of the `` parent '' 3d model . indeed , if @xmath95 , the following equations stand for the condensate ( @xmath96 ) : @xmath97 @xmath98 the last equation can be easily integrated , @xmath99 and solved relative to @xmath100 . here @xmath101 note that the medium nonlinearity parameter @xmath53 can be enhanced by the factor of the order of @xmath102 if the value of @xmath10 is less , but close to @xmath18 . ( for spatially localized solutions , @xmath103 and @xmath104 at @xmath105 , so that @xmath106 . ) if @xmath107 , we can always represent the solution of ( [ 11deq]),([11deqd ] ) in the following form @xmath108 ( indeed , the parameters of medium nonlinearity can be chosen as @xmath107 and @xmath109 @xcite . ) after substitution of ( [ dc ] ) into eq . ( [ 1deq ] ) , eqs . ( [ 1deq ] ) , ( [ 11deq ] ) can be rewritten as follows : @xmath110 @xmath111 where the interparticle interaction constants are renormalized as follows @xmath112 and more higher nonlinear terms are designated by @xmath113 . a small parameter in eq . ( [ kappa ] ) comes from the term @xmath114 where @xmath115 and @xmath116 is the mass of the crystal elementary cell . the effective two - particle interaction constant @xmath117 can be * negative * if the velocity of the condensate lies inside the interval @xmath118 , where @xmath119 outside this interval , @xmath120 @xcite and the velocity @xmath121 can be called the first ` critical ' velocity in the model . the meaning of this velocity can be clarified by rewriting ( [ kappa ] ) in the dimensionless form , @xmath122 if @xmath123 , @xmath124 where @xmath125 and @xmath126 are the characteristic energy and the cross - section of two - particle collisions in the exciton subsystem . the following inequalities are true for excitons in a crystal @xmath127 and , usually , the value of the parameter @xmath128 is @xmath129 . for para - excitons in cu@xmath00 , however , we assume the ( effective ) value of @xmath130 can be estimated as @xmath131 , whereas the value of @xmath132 . this makes the inequality ( [ negative ] ) valid at , say , @xmath133 , or @xmath134 . thus , within the effective 1d model , the critical factor @xmath135 is the following ratio @xmath136 and , for the substances with @xmath137 , the regime with @xmath138 can be obtained at velocities _ reasonably close _ but not equal to @xmath18 , for example , beginning from some @xmath139 , ( @xmath140 ) . on the other hand , the effective three - particle interaction constant @xmath141 is always positive for crystals with @xmath142 . it can be represented in the dimensionless form as follows @xmath143 here we estimated the `` bare '' vertex of the three - particle collisions as @xmath144 and the same @xmath145 can be taken as a characteristic energy of collisions . the effective vertex @xmath146 is enhanced by the medium nonlinearity , and the both terms in the r.h.s . of ( [ three - part ] ) can be equally important at @xmath147 . note that in the case of _ strongly nonlinear lattices _ with excitons , the effective interaction vertices in ( [ 1deqmod ] ) ( @xmath148 , @xmath149 , etc . ) depend on the velocity @xmath10 and the parameters of medium nonlinearity ( @xmath53 , @xmath150 , etc . ) . then the effective exciton - exciton interaction can be strongly renormalized at sufficiently large gamma - factors @xmath151 and the vertices may change their signs as it can happen with @xmath117 . in this article , however , we consider the case of _ weakly nonlinear medium _ with excitons ( e.g. , a crystal with long living excitons ) . more accurately , this means that at velocities @xmath152 the effective vertex @xmath117 became @xmath153 , while the more higher vertices , such as @xmath154 and @xmath155 , do not change their sign ; they remain @xmath156 at @xmath157 . finally , to describe the weakly nonlinear case , it is enough to take into account the parameters @xmath158 and @xmath107 and neglect more higher nonlinearities ( @xmath113 and @xmath159 in ( [ 1deqmod ] ) , ( [ dcnl ] ) ) . in this study , we will consider the case of @xmath118 in detail . indeed , in the case of @xmath160 and @xmath161 , some localized solutions of eqs . ( [ 1deqmod]),([dcnl ] ) do exist . for example , the so - called ` bright soliton ' solution of ( [ 1deqmod ] ) exists if the generalized chemical potential is negative , @xmath162 , and @xmath163 . here @xmath164 for @xmath165 and @xmath166 , we can roughly estimate the effective vertex @xmath154 as @xmath167 then @xmath168 , and the more is the value of @xmath169 the less is the value of @xmath170 . the ` bright soliton ' solution of eq . ( [ 1deqmod ] ) can be represented in the following form @xmath171 here @xmath172 is some dimensionless parameter , and the generalized chemical potential @xmath173 is given by the formula @xmath174 like the chemical potential @xmath175 , the amplitude of the bright soliton , @xmath176 , satisfies @xmath177 and @xmath178 . for @xmath179 , the following approximation is valid @xmath180 and this formula can be used up to @xmath181 . then we can _ approximate _ the solution of ( [ 1deqmod ] ) by the following formulas @xmath182 @xmath183 the amplitudes of the exciton and phonon parts of the condensate , the characteristic width of the condensate , and the value of the effective chemical potential @xmath184 depend on the normalization of the exciton wave function @xmath185 . we normalize it in 3d space assuming that the characteristic width of the packet in the @xmath186-plane is sufficiently large , i.e. the cross - section area of the packet @xmath187 can be made equal to the cross - section area s of a laser beam and @xmath188 then we can write this condition as follows : @xmath189 where @xmath190 is the number of condensed excitons , and , generally , @xmath191 . applying this normalization condition , we get the following results @xmath192 here we used the following notations , @xmath193 , @xmath194 , where @xmath195 and @xmath196 . the formula ( [ amp1 ] ) is valid for @xmath197 . we assume that , at @xmath198 ( this is the important parameter ! ) , we always have @xmath199 then , the following inequalities are valid : @xmath200 and @xmath201 the characteristic length of the packet can be estimated from eq . ( [ genform ] ) as follows ( @xmath202 ) : @xmath203 @xmath204 therefore , at @xmath205 , we can roughly estimate @xmath206 and , for the average concentration of condensed excitons in the packet , @xmath207 , we have @xmath208 recall that the second part of the condensate , the displacement field @xmath209 , is of the same importance as the first part , the exciton wave function @xmath185 . the displacement field @xmath210 can be represented as follows @xmath211 to estimate its amplitude , @xmath212 , we have to estimate the parameter @xmath213 first . for @xmath214 and @xmath215 ( i.e. @xmath216 ) , we obtain @xmath217 if this parameter is small enough , such as @xmath218 , we can neglect the nonlinear corrections to the amplitude @xmath219 and to the shape of @xmath220 as well , @xmath221 thus , due to the validity of @xmath222 , there is almost no difference between the approximation @xmath223 where @xmath224 and the exact solution of the weakly nonlinear case with @xmath225 and @xmath107 . for @xmath226@xmath227 , @xmath228@xmath227 , we estimate @xmath229 . although the approximate solutions we used in this study are valid for @xmath230 , they can be used at @xmath231 for estimates . note that the effective chemical potential is a rather small parameter in this model , @xmath232 that is why the characteristic length , @xmath233 , see ( [ width ] ) , can be estimated as @xmath234 within the validity of approximations ( [ amp1 ] ) , ( [ mu1 ] ) . moreover , @xmath235 and the parameter @xmath172 in ( [ genform ] ) can be estimated as @xmath236 . in this case , one can neglect it in eq . ( [ solitonmod ] ) . returning to the laboratory reference frame , we can write the condensate wave function in the form ( see fig . 1 ) : @xmath237 @xmath238 where we count the exciton energy from the bottom of the crystal valence band , ( @xmath239 ) , and @xmath240 is the amplitude of the phonon part of condensate , @xmath241 to calculate the energy of the moving condensate within the lagrangian approach , ( see eq . ( [ lagrangian ] ) ) , we have to integrate the zeroth component of the energy - momentum tensor @xmath242 over the spatial coordinates . consequently , we have the following formula @xmath243 @xmath244 here we do not take into account a small correction to this energy due to the quantum depletion of the condensate ( @xmath245 and @xmath246 ) . then the result reads @xmath247 @xmath248 we will disregard the terms @xmath249 in @xmath250 , and the corrections @xmath251 in @xmath252 . then we can write @xmath253 and @xmath254 where @xmath255 note that the parameter @xmath256 is a rather small one , @xmath257 , so that the value of @xmath258 can be @xmath259 , and , roughly , @xmath260 . one can see that the exciton - phonon condensate carries a non - zeroth momentum , @xmath261 : @xmath262 @xmath263 @xmath264 thus , we obtain @xmath265 and estimate the parameter @xmath266 at @xmath216 , @xmath267 . = 400pt = 300pt to consider the stability of the exciton - phonon condensate moving in a lattice , one has to couple the excitons with different sources of perturbation , such as impurities , thermal lattice phonons , surfaces , etc .. in this work , however , we will not specify any source . instead , we consider the stability conditions in relation to creation ( emission ) of the condensate excitations that can be found in the framework of investigation of the low - energy excitations of the condensate itself . although the condensate wave function @xmath268 was obtained in the framework of the effective 1d model , we normalized it in 3d space . therefore , we can use this solution as a classical part in the following decomposition of 3d field operators in the co - moving frame : @xmath269 @xmath270 where @xmath271 . substituting the field operators of the form ( [ hatpsi]),([hatu ] ) into the lagrangian density ( [ lagrangian ] ) , we can write the later in the following form : @xmath272 where @xmath273 stands for the classical part of @xmath55 , and @xmath274 is the bilinear form in the @xmath275-operators . in the simplest ( bogoliubov ) approximation @xcite,@xcite , @xmath276 and , hence , the bilinear form @xmath274 defines the equations of motion for the fluctuating parts of the field operators . as a result , these equations are _ linear _ and can be written as follows : @xmath277 @xmath278 and @xmath279 @xmath280 @xmath281 @xmath282 @xmath283 the same approximation can be performed within the hamiltonian approach . indeed , decomposition of the field operators near their nontrivial classical parts leads to the decomposition of the hamiltonian ( [ ham1 ] ) itself , and as it was done with the lagrangian only the classical part of @xmath284 , @xmath285 , and the bilinear form in the fluctuating fields , @xmath286 , are left for examination : @xmath287 in the co - moving frame , @xmath288 , i.e. @xmath289 and the standard commutation relation , @xmath290 $ ] , has the form @xmath291= [ \ , \delta{\hat u}_{j}({\bf x},t),\,\,\rho\,\partial_{t}\delta{\hat u}_{s}({\bf x}',t ) - \rho v\,\partial_{x}\delta{\hat u}_{s}({\bf x}',t)\ , ] = i\hbar\,\delta({\bf x}-{\bf x'})\,\delta_{js}. \label{comrel1}\ ] ] however , the hamiltonian ( [ ham2 ] ) can be diagonalized and rewritten in the form : @xmath292 here , @xmath293 is the quantum correction to the energy of the condensate and the indexes @xmath294 and @xmath295 label the elementary excitations of the system . we assume the operators @xmath296 , @xmath297 are the bose ones . these operators describe two different branches of the excitations , @xmath298 , and they can be represented by the following linear combinations of the `` delta- operators '' : @xmath299 @xmath300 note that by analogy with the exciton - polariton modes in semiconductors@xcite,@xcite the excitations of the condensate ( [ movingcond ] ) can be considered as a mixture of exciton- and phonon - type modes . however , in this model , the phonons are fluctuations of the @xmath301-part of the condensate . the commutation relations between @xmath302-operators are bose ones , so that @xmath303= \delta_{ss'}\ ] ] lead to the following orthogonality condition @xmath304 since the @xmath302-operators ( see eq . ( [ ham3 ] ) ) evolve in time as simply as @xmath305 these operators ( and the frequencies @xmath306 ) are the eigenvectors ( and , correspondingly , the eigenvalues ) of the equations of motion ( [ deltapsi]),([deltau ] ) obtained within the lagrangian method . then , the time dependent `` @xmath275-operators '' in ( [ deltapsi]),([deltau ] ) can be represented by the following linear combinations of the @xmath302-operators : @xmath307 @xmath308 @xmath309 @xmath310 for @xmath311 , one has to change @xmath312 to @xmath313 in ( [ uvtransform2 ] ) . note that this _ ansatz _ is , in fact , a generalization of the u - v bogoliubov transformation . then we can rewrite eqs . ( [ alpha]),([alpha1 ] ) as follows ( @xmath298 ) @xmath314 @xmath315 and one of the orthogonality relations has the form ( @xmath316 ) @xmath317 @xmath318 the question we want to clarify is whether coupling between excitonic excitations and phonon excitations is important for understanding the condensate excitations . substituting ( [ uvtransform1])-([uvtransform2 ] ) into eqs . ( [ deltapsi]),([deltau ] ) , we obtain the following coupled eigenvalue equations @xcite : @xmath319 @xmath320 @xmath321 @xmath322{c}_{j , s}^{r}({\bf x})=0 , \,\,\,\,r=2,3 , \label{eigfunction3}\ ] ] @xmath323 @xmath324 { c}_{j , s}^{x}({\bf x})=0 . \label{eigfunction4}\ ] ] here we used the following notations @xmath325 @xmath326 to simplify investigation of the characteristic properties of the different solutions of eqs . ( [ eigfunction1])-([eigfunction3 ] ) , we subdivide the excitations ( [ uvtransform1])-([uvtransform2 ] ) into two major parts , the _ inside_-excitations and the _ outside_-ones . inside_-excitations are localized merely inside the packet area , i.e. @xmath327 and @xmath328 , whereas the _ outside_-excitations propagate merely in the outside area , i.e. @xmath329 and @xmath330 . for the outside collective excitations , the asymptotics of the low - lying energy spectrum can be found easily . indeed , if we assume that @xmath331 and @xmath332 in the outside packet area , the equations ( [ deltapsi ] ) and ( [ deltau ] ) are ( formally ) uncoupled . then , eq . ( [ deltapsi ] ) describes the excitonic branch of the outside - excitations with the following dispertion low in the co - moving frame : @xmath333 and @xmath334 in the laboratory frame of reference . then the exciton field operator , which describes the exciton condensate with _ one _ long - wavelength outside - excitation , has the following form : @xmath335 @xmath336 it is easy to see that such a collective excitation , @xmath337 can be interpreted as a free exciton with the energy and the ( quasi)momentum @xmath338 note that the condition @xmath339 can be violated at the velocities close to @xmath121 , whereas @xmath340 is always positive . then the question is whether @xmath341 really means the condensate instability in relation to the creation of outside excitations . for example , being unstable , the condensate could continuously emit outside - excitations , which form a sort of ` tail ' behind the localized packet . recall that the particle number @xmath342 is not conserved in quantum states with a condensate , and @xmath343 . however , for @xmath344 and @xmath345 , the following estimate is valid @xmath346 therefore , we can compare the condensate energy @xmath347 and the energy of the condensate that emits excitons , or , equivalently , the condensate with outside excitations , @xmath348 . for simplicity s sake , we consider @xmath349 different wave vectors , @xmath350 , to be close to each other , so that the values of @xmath351 and @xmath352 are well - defined . ( this is a model of how the instability tail could be formed . ) we obtain ( see eqs . ( [ energy])-([theta ] ) ) @xmath353 @xmath354 @xmath355 for the momentum of the moving condensate with the outside excitations , we have @xmath356 @xmath357 note that the energy and the momentum of the phonon part of the condensate change after exciton emission . we hypothesize that the transformation @xmath358 ( with the emission of outside excitons , see figs . 1,2 ) corresponds to the case in which the outside exciton and the outside acoustic phonon appear _ together_. indeed , in the @xmath359 limit ( i.e. @xmath360 ) , we approximately considered the condensate collective excitations as being uncoupled . however , the phonon @xmath361 can be emitted with the energy compensating the changement of @xmath362 in the phonon part of the condensate energy . moreover , the order of value of @xmath363 is typical for the low - energy acoustic phonons , @xmath364mev . if @xmath365 , the emitted phonon can compensate the changement of @xmath366 as well . the most interesting case is the backward emission of excitons , i.e. @xmath367 in the co - moving frame . then we can rewrite ( [ cond_and_exc1 ] ) as follows @xmath368 @xmath369 the moving condensate can be considered as a stable one in relation to emission of the outside excitations ( @xmath370 ) if such an emission gains energy , @xmath371 this means that the following inequality has to be valid @xmath372 this condition can be rewritten in the dimensionless form as follows : @xmath373 we argue that , even for velocities close to @xmath121 ( where @xmath374 can be @xmath375 , and the instability could appear as @xmath376 ) , inequality ( [ stabout ] ) seems to be always true in the long - wavelength approximation , @xmath377 . on fig . [ l - outside ] , the stable ballistic condensate is shown with its long - wavelength outside - excitations . note that the stability against large-@xmath378 modes can not be properly described within approximation ( [ gap1]),([outside ] ) . however , we can discuss this case within the inside - approximation . = 400pt = 300pt to simplify the calculation of _ inside_-excitation spectrum ( see eqs . ( [ eigfunction1])-([eigfunction3 ] ) ) we will use the semiclassical approximation @xcite . in this approximation , the excitations can be labeled by the wave vector @xmath12 in the co - moving frame , and the following representation holds : @xmath379 where the phase @xmath380 , and @xmath381 , @xmath382 , and @xmath383 are assumed to be smooth functions of @xmath384 in the inside condensate area . notice that the @xmath12- and @xmath384-representations are mixed here . this means that the operator nature of the fluctuating fields is actually dismissed within the semiclassical approximation . howevever , the orthogonality relations among @xmath385 , @xmath386 , and @xmath387 , @xmath388 , and , hence , among @xmath389 , @xmath390 , and @xmath391 , @xmath392 come from the bose commutation relations between the operators @xmath393 and @xmath394 @xcite,@xcite . for example , eq . ( [ uvtransform1 ] ) is modified as follows : @xmath395 and the _ inside_-excitation part of the elementary excitation term in eq . ( [ ham3 ] ) , @xmath396 , can be written as @xmath397 note that the semiclassical energy @xmath398 of the inside - excitation mode @xmath399 is supposed to be a smooth function of @xmath91 as well , ( at least , as smooth as @xmath400 , which is taken constant in the inside - approximation ) . although the low - lying excitations can not be properly described within the semiclassical approximation , we apply it here to calculate the low - energy asymptotics of the spectrum . in fact , within the approximation @xmath401 , all the important properties of such excitations can be understood by use of the semiclassical approach . there are two different types of the inside - excitations , the longitudinal excitations and the transverse ones . the later have the wave vectors @xmath12 perpendicular to the @xmath402-direction . for the sake of simplicity , we choose @xmath403 . then the vector @xmath404 has one nonzeroth component for such transverse excitations , @xmath405 . substituting _ ansatz _ ( [ semiclassical ] ) with @xmath406 and @xmath407 into eqs . ( [ eigfunction1])-([eigfunction3 ] ) , we transform these differential equations into the algebraic ones ( within the inside - approximation @xmath408 ) : @xmath409 @xmath410 @xmath411 c_{j , k}^{y}({\bf x})=0 . \label{eig}\ ] ] after some straightforward algebra , we can write out the equation that defines the spectrum of transverse exciton - phonon excitations in the inside - approximation : @xmath412 @xmath413=\ ] ] @xmath414\phi_{\rm o}^{2 } ( x)+ % \nu_{1}\phi_{\rm o}^{4}(x ) l(- k_{\perp}^{2})- \nu_{0}\phi_{\rm o}^{2 } ( { x } ) - 2\nu_{1}\phi_{\rm o}^{4 } ( { x } ) \,\right)\ , \,\frac{2\,\sigma_{0}^{2}}{\rho\,c_{l}^{2}}\,\phi_{\rm o}^{2}(x)\,(c_{l}^{2 } k_{\perp}^{2 } ) . \label{spectrum0}\ ] ] taking into account the momentum cut - off @xmath415 , which is defined as @xmath416 we can rewrite eq . ( [ spectrum0 ] ) as follows @xmath417 @xmath418 + f(x)+ \epsilon_{+}\right)\!\cdot\ ! \left ( \frac{\hbar^{2 } } { 2 m } [ k_{\perp}^{2}- { k}_{0}^{2 } ] + f(x)+2\nu_{0}\phi^{2}_{\rm o}(x ) + \epsilon_{+}'\right ) \,\right]=\ ] ] @xmath419 + f(x)+ \epsilon_{+}\right)\,2 \frac{\sigma_{0}}{m c_{l}^{2}}\,(\sigma_{0 } a_{l}^{3})\,\phi_{\rm o}^{2 } \,(c_{l}^{2 } k_{\perp}^{2 } ) , \,\,\,\,\,\ , \vert x \vert < l_{0}. \label{spectrum0tt}\ ] ] here @xmath420 , i.e. @xmath421 inside the condensate , and @xmath422 although eq . ( [ spectrum0tt ] ) can be solved exactly for the transverse excitation spectrum @xcite , taking into account the coupling term in the r.h.s . of ( [ spectrum0tt ] ) changes the values of excitation energies slightly , and the excitations can be approximately considered as of the pure excitonic ( @xmath423 ) or the pure phonon ( @xmath424 ) types . it is also useful to investigate _ asymptotics _ of the transverse inside excitations . for definiteness sake , we investigate the left side asymptotics of these excitations here , @xmath425 @xmath426 here @xmath427 , @xmath428 , @xmath429 , and @xmath430 are smooth functions of @xmath384 at @xmath431 . note that we introduced two components of @xmath432 to make eqs . ( [ eigfunction1])-([eigfunction4 ] ) self - consistent . let the equalities @xmath433 be valid . then the system of differential equations ( [ eigfunction1])-([eigfunction4 ] ) can be reduced to a system of algebraic ones , which are analogous to eqs . ( [ eigfunction01])-([eig ] ) . consequently , we can write out the equation for @xmath434 valid at @xmath431 , @xmath435^{2}\,- \,c_{l}^{2 } [ k^{2}_{\perp}-(2/l_{0})^{2}]\,\right ) \times\ ] ] @xmath436 + \tilde{f}(x ) \right)\cdot \left ( \frac{\hbar^{2 } } { 2 m } [ { k}^{2}_{\perp } - \tilde{k}_{0}^{2 } ] + \tilde{f}(x)+2\nu_{0}(2\phi_{\rm o}\exp(x / l_{0})\,)^{2}\right ) \,\right]= % \phi_{\rm o}^{2}(x)\right ) \,\right]= % ( 2\phi_{\rm o}\exp(x / l_{0})\,)^{2}\right ) \,\right]=\ ] ] @xmath437 + \tilde{f}(x)\,\right)\ , 2 \frac{\sigma_{0}}{m c_{l}^{2}}\,(\sigma_{0 } a_{l}^{3})\ , \bigl(2\phi_{\rm o}\exp(x / l_{0 } ) \,\bigr)^{2 } \,c_{l}^{2 } [ \,k^{2}_{\perp } - ( 2/l_{0})^{2 } ] , \label{spectrum1t}\ ] ] where @xmath438 , @xmath439 , and @xmath440 ( we neglected the terms , such as @xmath441 and @xmath442 in eqs . ( [ eigfunction1])-([eigfunction2 ] ) , and the terms @xmath443 in eq . ( [ eigfunction4 ] ) as well . ) obviously , the structure of eqs . ( [ spectrum0tt ] ) and ( [ spectrum1 t ] ) is the same . as the coupling between exciton and phonon branches is weak for the transverse inside - excitations ( see the r.h . sides of eqs . ( [ spectrum0tt ] ) and ( [ spectrum1 t ] ) ) and the effect of the finite width @xmath444 can be taken into account as the spatial dependence of the important parameters in @xmath445 , we use the following formula to estimate the low - energy excitation spectrum : @xmath446 @xmath447 here we take the inside - condensate - asymptotics of @xmath448 , @xmath400 , and @xmath449 to estimate @xmath450 . note that , for the inside - condensate excitations , the the low - energy limit means @xmath451 then , in the co - moving frame , the low - energy excitation spectrum @xmath452 may develop a gap of the order of @xmath175 , ( see eq . ( [ outside ] ) for comparison ) . thus , inside the condensate , we obtain a strong deviation of the collective excitation spectrum from both the simple excitonic one , @xmath453 , and the bogoliubov - landau spectrum @xmath454 . the case of the longitudinal excitations , @xmath455 , @xmath456 , is more difficult to analyze because the mode interaction is non - negligible in the low - energy limit . ( on fig . 1 . , a longitudinal inside - excitation is shown with the two possible directions of the wave vector @xmath457 . ) recall that the `` bare '' phonon modes , which can be written in the laboratory frame as @xmath458 with @xmath459 and @xmath460 , will be considered in the co - moving frame , @xmath461 . then , within the inside - condensate approximation , the following equation stands for the excitation spectrum : @xmath462 @xmath463 + f(x)+ \epsilon_{+}\right)\cdot \left ( \frac{\hbar^{2 } } { 2 m } [ k_{x}^{2}- { k}_{0}^{2 } ] + f(x)+2\nu_{0}\phi^{2}_{\rm o}(x ) + \epsilon_{+}'\right ) \,\right]=\ ] ] @xmath464 + f(x)+ \epsilon_{+}\right)\,2 \frac{\sigma_{0}}{m c_{l}^{2}}\,(\sigma_{0 } a_{l}^{3})\,\phi_{\rm o}^{2 } \,(c_{l}^{2 } k_{x}^{2 } ) , \,\,\,\,\,\ , \vert x \vert < l_{0}. \label{spectrum0ll}\ ] ] it is important to note that , unlike the case of transverse excitations , the values of @xmath465 and @xmath466 are of the same order of value at @xmath467 , and the inequality @xmath468 is valid in the low - energy limit . moreover , the two cases , @xmath469 ( @xmath19-case ) and @xmath470 ( @xmath471-case ) , are different as it can be seen from the l.h.s . of eq . ( [ spectrum0ll ] ) . in the low - energy limit , we can write the approximate solution of ( [ spectrum0ll ] ) as follows @xmath472 + f(x)+ \epsilon_{+}\right)\times\ ] ] @xmath473 + f(x)+ 2\nu_{0}\,\phi^{2}_{\rm o}(x ) \pm 2 \,q_{\pm } \gamma(v)\,\frac{\sigma_{0}}{m c_{l}^{2}}\,(\sigma_{0 } a_{l}^{3})\,\phi_{\rm o}^{2}(x ) + \epsilon_{+}'\right),\ ] ] where @xmath474 and @xmath475 . note that @xmath476 , whereas , for the phonon - type branch , @xmath477 . for @xmath478 , we have the following inequaluty for the excitonic branch , @xmath479 + f(x)+ \epsilon_{+}\right)\times\ ] ] @xmath480 + f(x)+ 2\,\widetilde{\nu_{0}}(v)\,\phi^{2}_{\rm o}(x ) + \epsilon_{+}'\right ) < \bigl(\hbar \omega^{(-)}_{{\rm ex } , \,k_{x } } \bigr)^{2 } < \bigl(\hbar \omega_{{\rm ex } , \,k_{x}}^{(0 ) } \bigr)^{2 } , \label{inequality1}\ ] ] where @xmath481 within the inside - approximation , and , for the phonon - type branch , we obtain @xmath482 . to derive the formulas for the amplitudes @xmath483 , @xmath484 , and @xmath485 of the excitonic branch , we use the following approximations @xmath486 and @xmath487 is modified as @xmath488 then we can rewrite the formulas for the excitonic excitation spectrum as follows @xmath489 recall that the orthogonality relation ( [ import22 ] ) can be used to normalize the amplitudes . within the inside - approximation , eq . ( [ import22 ] ) can be rewritten as follows ( @xmath490 ) @xmath491 and we have for the excitonic amplitudes @xmath492 @xmath493 here the effective condensate volume @xmath494 is used to normalize the u- and v - wave functions of the inside excitations , and @xmath495 . subsiquently , we get for @xmath496 the following approximate formulas @xmath497 to estimate the characteristic value of @xmath498 , we use @xmath499 and obtain @xmath500 the parameters @xmath501 characterize the relative weight of excitonic degrees of freedom in the considered branch of excitations . as the parameter @xmath502 at @xmath503 , the parameter @xmath504 can be estimated as @xmath505 . for @xmath470 , we obtain the following equation from ( [ ortuvc1 ] ) , @xmath506 within the stability area ( see the next subsection for an extended discussion ) , we estimate the ratio @xmath507 at @xmath508 . then , @xmath509 and @xmath510 . to go beyond the inside - approximation , the effect of inhomogeneous behavior of the longitudinal excitations can be considered . we use the following _ ansatz _ for the left side asymptotics ( see fig . 2 ) @xmath511 @xmath512 where @xmath513 and @xmath514 . then , like the case of transversal excitations ( see eq . ( [ spectrum1 t ] ) ) , we can write the equation for @xmath515 valid at @xmath431 , @xmath516 @xmath517 + \tilde{f}(x ) \right ) \left ( \frac{\hbar^{2 } } { 2 m } [ \bar { k}^{2}_{x } - { k}_{0}^{2 } ] + \tilde{f}(x)+2\nu_{0}\bigl(2\phi_{\rm o}\exp(x / l_{0})\,\bigr)^{2}\right ) \,\right]= % % \phi_{\rm o}^{2}(x)\right ) \,\right]= % % ( 2\phi_{\rm o}\exp(x / l_{0})\,)^{2}\right ) \,\right]=\ ] ] @xmath518 + \tilde{f}(x)\,\right)\ , 2 \frac{\sigma_{0}}{m c_{l}^{2}}\,(\sigma_{0 } a_{l}^{3})\ , \bigl(2\phi_{\rm o}\exp(x / l_{0 } ) \,\bigr)^{2 } \,c_{l}^{2}\tilde{k}^{2}_{x } , \label{spectrum1ll}\ ] ] where @xmath519 in the phonon parts of this equation and @xmath520 in the exciton parts of it ( @xmath521 ) . it is easy to see that eqs . ( [ spectrum0ll ] ) and ( [ spectrum1ll ] ) are in the continuity correspondence , i.e. they describe the same object . for example , the ( left side ) asymptotic behavior of @xmath522 can be obtained from the inside - condensate formulas by the substitute @xmath523 and @xmath524 as a result , we obtain @xmath525 that corresponds to the outside - excitation spectrum , eq . ( [ gap1 ] ) . the excitonic input into @xmath526 , the quantum depletion of the moving condensate , can be calculated by @xmath527 @xcite . to estimate this value , one can approximate @xmath528 as follows @xmath529 however , the summation @xmath530 implies @xmath531 within the semiclassical approximation . assuming that such an integration makes the difference among @xmath532 , @xmath533 and @xmath534 not essentially important , we use the following estimate for @xmath535 , @xmath536 then , the integration @xmath531 can be reduced to @xmath537 , and the main input ( @xmath538 ) can be estimated from the following formula @xmath539 using this density , we speculate that the _ localized _ depletion of condensate i.e. the number of particles that are out of the condensate but move with it coherently seems to be a small value . we obtain the following estimate ( @xmath540 and @xmath541 ) : @xmath542 where the factor before @xmath190 can be estimated as @xmath543 ( here we used eqs . ( [ need1]),([need2 ] ) within the approximation @xmath544 . ) note that there is no small @xmath12 input to the estimate of @xmath530 because , first , such excitations belong to the outside - excitation branch in our model , and , second , we use the approximation @xmath545 for them . to investigate the stability of the moving condensate in relation to the creation of inside - excitations , we calculate the energy of the condensate with the _ one _ inside excitation , @xmath546 , described by the following set : @xmath378 , @xmath547 , and @xmath548 , @xmath549 , @xmath550 . in this study , we analyze the stability inside the excitonic sector of our model . although the excitations were defined in the co - moving frame , calculations should be done in the laboratory frame . returning to the lab frame , we represent the exciton and phonon field functions as follows : @xmath551 where @xmath552 and @xmath553 see eq . ( [ outside ] ) for comparison . in this analysis , the inside excitations are not considered as fluctuations , and the ( average ) number of particles in the condensate and its energy are changed as @xmath554 and @xmath555 , respectively . however , these changes are not important if the number of excitation in a system is less than @xmath556 . they could be important in the case of instability of the moving condensate . the zeroth component of the energy - momentum tensor can be represented in the form @xmath557 where the first part corresponds to the condensate energy @xmath558 and the second part gives the energy of inside - excitations , @xmath559 . after substitution of ( [ urfin]),([urfin1 ] ) into @xmath560 , we have for the total energy @xmath561 @xmath562 @xmath563 where @xmath564 @xmath565 see eqs . ( [ cond_and_exc1]),([cond_and_exc1 ] ) for comparison . in this study , we discuss qualitatively the stability of the condensate in relation to the backward emission of inside - excitations , ( i.e. @xmath566 in the co - moving frame ) . to begin with , we consider the standard criterion , @xmath567 where @xmath568 corresponds to the low - lying excitations . the value of @xmath569 is taken within the inside - approximation , see eq . ( [ inequality1 ] ) , so that @xmath570 and @xmath571 . then , it is easy to conclude that the following inequality @xmath572 is valid in the low - energy limit if the effective chemical potential ( [ need2 ] ) is large enough , see eq . ( [ stable ] ) for comparison . more precisely , the ballistic velocity @xmath10 and the number of particles in the condensate , @xmath190 , have to be large enough , f.ex . , @xmath573 and @xmath574 , in order to the inequality @xmath575 can be satisfied . thus , for @xmath576 where @xmath577ev , one can expect conditions ( [ stable1]),([real_param ] ) to be valid . despite the condensate can be formed near @xmath578 in theory , f.ex . , with @xmath579 and @xmath544 , such a ballistic state seems to be unstable against the creation of inside - excitations . note that the critical ( landau ) velocity , @xmath580 , can be found as a solution of eq . ( [ cr_mu ] ) and @xmath581 . in fact , the parameter @xmath582 controls the stability / instability of the condensate , see eqs . ( [ stable]),([stable1 ] ) . analyzing ( [ stable ] ) , we did not take into account eq . ( [ stableornot ] ) . however , if the instability regime takes place , more than @xmath583 inside - excitations can appear . as the changes in @xmath584 because of @xmath585 are nonlocal ( in spite of creation of the localized excitations , see figs . 1 and 3 ) , a free acoustic phonon can appear in the system lattice@xmath19excitons _ together _ with appearance of the localized excitation @xmath586 . like the case of outside - excitations , we assume that @xmath587 , see eq . ( [ stableornot ] ) . then , only the term @xmath588 is important . in fact , this term leads to some renormalization of the values of the critical parameters , @xmath589 and @xmath580 . = 410pt = 300pt in this section , we address the problem of interaction between two moving condensates . this problem is essentially nonstationary , especially if the initial ballistic velocities of packets are different . within the quasi-1d conserving model , the following equations govern the dynamics of the two input packets ( we choose the reference frame moving with the slow packet ) : @xmath590 @xmath591 then , the initial conditions can be written in the explicit 1d form by using the exact solution of the model ( [ 1deq]),([11deq ] ) . note that the amplitudes of the stationary ballistic state , @xmath592 , were defined from the normalization condition and depend on the values of @xmath10 and @xmath190 . hence , the amplitudes of the `` input '' condensates for eqs . ( [ timedep1d]),([timedep1dd ] ) may not have the same values . = 400pt = 300pt in this study , we approach the problem of strong interaction between the condensates . therefore , we choose the nonsymmetric initial conditions , i.e. the amplitude and the velocity of the `` input '' packets are different , for example , @xmath593 and @xmath594 , see fig . [ l - inside11 ] . here , we reply on the experimental observation @xcite that , at @xmath595 , the ballistic velocity of the condensate depends on the power of a laser beam irradiating the crystal . if the exciton concentration in the first packet , @xmath596 , is close to the value of the bose condensation threshold , the exciton concentration in the second packet , @xmath597 , the velocity difference between condensates can reach @xmath598 . then , in the reference frame moving with the first ( slow ) packet , the initial conditions can be taken as the following : @xmath599 @xmath600 where @xmath601 , @xmath602 , @xmath603 , and @xmath604 is the ( initial ) time delay . as the second packet moves in this frame of reference , the regime of strong nonlinear interaction between the condensates is ( theoretically ) unavoidable . note that , even before collision , a time - dependent interference term in @xmath605 begins to influence the packet dynamics , see fig . 4 . for example , the r.h.s . of ( [ timedep1dd ] ) contains @xmath606 where @xmath607 and @xmath608 . the ratio @xmath609 can be of the order of @xmath610 , and the characteristic scale of fringes ( [ fringes ] ) is @xmath611 that is they are of the long - wavelength nature . to answer the question which model ( conserving ( [ timedep1d]),([timedep1dd ] ) or kinetic @xcite,@xcite one ) is more adequate to describe the packet collision , we have to compare the estimate of interaction time , @xmath612 and characteristic time scales of the processes @xmath613 driven by phonons or by @xmath614 interaction . note that some thermal phonons have to be excited in the system to assist such transitions , and the value of @xmath615 is of the order of scattering time of the exciton - la - phonon interaction ( although without any macroscopical occupancy ) @xcite . if processes ( [ process ] ) are driven by the lattice phonons , two phonons are necessary to satisfy the laws of conservation . for instance , we choose @xmath616 in ( [ process ] ) and obtain ( see eqs . ( [ cond_and_exc1]),([cond - and - exc1 ] ) ) @xmath617 @xmath618 @xmath619 although the second packet moves faster , @xmath620 , this state can be considered as a more stable ( and , thus , more preferable ) one for the excitons of the slow packet . indeed , the following inequality for the effective difference between the generalized chemical potentials seems to be valid @xmath621 and the absolute value of the l.h.s . of ( [ difference ] ) is @xmath622 . thus , within the quantum kinetic model , the relevant transition probabilities have to be calculated at least in the second order of perturbation theory , f.ex . , @xmath623 as a result , the system of two quantum boltzmann equations will describe the interaction process . unlike the boltzmann equations , eqs . ( [ timedep1d],[timedep1dd ] ) contain information about the quantum coherence between two condensates explicitly and , moreover , can describe the case of strong interaction . unlike the 1d nls equation that supports many - soliton solutions , eqs . ( [ timedep1d]),([timedep1dd ] ) are , in fact , quasi-1d ones , and @xmath38 in ( [ timedep1d ] ) . therefore , it is an open question what happens with two ( excitonic ) solitons after they collide in the crystal . in this work , we assume that the dominant process(es ) of condensate interactions is that one(s ) leading to @xmath624 . then , at the time scales @xmath625 , one solitonic packet can appear as a result of these processes . such a resultant ballistic packet can be approximately described by the steady - state one - soliton solution of eqs . ( [ eq111]),([eq222 ] ) with @xmath626 and the low of energy conservation , @xmath627 if @xmath628 , all the approximate solutions having been found in this study are valid to describe the resultant packet . as we prescribed the value of @xmath629 , we have to estimate the value of @xmath630 from eq . ( [ energy ] ) , ( generally , @xmath631 ) . moreover , we have to assume that the total momentum of the condensates , @xmath632 , may not be conserved because of lattice participation in such a condensate `` merger '' . however , the challenging question of the results of coherent packet collision needs further theoretical and experimental efforts . in this study , we considered a model within which the inhomogeneous excitonic _ condensate _ with a nonzero momentum can be investigated . the important physics we include in our model is the exciton - phonon interaction and the appearance of a coherent part of the crystal displacement field , which renormalizes the x - x interaction vertices . then , the condensate wave function and its energy can be calculated exactly in the simplest quasi-1d model , and the solution is a sort of davydov s soliton @xcite . we believe that the transport and other unusual properties of the coherent para - exciton packets in cu@xmath0o can be described in the framework of the proposed model properly generalized to meet more realistic conditions . we showed that there are two critical velocities in the theory , namely , @xmath121 and @xmath580 . the first one , @xmath121 , comes from the renormalization of two particle exciton - exciton interaction due to phonons , and the bright soliton state can be formed if @xmath595 . then , the important parameter , which controls the shape and the characteristic width of the condensate wave function , is @xmath633 , see eqs . ( [ mu1]),([need2 ] ) . the second velocity , @xmath580 , comes from use of landau arguments @xcite for investigation of the dynamic stability / instability of the moving condensate . in fact , the important parameter , which controls the emission of excitations , is @xmath634 , see eq . ( [ cr_mu ] ) . then , within the semiclassical approximation for the condensate excitations , we found more close @xmath10 is to @xmath18 ( @xmath635 ) more stable the coherent packet is . it is interesting to discuss the possibility of observation of an instability when the condensate can be formed in the inhomogeneous state with @xmath636 , but with @xmath637 , or , better , @xmath638 . such a coherent packet has to disappear during its move through a single pure crystal used for experiments . as the shape of the moving packet depends on time , the form of the registered signal may depend on the crystal length changing from the solitonic to the standard diffusion density profile . we found that the excited states of the moving exciton - phonon condensate can be described by use of the language of elementary excitations . although the possibility of their direct observation is an unclear question itself ( and , thus , the question about the gap in the excitation spectrum is still an open one ) , the stability conditions of the moving condensate can be derived from the low - energy asymptotics of the excitation spectra at @xmath345 . however , the stability problem is not without difficulties @xcite,@xcite . one can easily imagine the situation when the condensate moves in a very high quality crystal , but with some impurity region prepared , f.ex . , in the middle of the sample . in this case , the excitonic superfluidity can be examined by impurity scattering of the ballistic condensate . indeed , such impurities could bound the noncondensed excitons , which always accompany the condensate , and could mediate , for instance , the emission of the outside excitations . the last process may lead to depletion of the condensate and , perhaps , some other observable effects , such as damping , bound exciton pl , etc .. one of the authors ( d.r . ) thanks i. loutsenko for helpful discussions , a. mysyrowicz for useful comments , and e. benson and e. fortin for providing the results of their work before publishing . a. l. ivanov , h. haug , and l. v. keldysh , phys . reports * 296 * , 237 ( 1998 ) ; h. haug , a. l. ivanov , and l. v. keldysh , _ nonlinear optical phenomena in semiconductors and semiconductor microstructures _ , ( world scientific , 1999 ) .

the problem of multiple kernel learning based on penalized empirical risk minimization is discussed . the complexity penalty is determined jointly by the empirical norms and the reproducing kernel hilbert space ( rkhs ) norms induced by the kernels with a data - driven choice of regularization parameters . the main focus is on the case when the total number of kernels is large , but only a relatively small number of them is needed to represent the target function , so that the problem is sparse . the goal is to establish oracle inequalities for the excess risk of the resulting prediction rule showing that the method is adaptive both to the unknown design distribution and to the sparsity of the problem . and .

1211.2998

let @xmath1 be independent copies of a random couple @xmath2 with values in @xmath3 , where @xmath4 is a measurable space with @xmath5-algebra @xmath6 ( typically , @xmath4 is a compact subset of a finite - dimensional euclidean space ) and @xmath7 is a borel subset of @xmath8 . in what follows , @xmath9 will denote the distribution of @xmath2 and @xmath10 the distribution of @xmath11 . the corresponding empirical distributions , based on @xmath12 and on @xmath13 , will be denoted by @xmath14 and @xmath15 , respectively . for a measurable function @xmath16 , we denote @xmath17 similarly , we use the notations @xmath18 and @xmath19 for the integrals of a function @xmath20 with respect to the measures @xmath10 and @xmath15 . the goal of prediction is to learn `` a reasonably good '' prediction rule @xmath21 from the empirical data @xmath22 . to be more specific , consider a loss function @xmath23 and define the risk of a prediction rule @xmath24 as @xmath25 where @xmath26 . an optimal prediction rule with respect to this loss is defined as @xmath27 where the minimization is taken over all measurable functions and , for simplicity , it is assumed that the minimum is attained . _ the excess risk _ of a prediction rule @xmath24 is defined as @xmath28 throughout the paper , the notation @xmath29 means that there exists a numerical constant @xmath30 such that @xmath31 . by `` numerical constants '' we usually mean real numbers whose precise values are not necessarily specified , or , sometimes , constants that might depend on the characteristics of the problem that are of little interest to us ( e.g. , some constants that depend only on the loss function ) . let @xmath32 be a reproducing kernel hilbert space ( rkhs ) associated with a symmetric nonnegatively definite kernel @xmath33 such that for any @xmath34 , @xmath35 and @xmath36 for all @xmath37 [ @xcite ] . if it is known that if @xmath38 and @xmath39 , then it is natural to estimate @xmath40 by a solution @xmath41 of the following empirical risk minimization problem : @xmath42 the size of the excess risk @xmath43 of such an empirical solution depends on the `` smoothness '' of functions in the rkhs @xmath44 . a natural notion of `` smoothness '' in this context is related to the unknown design distribution @xmath10 . namely , let @xmath45 be the integral operator from @xmath46 into @xmath46 with kernel @xmath47 . under a standard assumption that the kernel @xmath47 is square integrable ( in the theory of rkhs it is usually even assumed that @xmath4 is compact and @xmath47 is continuous ) , the operator @xmath48 is compact and its spectrum is discrete . if @xmath49 is the sequence of the eigenvalues ( arranged in decreasing order ) of @xmath48 and @xmath50 is the corresponding @xmath46-orthonormal sequence of eigenfunctions , then it is well known that the rkhs - norms of functions from the linear span of @xmath50 can be written as @xmath51 which means that the `` smoothness '' of functions in @xmath44 depends on the rate of decay of eigenvalues @xmath52 that , in turn , depends on the design distribution @xmath10 . it is also clear that the unit balls in the rkhs @xmath53 are ellipsoids in the space @xmath46 with `` axes '' @xmath54 . it was shown by @xcite that the function @xmath55,\ ] ] provides tight upper and lower bounds ( up to constants ) on localized rademacher complexities of the unit ball in @xmath44 and plays an important role in the analysis of the empirical risk minimization problem ( [ simple_erm ] ) . it is easy to see that the function @xmath56 is concave , @xmath57 and , as a consequence , @xmath58 is a decreasing function of @xmath59 and @xmath60 is strictly decreasing . hence , there exists unique positive solution of the equation @xmath61 . if @xmath62 denotes this solution , then the results of @xcite imply that with some constant @xmath63 and with probability at least @xmath64 @xmath65 the size of the quantity @xmath66 involved in this upper bound on the excess risk depends on the rate of decay of the eigenvalues @xmath52 as @xmath67 . in particular , if @xmath68 for some @xmath69 , then it is easy to see that @xmath70 and @xmath71 . recall that unit balls in @xmath53 are ellipsoids in @xmath46 with `` axes '' of the order @xmath72 and it is well known that , in a variety of estimation problems , @xmath73 represents minimax convergence rates of the squared @xmath0-risk for functions from such ellipsoids ( e.g. , from sobolev balls of smoothness @xmath74 ) , as in famous pinsker s theorem [ see , e.g. , @xcite , chapter 3 ] . sobolev spaces @xmath75 of smoothness @xmath76 is a well - known class of concrete examples of rkhs . let @xmath77 denote the @xmath78-dimensional torus and let @xmath10 be the uniform distribution in @xmath79 . it is easy to check that , for all @xmath76 , the sobolev space @xmath80 is an rkhs generated by the kernel @xmath81 , where the function @xmath82 is defined by its fourier coefficients latexmath:[\[\hat k_n=(|n|^2 + 1)^{-\alpha},\qquad n=(n_1,\ldots , n_d)\in{\mathbb z}^d,\qquad operator @xmath48 are the functions of the fourier basis and its eigenvalues are the numbers @xmath84 . for @xmath85 and @xmath86 , we have @xmath87 ( recall that @xmath49 are the eigenvalues arranged in decreasing order ) so , @xmath88 and @xmath89 , which is a minimax nonparametric convergence rate for sobolev balls in @xmath90 [ see , e.g. , @xcite , theorem 2.9 ] . more generally , for arbitrary @xmath91 and @xmath76 , we get @xmath92 and @xmath93 , which is also a minimax optimal convergence rate in this case . suppose now that the distribution @xmath10 is uniform in a torus @xmath94 of dimension @xmath95 . we will use the same kernel @xmath47 , but restrict the rkhs @xmath44 to the torus @xmath96 of smaller dimension . let @xmath97 . for @xmath98 , we will write @xmath99 with @xmath100 . it is easy to prove that the eigenvalues of the operator @xmath48 become in this case @xmath101 due to this fact , the norm of the space @xmath44 ( restricted to @xmath102 ) is equivalent to the norm of the sobolev space @xmath103 . since the eigenvalues of the operator @xmath48 coincide , up to a constant , with the numbers @xmath104 , we get @xmath105 [ which is again the minimax convergence rate for sobolev balls in @xmath103 ] . in the case of more general design distributions @xmath10 , the rate of decay of the eigenvalues @xmath52 and the corresponding size of the excess risk bound @xmath66 depends on @xmath10 . if , for instance , @xmath10 is supported in a submanifold @xmath107 of dimension @xmath108 , the rate of convergence of @xmath66 to @xmath109 depends on the dimension of the submanifold @xmath4 rather than on the dimension of the ambient space @xmath79 . using the properties of the function @xmath110 , in particular , the fact that @xmath111 is decreasing , it is easy to observe that @xmath112 . $ ] moreover , if @xmath113 denotes the smallest value of @xmath114 such that the linear function @xmath115 $ ] provides an upper bound for the function @xmath116 $ ] , then @xmath117 note that @xmath118 also depends on @xmath119 , but we do not have to emphasize this dependence in the notation since , in what follows , @xmath119 is fixed . based on the observations above , the quantity @xmath120 coincides ( up to a numerical constant ) with the slope @xmath118 of the `` smallest linear majorant '' of the form @xmath121 of the function @xmath122 . this interpretation of @xmath120 is of some importance in the design of complexity penalties used in this paper . instead of minimizing the empirical risk over an rkhs - ball [ as in problem ( [ simple_erm ] ) ] , it is very common to define the estimator @xmath41 of the target function @xmath40 as a solution of the penalized empirical risk minimization problem of the form @xmath123,\ ] ] where @xmath124 is a tuning parameter that balances the tradeoff between the empirical risk and the `` smoothness '' of the estimate and , most often , @xmath125 ( sometimes , ) . the properties of the estimator @xmath41 has been studied extensively . in particular , it was possible to derive probabilistic bounds on the excess risk @xmath43 ( oracle inequalities ) with the control of the random error in terms of the rate of decay of the eigenvalues @xmath49 , or , equivalently , in terms of the function @xmath126 [ see , e.g. , @xcite].=-1 in the recent years , there has been a lot of interest in a data dependent choice of kernel @xmath47 in this type of problems . in particular , given a finite ( possibly large ) dictionary @xmath127 of symmetric nonnegatively definite kernels on @xmath4 , one can try to find a `` good '' kernel @xmath47 as a convex combination of the kernels from the dictionary : @xmath128 the coefficients of @xmath47 need to be estimated from the training data along with the prediction rule . using this approach for problem ( [ aa ] ) with @xmath129 leads to the following optimization problem : @xmath130 this learning problem , often referred to as the multiple kernel learning , has been studied recently by @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite and @xcite among others . in particular [ see , e.g. , @xcite ] , problem ( [ ab ] ) is equivalent to the following : @xmath131\\[-8pt ] & & \hspace*{104.5pt}{}+\epsilon \sum_{j=1}^n \|f_j\|_{\mathcal{h}_{k_j } } \biggr),\nonumber\end{aligned}\ ] ] which is an infinite - dimensional version of lasso - type penalization . @xcite studied this method in the case when the dictionary is large , but the target function @xmath40 has a `` sparse representation '' in terms of a relatively small subset of kernels @xmath132 . it was shown that this method is adaptive to sparsity extending well - known properties of lasso to this infinite - dimensional framework . in this paper , we study a different approach to the multiple kernel learning . it is closer to the recent work on `` _ _ sparse additive models _ _ '' [ see , e.g. , @xcite and @xcite ] and it is based on a `` double penalization '' with a combination of empirical @xmath0-norms ( used to enforce the sparsity of the solution ) and rkhs - norms ( used to enforce the `` smoothness '' of the components ) . moreover , we suggest a data - driven method of choosing the values of regularization parameters that is adaptive to unknown smoothness of the components ( determined by the behavior of distribution dependent eigenvalues of the kernels ) . let @xmath133 . denote @xmath134 ( `` l.s . '' meaning `` the linear span '' ) , and @xmath135 note that @xmath136 if and only if there exists an additive representation ( possibly , nonunique ) @xmath137 where @xmath138 , @xmath139 . also , @xmath140 has a natural structure of a linear space and it can be equipped with the following inner product : @xmath141 to become _ the direct sum _ of hilbert spaces @xmath142 . given a convex subset @xmath143 , consider the following penalized empirical risk minimization problem : @xmath144\\[-8pt ] & & \hspace*{54.4pt}{}+\sum_{j=1}^n \bigl({\epsilon } _ j\| f_j\|_{l_2(\pi_n ) } + { \epsilon}_j^2\|f_j\|_{{\mathcal h}_j } \bigr ) \biggr].\nonumber\end{aligned}\ ] ] note that for special choices of set @xmath145 , for instance , for @xmath146 for some @xmath147 , one can replace each component @xmath148 involved in the optimization problem by its orthogonal projections in @xmath149 onto the linear span of the functions @xmath150 and reduce the problem to a convex optimization over a finite - dimensional space ( of dimension @xmath151 ) . the complexity penalty in the problem ( [ main_erm ] ) is based on two norms of the components @xmath148 of an additive representation : the empirical @xmath0-norm , @xmath152 , with regularization parameter @xmath153 , and an rkhs - norm , @xmath154 , with regularization parameter @xmath155 . the empirical @xmath0-norm ( the lighter norm ) is used to enforce the sparsity of the solution whereas the rkhs norms ( the heavier norms ) are used to enforce the `` smoothness '' of the components . this is similar to the approach taken in @xcite in the context of classical additive models , that is , in the case when @xmath156^n$ ] , @xmath157)$ ] for some smoothness @xmath86 and the space @xmath149 is a space of functions depending on the @xmath158th variable . in this case , the regularization parameters @xmath153 are equal ( up to a constant ) to @xmath159 . the quantity @xmath155 , used in the `` smoothness part '' of the penalty , coincides with the minimax convergence rate in a one component smooth problem . at the same time , the quantity @xmath153 , used in the `` sparsity part '' of the penalty , is equal to the square root of the minimax rate ( which is similar to the choice of regularization parameter in standard sparse recovery methods such as lasso ) . this choice of regularization parameters results in the excess risk of the order @xmath160 , where @xmath78 is the number of components of the target function ( the degree of sparsity of the problem ) . the framework of multiple kernel learning considered in this paper includes many generalized versions of classical additive models . for instance , one can think of the case when @xmath156^{m_1}\times\cdots\times[0,1]^{m_n}$ ] and @xmath161^{m_j})$ ] is a space of functions depending on the @xmath158th block of variables . in this case , a proper choice of regularization parameters ( for uniform design distribution ) would be @xmath162 ( so , these parameters and the error rates for different components of the model are different ) . it should be also clear from the discussion in section [ sec11 ] that , if the design distribution @xmath10 is unknown , the minimax convergence rates for the one component problems are also unknown . for instance , if the projections of design points on the cubes @xmath163^{m_j}$ ] are distributed in lower - dimensional submanifolds of these cubes , then the unknown dimensions of the submanifolds rather than the dimensions @xmath164 would be involved in the minimax rates and in the regularization parameters @xmath153 . because of this , data driven choice of regularization parameters @xmath153 that provides adaptation to the unknown design distribution @xmath10 and to the unknown `` smoothness '' of the components ( related to this distribution ) is a major issue in multiple kernel learning . from this point of view , even in the case of classical additive models , the choice of regularization parameters that is based only on sobolev type smoothness and ignores the design distribution is not adaptive . note that , in the infinite - dimensional lasso studied in @xcite , the regularization parameter @xmath114 is chosen the same way as in the classical lasso ( @xmath165 ) , so , it is not related to the smoothness of the components . however , the oracle inequalities proved in @xcite give correct size of the excess risk only for special choices of kernels that depend on unknown `` smoothness '' of the components of the target function @xmath40 , so , this method is not adaptive either . denote @xmath166 this @xmath167 gram matrix can be viewed as an empirical version of the integral operator @xmath168 from @xmath46 into @xmath46 with kernel @xmath169 . denote @xmath170 the eigenvalues of @xmath171 arranged in decreasing order . we also use the notation @xmath172 for the eigenvalues of the operator @xmath173 with kernel @xmath169 arranged in decreasing order . define functions @xmath174 , @xmath175 and , for a fixed given @xmath176 , let @xmath177 \biggr\}.\ ] ] one can view @xmath178 as an empirical estimate of the quantity @xmath179 that ( as we have already pointed out ) plays a crucial role in the bounds on the excess risk in empirical risk minimization problems in the rkhs context . in fact , since most often @xmath180 , we will redefine this quantity as @xmath181 \biggr\}.\ ] ] we will use the following values of regularization parameters in problem ( [ main_erm ] ) : @xmath182 , where @xmath183 is a sufficiently large constant . it should be emphasized that the structure of complexity penalty and the choice of regularization parameters in ( [ main_erm ] ) are closely related to the following bound on rademacher processes indexed by functions from an rkhs @xmath44 : with a high probability , for all @xmath184 , @xmath185.\ ] ] such bounds follow from the results of section [ sec : sam2pop ] and they provide a way to prove sparsity oracle inequalities for the estimators ( [ main_erm ] ) . the rademacher process is defined as @xmath186 where @xmath187 is a sequence of i.i.d . rademacher random variables ( taking values @xmath188 and @xmath189 with probability @xmath190 each ) independent of @xmath191 . we will use several basic facts of the empirical processes theory throughout the paper . they include symmetrization inequalities and contraction ( comparison ) inequalities for rademacher processes that can be found in the books of @xcite and @xcite . we also use talagrand s concentration inequality for empirical processes [ see , @xcite , @xcite ] . the main goal of the paper is to establish oracle inequalities for the excess risk of the estimator @xmath192 . in these inequalities , the excess risk of @xmath41 is compared with the excess risk of an oracle @xmath193 with an error term depending on the degree of sparsity of the oracle , that is , on the number of nonzero components @xmath138 in its additive representation . the oracle inequalities will be stated in the next section . their proof relies on probabilistic bounds for empirical @xmath0-norms and data dependent regularization parameters @xmath194 . the results of section [ sec : sam2pop ] show that they can be bounded by their respective population counterparts . using these tools and some bounds on empirical processes derived in section [ sec : empirical_process ] , we prove in section [ sec : mainres ] the oracle inequalities for the estimator @xmath195 . considering the problem in the case when the domain @xmath145 of ( [ main_erm ] ) is not bounded , say , @xmath196 , leads to additional technical complications and might require some changes in the estimation procedure . to avoid this , we assume below that @xmath145 is a bounded convex subset of @xmath140 . it will be also assumed that , for all @xmath139 , @xmath197 which , by elementary properties of rkhs , implies that @xmath198 because of this , @xmath199 denote @xmath200 we will allow the constants involved in the oracle inequalities stated and proved below to depend on the value of @xmath201 ( so , implicitly , it is assumed that this value is not too large ) . we shall also assume that @xmath202 is large enough , say , so that @xmath203 . this assumption is not essential to our development and is in place to avoid an extra term of the order @xmath204 in our risk bounds . we will formulate the _ assumptions on the loss function @xmath205_. the main assumption is that , for all @xmath206 , @xmath207 is a nonnegative convex function . in addition , we will assume that @xmath208 is uniformly bounded from above by a numerical constant . moreover , suppose that , for all @xmath206 , @xmath207 is twice continuously differentiable and its first and second derivatives are uniformly bounded in @xmath209 $ ] . denote @xmath210 and let @xmath211 we will assume that @xmath212 . denote @xmath213 clearly , for all @xmath206 , the function @xmath207 satisfies lipschitz condition with constant @xmath214 . the constants @xmath215 will appear in a number of places in what follows . without loss of generality , we can also assume that @xmath216 and @xmath217 ( otherwise , @xmath218 and @xmath214 can be replaced by a lower bound and an upper bound , resp . ) . the loss functions satisfying the assumptions stated above will be called _ the losses of quadratic type_. if @xmath205 is a loss of quadratic type and @xmath219 , then @xmath220 this bound easily follows from a simple argument based on taylor expansion and it will be used later in the paper . if @xmath221 is dense in @xmath46 , then ( [ excess_l_2 ] ) implies that @xmath222 the quadratic loss @xmath223 in the case when @xmath224 is a bounded set is one of the main examples of such loss functions . in this case , @xmath225 for all @xmath226 . in regression problems with a bounded response variable , more general loss functions of the form @xmath227 can be also used , where @xmath228 is an even nonnegative convex twice continuously differentiable function with @xmath229 uniformly bounded in @xmath230 , @xmath231 and @xmath232 . in classification problems , the loss functions of the form @xmath233 are commonly used , with @xmath228 being a nonnegative decreasing convex twice continuously differentiable function such that , again , @xmath229 is uniformly bounded in @xmath230 and @xmath232 . the loss function @xmath234 ( often referred to as the logit loss ) is a specific example . now we introduce several important geometric characteristics of dictionaries consisting of kernels ( or , equivalently , of rkhs ) . these characteristics are related to the degree of `` dependence '' of spaces of random variables @xmath235 and they will be involved in the oracle inequalities for the excess risk @xmath43 . first , for @xmath236 and @xmath237 $ ] , denote @xmath238 clearly , the set @xmath239 is a cone in the space @xmath140 that consists of vectors @xmath240 whose components corresponding to @xmath241 `` dominate '' the rest of the components . this family of cones increases as @xmath242 increases . for @xmath243 , @xmath244 coincides with the linear subspace of vectors for which @xmath245 . for @xmath246 , @xmath239 is the whole space @xmath140 . the following quantity will play the most important role : @xmath247 : \!&= & \inf\biggl\{\beta>0\dvtx \biggl(\sum_{j\in j}\|h_j\|_{l_2(\pi)}^2 \biggr)^{1/2 } \leq\beta\biggl\|\sum_{j=1}^n h_j \biggr\|_{l_2(\pi)},\\[-2pt ] & & \hspace*{18.17pt}\hspace*{117pt } ( h_1,\ldots , h_n)\in c_{j}^{(b ) } \biggr\}.\end{aligned}\ ] ] clearly , @xmath248 is a nondecreasing function of @xmath242 . in the case of `` simple dictionary '' that consists of one - dimensional spaces similar quantities have been used in the literature on sparse recovery [ see , e.g. , koltchinskii ( @xcite , @xcite , @xcite , @xcite ) ; bickel , ritov and tsybakov ( @xcite ) ] . the quantity @xmath248 can be upper bounded in terms of some other geometric characteristics that describe how `` dependent '' the spaces of random variables @xmath249 are . these characteristics will be introduced below . given @xmath250 , denote by @xmath251 the minimal eigenvalue of the gram matrix @xmath252 . let @xmath253 we will also use the notation @xmath254 the following quantity is the maximal cosine of the angle in the space @xmath46 between the vectors in the subspaces @xmath255 and @xmath256 for some @xmath257 : @xmath258 denote @xmath259 the quantities @xmath260 and @xmath261 are very similar to the notion of _ canonical correlation _ in the multivariate statistical analysis . there are other important geometric characteristics , frequently used in the theory of sparse recovery , including so called `` _ _ restricted isometry constants _ _ '' by candes and tao ( @xcite ) . define @xmath262 to be the smallest @xmath263 such that for all @xmath264 and all @xmath265 with @xmath266 , @xmath267 this condition with a sufficiently small value of @xmath268 means that for all choices of @xmath269 with @xmath266 the functions in the spaces @xmath270 are `` almost orthogonal '' in @xmath46 . the following simple proposition easily follows from some statements in koltchinskii ( @xcite , @xcite ) , ( @xcite ) ( where the case of simple dictionaries consisting of one - dimensional spaces @xmath149 was considered ) . for all @xmath236 , @xmath271 also , if @xmath266 and @xmath272 , then @xmath273 . thus , such quantities as @xmath274 or @xmath275 , for finite values of @xmath242 , are reasonably small provided that the spaces of random variables @xmath276 satisfy proper conditions of `` weakness of correlations . '' we are now in a position to formulate our main theorems that provide oracle inequalities for the excess risk @xmath43 . in these theorems , @xmath43 will be compared with the excess risk @xmath277 of an oracle @xmath278 . here and in what follows , @xmath279 . this is a little abuse of notation : we are ignoring the fact that such an additive representation of a function @xmath136 is not necessarily unique . in some sense , @xmath24 denotes both the vector @xmath280 and the function @xmath281 . however , this is not going to cause a confusion in what follows . we will also use the following notation : @xmath282 the error terms of the oracle inequalities will depend on the quantities @xmath179 related to the `` smoothness '' properties of the rkhs and also on the geometric characteristics of the dictionary introduced above . in the first theorem , we will use the quantity @xmath283 to characterize the properties of the dictionary . in this case , there will be no assumptions on the quantities @xmath284 : these quantities could be of different order for different kernel machines , so , different components of the additive representation could have different `` smoothness . '' in the second theorem , we will use a smaller quantity @xmath248 for a proper choice of parameter @xmath285 . in this case , we will have to make an additional assumption that @xmath286 are all of the same order ( up to a constant ) . in both cases , we consider penalized empirical risk minimization problem ( [ main_erm ] ) with data - dependent regularization parameters @xmath287 , where @xmath288 are defined by ( [ regul_par ] ) with some @xmath289 and @xmath290 for a numerical constant @xmath291 . [ co : homobd ] there exist numerical constants @xmath292 such that , for all all oracles @xmath278 , with probability at least @xmath293 , @xmath294\\[-8pt ] & & \qquad \le 2{\mathcal e}(\ell\circ f)+c_2\tau^2 \sum_{j\in j_f}\breve{\epsilon}_j^2 \biggl(\frac{\beta_{2,\infty}^2(j_f,\pi)}{m_{\ast}}+ \|f_j\|_{{\mathcal h}_j } \biggr).\nonumber\vadjust{\goodbreak}\end{aligned}\ ] ] this result means that if there exists an oracle @xmath278 such that : a. the excess risk @xmath277 is small ; b. the spaces @xmath295 are not strongly correlated with the spaces @xmath296 c. @xmath295 are `` well posed '' in the sense that @xmath297 is not too small ; d. @xmath298 are all bounded by a reasonable constant , then the excess risk @xmath43 is essentially controlled by @xmath299 . at the same time , the oracle inequality provides a bound on the @xmath46-distances between the estimated components @xmath300 and the components of the oracle ( of course , everything is under the assumption that the loss is of quadratic type and @xmath218 is bounded away from @xmath109 ) . not also that the constant @xmath301 in front of the excess risk of the oracle @xmath302 can be replaced by @xmath303 for any @xmath263 with minor modifications of the proof ( in this case , the constant @xmath304 depends on @xmath59 and is of the order @xmath305 ) . suppose now that there exists @xmath306 and a constant @xmath307 such that @xmath308 [ co : homobd-1 ] there exist numerical constants @xmath309 such that , for all oracles @xmath278 , with probability at least @xmath293 , @xmath310\\[-8pt ] & & \qquad \le2{\mathcal e}(\ell\circ f)+ c_2 \lambda\tau^2\breve \epsilon^2 \biggl(\frac{\beta_{2,b\lambda^2}^2(j_f,\pi)}{m_{\ast } } d(f ) + \sum_{j\in j_f}\|f_j\|_{{\mathcal h}_j } \biggr).\nonumber\end{aligned}\ ] ] as before , the constant @xmath301 in the upper bound can be replaced by @xmath303 , but , in this case , the constants @xmath304 and @xmath242 would be of the order @xmath311 . the meaning of this result is that if there exists an oracle @xmath278 such that : a. the excess risk @xmath277 is small ; b. the `` restricted isometry '' constant @xmath312 is small for @xmath313 c. @xmath298 are all bounded by a reasonable constant , then the excess risk @xmath43 is essentially controlled by @xmath314 . at the same time , the distance @xmath315 between the estimator and the oracle is controlled by @xmath316 . in particular , this implies that the empirical solution @xmath317 is `` approximately sparse '' in the sense that @xmath318 is of the order @xmath319 . it is easy to check that theorems [ co : homobd ] and [ co : homobd-1 ] hold also if one replaces @xmath202 in the definitions ( [ regul_par ] ) of @xmath178 and ( [ regul_par_01 ] ) of @xmath284 by an arbitrary @xmath320 such that @xmath321 ( a similar condition on @xmath202 introduced early in section [ sec : oracle ] is not needed here ) . in this case , the probability bounds in the theorems become @xmath322 . this change might be of interest if one uses the results for a dictionary consisting of just one rkhs ( @xmath323 ) , which is not the focus of this paper . \2 . if the distribution dependent quantities @xmath324 are known and used as regularization parameters in ( [ main_erm ] ) , the oracle inequalities of theorems [ co : homobd ] and [ co : homobd-1 ] also hold ( with obvious simplifications of their proofs ) . for instance , in the case when @xmath325^n$ ] , the design distribution @xmath10 is uniform and , for each @xmath139 , @xmath326 is a sobolev space of functions of smoothness @xmath86 depending only on the @xmath158th variable , we have @xmath327 . taking in this case @xmath328 would lead to oracle inequalities for sparse additive models is spirit of @xcite . more precisely , if @xmath329\dvtx\int_0 ^ 1 h(x)\,dx=0\}$ ] , then , for uniform distribution @xmath10 , the spaces @xmath326 are orthogonal in @xmath46 ( recall that @xmath326 is viewed as a space of functions depending on the @xmath158th coordinate ) . assume , for simplicity , that @xmath205 is the quadratic loss and that the regression function @xmath40 can be represented as @xmath330 , where @xmath269 is a subset of @xmath331 of cardinality @xmath78 and @xmath332 . then it easily follows from the bound of theorem [ co : homobd-1 ] that with probability at least @xmath333 @xmath334 note that , up to a constant , this essentially coincides with the minimax lower bound in this type of problems obtained recently by @xcite . of course , if the design distribution is not necessarily uniform , an adaptive choice of regularization parameters might be needed even in such simple examples and the approach described above leads to minimax optimal rates . in this section , the case of a single rkhs @xmath44 associated with a kernel @xmath47 is considered . we assume that @xmath335 . this implies that , for all @xmath336 , @xmath337 first , we study the relationship between the empirical and the population @xmath0 norms for functions in @xmath53 . [ th : empnorm ] assume that @xmath176 and @xmath338 . then there exists a numerical constant @xmath63 such that with probability at least @xmath339 for all @xmath336 @xmath340 where @xmath341\\[-8pt ] : \!&= & \inf\biggl\{\epsilon\ge\sqrt{a\log n\over n}\dvtx { { \mathbb{e}}\mathop{\sup_{\|h\|_{{\mathcal h}_k}=1}}_{\|h\|_{l_2(\pi)}\le\delta } } observe that the inequalities hold trivially when @xmath342 . we shall therefore consider only the case when @xmath343 . by symmetrization inequality , @xmath344 the definition of @xmath345 implies that @xmath346\\[-8pt ] & & \qquad\le { 8 { \mathbb{e}}\mathop{\sup_{\|h\|_{{\mathcal h}_k}=1}}_{\|h\|_{l_2(\pi)}\le 2^{-j+1 } } } |r_n(h ) | \le8 ( \bar{\epsilon}2^{-j+1}+\bar{\epsilon}^2 ) .\nonumber\end{aligned}\ ] ] an application of talagrand s concentration inequality yields @xmath347 with probability at least @xmath348 for any natural number @xmath158 . now , by the union bound , for all @xmath158 such that @xmath349 , @xmath350\\[-8pt ] & & \qquad \le32 \biggl(\bar{\epsilon}2^{-j}+\bar{\epsilon}^2 + 2^{-j}\sqrt{t+2\log j\over n}+{t+2\log j \over n } \biggr)\nonumber\end{aligned}\ ] ] with probability at least @xmath351\\[-8pt ] & \ge&1 - 2\exp(-t).\nonumber\end{aligned}\ ] ] recall that @xmath352 and @xmath353 . taking @xmath354 , we easily get that , for all @xmath184 such that @xmath355 and @xmath356 , @xmath357 with probability at least @xmath358 and with a numerical constant @xmath63 . in other words , with the same probability , for all @xmath184 such that @xmath359 , @xmath360 therefore , for all @xmath336 such that @xmath361 we have @xmath362 it can be now deduced that , for a proper value of numerical constant @xmath363 , @xmath364\\[-8pt ] \|h\|_{l_2(\pi_n ) } & \le & c \bigl(\|h\|_{l_2(\pi)}+ \bar{\epsilon}\|h\|_{{\mathcal h}_k } \bigr).\nonumber\end{aligned}\ ] ] it remains to consider the case when @xmath365 following a similar argument as before , with probability at least @xmath366 , @xmath367 under the conditions @xmath368 , @xmath369 then @xmath370 with probability at least @xmath358 , which also implies ( [ eq : empl2 - 1 ] ) and ( [ eq : empl2 - 2 ] ) , and the result follows . theorem [ th : empnorm ] shows that the two norms @xmath371 and @xmath372 are of the same order up to an error term @xmath373 . recall the definitions @xmath374,\ ] ] where @xmath49 are the eigenvalues of the integral operator @xmath48 from @xmath46 into @xmath46 with kernel @xmath47 , and , for some @xmath176 , @xmath375 \biggr\}.\ ] ] it follows from lemma 42 of @xcite [ with an additional application of cauchy schwarz inequality for the upper bound and hoffmann jrgensen inequality for the lower bound ; see also @xcite ] that , for some numerical constants @xmath376 , @xmath377\\[-8pt ] & \leq & c_2 \biggl(n^{-1}\sum_{k=1}^n ( \lambda_k \wedge\delta^2 ) \biggr)^{1/2}.\nonumber\end{aligned}\ ] ] this fact and the definitions of @xmath378 easily imply the following result . [ pr : bareps ] under the condition @xmath335 , there exist numerical constants @xmath292 such that @xmath379 if @xmath47 is the kernel of the projection operator onto a finite - dimensional subspace @xmath44 of @xmath46 , it is easy to check that @xmath380 ( recall the notation @xmath29 , which means that there exists a numerical constant @xmath30 such that ) . if the eigenvalues @xmath52 decay at a polynomial rate , that is , @xmath381 for some @xmath69 , then @xmath382 . recall the notation @xmath383 \biggr\},\hspace*{-35pt}\ ] ] where @xmath384 denote the eigenvalues of the gram matrix @xmath385 it follows again from the results of @xcite [ namely , one can follow the proof of lemma 42 in the case when the rkhs @xmath44 is restricted to the sample @xmath386 and the expectations are conditional on the sample ; then one uses cauchy schwarz and hoffmann jrgensen inequalities as in the proof of ( [ thirtythree ] ) ] that for some numerical constants @xmath376 @xmath387\\[-8pt ] & \leq & c_2 \biggl(n^{-1}\sum_{k=1}^n ( \hat\lambda_k \wedge\delta^2 ) \biggr)^{1/2},\nonumber\end{aligned}\ ] ] where @xmath388 indicates that the expectation is taken over the rademacher random variables only ( conditionally on @xmath386 ) . therefore , if we denote by @xmath389 \biggr\}\hspace*{-35pt}\ ] ] the empirical version of @xmath390 , then @xmath391 . we will now show that @xmath392 with a high probability . [ th : empeps ] suppose that @xmath176 and @xmath338 . there exist numerical constants @xmath292 such that @xmath393 with probability at least @xmath339 . let @xmath394 . it follows from talagrand concentration inequality that @xmath395 with probability at least @xmath348 . on the other hand , as derived in the proof of theorem [ th : empnorm ] [ see ( [ pi_n - pi ] ) ] @xmath396 with probability at least @xmath348 . we will use these bounds only for @xmath158 such that @xmath397 . in this case , the second bound implies that , for some numerical constant @xmath30 and all @xmath398 satisfying the conditions @xmath399 we have @xmath400 ( again , see the proof of theorem [ th : empnorm ] ) . combining these bounds , we get that with probability at least @xmath401 , @xmath402 where @xmath403 . applying now talagrand concentration inequality to the rademacher process conditionally on the observed data @xmath386 yields @xmath404 with conditional probability at least @xmath348 . from this and from the previous bound it is not hard to deduce that , for some numerical constants @xmath405 and for all @xmath158 such that @xmath397 , @xmath406 with probability at least @xmath407 . in obtaining the second inequality , we used the definition of @xmath408 and the fact that , for @xmath409 , @xmath410 , where @xmath411 is a numerical constant . now , by the union bound , the above inequality holds with probability at least @xmath412 for all @xmath158 such that @xmath349 simultaneously . similarly , it can be shown that @xmath413 with probability at least @xmath414 . for @xmath415 , we get @xmath416 for all @xmath417 , with probability at least @xmath418 now by the definition of @xmath345 , we obtain @xmath419 which implies that @xmath420 with probability at least @xmath421 . similarly one can show that @xmath422 for all @xmath417 , with probability at least @xmath421 , which implies that @xmath423 with probability at least @xmath421 . the proof can then be completed by the union bound . define @xmath424\\[-8pt ] & : = & \inf\biggl\{\epsilon\ge\sqrt{a\log n\over n}\dvtx \mathop{\sup_{\|h\|_{{{\mathcal h}}_k}=1 } } _ { \|h\|_{l_2(\pi ) } \le\delta}|r_n(h)|\le \epsilon\delta+ \epsilon^2 , \forall\delta\in(0,1 ] \biggr\}.\nonumber\end{aligned}\ ] ] the next statement can be proved similarly to theorem [ th : empeps ] . [ th : empeps-1 ] there exist numerical constants @xmath292 such that @xmath425 with probability at least @xmath339 . suppose now that @xmath426 is a dictionary of kernels . recall that @xmath427 , @xmath428 and @xmath429 . it follows from theorems [ th : empnorm ] , [ th : empeps ] , [ th : empeps-1 ] and the union bound that with probability at least @xmath430 for all @xmath139 @xmath431\\[-8pt ] \|h\|_{l_2(\pi_n)}&\le & c \bigl(\|h\|_{l_2(\pi)}+ \bar{\epsilon}_j\|h\|_{{\mathcal h}_k } \bigr),\qquad h\in\mathcal{h}_j,\nonumber \\ \label{conclusion_b } c_1\bar{\epsilon}_j&\le&\hat{\epsilon}_j\le c_2\bar{\epsilon}_j \quad\mbox{and}\quad c_1\bar{\epsilon}_j\le\check{\epsilon}_j\le c_2\bar{\epsilon}_j.\end{aligned}\ ] ] note also that @xmath432 provided that @xmath289 and @xmath433 . thus , under these additional constraints , ( [ conclusion_a ] ) and ( [ conclusion_b ] ) hold for all @xmath139 with probability at least @xmath434 . for an arbitrary set @xmath435 and @xmath436 , denote @xmath437 and let @xmath438\\[-8pt ] & & \hspace*{153pt } ( f_1,\ldots , f_n)\in\mathcal { k}_j^{(b ) } \biggr\}.\nonumber\end{aligned}\ ] ] it is easy to see that , for all nonempty sets @xmath269 , @xmath439 theorems [ co : homobd ] and [ co : homobd-1 ] will be easily deduced from the following technical result . [ th : main ] there exist numerical constants @xmath440 and @xmath441 such that , for all @xmath442 in the definition of @xmath443 and for all oracles @xmath444 , @xmath445 with probability at least @xmath293 . here , @xmath289 is a constant involved in the definitions of @xmath446 . recall that @xmath447,\end{aligned}\ ] ] and that we write @xmath448 hence , for all @xmath278 , @xmath449 by a simple algebra , @xmath450 and , by the triangle inequality , @xmath451 we now take advantage of ( [ conclusion_a ] ) and ( [ conclusion_b ] ) to replace @xmath194 s by @xmath452 s and @xmath453 by @xmath454 . specifically , there exists a numerical constant @xmath455 and an event @xmath456 of probability at least @xmath434 such that @xmath457 and , for all @xmath458 , @xmath459 taking @xmath460 , we have that , on the event @xmath456 , @xmath461 similarly , @xmath462 therefore , by taking @xmath183 large enough , namely @xmath463 , we can find numerical constants @xmath464 such that , on the event @xmath456 , @xmath465 we now bound the empirical process @xmath466 , where we use the following result that will be proved in the next section . suppose that @xmath467 , @xmath138 and @xmath468 ( we will need it with @xmath469 ) . denote @xmath470 [ th : uniep ] there exists a numerical constant @xmath63 such that for an arbitrary @xmath176 involved in the definition of @xmath471 with probability at least @xmath472 , for all @xmath473 the following bound holds : @xmath474 assuming that @xmath475 and using the lemma , we get @xmath476 for some numerical constant @xmath477 . by choosing a numerical constant @xmath291 properly , @xmath183 can be made large enough so that @xmath478 . then , we have @xmath479 which also implies @xmath480 we first consider the case when @xmath481\\[-8pt ] & & { } + ( c_2/2)\tau e^{-n}.\nonumber\end{aligned}\ ] ] then ( [ bound_x ] ) implies that @xmath482\\[-8pt ] & & \qquad\le6 c_2\sum_{j\in j_f}\tau\bar{\epsilon}_j\|f_j-\hat{f}_j\| _ { l_2(\pi)},\nonumber\end{aligned}\ ] ] which yields @xmath483 therefore , @xmath484 with @xmath485 . using the definition of @xmath486 , it follows from ( [ bound_y ] ) , ( [ bound_z ] ) and the assumption @xmath487 that @xmath488 recall that for losses of quadratic type @xmath489 then @xmath490 using the fact that @xmath491 , we get @xmath492 and @xmath493 therefore , @xmath494\\[-8pt ] & & \qquad \le{\mathcal e}(\ell\circ f)+100 \tau^2c_2 ^ 2m_{\ast}^{-1}\beta_b^2(j_f).\nonumber\end{aligned}\ ] ] we now consider the case when @xmath495\\[-8pt ] & & \qquad < { \mathcal e}(\ell\circ f)+ 2c_2\sum_{j\in j_f}\tau^2\bar{\epsilon}_j^2\|f_j\|_{{\mathcal h}_j}+(c_2/2)\tau e^{-n}.\nonumber\end{aligned}\ ] ] it is easy to derive from ( [ bound_y ] ) that in this case @xmath496\\[-8pt ] & & \qquad \le \biggl(\frac{3}{2}+\frac{c_1}{8c_2 } \biggr ) \biggl({\mathcal e}(\ell\circ f)+ 2c_2\sum_{j\in j_f}\tau^2\bar{\epsilon}_j^2\|f_j\|_{{\mathcal h}_j}+(c_2/2)\tau e^{-n } \biggr).\nonumber\end{aligned}\ ] ] since @xmath497 [ see the comment after the definition of @xmath486 ] , we have @xmath499 where we also used the assumptions that @xmath500 and @xmath289 . substituting this in ( [ eq : riskbddom2 ] ) and then combining the resulting bound with ( [ eq : riskbddom1 ] ) concludes the proof of ( [ main_oracle ] ) in the case when conditions ( [ case_1 ] ) hold . it remains to consider the case when ( [ case_1 ] ) does not hold . the main idea is to show that in this case the right - hand side of the oracle inequality is rather large while we still can control the left - hand side , so , the inequality becomes trivial . to this end , note that , by the definition of @xmath41 , for some numerical constant @xmath411 , @xmath501 [ since the value of the penalized empirical risk at @xmath41 is not larger than its value at @xmath502 and , by the assumptions on the loss , @xmath503 is uniformly bounded by a numerical constant ] . the last equation implies that , on the event @xmath456 defined earlier in the proof [ see ( [ condition_01 ] ) , ( [ condition_02 ] ) ] , the following bound holds : @xmath504 equivalently , @xmath505 as soon as @xmath506 , so that @xmath507 , we have @xmath508 note also that , by the assumptions on the loss function , @xmath509 where we used the lipschitz condition on @xmath205 , and also bound ( [ hatf ] ) and the fact that @xmath510 ( by its definition ) . recall that we are considering the case when ( [ case_1 ] ) does not hold . we will consider two cases : ( a ) when @xmath511 , where @xmath512 is a numerical constant , and ( b ) when @xmath513 . the first case is very simple since @xmath202 and @xmath119 are both upper bounded by a numerical constant ( recall the assumption @xmath338 ) . in this case , @xmath497 is bounded from below by a numerical constant . as a consequence of these observations , bounds ( [ hatf ] ) and ( [ excess_a ] ) imply that @xmath514 & & \qquad\leq c_2 \tau^2 \beta_b^2(j_f)\vspace*{-2pt}\end{aligned}\ ] ] for some numerical constant @xmath515 . in the case ( b ) , we have @xmath516 and , in view of ( [ hatf ] ) , this implies @xmath517 so , either we have @xmath518 or @xmath519 moreover , in the second case , we also have @xmath520 & \geq&(e^n/4)\sqrt { \frac { a\log n}{n}}.\vspace*{-2pt}\end{aligned}\ ] ] in both cases we can conclude that , under the assumption that @xmath521 and @xmath513 for a sufficiently large numerical constant @xmath522 , @xmath523 & & \qquad\leq c_1 + 2c_1 c^2 l_{\ast } \frac{1}{\tau}\sqrt{\frac{n}{a\log n}}+ 2c_1 c^2\\ & & \qquad\leq\frac{\tau^2 e^n}{4}\sqrt{\frac{a\log n}{n}}\leq\tau^2 \sum_{j\in j_f } \bar\epsilon_j^2 \|f_j\|_{\mathcal{h}_j}.\end{aligned}\ ] ] thus , in both cases ( a ) and ( b ) , the following bound holds : @xmath524\\[-8pt ] & & \qquad \leq c_2 \tau^2 \biggl(\sum_{j\in j_f } \bar\epsilon_j^2 \|f_j\| _ { \mathcal { h}_j}+\beta_b^2(j_f ) \biggr).\nonumber\end{aligned}\ ] ] to complete the proof , observe that @xmath525 note also that , by the definition of @xmath486 , for all @xmath441,@xmath526\\[-8pt ] & & \qquad\leq \tau\beta_b(j_f)\|\hat f - f\|_{l_2(\pi)}+ \tau\beta_b(j_f)\sqrt{\frac{n}{a\log n}}\sum_{j\notin j_f}\bar \varepsilon_j \|\hat f_j\|_{l_2(\pi ) } \nonumber\\ & & \qquad\leq \tau\beta_b(j_f)\|\hat f - f\|_{l_2(\pi)}+ \tau\beta_b(j_f ) \frac{2c_1c^2}{\tau } \sqrt{\frac{n}{a\log n}},\nonumber\end{aligned}\ ] ] where we used the fact that , for all @xmath158 , @xmath527 and also bound ( [ hatf ] ) . by an argument similar to ( [ quadr])([eq : riskbddom1 ] ) , it is easy to deduce from the last bound that @xmath528\\[-9pt ] & & { } + \frac{2c_1 ^ 2c^4}{\tau^2}\frac{n}{a\log n}.\nonumber\end{aligned}\ ] ] substituting this in bound ( [ konec1 ] ) , we get @xmath529 & & \qquad \leq c_2 \tau^2 \biggl(\sum_{j\in j_f } \bar\epsilon_j^2 \|f_j\| _ { \mathcal{h}_j}+ \beta_b^2(j_f ) \biggr ) \nonumber\\[-2pt ] & & \qquad\quad { } + \frac{3}{2}\frac{c_2 ^ 2 \tau^2}{m_{\ast } } \beta_b^2(j_f ) + \frac{1}{2}\mathcal{e}(\ell\circ f)+ \frac{2c_1 ^ 2c^4}{\tau^2}\frac{n}{a\log n } \\[-2pt ] & & \qquad \leq \frac{1}{2}\mathcal{e}(\ell\circ f)+ c_2^{\prime } \tau^2 \biggl(\sum_{j\in j_f } \bar\epsilon_j^2 \| f_j\| _ { \mathcal{h}_j}+ \frac{\beta_b^2(j_f)}{m_{\ast } } \biggr)\nonumber\\[-2pt ] & & \qquad\quad{}+\frac{2c_1 ^ 2c^2}{\tau^2}\frac{n}{a\log n},\nonumber\end{aligned}\ ] ] with some numerical constant @xmath530 . it is enough now to observe [ considering again the cases ( a ) and ( b ) , as it was done before ] , that either the last term is upper bounded by @xmath531 , or it is upper bounded by @xmath532 , to complete the proof . now , to derive theorem [ co : homobd ] , it is enough to check that , for a numerical constant @xmath30 , @xmath533 & \leq & c \biggl(\sum_{j\in j_{f}}\breve\epsilon_j^2 \biggr)^{1/2 } \beta_{2,\infty}(j_f),\end{aligned}\ ] ] which easily follows from the definitions of @xmath534 and @xmath535 . similarly , the proof of theorem [ co : homobd-1 ] follows from the fact that , under the assumption that @xmath536 we have @xmath537 where @xmath538 , @xmath539 being a numerical constant . this easily implies the bound @xmath540 where @xmath411 is a numerical constant . we now proceed to prove lemma [ th : uniep ] that was used to bound @xmath466 . to this end , we begin with a fixed pair @xmath541 . throughout the proof , we write @xmath542 . by talagrand s concentration inequality , with probability at least @xmath64 @xmath543 & & \qquad\le 2 \biggl({\mathbb{e}}\bigl[{\sup_{g\in{\mathcal g}(\delta_- , \delta_+,r ) } } & & \qquad\quad\hspace*{9.3pt}{}+\|\ell\circ g-\ell\circ f\|_{l_2(p ) } \sqrt{t\over n}+\|\ell\circ g-\ell\circ f\|_{l_{\infty}}{t\over n } \biggr).\end{aligned}\ ] ] now note that @xmath544 & \le & l_{\ast}\sum_{j=1}^n\|g_j - f_j\|_{l_2(\pi)}\\[-3pt ] & \le & l_{\ast } \bigl(\min_j\bar{\epsilon}_j \bigr)^{-1 } \sum_{j=1}^n\bar{\epsilon}_j\|g_j - f_j\|_{l_2(\pi)},\end{aligned}\ ] ] where we used the fact that the lipschitz constant of the loss @xmath205 on the range of functions from @xmath545 is bounded by @xmath214 . together with the fact that @xmath546 for all @xmath158 , this yields @xmath547 furthermore , @xmath548 & \le & l_{\ast}\sum_{j=1}^n \|g_j - f_j\|_{\mathcal{h}_j}\\[-3pt ] & \le & l_{\ast } \frac{n}{a\log n}\delta_{+}.\end{aligned}\ ] ] in summary , we have @xmath549 & & \qquad\le 2 \biggl({\mathbb{e}}\bigl[{\sup_{g\in{\mathcal g}(\delta_- , \delta_+,r ) } } \bigr]\\[-3pt ] & & \qquad\quad\hspace*{27pt}{}+ l_{\ast}\delta_-\sqrt{t\over a\log n}+l_{\ast}\delta _ + { t\over n}{n\over a\log n } \biggr).\end{aligned}\ ] ] now , by symmetrization inequality , @xmath550\nonumber\\[-8pt]\\[-8pt ] & & \qquad\le{2{\mathbb{e}}\sup_{g\in{\mathcal g}(\delta _ - , \delta_+,r ) } } |r_n(\ell\circ g-\ell\circ f ) |.\nonumber\end{aligned}\ ] ] an application of rademacher contraction inequality further yields @xmath551\nonumber\\[-8pt]\\[-8pt ] & & \qquad\le cl_{\ast}{\mathbb{e}}\sup_{g\in{\mathcal g}(\delta_- , \delta_+,r ) } numerical constant [ again , it was used here that the lipschitz constant of the loss @xmath205 on the range of functions from @xmath545 is bounded by @xmath214 ] . applying talagrand s concentration inequality another time , we get that with probability at least @xmath64 @xmath553 for some numerical constant @xmath63 . recalling the definition of @xmath554 , we get @xmath555 hence , with probability at least @xmath556 and with some numerical constant @xmath63 @xmath557 using ( [ conclusion_b ] ) , @xmath558 can be upper bounded by @xmath559 with some numerical constant @xmath30 on an event @xmath456 of probability at least @xmath434 . therefore , the following bound is obtained : @xmath560 it holds on the event @xmath561 , where @xmath562 . we will now choose @xmath563 and obtain a bound that holds uniformly over @xmath564 to this end , consider @xmath565 for any @xmath566 and @xmath567 satisfying ( [ eq : bddel ] ) , we have @xmath568 on the event @xmath569 . therefore , simultaneously for all @xmath566 and @xmath567 satisfying ( [ eq : bddel ] ) , we have @xmath570 on the event @xmath571 . the last intersection is over all @xmath572 such that conditions ( [ eq : bddel ] ) hold for @xmath573 . the number of the events in this intersection is bounded by @xmath574 . therefore , @xmath575 using monotonicity of the functions of @xmath576 involved in the inequalities , the bounds can be extended to the whole range of values of @xmath576 satisfying ( [ eq : bddel ] ) , so , with probability at least @xmath472 we have for all such @xmath576 @xmath577 if @xmath578 , or @xmath579 , it follows by monotonicity of the left - hand side that with the same probability @xmath580 which completes the proof . the authors are thankful to the referees for a number of helpful suggestions . the first author is thankful to evarist gin for useful conversations about the paper . koltchinskii , v. ( 2008 ) . oracle inequalities in empirical risk minimization and sparse recovery problems . ecole det de probabilits de saint - flour , lecture notes . koltchinskii , v. ( 2009a ) . sparsity in penalized empirical risk minimization . h. poincar probab . statist . _ * 45 * 757 . raskutti , g. , wainwright , m. and yu , b. ( 2009 ) . lower bounds on minimax rates for nonparametric regression with additive sparsity and smoothness . in _ advances in neural information processing systems _ ( nips 22 ) 15631570 . curran associates , red hook , ny . ravikumar , p. , liu , h. , lafferty , j. and wasserman , l. ( 2008 ) . spam : sparse additive models . in _ advances in neural information processing systems _ ( nips 20 ) 12011208 . curran associates , red hook , ny .

using the stellar population synthesis , we model the stellar contribution for a sample of 110 double - peaked broad - lines agns from the sloan digital sky survey ( sdss ) . the stellar velocity dispersions ( ) are obtained for 52 double - peaked agns with obvious stellar absorption features , ranging from 106 to 284 ^-1 km s . we also use multi - component profiles to fit$] and h emission lines . using the well - established relation , the black hole masses are calculated to range from to , and the eddington ratio from about 0.01 to about 1 . comparing with the known relation , we can get the factor , which indicates blrs geometry , inclination and kinematics . we find that far deviates from 0.75 , suggesting the non - virial dynamics of broad line regions . the peak separation is mildly correlated with the eddington ratio and smbh mass with almost the same correlation coefficients . it implies that it is difficult to detect obvious double - peak agns with higher eddington ratios . using the monochromatic luminosity at 5100 to trace the bolometric luminosity , we find that the external illumination of the accretion disk is needed to produce the observed strength of h emission line .

0706.2473

double - peaked broad balmer emission profiles have been detected in about 200 active galactic nuclei ( agns ) ( eracleous & halpern 1994 , 2003 ; strateva et al . 2003 , 2006 ) . a systematic survey of 110 sources ( mostly broad - line radio loud agns at @xmath12 ) by eracleous & halpern ( 1994 , 2003 ) suggested that about 20% sources show double - peaked broad balmer lines . they found some characteristics of double - peaked agns : large full width at half maximum ( fwhm ) of h@xmath11 line , ranging from about 4000 ^-1 km s@xmath1 up to 40000 ^-1 km s@xmath1 ( e.g. wang et al . 2005 ) ; about 50% starlight contribution in optical continuum around h@xmath11 ; large equivalent widths of low - ionization forbidden lines ; large [ o i]/@xmath2 $ ] ratios . strateva et al . ( 2003 ) found about 3% of the @xmath13 agns in the sloan digital sky survey ( sdss ) are double - peaked agns , the fraction is smaller than that in radio - loud agns sample . double - peaked lines have also been detected in some low - ionization nuclear emission line regions ( liners ) , e.g. ngc1097 , m81 , ngc4450 , ngc4203 , ngc4579 ( storchi - bergmann et al . 1993 ; bower et al . 1996 ; shields et al . 2000 ; ho et al . 2000 ; barth et al . it remains a matter as debate of the origin of the double - peaked profiles . there are mainly three models to interpret the origin of the double - peaked profiles : the accretion disk ( chen & halpern 1989 ; eracleous & halpern 1994 , 2003 ; gezari et al . 2007 ) ; the biconical outflow ( zheng et al . 1991 ; abajas et al . 2006 ) , and an anisotropic illuminated blrs ( goad & wanders 1996 ) . more recently , observational test and physical consideration preferentially suggested that double - peaked profiles originate from the accretion disk within a radius from a few hundreds @xmath14 to about thousands @xmath14 ( @xmath15 , @xmath16 is the smbh mass ) ( eracleous & halpern 1994 , 2003 ; eracleous et al . 1997 ; strateva et al . 2003 , 2006 ; gezari , et al . 2007 ) . in order to interpret the sparsity of double - peaked agns , the origin of the single - peaked lines from accretion disk have been discussed ( eracleous & halpern 2003 and the references therein ) : larger out radius of the line - emitting accretion disk ; face - on accretion disk ; and the accretion disk wind . very few of the double - peaked high - ionization line profiles ( e.g. civ ) is due to that these high ionization lines are thought to arise in a wind , not in the disk . the masses of central supermassive black holes ( smbhs ) can provide an important tool to understand the physics of double - peaked agns if we reliably estimate them ( e.g. lewis & eracleous 2006 ; lewis 2006 ) . during the last decade , there is a striking progress on the study of central supermassive black holes . the stellar and/or gaseous dynamics is used to derive the smbhs masses in nearby inactive galaxies . however , for agns , this method is very difficult because nuclei outshine their hosts . fortunately , we can use the broad emission lines from blrs ( e.g. h@xmath4 , h@xmath11 , mg @xmath17 , c@xmath18 ) to estimate smbh masses in agns by the reverberation mapping method and the empirical size - luminosity relation ( kaspi et al . 2000,2005 ; vestergaard 2002 ; mclure & jarvis 2002 ; wu et al . 2004 ; greene & ho 2006a ) . there is a scaling factor with larger uncertainty , which is due to the unknown geometry and dynamics of broad line regions , blrs ( e.g. krolik 2001 ; collin et al . nearby galaxies and agns seem to follow the same tight @xmath19 relation , where @xmath20 is the bulge velocity dispersion at eighth of the effective radius of the galaxy ( nelson et al . 2001 ; tremaine et al . 2002 ; greene & ho 2006a , 2006b ) , although it remains controversial for narrow - line seyfert 1 galaxies ( nls1s ) ( mathur et al . 2001 ; bian & zhao 2004 ; grupe & mathur 2004 ; watson et al . lewis & eracleous ( 2006 ) derived the black hole masses from @xmath0 through fitting absorption lines of the ca ii triplet ( @xmath21 ) for 5 double - peaked agns . therefore the determination of the black hole mass from independent method of reverberation mapping is an useful prober to explore mysteries of the double peaked agns . significant stellar contribution in double - peaked agns makes the measurement of @xmath0 possible and reliable . here we present our results on the @xmath0 measurements for the sample of 110 double - peaked agns from the sloan digital sky survey ( sdss ) ( strateva et al . 2003 ) . in section 2 , we introduce our working sample selected from strateva et al . section 3 is data analysis . we present the calculations of the smbh mass and the eddington ratio , and discuss their errors in section 4 . section 5 is contributed to the discuss of the blrs in double - peaked agns . section 6 is the relation between the peak separation and mass / eddington ratio . section 7 is the energy budget for double - peaked agns . our conclusion is given in section 8 . the last section is our conclusion . all of the cosmological calculations in this paper assume @xmath22 , @xmath23 , and @xmath24 . strateva et al . ( 2003 ) presented a sample of double - peaked agns ( @xmath13 ) selected from sdss . they used two steps to select candidates : ( 1 ) they selected the unusual ones from the symmetric lines using the spectral principal component analysis ( pca ) ( hao et al . 2003 ) ; and ( 2 ) they fitted the h@xmath11 region with a combination of several gaussians , and only selected agns better fitted by two gaussians . from the profiles of the broad components , several fitting parameters including the positions of the red and blue peaks ( @xmath25 , @xmath26 ) , the corresponding peak heights ( @xmath27 , @xmath28 ) , fwhms were given in their table 3 . they investigated the multi - wavelength properties of these double - peaked agns and suggested that eddington ratios could be large in these sdss double - peaked agns ( strateva et al . 2006 ) . sdss spectra cover the wavelength range of 3800 - 9200 with a spectral resolution of @xmath29 . the fiber diameter in the sdss spectroscopic survey is 3 " on the sky . at the mean redshift of 0.24 in the sample of strateva et al . ( 2003 ) , the projected fiber aperture diameter is 13.2 kpc , typically containing about 80% of the total host galaxy light ( kauffmann & heckman 2005 ) . the stellar absorption features in these sdss spectra provide us the possibility to measure the stellar velocity dispersion . we did not apply aperture corrections to the stellar velocity dispersions because this effect can be omitted for @xmath30 ( bernardi et al . 2003 ; bian et al . 2006 ) . there are a number of objective and accurate methods to measure @xmath20 , including two main different techniques , the `` fourier - fitting '' method ( sargent et al . 1977 ; tonry & davis 1979 ) , and the `` direct - fitting '' method ( rix & white 1992 ; greene & ho 2006b and reference therein ) . with the development of computing , the `` direct - fitting '' method become much more popular . for sdss spectra with significant stellar absorption features ( such as ca h+k @xmath31 3969 , 3934 , mg ib@xmath32 5167 , 5173 , 5184 triplet , and ca ii@xmath31 8498 , 8542 , 8662 triplet , etc . ) , @xmath20 can be measured by matching the observed spectra with the combination of different stellar template spectra broadened by a gaussian kernel ( e.g. kauffmann et al . 2003 ; cid fernandes et al . 2004 ; vanden berk et al 2006 ; greene & ho 2006b ; zhou et al . 2006 ; bian et al . the smbh mass can then be estimated from the @xmath33 relation if @xmath20 can be accurately measured from the spectrum of agn host galaxy . here we briefly introduce the method to measure @xmath20 . adopting the stellar library from bruzual & charlot ( 2003 ) , we used the stellar population synthesis code , starlight , ( cid fernandes et al . 2004 , 2005 ; bian et al . 2006 ) to model the observed spectrum @xmath34 . it models the spectrum @xmath35 by the linear combination of simple stellar populations ( ssp ) . the model spectrum @xmath35 is : @xmath36 \otimes g(v_*,\sigma_*)\ ] ] where @xmath37 is the @xmath38 template normalized at @xmath39 , @xmath40 is the population vector , @xmath41 is the synthetic flux at the normalization wavelength , @xmath42 is the reddening term , @xmath43 is the @xmath44-band extinction , and @xmath45 is the line - of - sight stellar velocity distribution , modelled as a gaussian centered at velocity @xmath46 and broadened by @xmath0 . the best fit is reached by minimizing @xmath47 , @xmath48 ^ 2\ ] ] where the weighted spectrum @xmath49 is defined as the inverse of the noise in observed spectra . we can obtained some parameters including @xmath0 by modelling the observed spectrum , which are important for our understanding the stellar populations in agns host galaxies . we download 110 double - peaked spectra from sdss dr5 in the sample of strateva et al . ( 2003 ) , as well as the extinction values at @xmath50 band . we perform the galactic extinction correction to observed spectra by using the extinction law of cardelli , clayton & mathis ( 1989 ) . then , the rest - frame spectra with errors and masks are obtained by use of the redshift given in their sdss fits headers . we normalize the spectra at 4020 and the signal - to - noise ratio ( s / n ) is measured between 4010 and 4060 . an additional power - law component is used to represent the agns continuum emission . we check visually our modelled spectra sorted by the equivalent width ( ew ) of caii k @xmath51 line . it is found that the fitting goodness depends on the s / n ( @xmath52 ) , the absorption lines equivalent widths ( ew ( caii k ) @xmath53 1.5 ) , and the contribution of stellar lights ( see also zhou et al . 2006 ) . at last , 54 double - peaked agns are selected , which are well fitted to derive reliable @xmath0 from their significant stellar absorption . using the two sample kolmogorov - smirnov ( k - s ) test , _ kolmov _ task in iraf , the distributions of the luminosity at 5100 in the total sample of strateva et al . ( 2003 ) and our sub - sample are drawn from the same parent population with the probability of 53% . therefore , this sub - sample is representative of the total sample of strateva et al . ( 2003 ) with respect to the luminosity at 5100 . 1 shows a fitting example for sdss j082113.71 + 350305.02 . the final results are presented in table 1 , arranging in order of increasing right ascension . using interactive data language ( idl ) , we also carefully model the profiles of @xmath2 $ ] and h@xmath4 lines . since the double - peaked profile of the h@xmath4 line and the asymmetric profiles of @xmath2$]@xmath544959 , 5007 lines , seven - gaussian profiles are used to model these lines carefully , two broad and one narrow components for h@xmath4 plus two sets of one broad and one narrow components for @xmath2$]@xmath55 . we take the same linewidth for each component of @xmath2$]@xmath55 , fix the flux ratio of @xmath2$]@xmath564959 to @xmath2$]@xmath565007 to be 1:3 , and set their wavelength separation to the laboratory value . and we also set the wavelength separation between the narrow component of h@xmath4 and the narrow @xmath2$]@xmath565007 to the laboratory value . we do the lines fitting for these 54 stellar - light subtracted spectra . for spectra without obvious stellar features , we do the emission lines fitting for the extinction - corrected rest - frame sdss spectra without stellar - light subtraction . at last we obtain the gaseous velocity dispersion , @xmath57 , from the core @xmath2$]@xmath58 line , the total flux of @xmath2$]@xmath58 line and the monochromatic flux at 5100 . our emission - line profile fitting for sdss j022014.57 - 072859.30 is shown in fig . from the partial sdss spectra with that for a sample of the local agns presented by greene & ho ( 2006a ) . the solid line denotes our best linear fit , the red dash lines are @xmath59 ^-1 km s@xmath1 . considering the error bar of @xmath0 presented by greeene & ho ( 2006a ) , most of local agns are located between two red dash lines.,width=340 ] in the synthesis , we focus on the strongest stellar absorption features , such as caii k , g - band , and ca ii@xmath31 8498 , 8542 , 8662 triplet , which are less affected by emission lines . we put twice more weight for these features during the stellar population synthesis . after correcting the template and the sdss instrumental resolution , we obtain the value of @xmath0 through the direct - fitting method . cid fernandes et al . ( 2005 ) applied the same synthesis method to a larger sample of 50,362 normal galaxies from the sdss data release 2 ( dr2 ) . their @xmath20 is consistent very well with that of the mpa / jhu group ( kauffmann et al . 2003 ) , the median of the difference is just 9 ^-1 km s@xmath1 . they also gave the @xmath0 uncertainty based on the s / n ( see table 1 in cid fernandes et al . 2005 ) . typically , the @xmath0 uncertainty based on s / n at 4020 is about : 24 ^-1 km s@xmath1 at s / n=5 ; 12 ^-1 km s@xmath1 at s / n=10 ; 8 ^-1 km s@xmath1 at s / n=15 . in order to use the typical errors suggested by cid fernandes et al . ( 2005 ) , we calculate the s / n and the starlight fraction at 4020 . the s / n is the mean flux divided by the root mean square ( rms ) of the flux in the range 4010 to 4060 . we also performed the method of cid fernandes et al . ( 2005 ) to compute the s / n by using the sdss error spectrum , the results are almost the same . considering the contribution from featureless continuum ( fc , represented by a power law ) , we use an effective s / n to show the typical error of @xmath0 , where the effective s / n is roughly the s / n multiplied by the stellar fraction . we use the featureless continuum fraction as the up limit of nuclei contribution , because the featureless continuum can be attributed either by a young dusty starburst , by an agn , or by these two combination ( cid fernandes et al . therefore , the effective s / n is the low limit . in our sample , the effective mean spectral s / n at 4020 for these objects are 7.2 ( see table 1 ) . thus the typical uncertainty in @xmath20 should be around 20 ^-1 km s@xmath1 . for 13 objects with effective s / n less than 5 , their @xmath0 in table 1 are preceded by colons . here , we also apply this synthesis method to a sample of the local agns presented by greene & ho ( 2006a ) . greene & ho ( 2006a ; 2006b ) performed a research on the systematic bias of @xmath20 derived from the regions around caii triplet , mgib triplet , and caii h+k , respectively ( barth et al . 2002 ) . they argued that the caii triplet provide the most reliable measurements of @xmath20 and there is a systematic offset between @xmath20 from caii k line and that from other spectral regions . we use our synthesis method to their sample in two manners : one is using whole spectrum , the other is just using partial spectrum between 3200 and 7500 at the rest frame . we put twice more weight for the strongest absorption features of ca h+k @xmath31 3969 , 3934 , g - band , and ca ii@xmath31 8498 , 8542 , 8662 triplet . we find that the values of @xmath0 in these two manners are similar by performing our synthesis method . by using the least - square regression , the best fit between the @xmath0 from these two manners ( @xmath60 and @xmath61 , respectively ) is : @xmath62 . the spearman coefficient @xmath63 is 0.97 , with a probability of @xmath64 for rejecting the null hypothesis of no correlation . because the sdss spectral coverage is from 3800 to 9200 , most of double - peaked agns with redshift larger than 0.083 will not cover the range of caii triplet . for those sources with redshift larger than 0.083 , we used the above formula to obtain the corrected velocity dispersion , i.e. @xmath65 . the corrected velocity dispersion is listed in col . ( 8) in table 1 . and we find that the corrected velocity dispersion is almost the same to the uncorrected one . the largest difference is about 10 ^-1 km s@xmath1 , which is less than the typical error of 20 ^ -1 km s@xmath1 . based on the sdss instrumental resolution , we take 60 ^-1 km s@xmath1 as a lower limit of @xmath0 ( e.g. bernardi 2003 ) . only for two objects , sdss j133433.24 -013825.41 and sdss j214555.03 + 121034.17 , the measured @xmath0 are below / near the lower limit ( 51 ^-1 km s@xmath1 and 62 ^-1 km s@xmath1 , respectively ) . these two objects are excluded from further analysis . the @xmath0 value from the partial sdss spectra is also used to compare our result with greene & ho ( 2006a ) , who fitted the region around caii triplet directly . in fig . 3 , we compared our @xmath0 with theirs . we found that the agreement is quite good , the @xmath0 difference ( @xmath66 ) distribution is -2.1 ^-1 km s@xmath1 with a standard deviation ( sd ) of 22.7 ^-1 km s@xmath1 . by using the least - square regression considering the errors of @xmath67 , the best fit between @xmath68 and @xmath61 is : @xmath69 ( solid line in fig . the spearman coefficient @xmath63 is 0.86 , the standard deviation is 2.11 , with @xmath70 . for these double - peaked agns with reliable @xmath71 , we use the @xmath19 relation to derive the smbh mass ( e.g. tremaine et al . 2002 and reference therein ) , @xmath72 we calculate the eddington ratio , e.g. , the ratio of the bolometric luminosity ( @xmath73 ) to the eddington luminosity ( @xmath74 ) , where @xmath75 . we use the monochromatic luminosity at 5100 ( @xmath76 at 5100 ) to estimate the bolometric luminosity , @xmath77(5100 ) , where @xmath78 ( kaspi et al . these results are presented in table 1 . for the typical uncertainty of 20 ^-1 km s@xmath1 for @xmath79 ^-1 km s@xmath1 , the error of @xmath80 would be about 0.05 dex , corresponding to 0.2 dex for @xmath81 . the error of @xmath82 is about 0.4 considering the error of 0.3 dex form the @xmath19 relation ( tremaine et al . richards et al . ( 2006 ) suggested a bolometric correction factor of @xmath83 at 5100 . therefore , the final eddington ratio , @xmath84 , has a large uncertainty , about 0.5 dex . .,width=340 ] in fig . 4 , we show the histograms of the black hole masses and the accretion ratios for these 52 double - peaked agns . the black hole masses range from @xmath6 to @xmath7 @xmath8 with a mean value of @xmath85 . using the h@xmath11 fwhm and the 5100 monochromatic luminosity , wu & liu ( 2004 ) estimated the smbh masses and the eddington ratios for an assembled double - peaked agns sample . lewis & eracleous ( 2006 ) noted that the bh masses from the h@xmath11 fwhm are not completely consistent with those from the stellar velocity dispersion . our bh masses derived from @xmath86 are indeed smaller than those from the h@xmath11 fwhm ( wu & liu 2004 ) by about an order of magnitude . this will lead to our @xmath84 larger than that from the h@xmath11 fwhm by almost an order of magnitude . = 0.75 for random blrs orbits ) . objects with effective s / n in 4020 less then 5 is denoted as open stars . the black dash line is the empirical size - luminosity relation , @xmath87(5100)@xmath88^{0.69 } ~\rm lt - days$ ] , found by kaspi et al . the black dot line is the empirical size - luminosity relation,@xmath89(5100)@xmath90 , found by bentz et al . the red solid line is our best fit with the fixed slope of 0.69 ( @xmath63=0.29 ) , lower by -0.63 dex respect to dash line.,width=340,height=264 ] the eddington ratio has a distribution with mean and the standard deviation of @xmath91 is @xmath92 . it is suggested that the accretion disk in double - peaked agns is in the mode of advection dominated accretion flow ( adaf ) ( e.g. eracleous & halpern 2003 ) . when the eddington ratio is below than the critical one @xmath93 ( mahadevan 1997 ) , the adaf appears , where @xmath94 is viscous coefficient . for our sdss sample , all objects have eddington ratios larger than this critical value of 0.0028 . the present results from fig . 4 clearly show that these double peaked agns have accretion disks in the standard regime . the black hole masses can be independently tested by the relation between the black holes and bulges . the appendix gives details of the test . we find the black hole masses are consistent from @xmath95 and @xmath5 relations . we also calculate the blrs sizes for these 52 double - peaked agns using the smbhs masses derived from @xmath96 and the h@xmath11 fwhm . we firstly transform the h@xmath11 fwhm to the h@xmath4 fwhm by ( greene & ho 2005b ) : @xmath97 from the smbh masses derived from the velocity dispersions , we can calculate the blrs sizes : @xmath98 where @xmath10 is the scaling factor related to the kinematics and geometry of the blrs , defined by @xmath99 , @xmath100 is the keplerian velocity in disk plane . for random orietation of blr cloud keplerian orbits , @xmath10 is 0.75 . in fig . 5 , we plot the blrs sizes versus the monochromatic luminosity at 5100 ( assuming @xmath10=0.75 for random blrs orbits ) . almost all objects are located below the empirical size - luminosity relation ( black dash line ) found by kaspi et al . the correlation between the blrs sizes and the monochromatic luminosity at 5100 is not too strong . the best fit is shown as red solid line in fig . 5 as @xmath101^{0.69}$ ] lt - days ( @xmath102 , @xmath103 ) , lower by -0.64 dex respecting to the empirical relation found by kaspi et al . ( 2005 ) . when excluding objects with effective s / n in 4020 less then 5 ( open stars in fig . 5 ) , for fixed slope of 0.69 , the best fit gives almost the same line but with a smaller @xmath63 of @xmath104 . if we used @xmath10=0.52 ( table 2 in collin et al . 2006 ) , @xmath105 will increased by 0.16 dex , the blrs sizes of double - peaked agns are still deviated from the empirical relation found by kaspi et al . ( 2005 ) ( about 0.48 dex ) . fwhms of the double - peaked agns are about twice as broad as other agn of similar luminosity ( eracleous & halpern 1994 , 2003 ; strateva et al . if the double - peaked agns are not systematically more massive than other agns , equation 5 suggested that blrs radii can be smaller than for a similarly massive `` normal '' agns by a factor of 4 . we have to point out that here we are not really testing the @xmath9 relation in double peaked agns , but the comparison with @xmath9 relation allows us to derive the factor @xmath10 . if we use the empirical @xmath9 relation ( kaspi et al . 2005 ) to derive the @xmath106 and the smbh masses , @xmath107 , to do the calibration of the factor @xmath10 ( onken et al . 2004 ) , we find that the distribution of @xmath10 is 0.179 with a standard deviation of 0.171 . it is not consistent with the value of @xmath108 presented by onken et al . ( 2004 ) and it is about 1/3 of @xmath109 suggested by collin et al . our results suggest that the empirical relation between the blrs sizes and the luminosity at 5100 does not hold for double - peaked agns ( see also zhang et al . 2007 ) , otherwise the calibration factor @xmath10 should be as low as 0.179 . an important consequence of the breakdown in the size - luminosity relationship is that using mass estimation methods based on the size - luminosity relationship and the calibration factor of 0.75 can lead to an order of magnitude over - estimate of the smbh mass ( wu & liu 2004 ) . if blrs are disk - like with an inclination of @xmath110 , the relation between the h@xmath4 fwhm and the keplerian disk plane velocity , @xmath111 , is given by ( wills & browne 1986 ) @xmath112 where @xmath113 is the random isotropic component . we may derive the scaling factor as @xmath114 $ ] . ignoring @xmath113 , @xmath115 , and the minimum of @xmath10 is 0.25 . for only ten object with @xmath116 , we can derive the inclination of @xmath110 by the above formula . the mean inclination is 56 degrees . it is suggested that double - peaked agns are not preferentially edge - on ( eracleous & halpern 1994 , 2003 ; strateva et al . 2003 ) , most have an inclination of less than 50 degrees . if all these objects were nearly edge - on , the obscuring torus would prevent us from seeing the broad lines . collin et al . ( 2006 ) suggested that objects with large fwhms , inclination effects are not really important . in these 52 double - peaked agns , 42 objects have smaller @xmath10 less than 0.25 . @xmath113 ca nt be omitted , we need to consider the random isotropic component , implying the non - virial dynamics of blrs in double - peaked agns . if we use @xmath10 to trace the non - virial effect , smaller @xmath10 means strong non - virial effect . we find @xmath10 have strong correlations with @xmath81 and @xmath117 ( see fig . 6 ) . using the least - square regression , we derive the correlations are : @xmath118 where @xmath119 , @xmath64 ; and @xmath120 where @xmath121 with @xmath122 . when we exclude objects with effective s / n in 4020 less then 5 ( open stars in fig . 6 ) , the best fits for @xmath10 relations between smbh mass and the eddington ratio give a little larger @xmath63 ( @xmath123 , respectively ) . we should note that the reason of these strong correlations are that @xmath10 is derived from @xmath16 , @xmath106 [ from @xmath124 ] , and @xmath125 . these strong correlations suggest that objects with larger eddington ratios have strong non - virial effect . versus the eddington ratio ( left panel ) and the smbh masses ( right panel ) . objects with effective s / n in 4020 less then 5 is denoted as open stars . the dash lines are @xmath126 . the correlation coefficient for the relation between the peak separation and the eddington ratio is @xmath119 and that between the separation and the black hole mass is @xmath121.,width=340 ] 3c 390.3 is the only double peaked object with the direct measurements of the time delay and the host stellar velocity dispersion . and it locates in the empirical relation between blrs sizes and the luminosity ( e.g. kaspi et al . 2005 ) . by @xmath0 of 240 ^-1 km s@xmath1 ( onken et al . 2004 ) , we find that @xmath10 is about 1.3 , consistent with the assumption of random blrs orbits in 3c390.3 . its small eddington ratio ( lewis & eracleous 2006 ) , @xmath127 , is also consistent with the virial blrs dynamics ( see fig , which is almost the same for that between the separation and the black hole mass ( @xmath128).,width=340 ] for double - peaked agns , the peak separation , @xmath129 , has a large variance ranging from about 1000 ^-1 km s@xmath1 to about 10000 ^-1 km s@xmath1 ( table 3 in strateva et al . 2003 ) . in fig . 7 , we show the relations between @xmath130 and @xmath117 , @xmath82 . using the least - square regression , we derive the correlation between @xmath131 and @xmath132 to be : @xmath133 where @xmath134 , @xmath135 ( see fig . 7 ) . for the relation between @xmath130 and @xmath81 , @xmath136 where @xmath63 is 0.35 with @xmath137 . when excluding objects with effective s / n in 4020 less then 5 ( open stars in fig . 7 ) , the best fits for @xmath131 relations between smbh mass and the eddington ratio give smaller @xmath63 ( @xmath138 , respectively ) . we also do the multiple regression for the dependence of @xmath139 on @xmath117 and @xmath81 , @xmath140 where the @xmath141square correlation coefficient is 0.15 . wu & liu ( 2004 ) also studied this correlation and obtained an apparent stronger correlation ( @xmath142 ) . their derived strong correlation is mainly due to the very strong correlation between the separation and the h@xmath11 fwhm ( @xmath143 ) . when the h@xmath11 fwhm is fixed , they found that the partial correlation coefficient is only 0.05 . in this paper , we use the stellar velocity dispersion to estimate the black hole masses , and we also find a mild correlation between @xmath139 and @xmath144 , which implies that the peak separation would be smaller for agns with higher eddington ratios . it provides clues to why previous double - peaked agns have lower eddington ratios . there is increasing evidence for a disk geometry of the blr ( see a review of laor 2007 ) . if we assume that peak separation is created by the doppler shift of the movement of a thin annulus and the annulus radius corresponds the location where the self - gravitation domiantes ( bian & zhao 2002 ) , we have the radius @xmath145 ( their eq . 18 in laor & netzer 1989 ) , where @xmath146 cm , @xmath147 , @xmath148 is the eddington ratio and @xmath11 is the viscosity . the maximum separation of the double peaks in units of under the edge - on orientation to an observer is given @xmath149 where we use the peak separation in term of separation velocity , @xmath150 where @xmath94 and h@xmath11 wavelength @xmath151 . considering the uncertainties of the fitting results for the peak separation correlations with smbh mass and the eddington ratio , the slopes are consistent with the simple theoretical expectation . we have to stress that equation ( 11 ) is for the maximum separation ( edge - on orientation ) when the disk structure is given . more sophisticated model is needed for explanations of the dependence of peak separations and eddington ratios . the scatters in fig 7 may be caused by different orientation of the blr in the sample . we also find no significant correlations between the ratio of the red peak height to the blue height and the eddington ratio , the black hole mass , the peak separation . since the keplerian velocity is much below the light speed , the doppler boosting effects could be hidden by complex situations of the blr . here we discuss the energy budget of the line - emitting accretion disk . based on the standard accretion disk model , the disk radiation as a function of radius @xmath152 is ( chen et al . 1989 ) : @xmath153~ \rm ergs~ cm^{-2}~s^{-1}\ ] ] where @xmath147 , @xmath154 is the accretion rate in units of @xmath155 , and the dimensionless radius @xmath156 is @xmath157 . using @xmath158 where the efficiency @xmath159 is 0.1 , the gravitational power output of the line - emitting disk annulus between @xmath160 and @xmath161 is ( eracleous & halpern 1994 ) : @xmath162 { \rm ergs ~s^{-1}},\end{aligned}\ ] ] where @xmath73 is in units of @xmath163 . it is noted that @xmath164 is independent of the black hole mass , when we use the typical radius in units of @xmath165 . we use the luminosity at 5100 to calculate the bolometric luminosity . from the work of eracleous & halpern ( 2003 ) and strateva et al . ( 2003 ) , the inner radius is about hundreds of @xmath166 , and the outer radius is about thousands of @xmath166 . the outer radius of line - emitting accretion disk is about near the inner position of torus . we can assume typical inner and outer radii of @xmath167 and @xmath168 to calculate the energy output for these double - peaked sdss agns , i.e. @xmath169 . the distribution of @xmath170 is @xmath171 with the standard deviation of 0.57 ( see also fig . 8) . if assuming as much as @xmath172 of the power was radiated as h@xmath11 line ( dashed line in fig . 8) , our results show that only 36 out of 105 double - peaked agns would generate enough power to produce observed strength of h@xmath11 emission . if we adopt the value of 10% , more objects ( 83 out of 105 objects ) showed the energy problem . it implied that the majority of double - peaked agns need external illumination of the disk ( e.g. an inner iron torus or corona ) to produce the observed strength of h@xmath11 line ( strateva et al . 2006 ; cao & wang 2006 ) . we use the simple population synthesis to model the stellar contributions in double - peaked sdss agns . the reliable stellar velocity dispersions are obtained for 52 medium - luminous double - peaked sdss agns with obvious stellar features . we find that : 1 ) the black hole mass is from @xmath173 to @xmath174 and the eddington ratio is from about 0.01 to about 1 ; 2 ) the factor @xmath10 far deviates from the virialized value 0.75 , suggesting the non - virial dynamics of blrs ; 3 ) the peak separation is mildly correlated with the eddington ratio and smbh mass with almost the same correlation coefficients , which can be interpreted in the doppler shift of thin annulus of blrs created by gravitational instability ; 4 ) based on the line - emitting accretion disk model , we need external illumination of the accretion disk to produce the observed strength of h@xmath11 line . in the future , using different models , we would fit the double - peaked profiles to constrain the nature of double - peaked agns . we can also use the double - peaked agns to constrain the blrs origin ( nicastro 2000 ; laor 2003 ; bian & gu 2007 ) . we are very grateful to the anonymous referee and ari laor for their thoughtful and instructive comments which significantly improved the content of the paper . we thank luis c. ho for his very useful comments , and thank discussions among people in ihep agn group . this work has been supported by the nsfc ( nos . 10403005 , 10473005 ) , the science - technology key foundation from education department of p. r. china ( no . 206053 ) , and and the china postdoctoral science foundation ( no . 20060400502 ) . qsg would like to acknowledge the financial supports from china scholarship council ( csc ) and the nsfc under grants 10221001 and 10633040 . jmw thanks nsfc grants via no . 10325313 and 10521001 and supports from cas key project via kjcx2-yw - t03 . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the u.s . department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society , and the higher education funding council for england . the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , cambridge university , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . for the sample of 52 double - peaked sdss agns , assuming @xmath175 is translated to @xmath176 . the host luminosity in v band is the value of @xmath177 multiplied by the stellar fraction at 5530 . we assume that the host luminosity in v band approximates the bulge luminosity @xmath178 in v - band . we then use the following formula to calculate the bulge mass : @xmath179 ( magorrian et al . 9 shows this bulge mass versus the bh mass from @xmath0 . we fit with fixed slope as 1 , intercept is @xmath180 , correlation coefficient is 0.53 . the null hypothesis is less then @xmath181 . we find that the @xmath182 versus the @xmath183 relation is consistent with fig.4 of faber et al . therefore , the smbh mass from @xmath184 and the magorrian relation agrees with that from @xmath0 . when we exclude objects with effective s / n in 4020 less then 5 ( open stars in fig . 9 ) , for fixed slope of 1 , the best fit gives that intercept is @xmath185 , correlation coefficient is 0.49 . it is suggested that the mass from the @xmath19 relation could underestimate the smbh mass for the massive ellipticals with smbh mass larger than @xmath186 ( fig . 2 in lauer et al . 2007 ) . for our 52 double - peaked agns , their smbh masses are all less than @xmath186 from @xmath187 relation . the present results are less affected by lauer et al s findings . abazajian , k. , et al . 2005 , aj , 129 , 1755 barth , a. j. , ho , l. c. , filippenko , a. v. , rix , h. , sargent , w. l. w. 2001 , apj , 546 , 205 bentz et al . 2006 , apj , 644 , 133 bernardi , m. , et al . 2003 , aj , 125 , 1817 bian , w. , zhao , y. 2002 , a&a , 395 , 465 bian , w. , zhao , y. 2004 , mnras , 347 , 607 bian , w. , yuan , q. , zhao , y. 2005 , mnras , 364 , 187 bian , w. , gu , q , zhao . y , chao , l. , cui , q. 2006 , mnras , 372 , 876 bian , w. , gu , q. 2007 , apj , 657 , 159 bower , g. a. , wilson , a. s. , heckman , t. m. , richstone , d. o. 1996 , aj , 111 , 1901 bruzual , g. , charlot , s. 2003 , mnras , 344 , 1000 cao , x.w . , wang , t.g . 2006 , apj , 652 , 112 cardelli , j. a. , clayton , g. c. , mathis , j. s. 1989 , apj , 345 , 245 chen , k. , halpern , j. p. 1989 , apj , 344 , 115 cid fernandes r. , gu q. , melnick j. , et al . 2004 , mnras , 355 , 273 cid fernandes r. , mateus a. , sodre l. , stasinska g. , gomes j. 2005 , mnras , 358 , 363 collin , s. et al . 2006 , a&a , 456 , 75 dietrich , m. , et al . 1998 , apjs , 115 , 185 eracleous , m. , halpern , j. p. 1994 , apjs , 90 , 1 eracleous , m. , halpern , j. p. 2003 , apj , 599 , 886 eracleous , m. , et al . 1997 , apj , 490 , 216 gezari , s. , halpern , j. p. , eracleous , m. 2007 , apjs , 169 , 167 goad , m. , wanders , i. 1996 , apj , 469 , 113 greene , j. e. , ho , l. c. 2005a , apj , 627 , 721 greene , j. e. , ho , l. c. 2005b , apj , 630 , 122 greene , j. e. , ho , l. c. 2006a , apjl , 641 , l21 greene , j. e. , ho , l. c. 2006b , apj , 641 , 117 grupe , d. , mathur , s. 2004 , apj , 606 , l41 hao , l. , et al . 2003 , aj , 129 , 1783 heckman , t. m. , et al . 2004 , apj , 613 , 109 ho , l. c. , rudnick , g. , rix , h. , shields , j. c. , mcintosh , d. h. , filippenko , a. v. , sargent , w. l. w. , eracleous , m. 2000 , apj , 541 , 120 kaspi , s. , maoz , d. , netzer , h. , peterson , b.m . , vestergaard , m. , & jannuzi , b.t . 2005 , apj , 629 , 61 kaspi , s. , smith , p.s . , netzer , h. , maoz , d. , jannuzi , b.t . , giveon , u. 2000 , apj , 533 , 631 kauffmann , g. , et al . 2003 , mnras , 346 , 1055 kauffmann , g. , heckman , t. m. 2005 , philos . london , a363 , 621 lauer , t. r. , et al . 2007 , apj , 662 , 808 laor , a. 2003 , apj , 590 , 86 laor , a. 2007 , in proceedings of the central engine of active galactic nuclei , ed . l. c ho & j .- wang ( san francisco : asp ) in press ( astro - ph/0702577 ) laor , a. & netzer , h. 1989 , mnras , 238 , 897 kormendy , j. , & gebhardt , k. 2001 , in proc . 20th texas symposium , ed . h. martel & j. c. wheeler ( austin : aip ) , 363 krolik , j. h. 2001 , apj , 551 , 72 lewis , k. t. , eracleous , m. 2006 , apj , 642 , 711 lewis , k. t. 2006 , in asp conference series , the central engine of active galactic nuclei . mahadevan , r. 1997 , apj , 477 , 585 magorrian , j. , et al . 1998 , aj , 115 , 2285 mathur , s. , kuraszkiewicz , j. , czerny , b. 2001 , newa , 6 , 321 mclure , r. j. , jarvis , m. j. 2002 , mnras , 337 , 109 nelson , c. h. 2001 , apj , 544 , l91 nicastro , f. 2000 , apj , 530 , l65 onken , c. a. , et al . 2004 , apj , 615 , 645 peterson , b. m. apj , 2004 , 613 , 682 richards , g. t. , et al . 2006 , apjs , 166 , 470 rix , h. w. , white , s. d. m. 1992 , mnras , 254 , 389 sargent , w. l. w. , schechter , p. l. , boksenberg , a. , shortridge , k. 1977 , apj , 212 , 326 shields , j. c. , rix , h. , mcintosh , d. h. , ho , l. c. , rudnick , g. , filippenko , a. v. , sargent , w. l. w. , sarzi , m. 2000 , apj , 534 , l27 storchi - bergmann , t. , baldwin , j. a. , wilson , a. s. 1993 , apj , 410 , l11 strateva , i. v. , et al . 2003 , aj , 126 , 1720 strateva , i. v. , et al . 2006 , apj , 651 , 749 tonry , j. , & davis , m. 1979 , apj , 84 , 1511 tremaine , s. , et al . 2002 , ap j , 574 , 740 vanden berk , d. e. , et al . 2006 , aj , 131 , 84 vestergaard , m. 2002 , apj , 571 , 733 wang , t .- g . , dong , x .- b . , zhang , x .- , zhou , h .- y . , wang , j .- x . , lu , y .- 2005 , apj , 625 , l35 watson , l. c. , grupe , d. , mathur , s. 2007 , apj , 133 , 2435 wills , b.j . , browne , i.w.a . 1986 , apj , 302 , 56 woo , j. h. et al . 2006 , apj , 645 , 900 wu x .- b.,wang r. , kong m. z. , liu f. k. , han j. l. 2004 , a&a , 424 , 793 wu , x .- b . , liu , f. k. 2004 , apj , 614 , 91 zhang , x. g. , dultzin - hacyan d. , wang , t.g . 2007,mnras , 377 , 1215 zheng , w. , veilleux , s. , grandi , s. a. 1991 , apj , 381 , 418 zhou , h. y. et al . 2006 , apjs , 166 , 128 lllllllllllll name & z & ew(ca k ) & @xmath47 & fc & s / n & @xmath20 & @xmath188 & @xmath189 & @xmath124 & @xmath16 & @xmath190 & @xmath191 + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6)&(7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11)&(12)&(13 ) + sdss [email protected] & 0.199 & @xmath193 & 0.86 & 0.45 & 4.12 & : 148.30 & 146.37 & @xmath194 & 43.71 & 7.59 & -1.02 & 3100 + sdss [email protected] & 0.207 & @xmath195 & 0.91 & 0.58 & 3.49 & : 110.93 & 110.87 & @xmath196 & 43.77 & 7.10 & -0.48 & 2300 + sdss [email protected] & 0.070 & @xmath197 & 1.07 & 0.07 & 9.47 & 121.14 & 121.14@xmath198 & @xmath199 & 42.95 & 7.25 & -1.45 & 3800 + sdss [email protected] & 0.210 & @xmath200 & 1.05 & 0.64 & 2.78 & : 125.25 & 124.48 & @xmath201 & 43.67 & 7.30 & -0.77 & 2800 + sdss [email protected] & 0.265 & @xmath202 & 0.86 & 0.30 & 11.00 & 186.03 & 182.22 & @xmath203 & 43.97 & 7.97 & -1.14 & 3200 + sdss [email protected] & 0.247 & @xmath204 & 1.31 & 0.67 & 7.52 & 121.61 & 121.02 & @xmath205 & 44.22 & 7.25 & -0.18 & 3100 + sdss [email protected] & 0.333 & @xmath206 & 0.81 & 0.62 & 3.41 & : 293.48 & 284.29 & @xmath207 & 44.08 & 8.74 & -1.81 & 8100 + sdss [email protected] & 0.184 & @xmath208 & 0.72 & 0.38 & 6.26 & 155.34 & 153.06 & @xmath209 & 43.62 & 7.66 & -1.19 & 2700 + sdss [email protected] & 0.170 & @xmath211 & 1.01 & 0.50 & 6.81 & 174.65 & 171.41 & @xmath212 & 43.82 & 7.86 & -1.18 & 2600 + sdss [email protected] & 0.102 & @xmath213 & 0.90 & 0.62 & 4.51 & : 137.38 & 136.00 & @xmath214 & 43.28 & 7.46 & -1.33 & 3600 + sdss [email protected] & 0.287 & @xmath215 & 0.98 & 0.61 & 4.32 & : 148.31 & 146.38 & @xmath216 & 44.09 & 7.59 & -0.64 & 5300 + sdss [email protected] & 0.062 & @xmath217 & 1.33 & 0.16 & 9.35 & 172.92 & 172.92@xmath198 & @xmath218 & 43.08 & 7.88 & -1.94 & 6300 + sdss [email protected] & 0.223 & @xmath219 & 1.06 & 0.61 & 4.84 & : 173.85 & 170.65 & @xmath220 & 43.97 & 7.85 & -1.03 & 5400 + sdss [email protected] & 0.128 & @xmath221 & 0.90 & 0.19 & 9.47 & 171.11 & 168.05 & @xmath222 & 43.58 & 7.83 & -1.39 & 5200 + sdss [email protected] & 0.260 & @xmath223 & 0.77 & 0.33 & 6.52 & 106.66 & 106.82 & @xmath224 & 43.82 & 7.04 & -0.36 & 5600 + sdss [email protected] & 0.198 & @xmath225 & 0.80 & 0.33 & 11.38 & 156.94 & 154.59 & @xmath226 & 43.97 & 7.68 & -0.86 & 7400 + sdss [email protected] & 0.186 & @xmath227 & 0.79 & 0.26 & 8.78 & 203.50 & 198.82 & @xmath228 & 43.85 & 8.12 & -1.42 & 3400 + sdss [email protected] & 0.199 & @xmath229 & 1.00 & 0.52 & 3.92 & : 172.93 & 169.77 & @xmath230 & 43.79 & 7.84 & -1.20 & 6600 + sdss [email protected] & 0.141 & @xmath231 & 1.50 & 0.26 & 12.21 & 221.51 & 215.92 & @xmath232 & 43.85 & 8.26 & -1.56 & 9000 + sdss [email protected] & 0.294 & @xmath233 & 0.82 & 0.55 & 5.30 & 179.60 & 176.11 & @xmath234 & 43.98 & 7.91 & -1.08 & 3700 + sdss [email protected] & 0.230 & @xmath235 & 1.04 & 0.43 & 6.20 & 205.50 & 200.72 & @xmath236 & 43.82 & 8.14 & -1.47 & 8500 + sdss [email protected] & 0.182 & @xmath237 & 0.91 & 0.27 & 9.75 & 185.68 & 181.88 & @xmath238 & 43.78 & 7.96 & -1.33 & 5700 + sdss [email protected] & 0.303 & @xmath239 & 1.41 & 0.47 & 10.42 & 179.70 & 176.20 & @xmath240 & 44.24 & 7.91 & -0.81 & 4500 + sdss [email protected] & 0.327 & @xmath241 & 0.99 & 0.42 & 6.12 & 205.09 & 200.32 & @xmath242 & 44.18 & 8.13 & -1.10 & 3700 + sdss [email protected] & 0.132 & @xmath243 & 0.98 & 0.31 & 10.85 & 148.10 & 146.19 & @xmath244 & 43.73 & 7.58 & -1.00 & 2100 + sdss [email protected] & 0.239 & @xmath245 & 1.18 & 0.74 & 6.41 & 143.45 & 141.76 & @xmath246 & 44.22 & 7.53 & -0.45 & 2500 + sdss [email protected] & 0.132 & @xmath247 & 0.78 & 0.41 & 6.43 & 134.19 & 132.97 & @xmath248 & 43.50 & 7.42 & -1.06 & 9100 + sdss [email protected] & 0.149 & @xmath249 & 0.89 & 0.46 & 5.98 & 137.15 & 135.78 & @xmath250 & 43.69 & 7.45 & -0.91 & 4900 + sdss [email protected] & 0.288 & @xmath251 & 0.96 & 0.49 & 6.70 & 186.09 & 182.28 & @xmath252 & 44.02 & 7.97 & -1.09 & 3100 + sdss [email protected] & 0.267 & @xmath253 & 1.30 & 0.36 & 10.66 & 199.92 & 195.42 & @xmath254 & 43.95 & 8.09 & -1.28 & 3000 + sdss [email protected] & 0.116 & @xmath255 & 1.12 & 0.45 & 3.20 & : 145.40 & 143.62 & @xmath256 & 42.98 & 7.55 & -1.72 & 9700 + sdss [email protected] & 0.151 & @xmath257 & 1.41 & 0.60 & 4.70 & : 154.22 & 152.00 & @xmath258 & 43.64 & 7.65 & -1.16 & 3600 + sdss [email protected] & 0.217 & @xmath259 & 1.03 & 0.55 & 5.18 & 105.37 & 105.59 & @xmath260 & 43.71 & 7.01 & -0.45 & 3400 + sdss [email protected] & 0.202 & @xmath261 & 0.97 & 0.36 & 7.34 & 206.10 & 201.29 & @xmath262 & 43.82 & 8.14 & -1.47 & 4100 + sdss [email protected] & 0.116 & @xmath263 & 1.37 & 0.64 & 7.94 & 153.91 & 151.71 & @xmath264 & 43.73 & 7.65 & -1.07 & + sdss [email protected] & 0.331 & @xmath265 & 1.05 & 0.83 & 5.08 & 204.44 & 199.71 & @xmath266 & 44.59 & 8.13 & -0.69 & 5300 + sdss [email protected] & 0.269 & @xmath267 & 1.15 & 0.29 & 10.97 & 218.93 & 213.47 & @xmath268 & 44.03 & 8.24 & -1.36 & 2700 + sdss [email protected] & 0.158 & @xmath269 & 1.02 & 0.54 & 6.70 & 212.96 & 207.80 & @xmath270 & 43.81 & 8.20 & -1.54 & 5900 + sdss [email protected] & 0.148 & @xmath271 & 1.18 & 0.55 & 8.11 & 149.36 & 147.38 & @xmath272 & 43.85 & 7.60 & -0.90 & 2700 + sdss [email protected] & 0.145 & @xmath245 & 1.39 & 0.70 & 7.33 & 159.98 & 157.48 & @xmath273 & 43.86 & 7.71 & -0.99 & 4500 + sdss [email protected] & 0.175 & @xmath274 & 1.65 & 0.51 & 10.09 & 191.25 & 187.18 & @xmath275 & 44.00 & 8.01 & -1.17 & 3100 + sdss [email protected] & 0.268 & @xmath276 & 0.92 & 0.46 & 9.34 & 193.04 & 188.88 & @xmath277 & 44.12 & 8.03 & -1.05 & 2700 + sdss [email protected] & 0.094 & @xmath278 & 1.62 & 0.75 & 3.79 & : 127.55 & 126.66 & @xmath279 & 43.28 & 7.33 & -1.20 & 1600 + sdss [email protected] & 0.192 & @xmath280 & 0.78 & 0.35 & 4.45 & : 125.66 & 124.86 & @xmath281 & 43.60 & 7.31 & -0.85 & 3400 + sdss [email protected] & 0.179 & @xmath282 & 0.85 & 0.32 & 7.52 & 177.42 & 174.04 & @xmath283 & 43.75 & 7.89 & -1.29 & 3700 + sdss [email protected] & 0.239 & @xmath284 & 0.85 & 0.56 & 1.92 & : 178.69 & 175.24 & @xmath285 & 43.66 & 7.90 & -1.38 & 10800 + sdss [email protected] & 0.288 & @xmath286 & 0.84 & 0.56 & 5.01 & 190.63 & 186.59 & @xmath287 & 44.12 & 8.01 & -1.03 & 8200 + sdss [email protected] & 0.058 & @xmath288 & 1.68 & 0.37 & 10.05 & 121.19 & 121.19@xmath198 & @xmath289 & 43.09 & 7.26 & -1.31 & 2500 + sdss [email protected] & 0.174 & @xmath290 & 0.96 & 0.36 & 9.12 & 145.42 & 143.64 & @xmath291 & 43.58 & 7.55 & -1.12 & 3100 + sdss [email protected] & 0.269 & @xmath292 & 0.93 & 0.35 & 5.52 & 208.17 & 203.25 & @xmath293 & 44.00 & 8.16 & -1.30 & 8200 + sdss [email protected] & 0.046 & @xmath294 & 2.10 & 0.45 & 18.59 & 221.87 & 221.87@xmath198 & @xmath295 & 43.45 & 8.31 & -2.01 & 4400 + sdss [email protected] & 0.096 & @xmath296 & 1.21 & 0.66 & 6.96 & 155.30 & 153.03 & @xmath297 & 43.66 & 7.66 & -1.15 & 3700 +

in this paper we propose , discuss and illustrate a computationally feasible definition of chaos which can be applied very generally to situations that are commonly encountered , including attractors , repellers and non - periodically forced systems . this definition is based on an entropy - like quantity , which we call `` expansion entropy '' , and we define chaos as occurring when this quantity is positive . we relate and compare expansion entropy to the well - known concept of topological entropy , to which it is equivalent under appropriate conditions . we also present example illustrations , discuss computational implementations , and point out issues arising from attempts at giving definitions of chaos that are not entropy - based . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ toward the end of the 19th century , poincar demonstrated the occurrence of extremely complicated orbits in the newtonian dynamics of three gravitationally attracting bodies . this complexity is now called chaos and has received a vast amount of attention since poincar s early discovery . in spite of this abundant past and current work , there is still no broadly applicable , convenient , generally accepted definition of the term chaos . in this paper , we advocate a particular entropy - based definition that appears to be very simple , while , at the same time , is readily accessible to numerical computation , and can be very generally applied to a variety of often - encountered situations , including attractors , repellers , and non - periodically forced systems . we also review and compare various previous definitions of chaos . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

1501.07896

while the word chaos is widely used in science and mathematics , there are a variety of ways of defining it . thus , for this 25th anniversary issue of the journal chaos , we are motivated to review issues that arise when attempting to formulate a generally applicable definition of chaos , and to advocate a particular entropy - based definition that seems to us to be especially apt . we also relate our proposed definition to previous definitions . intuitively , perhaps the two most prominent ( not necessarily independent ) attributes of what scientists commonly think of as chaos are the presence of complex orbit structure and extreme sensitivity of orbits to small perturbations . indeed , in the paper by li and yorke@xcite where the term chaos was introduced in its now widely accepted nonlinear dynamics context , the term was motivated by the simultaneous presence of unstable periodic orbits of all periods , as well as an uncountable infinity of non - periodic orbits . thus , li and yorke s introduction of this terminology was motivated by the chaos attribute of complex orbit structure . on the other hand , lorenz@xcite was concerned with weather forecasting and accordingly focused on the chaos attribute of temporally exponential increase of the sensitivity of orbit locations to small initial perturbations . as we will discuss , these two attributes can be viewed as `` two sides of the same coin '' . we think of a definition of chaos as being `` good '' if it conforms to common intuitive notions of chaos ( such as complex orbit structure and orbit sensitivity ) and , at the same time , has the following three desirable features : * _ generality _ : the definition should work for almost all the examples that typical readers of this journal are likely to judge as chaotic . * _ simplicity _ : the definition should be fairly concise and not too technical . * _ computability _ : the definition should allow a practical , straightforward computational implementation for discerning the existence of chaos in a model . considering the issue of generality , one would like a definition of chaos to be applicable not only to attractors , but also to non - attracting sets , often called repellers . with respect to chaotic repellers@xcite , we note that they are central to the physically relevant topics of fractal basin boundaries@xcite , chaotic transients , and chaotic scattering@xcite , occurring , for example in fluid dynamics@xcite , celestial mechanics@xcite , chemistry@xcite , and atomic physics@xcite . furthermore , again considering the issue of generality , due to their common occurrence in applications , we desire that our definition of chaos be applicable to non - autonomous dynamical systems ( i.e. , systems that are externally forced by time dependent inputs ) , including external inputs that are temporally quasi - periodic@xcite , stochastic@xcite , or are themselves chaotic . here physical examples include quasi - periodic forcing of atmospheric jets@xcite , quasi - periodic forcing of stellar luminosity variation by two superposed stellar modal oscillations@xcite , advective transport in fluids with temporally and spatially irregular flow fields@xcite , and phase synchronism of chaos by noisy or chaotic drives@xcite . we emphasize that , when considering externally forced systems , we are interested in identifying chaos in the response of the system to a particular realization of the forcing , not in characterizing whether the forcing is chaotic . an important point for consideration of non - periodically forced chaotic systems is that the notion of a compact invariant set , which is typically used in definitions of chaos for autonomous systems ( including poincar maps of periodically forced systems ) , may not be appropriate or convenient for situations with non - periodic forcing . furthermore , in practice , it may be difficult to locate or detect an invariant set that is not an attractor . thus , rather than defining chaos for an invariant set , we will instead consider a notion of chaos for the dynamics within any given bounded positive - volume subset @xmath0 of the state space . we call such a set @xmath0 a _ restraining region_. for autonomous systems , chaos for an invariant set can be detected by taking @xmath0 to be a neighborhood of the desired invariant set . in our opinion , the currently most satisfactory way of defining chaos for autonomous systems is by the existence of positive topological entropy or metric entropy . we note , however , that the standard definitions of these entropies are quite difficult to straightforwardly implement in a numerical procedure . in addition , while generalizations to the original definitions of topological and metric entropy have been proposed , we view it as desirable to have a relatively simple definition that is applicable very broadly . however , we do not address here the question of identifying chaos in experimental data , which presents additional challenges , especially in the cases of non - attracting sets and externally forced systems . motivated by the considerations above , in sec . [ sec2 ] we introduce and discuss the definition of an alternate entropy quantity that we call `` expansion entropy '' . the expansion entropy of an @xmath1-dimensional dynamical system on a restraining region @xmath0 is the difference between two asymptotic exponential rates : first , the maximum over @xmath2 of the rate at which the system expands @xmath3-dimensional volume within @xmath0 ; and second , the rate at which @xmath1-dimensional volume leaves @xmath0 ( this rate is @xmath4 for an invariant set ) . _ we define chaos as the existence of positive expansion entropy on a given restraining region . _ expansion entropy generalizes ( to nonautonomous systems and noninvariant restraining regions ) a quantity that was formulated by sacksteder and shub@xcite in the case of an autonomous system on a compact manifold . in this restricted case , by the results of kozlovski@xcite , expansion entropy is equal to topological entropy for infinitely differentiable maps . in sec . [ sec2.6 ] we present examples of the application of our definition of expansion entropy to various systems , and also provide illustrative numerical evaluations of expansion entropy for some of these examples . section [ sec3 ] discusses topological entropy and previous work on computation of this quantity . section [ sec4 ] discusses issues that arise in previous non - entropy - based definitions of chaos . our definition of expansion entropy , which we denote @xmath5 , is closely related to previous definitions of topological entropy , to which it is equivalent under appropriate conditions ( see sec . [ sec3 ] ) . expansion entropy uses the linearization of the dynamical system and a notion of volume on its state space ; thus , unlike topological entropy , it is defined only for smooth dynamical systems . on the other hand , expansion entropy does not require the identification of a compact invariant set . as we will discuss , the differences in the definitions may make the criterion @xmath6 attractive as a general definition of chaos in smooth dynamical systems . we assume that the state space of the dynamical system is a finite - dimensional manifold @xmath7 . ( for example , if the manifold @xmath7 is @xmath1-dimensional euclidean space , the state @xmath8 at a given time is a vector @xmath9 $ ] where each @xmath10 is a real number . if some of the coordinates are angle variables , they can be taken modulo @xmath11 , resulting in manifolds such as a circle , cylinder or torus . ) we write @xmath12 for the volume@xcite of a subset @xmath0 of @xmath7 , and write @xmath13 for integration with respect to this volume ; for @xmath1-dimensional euclidean space , one can take @xmath14 . given a dynamical system on @xmath7 , we will use an integral to define its expansion entropy on a closed subset @xmath0 ( the restraining region ) that has positive , finite volume . the set @xmath0 need not be invariant under the system . we consider a deterministic dynamical system to be defined by an evolution operator , by which we mean a family @xmath15 of maps @xmath16 , with the interpretation that if @xmath8 and @xmath17 are the states of the system at times @xmath18 and @xmath19 , respectively , then @xmath20 . ( for example , @xmath21 could represent the solution from time @xmath18 to @xmath19 of a system of differential equations . ) the family @xmath15 must satisfy the identities @xmath22 and @xmath23 . the maps @xmath21 are defined for @xmath24 being integer - valued ( discrete time ) or being real numbers ( continuous time ) , with the restriction that @xmath25 if the system is noninvertible . we allow the system to be nonautonomous , including the case where @xmath21 is a realization of a stochastic dynamical system.@xcite if the system is autonomous , then @xmath21 depends only on @xmath26 , and in this case we will often write @xmath27 where @xmath28 . regardless , we assume that @xmath21 is a differentiable function of @xmath8 . recall that the singular values of a matrix @xmath29 are the square roots of the eigenvalues of @xmath30 . thinking of @xmath29 as a linear transformation , the image of the unit ball under @xmath29 is an ellipsoid , and the singular values of @xmath29 are the semiaxes of this ellipsoid . let @xmath31 be the product of the singular values of @xmath29 that are greater than @xmath32 ; if none of the singular values are greater than @xmath32 , let @xmath33 . then @xmath31 is roughly the number of @xmath34-balls needed to cover the image of an @xmath34-ball under @xmath29 . if the matrix @xmath29 is @xmath35 , consider a @xmath3-dimensional plane @xmath36 in @xmath1-dimensional euclidean space , where @xmath2 . let @xmath37 be a @xmath3-dimensional ball in @xmath36 , let @xmath38 denote the image of @xmath37 under @xmath29 , and let @xmath39 denote @xmath3-dimensional volume . then @xmath31 is the maximum over orientations of @xmath36 and over @xmath3 of @xmath40 . @xmath31 is the largest possible growth ratio of @xmath3-dimensional volumes under @xmath29_. below we will apply @xmath41 to the derivative matrix @xmath42 , in which case it represents a volume growth ratio for the ( typically nonlinear ) map @xmath21 . let @xmath43 be the set of @xmath8 such that @xmath44 for all @xmath45 between @xmath18 and @xmath19 ( that is , the trajectory of @xmath8 from @xmath18 to @xmath19 under @xmath15 never leaves @xmath0 ) . let @xmath46 * definition of expansion entropy * : we define the expansion entropy @xmath5 to be @xmath47 we consider @xmath5 and other limiting quantities below to be well - defined only if the limit involved exists.@xcite we remark that if the system @xmath15 is nonautonomous and the restraining region @xmath0 is not invariant , @xmath48 could potentially depend on the starting time @xmath18 in addition to @xmath15 and @xmath0 . also , it can be shown that the value of @xmath5 is invariant under differentiable changes of coordinates that are nonsingular on @xmath0 . to help interpret the definition of @xmath48 , we now replace @xmath49 with @xmath50[\mu(s_{t',t})/\mu(s)]$ ] in eq . ( [ eq1 ] ) . the definition ( [ eq2 ] ) can then be expressed as @xmath51 where @xmath52}{t'-t}\ ] ] and @xmath53 thus , we can view @xmath48 as the difference of two exponential rates , given by the limits in eqs . ( [ eq2a ] ) and ( [ eq2b ] ) , with the following interpretations . imagine that @xmath54 initial conditions are uniformly sprinkled throughout the volume of @xmath0 at time @xmath18 , and that @xmath54 is very large ( @xmath55 ) . the second term @xmath56 in eq . ( [ eq2a ] ) is then the exponential decay rate , as @xmath19 increases , of the number of trajectories from these initial conditions that remain in @xmath0 for all times between @xmath18 and @xmath19 . the quantity @xmath57 is the average over these remaining trajectories of the maximum local @xmath3-dimensional volume growth ratio along the trajectory . thus , the first term in eq . ( [ eq2a ] ) is the exponential growth rate of this average . it can be shown that the exponential growth rate of @xmath58 as @xmath59 is the sum of the positive lyapunov exponents of the trajectory starting at @xmath8 at time @xmath18 . thus , the limit in eq . ( [ eq2a ] ) is , in a sense , an average of this sum of positive lyapunov exponents over trajectories that remain in @xmath0 for all ( forward ) time . the criterion @xmath6 , which we propose for defining chaos , requires that this exponential volume growth rate strictly exceed the exponential rate @xmath56 at which trajectories leave @xmath0 . note that if @xmath0 is forward invariant , e.g. , an absorbing neighborhood of an attractor , then @xmath60 . some points of interest for this entropy definition are that 1 . it applies to non - autonomous systems , 2 . it assigns an entropy value @xmath5 to every restraining region @xmath0 in the manifold @xmath7 , and 3 . it directly suggests a computational technique for numerically estimating @xmath5 ( see sec . [ sec2.3 ] ) . finally , notice that @xmath61 this property follows from the fact that @xmath62 if @xmath63 , and consequently @xmath64 . thus , if there is chaos according to the definition @xmath6 with respect to a restraining region @xmath65 , then there is also chaos with respect to every restraining region @xmath0 that contains @xmath65 . in particular , as illustrated by the example in sec . [ sec2.6.6 ] , this implies that the expansion entropy will detect chaos ( @xmath6 ) within a restraining region @xmath0 when @xmath0 also contains a nonchaotic attractor , even when the chaos exists only on a repeller . in this section , we show that the expansion entropy of an autonomous , invertible system is the same as the expansion entropy of the inverse system . ( note that this is also true for the topological entropy ; see sec . [ sec3 ] . ) this equality results from the following identity for all invertible systems ( not necessarily autonomous ) : @xmath66 to verify this identity , notice that @xmath67 is the inverse of @xmath21 . below we use the notation @xmath20 , and consequently @xmath68 . then @xmath69 and @xmath70 are inverses , and hence the singular values of @xmath69 are the inverses of the singular values of @xmath70 . since the product of the singular values of a square matrix is the absolute value of its determinant , if @xmath29 is invertible then @xmath71 . in particular , @xmath72 . also , @xmath73 . writing @xmath74 as an integral over @xmath17 and then making the change of variables @xmath20 , @xmath75 if @xmath15 is autonomous and invertible , we write @xmath76 and @xmath77 = @xmath78 . then @xmath79/{t}\\ & = & \lim_{{t}\to\infty } \ln [ e_{-{t},0}(f , s)]/{t}\\ & = & \lim_{{t}\to\infty } \ln [ e_{0,-{t}}(f , s)]/{t}\\ & = & \lim_{{t}\to\infty } \ln [ e_{{t},0}(f , s)]/{t}= h_0(f , s)\end{aligned}\ ] ] here the first equality is by definition , the second equality is a change of notation , the third equality follows from the time - reversal identity for @xmath80 derived above , and the fourth equality uses the fact that @xmath15 is autonomous . with respect to point ( iii ) in sec . [ sec2.1 ] , we can imagine a computation of @xmath5 proceeding as follows . first , randomly sprinkle a large number of initial conditions @xmath81 uniformly in @xmath0 . then evolve each trajectory @xmath82 and the corresponding tangent map @xmath83 forward in time , continuing to evolve only as long as the trajectory remains in @xmath0 . at a discrete sequence of times @xmath84 , compute @xmath85 where the prime on the summation symbol signifies that only those @xmath86 values for which @xmath82 remains in @xmath0 up to time @xmath18 are included in the sum . from our definition of @xmath80 in eq . ( [ eq1 ] ) , we see that @xmath87 is an estimate of @xmath88 . plotting @xmath89 versus @xmath84 , for sufficiently large @xmath54 and @xmath84 , we expect to find an approximately linear relationship . accordingly , we can estimate @xmath5 as the slope of a straight line fitted to such data ( see also jacobs et al.@xcite for a similar approach in two dimensions ) . as in other such procedures , judgment and experimentation are called for in determining reliable choices of @xmath54 and the range of @xmath84 over which to do the fit , and such choices will be constrained by computer resources . in practice , we find it useful to choose a number , say @xmath90 , of different samples of size @xmath54 , compute @xmath89 for each sample , and take the mean and standard deviations of these logarithms . not only does this allow us to estimate the sampling error , it also produces a more reliable mean estimate than computing @xmath91 for a single sample of @xmath92 points . example illustrations of this computational approach are given in secs . [ sec2.6.6 ] , [ sec2.6.2 ] , and [ sec2.6.4 ] . specializing to the case of an autonomous invertible system @xmath15 , since sec . [ sec2.2 ] shows that the expansion entropy of @xmath15 and @xmath93 are the same , one could do a numerical computation of @xmath5 using either @xmath15 or @xmath93 . the question then arises as to which of these two alternatives is preferable from the point of view of computational cost and accuracy . in the remainder of this section , we argue that it is computationally preferable to calculate @xmath5 from @xmath15 if @xmath15 is volume contracting in @xmath0 , while calculation from @xmath93 is preferable if @xmath15 is volume expanding in @xmath0 . in order to see this , we generalize eq . ( [ eq2b ] ) to define both forward and backward exponential decay rates @xmath94\ ] ] where @xmath95 is , as in sec . [ sec2.1 ] , the set of initial conditions at time @xmath4 whose trajectories under @xmath15 remain in @xmath0 between times @xmath4 and @xmath96 , respectively . that is , in terms of the previously stated numerical procedure for calculating the expansion entropy , @xmath56 is the exponential temporal decay rate of the number of initial conditions sprinkled uniformly throughout @xmath0 at time @xmath4 that lead to orbits that never leave @xmath0 up to time @xmath84 , while @xmath97 is the analogous quantity taking the initially sprinkled points backward from time @xmath4 to time @xmath98 . since , to estimate the integral in eq . ( [ eq1 ] ) , we need to compute the average expansion rates only from those initial conditions that have not left @xmath0 , statistics at any given @xmath84 are improved when the number of such orbits is largest . further , the estimate of the limit @xmath99 dictates that we make @xmath84 large . these two considerations indicate that the forward ( respectively , backward ) calculation of @xmath5 will be computationally more efficient if @xmath100 ( respectively , @xmath101 ) . subtracting the definition ( [ eqx ] ) of @xmath56 from the definition of @xmath97 , and using the fact that @xmath102 for an autonomous invertible system , we obtain @xmath103 the right hand side is positive ( respectively , negative ) when the map is volume expanding ( respectively , contracting ) in @xmath0 . thus , @xmath101 if the map is volume expanding , while @xmath100 if the map is volume contracting . in particular , if @xmath0 is a neighborhood of an attractor , it is best to employ a forward time calculation . we note that the common examples of the hnon map and the lorenz system are uniformly volume contracting at all points in state space ( implying that @xmath104 ) , while hamiltonian systems are volume preserving ( implying that @xmath105 ) . in past work on fractal dimension , the box - counting dimension has been generalized to a spectrum of dimensions often denoted @xmath107 , where the box - counting dimension corresponds to @xmath108 , and the index @xmath106 can be any nonnegative number@xcite . in addition , a spectrum of entropy - like quantities , again depending on an index @xmath109 , has been introduced by grassberger and procaccia@xcite , where @xmath108 corresponds to the topological entropy , and @xmath110 corresponds to the metric entropy . thus , motivated by these past works , it is natural to introduce an analogous spectrum of @xmath106-order expansion entropies , @xmath111 , and to consider whether they are useful with respect to the issue of defining chaos . in appendix [ appa ] , we introduce and discuss a natural way of specifying @xmath111 . in particular , the form defining @xmath111 is specified so that it gives eqs . ( [ eq1 ] ) and ( [ eq2 ] ) when @xmath108 , gives an expansion entropy analogue of the entropy of grassberger and procaccia@xcite , and also gives a correspondence for @xmath110 with previous results for the metric entropy of repellers@xcite and with pesin s formula@xcite for the metric entropy for attractors . however , as we will argue in appendix [ appa ] , @xmath108 is special in regard to defining chaos . in particular , appendix [ appa ] will consider @xmath111 for an example in which @xmath0 contains an attracting fixed point and a chaotic repeller ( see sec . [ sec2.6.6 ] ) . for this example , it is shown that @xmath6 , while @xmath111 for @xmath112 can be zero . thus , @xmath5 successfully detects the chaos within @xmath0 , but @xmath111 for @xmath112 may not . consider a one - dimensional differentiable map @xmath15 with a fixed point @xmath113 ; for simplicity , we assume that @xmath114 . let the restraining region @xmath0 be an interval containing @xmath113 on which @xmath115 ; that is , @xmath15 is either uniformly expanding or uniformly contracting on @xmath0 . in either case , we show below that the expansion entropy @xmath48 is zero , i.e. , that fixed points are not chaotic according to our definition . in the case of an attracting fixed point , @xmath116 , and hence @xmath117 , for all @xmath118 and @xmath119 . also , @xmath120 for @xmath119 . then from eq . ( [ eq1 ] ) and ( [ eq2 ] ) we have @xmath121 for @xmath119 and @xmath122 . in the expanding case , @xmath123 for trajectories that remain in @xmath0 from time @xmath18 to @xmath19 . also , @xmath43 is a subinterval of @xmath0 whose endpoints map to the endpoints of @xmath0 under @xmath21 . thus , @xmath124 once again , @xmath122 . for fixed points ( or periodic orbits ) of higher - dimensional systems , similar arguments can be made , though they are more complicated in the case when the fixed point has both stable and unstable directions . the essence of these calculations is that any growth in the integrand @xmath41 of eq . ( [ eq1 ] ) as @xmath26 increases is balanced ( up to a time - independent multiplicative constant ) by a reduction in the volume of @xmath43 . the conclusion remains that @xmath125 , i.e. , that isolated period orbits are not chaotic . consider the one - dimensional map @xmath15 shown in fig . [ fig1d ] and the two restraining regions @xmath0 and @xmath126 , where we take @xmath127 $ ] and @xmath128 $ ] . the map is linear with derivative @xmath129 on @xmath130 $ ] and linear with derivative @xmath131 on @xmath132 $ ] , mapping each of these intervals to @xmath133 $ ] . for this example , the fixed point @xmath134 attracts almost all initial conditions with respect to lebesgue measure in @xmath0 . a map illustrating a case where @xmath127 $ ] contains a chaotic repeller ( in @xmath128 $ ] ) and an attracting fixed point ( @xmath134).,width=312 ] on the other hand , @xmath65 contains an invariant cantor set that is commonly called a chaotic repeller . we now show that according to our definition , @xmath15 exhibits chaos by having positive expansion entropy @xmath5 in both @xmath0 and @xmath65 . the invariant cantor set consists of all initial conditions in @xmath133 $ ] whose trajectories never land in the interval @xmath135 , i.e. , those whose base @xmath136 expansion does not contain the digit @xmath137 . the set @xmath138 of initial conditions whose trajectories remain in @xmath133 $ ] from time @xmath4 to time @xmath84 consists of @xmath139 intervals corresponding to all possible strings of length @xmath84 of the letters @xmath140 and @xmath141 , where @xmath140 denotes an iteration in the interval @xmath130 $ ] and @xmath141 denotes an iteration in the interval @xmath132 $ ] . a string with @xmath142 @xmath140 s and @xmath143 @xmath141 s corresponds to an interval of length @xmath144 , on which @xmath145 . then the integral of @xmath146 on each such interval is @xmath32 . thus , @xmath147 , and hence @xmath148 . in accordance with property ( [ eqz ] ) , since @xmath0 contains @xmath65 , we also have @xmath149 . in appendix [ appa ] , in addition to defining the quantity @xmath111 discussed in sec . [ sec2.5 ] , we also evaluate @xmath111 for the map in fig . [ fig1d ] . we find for the smaller restraining region @xmath65 that @xmath150 for all @xmath109 , but for the larger restraining region @xmath0 there is a critical value @xmath151 such that @xmath152 for @xmath153 . for @xmath110 , we have @xmath154 , while @xmath155 . our interpretation is that @xmath156 is dominated by the dynamics of lebesgue almost every initial condition whose trajectory approaches the fixed point attractor , while @xmath48 is dominated by the chaotic saddle in @xmath65 . we therefore conclude that @xmath157 ( and similarly @xmath111 for @xmath112 ) is not an appropriate tool for detecting non - attracting chaos in a restraining region . next we use this example to illustrate the numerical computation of @xmath5 . our procedure , as explained in sec . [ sec2.3 ] , is to choose a sample size @xmath54 and range of @xmath158 values and to do the following for each @xmath158 . using eq . ( [ eqw ] ) , we compute an estimate @xmath159 of @xmath160 for each of @xmath90 different samples of @xmath54 points each in the restraining region , and compute the mean and standard deviation of the @xmath90 samples . the results for @xmath161 and @xmath162 are shown in fig . [ fig1d1 ] for @xmath65 and in fig . [ fig1d2 ] for @xmath0 . the estimated value of @xmath5 is the slope of the solid curve in an appropriate scaling interval . the scaling interval for a given @xmath54 can be judged by consistency of the results with a larger value of @xmath54 , in addition to smallness of the error bars and straightness of the curve . notice that the somewhat arbitrarily chosen restraining region @xmath0 yields nearly as long a scaling interval as the restraining region @xmath65 that is chosen with knowledge of the invariant cantor set . computation of @xmath163 versus @xmath158 for the map of fig . [ fig1d ] with restraining region @xmath65 . for each @xmath158 , we computed @xmath164 for @xmath90 different samples , with @xmath165 randomly chosen initial conditions in each sample ( top figure ) and @xmath166 randomly chosen initial conditions in each samples ( bottom figure ) . the solid curve shows the mean of the @xmath90 samples , and the error bars show their standard deviation . as we discussed in sec . [ sec2.3 ] , the slope of the solid curve should , in the limit of large @xmath54 and @xmath158 , approximate @xmath167 . the dashed line has slope @xmath168 , which is the value we obtained analytically for @xmath167.,title="fig:",width=312 ] computation of @xmath163 versus @xmath158 for the map of fig . [ fig1d ] with restraining region @xmath65 . for each @xmath158 , we computed @xmath164 for @xmath90 different samples , with @xmath165 randomly chosen initial conditions in each sample ( top figure ) and @xmath166 randomly chosen initial conditions in each samples ( bottom figure ) . the solid curve shows the mean of the @xmath90 samples , and the error bars show their standard deviation . as we discussed in sec . [ sec2.3 ] , the slope of the solid curve should , in the limit of large @xmath54 and @xmath158 , approximate @xmath167 . the dashed line has slope @xmath168 , which is the value we obtained analytically for @xmath167.,title="fig:",width=312 ] computation of @xmath163 versus @xmath158 for the map of fig . [ fig1d ] with restraining region @xmath0 . this is the analogue of fig . [ fig1d1 ] with @xmath65 replaced by @xmath0.,title="fig:",width=312 ] computation of @xmath163 versus @xmath158 for the map of fig . [ fig1d ] with restraining region @xmath0 . this is the analogue of fig . [ fig1d1 ] with @xmath65 replaced by @xmath0.,title="fig:",width=312 ] consider the one - dimensional random map @xmath169 \text { mod } 2\pi\ ] ] where @xmath170 and @xmath171 are independent random variables that are uniformly distributed in the circle @xmath172 . we take the restraining region @xmath0 to be the entire circle . notice that @xmath173 and that @xmath174 where @xmath175 denotes an average over @xmath176 . if @xmath177 is uniformly distributed , then @xmath178 are _ independent _ and uniformly distributed , so that @xmath179 this suggests that for a typical realization of the random inputs , @xmath180 , where @xmath181 for @xmath182 , @xmath183 while for @xmath184 , @xmath185 for @xmath182 , each map is a diffeomorphism , so @xmath186 , and @xmath187 thus , @xmath188 is not exponentially increasing , so indeed @xmath125 for @xmath182 . numerical experiments agree with the argument above that @xmath6 for @xmath184 , though establishing that the transition to chaos ( according to our definition ) occurs exactly at @xmath189 would require a more definitive study . in fig . [ fig0 ] , we show the computed @xmath163 ( see sec . [ sec2.3 ] and [ sec2.6.6 ] ) versus @xmath158 for @xmath190 . computation of @xmath163 versus @xmath158 for the random circle map ( [ eqk ] ) with @xmath190 . for each @xmath158 , we computed @xmath191 for @xmath90 different samples , with @xmath192 randomly chosen initial conditions in each sample . each initial condition used a different realization of the random sequence of maps . the solid curve shows the mean of the @xmath90 samples , and the error bars show their standard deviation . the dashed line has slope @xmath193 ; this estimate of the expansion entropy appears to be slightly larger than the slope of the computed data.,width=312 ] this example illustrates a case where orbits are dense and , as in chaos , typical nearby orbits separate from each other with increasing time . however , the rate of separation is linear , rather than exponential in time . the example is the following map of the 2-torus : @xmath194 \text { mod } 2\pi , \quad \theta_{t+1 } = [ \theta_t + \omega ] \text { mod } 2\pi,\ ] ] where @xmath195 is irrational@xcite . as shown in fig . [ fig1 ] , the image under this map of a curve @xmath196 looping around the @xmath197-torus once in the @xmath198 direction is a curve @xmath199 that loops once in the @xmath200 direction , as well as once in the @xmath198 direction , with the number of @xmath200 loops increasing by one on each subsequent iterate . orbits are dense and nearby initial conditions with different values of @xmath198 separate linearly with @xmath18 . to evaluate @xmath5 for this map from the definition eq . ( [ eq2 ] ) , with the restraining region @xmath0 being the entire torus , we note that @xmath201 for all @xmath200 and @xmath198 . for @xmath202 , the singular values of @xmath42 are approximately @xmath26 and @xmath203 . thus , for large @xmath26 , @xmath204 and @xmath125 ( since @xmath205 as @xmath59 ) . hence this example is not chaotic according to our definition . image @xmath199 of a circle @xmath196 given by @xmath206 under the shear map ( [ eqy]).,width=312 ] figure [ fig2 ] shows the action of a horseshoe map in the plane on a unit square @xmath0 , which we also take to be the restraining region . the step ( a ) @xmath207 ( b ) represents a uniform horizontal compression and vertical stretching of the square . let @xmath208 be the ratio by which the vertical length of the square is stretched , and assume that bending deformations in the step ( b ) @xmath207 ( c ) take place only in the shaded region . the fraction of the original square that remains in the square after one iterate is @xmath209 ( see fig . [ fig2](d ) ) , and after @xmath26 iterates , the fraction is @xmath210 . also , @xmath211 , yielding @xmath212 , and @xmath213 . thus , by our definition the horseshoe map is chaotic in @xmath0 . a horseshoe map @xmath15 that is linear for trajectories that remain in the restraining region @xmath0.,width=312 ] for the hnon map @xmath214 with @xmath215 , the results of devaney and nitecki@xcite imply that for @xmath216 , the map has a topological horseshoe , and for @xmath217 , the nonwandering set is contained in the square @xmath218 , which we take to be the restraining region @xmath0 . [ fig2n ] shows the results of a numerical computation for @xmath219 of @xmath163 ( see sec . [ sec2.3 ] and [ sec2.6.6 ] ) versus @xmath158 , which agrees well with the value @xmath213 . computation of @xmath163 versus @xmath158 for the hnon map with @xmath219 and @xmath215 . for each @xmath158 , we computed @xmath191 for @xmath90 different samples , with @xmath162 randomly chosen initial conditions in each sample . the solid curve shows the mean of the @xmath90 samples , and the error bars show their standard deviation . the dashed line has slope @xmath213.,width=312 ] in this section , we define _ topological entropy _ , and discuss its relation to ( and equivalence with , in appropriate circumstances ) both expansion entropy and the related notion of _ volume growth_. the original definition of topological entropy , by adler , konheim , and mcandrew@xcite , was for a continuous map @xmath15 on a compact topological space @xmath220 . if @xmath220 is a metric space , an equivalent definition of topological entropy due to dinaburg@xcite and bowen@xcite is as follows . ( equivalence to the original definition was proved by bowen@xcite . ) for @xmath221 , two points @xmath8 and @xmath222 in @xmath220 are called @xmath223-separated if the distance between their @xmath142th iterates satisfies @xmath224 for some @xmath225 . a finite set of points @xmath226 is said to @xmath223-span @xmath220 if there is no point in @xmath220 that is @xmath223-separated from every point in @xmath227 . let @xmath228 be the minimum number of points needed to @xmath223-span @xmath220 , and let @xmath229 be the maximum number of points in @xmath220 that can be pairwise @xmath223-separated . let @xmath230 and @xmath231 it is not hard to show that @xmath232 . this implies the analogous relation between @xmath233 and @xmath234 , which implies that they have the same limit as @xmath235 . define the topological entropy @xmath236 of @xmath15 on @xmath220 by @xmath237 the notions of expansion entropy @xmath5 and topological entropy @xmath236 are both well - defined in the case when @xmath15 is a smooth , autonomous system on a compact manifold @xmath7 and the restraining region @xmath0 is all of @xmath7 . in this case , sacksteder and shub@xcite defined a quantity they called @xmath238 that is equivalent to expansion entropy . subsequently , przytycki@xcite proved that @xmath238 is an upper bound on @xmath236 if @xmath15 is a @xmath239 diffeomorphism for @xmath240 ; this proof was extended to noninvertible maps by newhouse@xcite . though there are examples@xcite for which the two quantities differ , kozlovski@xcite proved that @xmath241 for @xmath242 maps . thus , @xmath243 for a sufficiently smooth map @xmath15 on a compact manifold @xmath7 . from our point of view , these results leave open consideration of important issues regarding nonautonomous systems and the role of restraining regions . for example , suppose now that @xmath244 is a compact invariant set of an autonomous system @xmath15 on a ( not necessarily compact ) manifold @xmath7 . if @xmath244 has volume zero , @xmath245 is undefined , but we can define @xmath5 for a neighborhood @xmath0 of @xmath244 that contains no other invariant sets . in this case , we conjecture that @xmath246 if @xmath15 is @xmath242 . more generally , when the restraining region @xmath0 contains multiple invariant sets , we conjecture ( consistent with eq . ( [ eqz ] ) ) that @xmath48 is the maximum topological entropy of @xmath15 on an invariant subset of @xmath0 . our notion of expansion entropy is related to the notion of volume growth defined by yomdin@xcite and newhouse@xcite . yomdin defines the exponential rate @xmath247 of @xmath3-dimensional volume growth of a smooth map @xmath15 on a compact manifold @xmath7 , and proves that @xmath248 if @xmath15 is @xmath242 . newhouse defines the volume growth rate more generally for a neighborhood @xmath249 of a compact invariant set @xmath250 , and proves that @xmath251 is bounded above by the maximum over @xmath3 of the @xmath3-dimensional volume growth rate on @xmath249 . see also gromov@xcite for a discussion of these results . based on these results , newhouse and pignataro@xcite proposed and implemented algorithms for computing entropy of two - dimensional diffeomorphisms ( including poincar sections of three - dimensional differential equations ) by computing the exponential growth rate of the length of an iterated curve . other algorithms@xcite compute the growth rate of the number of disconnected arcs resulting from the iteration of an initial line segment within a neighborhood of a two - dimensional chaotic saddle or repeller . of course , these methods could be extended to higher dimensions by considering growth of surface areas , etc . expansion entropy , by estimating volume growth locally , allows an analogous computation to be done without having to compute and measure multidimensional surfaces . it is analogous to the approach used by jacobs et al.@xcite for two - dimensional maps . another approach to computing entropy is by symbolic dynamics : partition the state space into numbered subsets , and estimate the exponential growth rate ( as time increases ) of the number of different sequences of subsets that can be visited by a finite - time trajectory . this approach can yield good estimates with well - chosen partitions , but inadequate partitions may lead to underestimation , and in some cases symbolic dynamics indicates positive entropy when the topological entropy is actually zero.@xcite we conclude this section with a brief discussion of the connection between the definitions of expansion entropy and topological entropy in the case of a smooth , autonomous system on a compact manifold @xmath7 , with restraining region @xmath252 . in appendix [ appb ] , we argue that for sufficiently small @xmath221 , the quantity @xmath253 of eq . ( [ eq1 ] ) approximates @xmath254 , where @xmath255 is the maximum number of trajectories that are a distance @xmath34 apart at either time @xmath4 or at time @xmath158 . note that @xmath255 is a lower bound on @xmath256 , because the latter distinguishes between trajectories that are @xmath34 apart at some time between @xmath4 and @xmath158 ; however , at least for hyperbolic systems the difference between @xmath257 and @xmath54 should be inconsequential . ( [ eq00 ] ) and ( [ eq0 ] ) first take a limit with respect to @xmath158 and then @xmath34 . normalizing by @xmath258 does not change the limit : @xmath259 we have argued above that @xmath260 thus , by eq . ( [ eq2 ] ) , the definition of @xmath5 differs from the definition of @xmath236 primarily because it uses @xmath261 , and because the limits with respect to @xmath158 an @xmath34 are taken in the reverse order . a concept often associated with chaos is that of sensitive dependence , the idea that the orbits from two nearby initial conditions can move far apart in the future . in the mathematical literature the most common definition of `` sensitive dependence '' is as follows . ( this definition is also sometimes called `` weak sensitive dependence . '' ) * definition * : a continuous map , @xmath262 , on the compact metric space @xmath7 has _ sensitive dependence _ if there exists a @xmath263 such that for each @xmath264 ( no matter how small ) and each @xmath265 , there is a @xmath266 that is within the distance @xmath267 of @xmath8 and for which at some later time @xmath18 , @xmath268 . that is , no matter how close together the initial conditions @xmath8 and @xmath222 are , if we wait long enough , the orbits from these initial conditions will separate by more than some fixed value @xmath269 . notice that this definition of sensitive dependence does not say anything about the rate at which these orbits diverge from each other : this rate might , for example , be exponential ( e.g. , as for situations with a positive lyapunov exponent ) , or linear ( e.g. , as for the example in sec . [ sec2.6.3 ] ) . another often used concept assigns sensitive dependence to the dynamics on a compact invariant set ( the space @xmath7 is now not necessarily compact ) as follows . * definition * : a continuous map @xmath15 has sensitive dependence on a compact invariant set @xmath244 of a metric space @xmath7 if for every @xmath264 ( no matter how small ) and every point @xmath270 , there is a point @xmath271 within a distance @xmath267 of @xmath8 such that , at some later time @xmath18 , @xmath268 for some fixed value @xmath263 . the following is a definition of chaos , based on that given by devaney@xcite . * definition of devaney - chaos * : a continuous map @xmath262 , with @xmath7 a compact metric space , is chaotic if it satisfies the following three conditions . 1 . @xmath15 has sensitive dependence . @xmath15 has periodic orbits that are dense in @xmath7 . @xmath15 has an orbit that is dense in @xmath7 , i.e. , there exists an initial condition @xmath272 such that for each @xmath273 and each @xmath264 ( no matter how small ) , at some time @xmath18 , the orbit from @xmath272 will be within the distance @xmath267 from @xmath222 : @xmath274 . this definition can be converted to define devaney chaos for a compact invariant set , @xmath275 , by replacing @xmath7 in conditions ( ii ) and ( iii ) by @xmath244 , and condition ( i ) by `` @xmath15 has sensitive dependence on @xmath244 '' . it was pointed out by banks et al . @xcite that conditions ( ii ) and ( iii ) of devaney s definition , imply his condition ( i ) . thus condition ( i ) for devaney - chaos can be omitted . a serious drawback of devaney s definition of chaos is that it excludes significant cases that are sometimes considered and that most would regard as chaotic . for example , consider a map with quasi - periodic forcing , @xmath276 \text { mod } 2\pi,\ ] ] where @xmath195 is an irrational number . regarding this as a dynamical system with a state @xmath277 , we see that , because of the quasi - periodic behavior of @xmath198 , there are no periodic orbits of this system . hence , the system ( [ eq5 ] ) fails condition ( ii ) for devaney chaos . thus according to the definition of devaney chaos , a system like ( [ eq5 ] ) can never exhibit chaos . yet quasi - periodically forced systems are of practical interest and can have attractors with a positive lyapunov exponent , a situation generally thought of as chaotic . another point to make in connection with the example ( [ eq5 ] ) is that it presents a problem for the devaney definition of chaos even when @xmath41 is independent of @xmath198 : in that case @xmath278 , on its own , might indeed satisfy the conditions for devaney - chaos ; however , by considering the state to be @xmath279 with @xmath280 quasi - periodic , the devaney chaos condition ( ii ) is not satisfied , even though there is no change in the chaotic dynamics of @xmath281 . according to banks et al . , devaney - chaos only requires satisfaction of the two conditions that there be a dense orbit and a dense set of periodic orbits . robinson@xcite , on the other hand , notes that of the three conditions originally specified by devaney , the requirement of a dense set of periodic orbits does not seem as `` central to the idea of chaos '' as the other two conditions ( sensitive dependence and a dense orbit ) . thus he ( and also , independently , wiggins@xcite ) proposes the following definition . * definition of robinson - chaos * : the same as devaney - chaos except that condition ( ii ) is deleted . this definition , by not requiring periodic orbits , has the benefit of potentially allowing more consistent treatment of forced systems , like ( [ eq5 ] ) . however , there is still , in our opinion , a drawback . this occurs , e.g. , with reference to the shear map example ( [ eqy ] ) of sec . [ sec2.6.3 ] , which was considered by robinson@xcite ( see also hunt and ott@xcite ) . as discussed in sec . [ sec2.6.3 ] , orbits are dense and nearby points typically separate linearly with @xmath18 . thus , this example is robinson chaotic . however , the two lyapunov exponents are zero . while this example satisfies the robinson - chaos definition , due to the slow , linear - in - time , separation of orbits , such dynamics has previously been classified as nonchaotic ( see literature on so - called strange nonchaotic attractors@xcite ) . indeed , this linear - in - time separation of nearby orbits presents comparatively little prediction difficulty as compared to the exponential divergence emphasized by lorenz . li and yorke@xcite define the notion of a `` scrambled set '' , and the presence of a scrambled set can be taken as another definition of chaos . while this works well in the original context of one - dimensional maps considered by li and yorke , as we will see , it is not as appropriate for higher dimensional systems . * definition of a scrambled set * : for @xmath262 with @xmath7 a compact metric space , an uncountably infinite subset @xmath244 of @xmath7 is scrambled if , for every pair @xmath282 with @xmath283 , @xmath284 thus , by the second li - yorke condition , the orbits from @xmath8 and @xmath222 come arbitrarily close to each other an infinite number of times , while by the first li - yorke condition , the distance between the orbits from @xmath8 and @xmath222 also exceeds a fixed positive amount an infinite number of times . an attractive aspect of scrambling is that it excludes some cases that have sensitive dependence but are usually not considered chaotic . in particular , the shear map example ( [ eqy ] ) discussed above does not have a scrambled set because the @xmath198-distance ( or @xmath200-distance , if the @xmath198-distance is @xmath4 ) between a pair of orbits remains constant , thus violating the second li - yorke condition for scrambling . nevertheless , as with robinson - chaos , the definition of chaos as having an uncountable scrambled set includes cases that are generally regarded as nonchaotic . one example is a two - dimensional flow with an attracting homoclinic orbit , considered by robinson@xcite and w. ott and yorke@xcite ( see figure 1 in either paper ) ; on a trajectory converging to the homoclinic orbit , a finite piece of the trajectory forms an uncountable scrambled set . thus , the compact invariant set formed by the homoclinic orbit and its interior exhibits scrambling . from the discussion above , we see that using notions related to the common definition of sensitive dependence presents problems when attempting to use them to give a generally applicable definition of chaos . on the other hand , another type of dynamical characterization , namely that of lyapunov exponents , seems better suited to defining chaos . indeed , it can be quite useful to define a chaotic attractor using lyapunov exponents . if one excludes certain cases of milnor attractors ( see below ) and concentrates on a definition of an attractor of a map @xmath15 as a bounded set @xmath29 with a dense orbit such that there is an @xmath34-neighborhood @xmath285 for which @xmath286 , then it seems that a good definition of a chaotic attractor of the map @xmath15 is simply an attractor that has a positive lyapunov exponent . now , however , consider milnor s definition@xcite of an attractor : @xmath29 is an attractor for @xmath262 if there is a positive lebesgue measure of points @xmath265 such that @xmath29 is the forward time limit set of @xmath29 . figure [ fig5 ] shows an example demonstrating that the definition of a chaotic attractor as an attractor with a positive lyapunov exponent can be problematic , if milnor s definition of an attractor is used . a one - dimensional map for which the unstable fixed point at @xmath287 is a milnor attractor . trajectories that reach @xmath288 $ ] map to @xmath4 two iterates later.,width=312 ] the function @xmath289 in fig . [ fig5 ] goes to zero at @xmath290 and remains zero for @xmath291 . there is a positive measure of initial conditions @xmath113 that go to @xmath287 and stay there ( e.g. , if @xmath292 , then @xmath293 and @xmath294 ) . thus , the unstable fixed point @xmath287 is a milnor attractor with a positive lyapunov exponent ( because @xmath295 at @xmath287 ) yet it would be , we think , unacceptable to call the set @xmath287 chaotic . this example is rather special and contrived . thus , in practice , it is still very useful to think of a chaotic attractor as one with a positive lyapunov exponent . however , a main concern of this paper is a definition of chaos that works fairly generally , including being applicable to both attractors and repellers . in the case of repellers , basing the existence of chaos on lyapunov exponents presents a problem , since a repelling fixed point with a positive lyapunov exponent could not reasonably be considered chaotic . on the other hand , the anomaly for the fixed - point repeller example and the milnor example of fig . [ fig5 ] is removed if we define chaos by positive expansion entropy ( see , e.g. , sec . [ sec2.6.1 ] ) . in this paper we have introduced a quantity , the _ expansion entropy _ ( [ eq1 ] ) and ( [ eq2 ] ) ) , and we have argued that expansion entropy provides a `` good '' definition of chaos in that it possesses several desirable properties . we also compare this definition with other past definitions of chaos ( secs . [ sec3 ] and [ sec4 ] ) . in particular , the expansion entropy @xmath5 enjoys the properties of _ generality _ , _ simplicity _ , and _ computability _ discussed in sec . [ sec1 ] . one important feature of @xmath5 is that it assesses the presence of chaos in any given bounded region @xmath0 in state space , rather than in an invariant set . as such , it applies naturally in cases where the invariant sets are unknown or ( e.g. , in both deterministically and randomly forced systems ) do not exist . section [ sec2.6.2 ] presents examples illustrating various issues and features of expansion entropy , perhaps most importantly its numerical computation . it is our hope that our paper will lead to the use of expansion entropy in applications and to further study of its properties . the work of e. ott was supported by the u.s . army research office under grant w911nf-12 - 1 - 0101 . we thank s. newhouse for pointing out earlier work related to expansion entropy , and j. yorke and the reviewers for helpful comments . as discussed in sec . [ sec2.5 ] , we here generalize our definition of @xmath5 to a definition of a @xmath106-order entropy @xmath111 : @xmath296 where the argument of @xmath41 is the same as in eq . ( [ eq1 ] ) . comparing eqs . ( [ eq1 ] ) and ( [ eq2 ] ) with ( [ eq3 ] ) , we see that ( [ eq3 ] ) reduces to ( [ eq1 ] ) and ( [ eq2 ] ) for @xmath108 . furthermore , letting @xmath297 and assuming that the @xmath297 and @xmath298 limits can be interchanged , we obtain @xmath299 - \frac{1}{\tau_+},\ ] ] where , as in sec . [ sec2.3 ] , @xmath300 the quantity ( [ eq4 ] ) can be viewed as bearing a relationship to metric entropy that is analogous to the relationship between @xmath5 and topological entropy . in the case where @xmath0 contains an attractor , @xmath60 . in the case where @xmath0 contains a repeller , we call @xmath301 the average lifetime of repeller orbits . in the case where @xmath0 is a neighborhood if an invariant set with a `` natural measure''@xcite , we can identify the first term in eq . ( [ eq4 ] ) with the sum of the positive lyapunov exponents @xmath302 : @xmath303 which agrees with the results for metric entropy of kantz and grassberger@xcite for chaotic repellers and of pesin@xcite for @xmath60 . we now obtain @xmath304 and @xmath305 for the example in sec . [ sec2.6.6 ] , where @xmath0 is the large interval @xmath306 $ ] containing both the attracting fixed point and the chaotic repeller , while @xmath63 is the smaller restraining region @xmath133 $ ] containing only the chaotic repeller . we begin by finding @xmath305 . as discussed in sec . [ sec2.6.6 ] , the set @xmath138 of initial conditions that remain in @xmath65 from time @xmath4 to time @xmath158 consists of @xmath139 initial intervals of varying widths @xmath144 ( where @xmath307 ) on each of which @xmath308 . in addition , the number of such intervals with a given @xmath142 is the binomial coefficient @xmath309 $ ] . thus , the integral of @xmath310 that appears in eq . ( [ eq3 ] ) is @xmath311^{1-q } 3^{-k } 2^{k - { t } } \nonumber\\ & = & ( 3^{-q } + 2^{-q})^{t}.\end{aligned}\ ] ] furthermore , the total length of @xmath138 decreases by the ratio @xmath312 upon increase of @xmath84 by one , so that @xmath313 using eqs . ( [ eqb ] ) and ( [ eqc ] ) in eq . ( [ eq1 ] ) , we obtain @xmath314.\ ] ] note that this quantity is positive for all @xmath109 , and decreases monotonically with increasing @xmath106 ( dashed curve in fig . [ fighq ] ) . for @xmath108 , this agrees with our previous result of sec . [ sec2.6.6 ] that @xmath315 , while taking the limit @xmath297 yields @xmath316 . as @xmath317 , @xmath318 , so that @xmath319 . we now turn to the evaluation of the @xmath106-order expansion entropy for the larger restraining region @xmath127 $ ] . the main difference from @xmath65 is that @xmath320 for all @xmath321 , so @xmath322 , in contrast with eq . ( [ eqc ] ) . to estimate the integral of @xmath323 over @xmath324 , note first that by eq . ( [ eqb ] ) , @xmath325 the contribution to the integral of @xmath323 from initial conditions in @xmath324 but not in @xmath138 can be bounded above by @xmath326 $ ] for a constant @xmath196 independent of @xmath84 ; the factor of @xmath84 in the upper bound comes from considering trajectories that first leave @xmath65 at time @xmath18 , for each of the values @xmath327 . also , since @xmath328 for initial conditions in the interval near @xmath134 on which @xmath329 , such initial conditions contribute at least @xmath330 to the integral , where @xmath331 is the length of the contracting interval . thus , @xmath332.\end{aligned}\ ] ] from eq . ( [ eq3 ] ) , recalling that @xmath333 , we conclude that for @xmath334 , @xmath335 note that there is a critical value @xmath336 for which @xmath337 ; then @xmath152 for @xmath153 . in particular , @xmath155 by taking the limit @xmath297 . comparing eqs . ( [ eqd ] ) and ( [ eqh ] ) , we see that @xmath338 ( as we argued in sec . [ sec2.6.2 ] ) , but @xmath339 for @xmath112 ; see fig . [ fighq ] . note also that if the slopes @xmath129 and @xmath137 were increased , the critical value @xmath340 beyond which @xmath152 could be made arbitrarily close to @xmath4 . we conclude that @xmath111 for @xmath112 does not always detect chaos ( i.e. , @xmath111 may be zero ) in a restraining region containing an invariant set that is chaotic by all the definitions we reviewed in sec . [ sec4 ] , as well as by our definition @xmath6 . @xmath305 ( dashed curve ) and @xmath304 ( solid curve ) versus @xmath106 . the dashed curve decreases slightly from @xmath341 at @xmath108 to about @xmath342 at @xmath343.,width=312 ] here we justify the claim in section [ sec3 ] that the integral @xmath253 of eq . ( [ eq1 ] ) used to define expansion entropy approximates , for small @xmath34 , the ratio @xmath254 related to the definition of topological entropy . below , when we say that two quantities have the `` same order of magnitude '' , we mean that their ratio is bounded above and below by positive constants that are independent of @xmath34 and @xmath158 . assume that @xmath221 is small enough that the remainder term in the first order taylor expansion of @xmath344 is much smaller than @xmath34 for points within @xmath34 of each other : @xmath345 for @xmath346 with @xmath347 . cover @xmath0 with a grid of @xmath348 boxes whose diameters are @xmath34 ; then @xmath348 has the same order of magnitude as the maximum number @xmath258 of @xmath34-separated points in @xmath0 . each box @xmath349 is contained in a ball of radius @xmath34 , and contains a ball whose radius has the same order of magnitude as @xmath34 . notice also that @xmath350 for small @xmath34 . let @xmath351 be the center of @xmath349 , and let @xmath352 be the singular values of @xmath353 . then the image of @xmath349 under @xmath344 contains and is contained in ellipses whose semiaxes have the same order of magnitude as @xmath354 . let @xmath3 be the largest index for which @xmath355 . then the maximum number of @xmath34-separated points in the image of @xmath349 has the same order of magnitude as @xmath356 . summing over all @xmath349 , the maximum number @xmath255 of trajectories that are @xmath34-separated at either time @xmath4 or at time @xmath158 has the same order of magnitude as @xmath357 since we assumed that @xmath0 is invariant , @xmath358 is the same as @xmath0 . comparing with eq . ( [ eq1 ] ) , the discussion above constitutes an outline of a proof that @xmath254 has the same order of magnitude as @xmath253 . ( in fact , the same is true when @xmath0 is not invariant , if we define @xmath257 to count only trajectories that remain in @xmath0 between times @xmath4 and @xmath158 . ) e.g. , r. e. gillian and g. s. ezra , transport and turnstiles in multidimensional hamiltonian mappings for unimolecular fragmentation : application to van der waals predissociation , j. chem . 94 , 26482668 ( 1991 ) . i. i. rypina , f. j. beron - vera , m. g. brown , h. kocak , m. j. olascoaga and i. a. udovydchenkov , on the lagrangian dynamics of atmospheric zonal jets and the permeability of the stratospheric polar vortex , j. atmos . 64 , 35953610 ( 2007 ) . if @xmath7 is a riemannian manifold , then it has a canonical volume that is equivalent to lebesgue measure in appropriate local coordinates . furthermore , when we treat the derivative of a map as a matrix , or use ( small ) distances in @xmath7 , we assume the use of `` normal coordinates '' , which exist at least for @xmath359 riemannian manifolds . though we define expansion entropy only for a particular realization of a stochastic system , we expect that under appropriate hypotheses it has the same value for almost every realization . at a minimum , this should be the case when @xmath0 is invariant for all realizations of a system forced by an iid process , because in this case the expansion entropy depends only on the tail of the iid process . a suitable setting for the rigorous study of expansion entropy in random systems would be that of ledrappier and young @xcite . if @xmath15 is a @xmath242 map on a compact manifold , and @xmath0 is the entire manifold , then the limit has been proved to exist@xcite . more generally , @xmath5 could be defined as the @xmath360 , as in other definitions of entropy ( see sec . [ sec3 ] ) . ya . b. pesin , lyapunov characteristic exponents and ergodic properties of smooth dynamical systems with an invariant measure , dokl . nauk sssr 226 , 774777 ( 1976 ) . translated in sov . 17 , 196199 ( 1976 ) .

we develop equations and obtain solutions for the structure and evolution of a protodisc region that is initially formed with no radial motion and super - keplerian rotation speed when wind material from a hot rotating star is channelled towards its equatorial plane by a dipole - type magnetic field . its temperature is aroundk because of shock heating and the inflow of wind material causes its equatorial density to increase with time . the centrifugal force and thermal pressure increase relative to the magnetic force and material escapes at its outer edge . the protodisc region of a uniformly rotating star has almost uniform rotation and will shrink radially unless some instability intervenes . in a star with angular velocity increasing along its surface towards the equator , the angular velocity of the protodisc region decreases radially outwards and magnetorotational instability ( mri ) can occur within a few hours or days . viscosity resulting from mri will readjust the angular velocity distribution of the protodisc material and may assist in the formation of a quasi - steady disc . thus , the centrifugal breakout found in numerical simulations for uniformly rotating stars does not imply that quasi - steady discs with slow outflow can not form around magnetic rotator stars with solar - type differential rotation . [ firstpage ] circumstellar matter stars : early type stars : emission - line , be stars : magnetic fields stars : rotation stars : winds , outflows

0811.4768

we use the term _ protodisc _ to refer to the disc region that is initially formed when wind material from a star is channelled by dipole - type magnetic fields towards the equatorial plane . the material in a protodisc region has no radial motion and rotates with the magnetic field . it is bounded above and below by shock boundaries . we develop models for the structure and evolution of protodiscs around magnetic rotator stars . we study protodiscs as separate entities while recognising that they may be precursors of quasi - steady discs , such as those of be stars . we do not construct models of quasi - steady discs in this paper but indicate how protodiscs relate to models of be star discs . for example , a protodisc region could initially be present in the inner part of the quasi - steady disc model with slow outflow studied by @xcite . also , a more detailed study of protodiscs is important to verify how centrifugal outflow in protodiscs of rapidly rotating stars ( e.g. , * ? ? ? * ; * ? ? ? * ) affects the formation of quasi - steady discs around magnetic rotator stars . in this paper we consider protodiscs around stellar models with both uniform and differential rotation . we study how their radial extents change with time and we determine the conditions under which mri will occur to affect the angular velocity distribution and the radial outflow in the disc region . there has been a tremendous increase in observations of emission line b stars , including polarisation , interferometry and magnetic diagnostics , but the question of how discs with near keplerian rotation speeds are formed around them is has not been fully answered . there can be other mechanisms that are responsible for disc formation in different stars depending on their properties . for example , @xcite have proposed a decretion disc model for be stars with near maximal rotation speeds . there is no magnetic force in their model and angular momentum is transported by viscous forces . be stars are known to rotate rapidly ( e.g. , see * ? ? ? * ) and the presence of significant magnetic fields in ob stars has been reported by several authors ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the concept of a wind compressed disc was initially developed by @xcite . subsequently , @xcite proposed a magnetically torqued disk(mtd ) model in which a sufficiently strong magnetic field channels a flow of wind material towards the equatorial plane to form a disc and provides the torque required by the disc to maintain quasi - keplerian rotation speeds . @xcite developed the magnetic rotator wind disk(mrwd ) model in which keplerian discs may be formed with stellar magnetic fields that do not have to be as strong as those required in the mtd model . @xcite presented an amended version of the mtd model to include stellar gravity darkening and @xcite suggested that this should be further modified to include the effect of gravity arising from finite thickness of the disc . @xcite have studied the properties of discs with slow radial outflow and address criticisms of the mtd model . in the mrwd model for rotating stars with winds , when magnetic field lines that emerge from the stellar surface loop back to the star , a transient disc that has no meridional motion and corotates with the star is initially formed . after the local density in the disc has increased to a value where the centrifugal force exceeds the sum of gravity and the magnetic force , material diffuses radially outwards while stretching out the magnetic field lines . @xcite suggested that such material can diffuse into quasi - steady discs and that the presence of viscosity during this stage would enable the formation of keplerian discs of larger radial extent . on the other hand , numerical simulations by @xcite for the flow of wind material along dipole magnetic field lines display centrifugal breakout in the form of transient bursts and they concluded that mtd type models were not appropriate to explain disc formation . however , there are problems with their simulations : ( a ) as pointed out by @xcite they did not follow changes for a sufficiently long duration of time . ( b ) the simulations are essentially two dimensional . ( c ) numerical mhd simulations do not always provide accurate results because discretization of the mhd induction equation leads to the introduction of artificial magnetic diffusivity that is several orders of magnitude larger than the real magnetic diffusivity in discs . this can yield spurious results for magnetic reconnection that may not be realistic . ( d ) their models fail to consider differentially rotating stars and , therefore , do not obtain differentially rotating disc regions that are subject to the onset of mri . the occurrence of mri can significantly affect the rotation speed in a disc region . we carry out an analytical study and obtain time - dependent solutions for uniform and differentially rotating stars . we find that when the stellar angular velocity increases along the surface towards the equator there is a broad range of stellar parameters for which mri occurs in a sufficiently large part of the protodisc region . we consider a stellar model in which the magnetic field is dipole - type over some region of the stellar surface , so that field lines leaving the stellar surface pass through the equatorial plane and return to the star . wind material flowing along such lines is channelled into a dense equatorial protodisc region . since the flow of supersonic wind material into this region from above and below are in opposite directions , this material will cross shock surfaces that form the upper and lower boundaries of the disc region . we refer to the region between the sonic surface of the wind and the shock boundary of the disc as the _ wind zone_. thus , we need not be concerned about gas pressure effects on the wind flow , which is treated in an empirical fashion in that the velocity is chosen to obey a beta law and the density is assumed to follow from a global mass conservation relation . in section [ sec - windzone ] , we set up the relevant properties of axially symmetric wind zones that enable us to determine values of physical quantities at points on the wind side of the shock boundary from known values at the stellar surface . we develop formalism to account for the presence of differential rotation along the region of the stellar surface from which wind material flows into the disc region . the material in a protodisc region has no radial motion and rotates with magnetic field lines because the centrifugal and pressure gradient forces that are directed outwards are balanced by gravity and magnetic force . in section [ sec - proto - region ] we describe the general concept of a protodisc , which is a more generalised version of the pre - keplerian disc developed in the mrwd model . also , we describe the physical processes that are involved in their formation and evolution . in section [ sec - shock - proto - jump - conditions ] , we develop mhd generalisations of the rankine - hugoniot jump conditions that are appropriate for shock boundaries of protodiscs so that we can compute changes in values of quantities across shock surfaces . we take into account the vertical motion of shock surfaces caused by continuous filling up of a protodisc region by wind material . in section [ sec - disc - proto - eqns ] , we develop the equations that govern the structure of the models of protodiscs that we consider . we explicitly include the effect of the component of stellar gravity perpendicular to discs and , because of shock heating of the disc region , we include the vertical and radial components of the thermal pressure gradient in the respective momentum equations for protodiscs . even though the heights of protodiscs may be relatively small , the presence of a vertical component of stellar gravity leads to a vertical gradient of pressure and density . the equatorial density of material in the protodisc increases with time as the disc fills up . this has the effect of increasing the centrifugal force in relation to the magnetic force as well as increasing the thermal pressure in relation to magnetic pressure . the first of these effects tends to cause material at the outer edge of a protodisc to move radially outwards whereas the latter leads to the onset of mri within the protodisc . to determine the relative importance of these competing effects , in section [ sec - onset - mri - protodiscs ] we derive criteria that influence the evolution of protodiscs . also , we discuss the conditions under which the shrinking of a protodisc caused by centrifugal outflow is halted by the onset of mri . this leads to viscous readjustment of the rotation speed of the disc region . in section [ sec - disc - proto - applics - results ] , we apply our equations and jump conditions to protodiscs of different stellar models and present the results of our computations . in section [ sec - evln - protodisc - after - mri ] , we discuss important factors that affect the evolution of protodisc material after the onset of mri and refer to some relevant quasi - steady models . our conclusions regarding the formation and evolution of protodiscs around rotating magnetic stars with winds are summarised in section [ sec - discussion ] . an important objective of this paper is to relate the properties of protodisc regions to values of quantities at the stellar surface . therefore , we include all the equations that are necessary to solve for the different physical quantities in the wind zone , at the shock surface and in the protodisc region . in this section we present the equations that are used to determine the values of different physical quantities at points on the wind side of the shock boundary in terms of their values at the stellar surface . we assume that the wind zone is in a steady state and is isothermal . the mass , luminosity , radius and effective temperature of the star are @xmath1 and @xmath2 respectively . we use an inertial frame of reference with spherical coordinates @xmath3 in the wind zone and cylindrical coordinates @xmath4 in the disc region . we assume that the star , wind and magnetic field are symmetrical about the axis of rotation . let the angular velocity at any point @xmath5 on the stellar surface be @xmath6 , which may vary with latitude along the stellar surface . we show in section [ sec - proto - region ] that stellar models with appropriate differential rotation will have protodiscs that satisfy a condition that is necessary for the occurrence of mri . all the equations that we set up for the wind zone are valid even when @xmath6 varies along the stellar surface provided that the model is axially symmetric . we define an associated rotation rate parameter @xmath7 for the point @xmath5 by the equation @xmath8^{1/2 } \ , { \mathcal s}_\star$ ] . if the star has uniform rotation , then @xmath6 is constant for the star and @xmath7 gives the ratio of the equatorial rotation speed of the star to the keplerian rotation speed . otherwise , @xmath7 will vary with @xmath6 along the stellar surface . let @xmath9 and @xmath10 denote density and magnetic field , respectively . we use @xmath11 and @xmath12 , respectively , to denote velocities in the wind zone and the disc region . subscript @xmath13 denotes values of meridional components . then , @xmath14 and @xmath15 . let @xmath16 be a point of impact of wind material on the shock boundary . subscripts @xmath17 and @xmath18 , respectively , denote values of quantities on the wind side and the disc side at the point @xmath16 . let the meridional magnetic field line that emerges from the point @xmath5 on the star pass through the point @xmath16 on the shock surface . equations for the wind zone were set up in @xcite using the steady state equations given by @xcite . initial values of different quantities in the wind zone are specified at the sonic surface and are denoted using subscript @xmath19 . the mhd induction gives @xmath20 , which implies that the meridional vectors @xmath21 and @xmath22 must be parallel to each other . hence , the streamlines of flow in a meridional plane in the wind zone coincide with magnetic field lines . this , together with the mass and magnetic flux conservation conditions @xmath23 and @xmath24 require that @xmath25 where @xmath26 is constant along meridional streamlines and represents the ratio of mass flux to magnetic flux . at the sonic surface , @xmath27 will be equal to the isothermal sound speed . in general , @xmath28 and the density @xmath29 at the sonic surface will be functions of @xmath30 . in our models , we take the sonic surface to be approximately spherical so that @xmath28 is constant . if the stellar mass loss rate @xmath31 , is known only as an average value for the star rather than as a function of @xmath30 , we can write @xmath32 where @xmath29 is constant and represents the average density over the sonic surface . the azimuthal components of wind velocity and magnetic field satisfy the condition @xmath33 equations ( [ eqn - wind - rho - vm - bm ] ) and ( [ eqn - wind - v - phi - omega ] ) allow us to visualise the magnetic field line that is anchored at the point @xmath5 in the star to be rotating in the wind zone with the same angular velocity @xmath6 as @xmath5 . the azimuthal field @xmath34 is negative in the northern hemisphere of the wind zone so that field lines curve in the direction opposite to that of rotation . since wind material flows along field lines , the rotation speed of the wind material is less than the corotation speed as verified in equation ( [ eqn - wind - v - phi - omega ] ) . in our stellar models , the material flowing into the disc region encounter the shock boundary before it could reach the alfvn point . the azimuthal torque equation for the wind can be written in the form @xmath35 where @xmath36 is constant along a streamline and may be interpreted as total angular momentum per unit mass . each field line in the wind zone is associated with a characteristic density @xmath37 and a characteristic distance @xmath38 from the rotation axis . these are given by @xmath39 and @xmath40 in general , the values of @xmath37 and @xmath38 will vary with the latitude of the point @xmath5 on the stellar surface from which the corresponding field line emerges . note that a streamline that does not enter the disc region but travels with the wind to a large distance will have an alfvn point where @xmath41 and @xmath42 . however , for streamlines that enter the shock boundary of a protodisc region , there is no point in the wind zone where @xmath9 and @xmath43 are equal to these alfvn values but we define them as streamline constants because we can use them to compute the rotation speed and azimuthal component of the magnetic field at any point in the wind zone . we solve equations ( [ eqn - wind - v - phi - omega ] ) and ( [ eqn - wind - azimuthal ] ) to obtain the values of the @xmath44-components of the wind velocity and magnetic field . then , using equations ( [ eqn - wind - rho - c ] ) and ( [ eqn - wind - ell ] ) we obtain @xmath45 \left(1-{\rho_{a}}/{\rho_\circ}\right ) } { \left[1-\left(\rho_{a } \varpi_{a}^2\right)/\left(\rho_\circ \varpi_\circ^2\right)\right ] \left(1-{\rho_{a}}/{\rho}\right)}\right\ } \,\ ] ] and @xmath46 we express the meridional magnetic field in the wind zone in the form for dipole - type fields developed in @xcite . we have @xmath47 and @xmath48 , where @xmath49 is a constant such that @xmath50 . we should point out that @xmath51 represents a strict dipole field and that the the meridional component of the magnetic force will be zero in this case . however , for @xmath52 there is a nonzero meridional component of the magnetic force . as was pointed in the earlier paper , even if @xmath51 at the stellar surface its value will be less than 3 in the wind zone because the field lines will be stretched out to some extent by the wind material . we obtain the meridional component of the magnetic field in the wind zone as @xmath53^{1/2 } \ , .\ ] ] in the case of disc regions with relatively small height , we have @xmath54 at points along the shock boundary . then , equation ( [ eqn - wind - bm ] ) gives an approximation for @xmath55 in the form @xmath56 to obtain the meridional speed @xmath57 of the wind material as it enters the shock boundary , we use the beta velocity law for line driven winds given by @xmath58 where @xmath59 is a constant that is in the range @xmath60 ( see e.g. , * ? ? ? * ) . also , @xmath61 is a terminal speed that is constant for each streamline but may be different for different streamlines . in our computations we take @xmath61 to be the same for all streamlines and equal to the velocity at infinity of the wind . the density @xmath62 of the wind at the shock surface is obtained in terms of the density at the initial surface given by equation ( [ eqn - wind - initial - condn - rho - vm ] ) by using the mass and magnetic flux continuity equation ( [ eqn - wind - rho - vm - bm ] ) together with the velocity law ( [ eqn - wind - beta - law ] ) . in stellar models where the ratio @xmath63 of the stellar luminosity to the eddington luminosity becomes important , we replace @xmath64 by @xmath65 in the appropriate equations . we introduce the variable @xmath66 to denote the non - dimensional distance @xmath67 of the point @xmath68 on the equatorial plane , whose distance from the centre of the star is @xmath43 . when a subscript is added to @xmath68 we add the same subscript to the corresponding @xmath67 . the wind material that is channelled along flux tubes of a magnetic field crosses a shock surface and moves into an equatorial region between the stellar equator and a point @xmath69 , which is the outermost point on the equatorial plane to which the magnetic field is able to channel the flow of wind material . let @xmath70 be the innermost point on the equatorial plane where local keplerian rotation speed equals the corotation speed . then , @xmath71 , where @xmath72 is the rotation rate of the star at the point @xmath73 on the stellar surface that is linked to @xmath70 by a magnetic field line @xcite . the conditions under which the stellar magnetic field will be able to channel the wind into a disc region and relevant parameters are discussed in detail in sections 3 , 4 and 10 through 13 in @xcite . in particular , the value of the meridional alfvn parameter @xmath74 must be larger than unity at all points along a streamline from the star to the shock surface . hence , the strength of the meridional magnetic field at the initial surface must be larger than a minimum value given by @xmath75 for the formation of protodiscs . values of @xmath75 are approximately 1 g for b5 or b9 stars , 10 g for b2 stars , 60 g to 80 g for b0 stars and of the order of 100 g for o type stars . our interest is in stars with magnetic fields that are only slightly stronger than these minimum values so that they are able to channel wind flow into a protodisc region . in this paper we consider models of o and b type stars shown in table [ table - star - properties ] with rapid rotation such that @xmath76 . here , we describe the physical properties of protodiscs and the processes involved in their formation and evolution . a magnetic field of sufficient strength in a hot rotating star channels the flow of wind material from the star into an equatorial region . furthermore , in some part of this region , the field lines initially loop back to the star and the magnetic force is strong enough to provide support against centrifugal force so that the material accumulates to form a protodisc with no meridional motion . it is bounded above and below by shock surfaces as a result of the flow of wind material towards the equatorial plane from opposite hemispheres . figure [ fig1 ] gives a schematic meridional cross - section of a protodisc region in an axially symmetric star . let @xmath77 denote the angle that the tangent line to the shock surface at the point @xmath16 makes with the equatorial plane . we use subscript @xmath78 to denote the normal component of a vector at the shock boundary in the direction from the wind zone into the disc region . let @xmath79 be the height of the point @xmath16 above the equatorial plane . we assume that the material in the disc region satisfies the ideal gas law @xmath80 , where @xmath81 is the gas constant and @xmath82 is the mean molecular mass . for convenience , we use the substitution @xmath83 , where @xmath84 . note that the adiabatic sound speed is @xmath85 , where @xmath86 is the usual ratio of specific heats . let @xmath87 and @xmath88 shown in figure [ fig1 ] be the innermost and outermost points of the region in the equatorial plane where the magnetic force is strong enough to prevent radial motion against the effects of the centrifugal , gravitational and thermal pressure forces . the point @xmath87 will be located very close to the point @xmath70 and its position will be influenced by the radial components of the magnetic and thermal pressure force in addition to gravity and centrifugal force . a protodisc will form in the region between the points @xmath87 and @xmath88 . wind material arriving between the stellar equator and @xmath87 will tend to fall back towards the stellar surface because the sum of gravity and magnetic force dominates the sum of centrifugal and pressure forces . material arriving between @xmath88 and @xmath69 will flow outwards because the centrifugal force dominates in this region . meridional magnetic field lines emerging from the points @xmath89 and @xmath90 on the stellar surface are connected to the points @xmath87 and @xmath69 , respectively , on the equatorial plane . the protodisc region has super - keplerian rotation speed . as this region fills up , the density at points along the equatorial plane increases and the centrifugal force increases relative to the sum of gravity and magnetic force . while the location of @xmath87 remains fixed , centrifugal outflow causes the outer edge of a protodisc @xmath88 to move towards @xmath87 as time @xmath91 increases , provided that factors such as mri do not intervene . at the same time , the continuing impact of wind material on the shock boundary causes the protodisc to heat up . the thermal pressure can rise to values where it exceeds the local magnetic pressure and the ratio @xmath92 can become larger than unity . let the angular velocity of the disc material at a point @xmath68 on the equatorial plane be @xmath93 so that @xmath94 . there will be an onset of mri when the two conditions @xmath95 are satisfied @xcite . then , viscosity will become an important factor in the rotation of the disc region and the magnetic field will be affected by turbulence resulting from mri . in general , the angular velocity @xmath6 of the region of the stellar surface from which material flows to a protodisc region will vary with latitude . from section [ sec - windzone ] , we know that @xmath6 is constant along a meridional magnetic field line in the wind zone . in our model of a protodisc , each magnetic field line from the stellar surface that enters the protodisc region is continuous across the protodisc and returns to the stellar surface . in the case of stars that have differential rotation in which @xmath6 increases along the stellar surface from the point @xmath90 to the point @xmath89 , if the shock boundary of the protodisc is stationary , the protodisc rotation theorem established in @xcite shows that there will be a corresponding decrease in the angular velocity @xmath93 of the protodisc region from @xmath87 to @xmath88 . that is , the second of the two necessary conditions ( [ eqn - disc - proto - mri - conditions ] ) , @xmath96 , for the onset of mri in protodiscs is satisfied when the angular velocity @xmath6 increases along the stellar surface towards the stellar equator and the shock boundary of the protodisc is stationary . in section [ sec - onset - mri - protodiscs ] , we consider the case where the shock boundary of the protodisc is not stationary but moves so that its height changes with time . here too , results of numerical computations presented in section [ sec - disc - proto - applics - results ] show that for appropriate models of differential rotation of the star where @xmath6 increases towards the equator , the condition @xmath96 is satisfied in the protodisc . the first condition for the onset of mri will be satisfied when the increase in density and shock heating of the protodisc region cause @xmath97 to become larger than unity . @xcite have also shown that the rapidly growing modes increase exponentially on a time scale of order @xmath98 seconds . the rotation time scale for the protodisc region is @xmath99 seconds . thus , mri will occur on a time scale that is even shorter than the rotation time scale when the necessary conditions for the onset of mri are satisfied . although we focus on the onset of mri , it is possible that other types of mhd instability could occur in protodiscs . our interest is in identifying at least one mechanism that will prevent a protodisc from shrinking completely and redistribute its angular momentum so that a quasi - steady disc may be formed . we consider the protodisc phase of a disc to end at the time that mri occurs in that region . in an axially symmetric protodisc , the meridional component of the velocity is negligible and we have @xmath100 in cylindrical coordinates . the magnetic field is given by @xmath101 . when @xmath102 is relatively small , the slope @xmath103 of the shock surface is also small and the normal component wind velocity @xmath104 is a function of @xmath43 only along the shock boundary . then , because the ram pressure from inflowing wind material on the shock boundary remains steady , the temperature @xmath105 of the protodisc at a point @xmath16 of the shock will also be function of @xmath43 only and does not change as @xmath79 increases . thus , if the radiative cooling time of the protodisc is much longer than the flow time of wind material into the shock surface , the temperature @xmath106 in the protodisc will have the same value along a vertical line because there is no meridional flow of material in a protodisc region . that is , we can take @xmath106 to be independent of @xmath107 so that @xmath108 and @xmath109 . in models where @xmath102 becomes a large fraction before the onset of mri it will be necessary to solve for @xmath106 as a function of @xmath43 and @xmath107 . in this section we set up the equations that are necessary to determine values of quantities at points on the disc side of the shock surface from the values on the wind side that we obtain in section [ sec - windzone ] . as material flows from the wind zone into a protodisc region , the height @xmath79 of the shock surface forming the boundary of the protodisc increases . let @xmath110 denote the normal component of the velocity of the point @xmath16 on the shock surface in the direction of @xmath78 . to solve for the properties of a protodisc we need the values of @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 and @xmath116 at points on the disc side of the shock boundary . these are obtained by solving the six rankine - hugoniot jump conditions ( [ eqn - shock - disc - proto - mass - continuity ] ) through ( [ eqn - shock - disc - proto - energy ] ) . the continuity of mass across the shock boundary gives @xmath117 because the normal component of the magnetic field and the tangential component of @xmath118 must be continuous across the shock boundary , we have @xmath119 and @xmath120 continuity of the normal component of momentum gives the jump condition @xmath121 and the equation for the continuity of the @xmath44-component of momentum is @xmath122 the jump condition for energy across the shock surface is obtained by using the condition that the sum of the kinetic energy , internal energy , thermal energy and electromagnetic energy must be conserved . then , @xmath123 - \frac{v_{\phi , w } b_{n , w } b_{\phi , w}}{4 \pi } \\ = \rho_d u_s \left[\frac{u_s^2 + u_{\phi , d}^2}{2 } + \frac{\gamma \ , p_d}{(\gamma -1 ) \ , \rho_d } + \frac{b_{\phi , d}^2}{4\pi \rho_d}\right ] - \frac{u_{\phi , d } b_{n , d } b_{\phi , d}}{4 \pi } \end{array}\ ] ] for the flow of energy per unit area normally across the shock surface . when both sides of this equation are expanded to remove the brackets , we obtain four terms on each side . the first term represents the flow of kinetic energy of the material and the second term is the sum of the flows of internal energy and thermal energy . the sum of the third and fourth terms gives the normal component of the poynting vector that represents the flux of electromagnetic energy of the plasma . the rankine - hugoniot equations that we have set up above have general validity for protodiscs . however , in the case of the protodisc models that we consider here , the changes in values of @xmath124 and @xmath34 across the shock surface are not large . thus , we find that the influx of mass , momentum and kinetic energy of the wind material go mainly into increasing the the density and temperature of the protodisc . equations ( [ eqn - shock - disc - proto - vxb - tangential ] ) and ( [ eqn - shock - disc - proto - mom - phi ] ) give the azimuthal components of velocity and magnetic field on the disc side of the shock surface as @xmath125 and @xmath126 where @xmath127 is the normal component of the alfvn velocity at a point on the disc side of the shock boundary . in our protodisc models we have @xmath128 larger than @xmath129 . this is a consequence of the condition that the stellar magnetic field is able to channel the flow of wind material into the disc region . since @xmath130 , equations ( [ eqn - shock - disc - proto - uphid ] ) and ( [ eqn - shock - disc - proto - bphid ] ) show that the rotation speed and azimuthal component of the magnetic field increase across the shock boundary . in our protodiscs we find that @xmath114 is at most 2% larger than the corotation speed @xmath131 . in protodiscs where the shock surface is stationary , that is @xmath132 , we find that @xmath133 , which is consistent with the protodisc rotation law given in @xcite . also , because @xmath134 , the strength of @xmath135 decreases slightly across the shock boundary . here , we set up equations for the structure of protodiscs and obtain analytic solutions . we do not consider a disc region to be a protodisc after the onset of mri . since material flows continuously into a protodisc region along magnetic flux tubes and does not flow outwards , its height @xmath79 increases with time @xmath91 . the accumulation of mass can be expressed in the form @xmath136 let @xmath68 be a point on the equatorial plane at distance @xmath43 from the centre of the star and let @xmath16 be on the shock surface vertically above @xmath68 . to obtain values of the magnetic field components at the shock boundary , we assume that during the protodisc phase @xmath79 is relatively small compared with @xmath43 and that the angle @xmath77 is small . the computational results that we obtain for our models confirm that this assumption is appropriate . then , we have @xmath137 and @xmath138 . in such protodisc regions , the meridional magnetic field , which is a continuation of the dipole - type field in the wind zone , is almost vertical and we may take @xmath139 and @xmath140 . the general form of the azimuthal component of the momentum equation for an axially symmetric disc region is @xmath141 where @xmath142 is the viscosity coefficient developed by @xcite , @xmath143 is the local sound speed and @xmath144\ ] ] is the @xmath44-component of the magnetic force . before the onset of mri , material in a protodisc region and the magnetic field rotate steadily with super - keplerian speed and @xmath145 . since @xmath100 , we require solutions that satisfy the conditions @xmath146 , @xmath147 and @xmath148 . then , equation ( [ eqn - disc - proto - mom - phi-0 ] ) requires that @xmath149 , which says that the azimuthal component of the magnetic field between the equatorial plane and the upper shock boundary is constant along a vertical line . applying the boundary condition at the shock surface we have @xmath150 . note that , in the part of the protodisc below the equatorial plane , @xmath34 will be positive and have the same magnitude as in the upper part where it is negative . this implies a discontinuity in @xmath34 across the equatorial plane , which is physically plausible and is accounted for by the presence of a current sheet . the condition that @xmath147 together with the mhd induction equation imply that @xmath151 . when we substitute the solutions that we have already obtained for @xmath152 , @xmath34 and @xmath153 we find that @xmath124 satisfies the condition @xmath154 . therefore , as in the case of @xmath34 , the rotation speed @xmath124 is the same at all points along a vertical line in the protodisc and using the boundary condition at the shock surface , we obtain @xmath155 . thus , we have a physically valid model of a protodisc that is described by the self - consistent solutions @xmath156 and @xmath157 . also , we find that the vertical component of the magnetic force @xmath158 in the protodisc region of our model and the radial component @xmath159 at the point @xmath68 on the equatorial plane . since @xmath160 is negligible in the protodisc region , the general form of the @xmath107-component of the momentum equation for such a region is @xmath161\ , , \ ] ] where @xmath162 and @xmath9 are functions of time @xmath91 . this equation implies that the pressure gradient in the @xmath107-direction at any point is instantaneously balanced by the @xmath107-components of stellar gravity and magnetic force . in the case of our protodisc models in which @xmath107 is much smaller than @xmath163 we may take @xmath164 . since @xmath165 and @xmath166 is independent of @xmath107 , equation ( [ eqn - disc - proto - mom - z-0 ] ) gives @xmath167 where @xmath168 is the density scale height in the protodisc . this equation implies that the density gradient of the protodisc in the vertical direction is influenced by the vertical component of gravity . the vertical component of the magnetic force is zero in our model and does not contribute to vertical equilibrium . we solve this equation to write the density of the protodisc at height @xmath107 above @xmath68 , in the form @xmath169 where @xmath111 is the density of the protodisc at the point @xmath16 of the shock boundary . using equation ( [ eqn - disc - proto - mass - accum ] ) and the expression for @xmath9 in equation ( [ eqn - disc - proto - rho - z ] ) we obtain an equation connecting the height @xmath79 of the protodisc and time @xmath91 in the form @xmath170 \ , , \ ] ] where @xmath171 denotes the error function value at @xmath172 . equations ( [ eqn - disc - proto - rho - z ] ) and ( [ eqn - disc - proto - height - time - t ] ) show that the density at any point @xmath68 on the equatorial plane of the protodisc region increases with time . then , unless mri intervenes and terminates the protodisc phase , the density at @xmath68 will increase to the value @xmath173 that is required for the centrifugal force to balance the sum of gravity , magnetic force and thermal pressure . that is , @xmath173 is obtained as the solution of the equation @xmath174 we denote the time at which this occurs by @xmath175 and let @xmath176 be the height of the protodisc at @xmath88 at this time . the value of @xmath176 is obtained by putting @xmath177 in equation ( [ eqn - disc - proto - rho - z ] ) and solving for @xmath79 . the time @xmath175 at which this occurs is found by using equation ( [ eqn - disc - proto - height - time - t ] ) . in addition to giving us the values of the properties at the outer edge , the importance of these equations is that they give the rate at which the radius @xmath178 of the outer edge of the protodisc decreases . thus , we are able to compute the radial extent of a protodisc at the time at which mri occurs . numerical values are shown in the tables presented in section [ sec - disc - proto - applics - results ] . in our model of a protodisc , a magnetic field with a sufficiently strong meridional component channels the flow of wind material into an equatorial disc region . a much weaker azimuthal component of the magnetic field acts in concert with the meridional component to apply a torque on the wind material so that it enters the protodisc region with super - keplerian angular velocity after crossing its shock boundary . the radial component of the magnetic force in the protodisc is initially able to assist gravity in maintaining radial equilibrium against the centrifugal force and thermal pressure . the protodisc region rotates steadily with no meridional motion when @xmath124 and @xmath34 are constant along vertical lines . consequently , the vertical and azimuthal components of the magnetic force are zero in the protodisc region and the density gradient in the vertical direction is influenced only by stellar gravity . increase in equatorial density causes centrifugal outflow at the outer edge and causes the radial extent of the protodisc to decrease . it will continue to do so unless mri or some other mhd instability occurs . we discuss this in section [ sec - onset - mri - protodiscs ] . as we pointed out in sections [ sec - proto - region ] and [ sec - shock - proto - jump - conditions ] , continuous flow of wind material causes continuous shock heating of the protodisc . the inflow time of the wind into the protodisc region is @xmath179 the temperature of a protodisc will remain high if the shock heating is not offset by radiative cooling . the radiative cooling time for the disc material is given by @xmath180 s , where @xmath181 @xmath182 is the internal energy per unit volume and @xmath183 @xmath184 is the cooling rate per unit volume , where @xmath185 denotes the number of particles per cubic centimetre . @xcite gives @xmath186 . from this we obtain the expression @xmath187 ergs @xmath188/s for the cooling function that is appropriate for the postshock temperature ranges of our protodiscs . we assume that the disc material consists of hydrogen and helium with mass fractions @xmath189 and @xmath190 , respectively . then the number density of electrons is given by @xmath191 per cubic centimetre and the cooling time is @xmath192 our computations show that the radiative cooling rate in our models is slower than the rate of shock heating so that protodiscs continue to maintain their high temperatures . there are several values of the density and time scales that play critical roles in the evolution of protodiscs . the requirement @xmath193 for the onset of mri stipulated in condition ( [ eqn - disc - proto - mri - conditions ] ) will be satisfied when the density @xmath194 at a point @xmath68 on the equatorial plane exceeds the critical value @xmath195 given by @xmath196 because @xmath197 . if @xmath198 at a point @xmath68 in a protodisc , the value of @xmath194 will reach the value of the density required for mri to set in at @xmath68 before the outer edge of the protodisc reaches the point @xmath68 . let @xmath199 denote the time required for mri to set in at the point @xmath68 . let the point @xmath5 on the stellar surface be linked to the point @xmath68 on the equatorial plane of a protodisc region by a meridional magnetic field line . let @xmath6 be the angular velocity of the stellar surface at @xmath5 and let @xmath93 be that of the protodisc at @xmath68 . in section [ sec - proto - region ] , we pointed out that in models of protodiscs where the shock surface is stationary , the condition that @xmath6 should increase towards the equator along the stellar surface will ensure that the second necessary condition , @xmath96 , for the onset of mri is satisfied . however , in our models of protodiscs , the shock surface moves so that its height increases with time . in this case , equations ( [ eqn - wind - v - phi - omega ] ) , ( [ eqn - shock - disc - proto - bn ] ) and ( [ eqn - shock - disc - proto - uphid ] ) give @xmath200 where we have used the equations @xmath201 . we interpret this to be a generalised form of the protodisc rotation law . note that we recover the protodisc rotation theorem in @xcite when @xmath132 . for stars with differential rotation , we need the distribution of @xmath6 along the stellar surface to compute values of @xmath93 . to determine whether @xmath93 will decrease with increasing @xmath43 in a protodisc for realistic values of @xmath6 , we use the angular velocity model for the sun ( e.g. , see solar rotation in allen s astrophysical quantities 2000 ) , which has the form @xmath202 \ , , \ ] ] where @xmath203 is the angular velocity of the star at the equator , @xmath204 denotes colatitude of the point @xmath5 on the stellar surface , @xmath67 is the non - dimensional distance of the point @xmath68 from the rotation axis and @xmath205 is a constant such that @xmath206 . we have used the relation @xmath207 for dipole - type fields @xcite . when @xmath6 varies along the stellar surface , so does the rotation rate @xmath72 . we use @xmath208 to denote the value of @xmath72 corresponding to @xmath203 . in section [ sec - disc - proto - applics - results ] , we present results of numerical computations for a variety of models . we find that the condition @xmath96 is satisfied in protodisc regions for a wide range of permissible values of @xmath205 . [ cols="<,<,>,^,^,^,>,^,^,^ " , ] tables [ table - protodisc - critical - ratios - alpha - ara-07-uniform ] shows critical values of properties of a protodisc when a point @xmath68 on the equatorial plane becomes the outer edge at @xmath68 for a uniformly rotating stellar model having the properties of @xmath142 ara with @xmath209 . tables [ table - protodisc - critical - ratios - alpha - ara-07-diff ] and [ table - protodisc - critical - ratios - alpha - ara-05-diff ] show the corresponding values when @xmath68 becomes the outer edge or when mri occurs at @xmath68 for differentially rotating models with @xmath210 and @xmath211 , respectively , @xmath212 and having the same basic stellar properties . suppose that we consider the same basic model with different rotation rates and different magnetic field strengths . then , when the rotation rate is fixed and the magnetic field strength is increased , the value of @xmath213 increases slightly . the results for stellar models with the properties of @xmath142 ara are shown in table [ table - protodisc - properties - alpha - ara - uni - diff - rot - mag ] for uniform rotation and for differential rotation with @xmath212 . also , when the stellar rotation rate is increased while the magnetic field strength is fixed , the value of @xmath213 decreases . we note that the values of @xmath213 for models with differential rotation are somewhat larger than those for the corresponding models with uniform rotation . our results show that when stars have solar - type differential rotation , there are broad ranges of stellar magnetic field strengths and rotation rates such that @xmath214 for which mri sets in over a significant region from @xmath87 to @xmath215 before the protodiscs shrink radially to that region . however , when a star rotates with near critical rotation speed such that @xmath216 , we find that the time for the onset of mri is longer than the time for the protodisc region to shrink to a thin ring , unless some other mhd instability occurs . lcrrrrrrrrrrrc & & & + + @xmath208 & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 & & @xmath219 & @xmath220 & @xmath221 & @xmath213 & @xmath223 + & ( g ) & ( g ) & & & & ( days ) & & ( g ) & & & & ( days ) + @xmath224 & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & & @xmath231 & @xmath232 & @xmath233 & @xmath234 & @xmath235 + @xmath236 & @xmath225 & @xmath226 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & & @xmath241 & @xmath242 & @xmath243 & @xmath244 & @xmath245 + @xmath246 & @xmath225 & @xmath226 & @xmath247 & @xmath248 & @xmath249 & @xmath250 & & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 + @xmath256 & @xmath225 & @xmath226 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & & @xmath261 & @xmath262 & @xmath263 & @xmath264 & @xmath265 + @xmath266 & @xmath225 & @xmath226 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & & @xmath271 & @xmath272 & @xmath273 & @xmath274 & @xmath275 + @xmath276 & @xmath225 & @xmath226 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & & @xmath281 & @xmath282 & @xmath254 & @xmath282 & @xmath283 + @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 & @xmath289 & @xmath290 & & @xmath291 & @xmath292 & @xmath293 & @xmath228 & @xmath294 + @xmath284 & @xmath295 & @xmath296 & @xmath297 & @xmath288 & @xmath298 & @xmath299 & & @xmath300 & @xmath292 & @xmath301 & @xmath302 & @xmath303 + @xmath284 & @xmath225 & @xmath226 & @xmath304 & @xmath288 & @xmath305 & @xmath306 & & @xmath307 & @xmath292 & @xmath308 & @xmath309 & @xmath310 + @xmath284 & @xmath311 & @xmath312 & @xmath313 & @xmath288 & @xmath314 & @xmath315 & & @xmath316 & @xmath292 & @xmath317 & @xmath318 & @xmath319 + a detailed study of the evolution of disc material after the onset of mri is beyond the goals of the present paper . however , we consider several factors that will play a role in the evolution from the protodisc phase to a quasi - steady disc . because the electrical conductivity is extremely high in a protodisc region , the magnetic diffusivity is negligible and the ohmic diffusion time is extremely long @xcite . thus , before the onset of mri , the magnetic field within a protodisc region remains frozen in the material as it rotates . at the outer edge , the sum of the centrifugal force and thermal pressure exceeds the sum of gravity and the magnetic force , so that disc material will flow outwards and draw out magnetic field lines with it . we focus on stellar models where magnetic fields are sufficiently strong to channel wind flow into disc regions . in the case of b type stars , field strengths of order 1 g to 10 g would be appropriate . we wish to understand how the onset of mri in protodiscs of such stars helps in the formation of quasi - steady discs that survive over a period of several decades . turbulence arising from mri may persist for a long time and will cause changes in the magnetic field and the magnetic diffusivity of the disc material will be affected . the occurrence of mhd dynamo action is possible . a study of the evolution of the magnetic field will be complicated . in a quasi - steady state we expect that the structure and strength of the field will allow a slow outflow of the disc material . after the onset of mri , viscosity becomes important and the shakura - sunyaev viscosity coefficient is given by @xmath320 subject to the condition that @xmath321 . the azimuthal viscous force is negative and will tend to decelerate the disc rotation speed . also , in general , @xmath322 will no longer be zero and the azimuthal magnetic force @xmath323 may play a role in the rotation of the disc material . the importance of viscosity has been invoked in the modelling of decretion discs ( e.g. , * ? ? ? * ; * ? ? ? * ) and accretion discs ( e.g. , * ? ? ? * ; * ? ? ? * ) . in decretion discs , material is fed into the disc region from the stellar equator and drifts outwards due to the effect of viscosity . in accretion discs , material in keplerian discs fall towards the central star due to viscous loss of angular momentum . these studies consider thin discs and set up equations for physical quantities averaged over the heights of discs . we have not used such an approach in our models of protodiscs . however , because the heights of protodiscs are relatively small at the time that mri occurs , thin disc models may be a convenient alternative in a study of the evolution of disc material after the onset of mri . @xcite have used a shearing box model for thin accretion discs and found that there will be an outward flux of angular momentum . we are interested in scenarios in which the angular velocity distribution of the disc material readjusts so that a quasi - steady disc may be formed , possibly with inflow and slow outflow . a model conforming to such a description has been developed by @xcite where wind material is fed into a disc region by a magnetic field and flows slowly outwards with quasi - keplerian rotation speed . another possible quasi - steady model may be a magnetic rotator decretion disc model in which a protodisc that suffers mri will provide material with super - keplerian rotation speed to a decretion disc region . such a model will be more realistic than the decretion disc model studied by @xcite and @xcite because the nonmagnetic models of lee et al and okazaki require the extremely special condition that keplerian rotation must occur at the stellar equator . in the case of a quasi - steady disc , continuity of mass and magnetic flux require that @xmath324 , @xmath325 and an appropriate mhd induction equation be satisfied . the radial component of the momentum equation for axially symmetric quasi - steady discs with @xmath326 can be written as @xmath327 where @xmath328\ ] ] is the radial component of the magnetic force . the thermal pressure term in equation ( [ eqn - qsdisk - radial ] ) may be negligible in a quasi - steady state where the disc material has cooled down and a radiative force may be included when appropriate . we are interested in quasi - steady models in which the radial speed @xmath329 is small and the rotation speed @xmath124 is approximately equal to the magnetically modified keplerian speed @xmath330^{1/2}$ ] defined in @xcite , where @xmath331 is the ratio of the radial component of the magnetic force to gravity and is small . the azimuthal component of momentum satisfies @xmath332 the last term in this equation represents the @xmath142-viscosity force and will be negative . when @xmath329 is positive , the excess angular momentum brought in by the wind material is removed by the outward drift of the disc material . we expect that @xmath329 will be small as shown by @xcite . this is because the continuity of mass in a quasi - steady state requires that @xmath333 across the shock boundary . if the disc region is isothermal in a quasi - steady state , we have @xmath334 @xcite so that we obtain @xmath335 . because @xmath336 for the types of stars that we consider , we find that @xmath337 . the @xmath107-component of the momentum equation remains the same as ( [ eqn - disc - proto - mom - z-0 ] ) . in a detailed study we will have to include appropriate jump conditions at the shock boundary of the disc . at the inner end of a protodisc , if the azimuthal viscous force is sufficiently strong to decelerate the rotation speed , disc material will tend to flow back towards the star as in the case of an accretion disc . however , this can compress the field lines of a dipole type magnetic field causing an increase in the magnetic pressure that will oppose the backward flow . also , continuing inflow of wind material will supply angular momentum and it may be possible for a quasi - steady state to exist for some length of time despite any backward flow that may occur at the inner boundary of the disc region . obviously , a detailed study will be required to determine the nature of the dynamics in this region . although there may be other types of mhd instability that could affect the angular velocity distribution in a protodisc , we focus on mri because we are able to determine the conditions that must be satisfied by the central star for its occurrence in a protodisc . an important aspect of our study of protodiscs of differentially rotating stars is that scenarios may exist in which mri can intervene to redistribute angular momentum of the disc region and , possibly , permit the formation of quasi - steady disc regions . in this paper , we set up equations and jump conditions to study the formation , structure and evolution of protodiscs around rotating magnetic stars with winds . we obtain solutions for a variety of stellar models with different rotation rates and magnetic field strengths . when the magnetic field of a rotating hot star is strong enough to channel the flow of wind material towards the equatorial plane and field lines from the stellar surface cross the equatorial plane and return to the star , a protodisc with super - keplerian rotation and no radial motion will be formed in an equatorial region . initially , the sum of the magnetic force and gravity in this region is sufficient to prevent any radial outflow that may be caused by the centrifugal force and thermal pressure . as wind material continues to flow into the protodisc , its equatorial density and height increase with time . the resulting increase in centrifugal force relative to the magnetic force causes the material at the outer edge of the protodisc to move outwards so that the radial extent of the protodisc region decreases . shock heating causes protodisc temperatures rise to values of order @xmath338k . because the radiative cooling time is longer than the heating up time , protodiscs retain their high temperatures . there is no meridional flow of material within a protodisc and the continuing increase in density causes the thermal pressure in the protodisc to increase in relation to the magnetic pressure . when the rotation rate of the central star satisfies the condition @xmath214 and the angular velocity of the protodisc region decreases radially outwards , mri will occur before the radial extent of the protodisc can decrease significantly . in the case of most stars with near critical rotation rates such that @xmath216 , the time taken for the onset of mri is longer than the time by the protodisc to shrink to a thin ring . the evolution of protodiscs is strongly influenced by the inflow of wind material , centrifugal outflow and the occurrence of mri . the protodisc phase of a disc region terminates either when it shrinks completely or when some instability such as mri sets in . we derive conditions for centrifugal outflow at the outer edge of a protodisc and for the onset of mri in a protodisc . we perform numerical computations to study the structure and properties of protodiscs of several different stellar models and to determine how the protodisc phase terminates . our results show that there are situations involving different field strength domains that lead to the formation of protodiscs that suffer mri soon after their formation or shrink to a thin ring when no instability intervenes . equation ( [ eqn - proto - relation - omega - omega ] ) gives the angular velocity of a protodisc in terms of that of the central star . when the central star has uniform rotation , the protodisc has approximately uniform rotation and if no instability intervenes , its radial extent continues to decrease . a protodisc region may take a long time to shrink completely and during most of this time it would appear like a ring . however , it is not clear whether steady uniform rotation is possible in the surface regions of rapidly rotating early type stars because large scale meridional circulations must occur in the radiative envelopes of such stars . meridional circulation will transport angular momentum and disrupt the angular velocity distribution to cause differential rotation . studies by @xcite showed that steady rotation may be possible in the interior of the radiative envelope with angular velocity constant along field lines and having about @xmath339 percent variation from uniform rotation . however , because of the sensitivity of meridional circulation near the stellar surface to perturbations in angular velocity , steady uniform rotation may not possible in the surface region . other efforts to develop models of steady uniform or differential rotation in early type stars with magnetic fields have been discussed by @xcite and @xcite . we are not aware of any conclusive theoretical or observational evidence that early type stars with fast rotation and magnetic fields will have steady unform rotation along the surface . when the stellar surface has differential rotation with angular velocity increasing towards the equator , equation ( [ eqn - proto - relation - omega - omega ] ) shows the angular velocity of the protodisc decreases radially outwards . in this case , mri occurs in the protodisc within a few hours or days . we know that such rotation occurs in the sun , where the differential rotation is maintained by meridional circulation and convection ( see e.g. , * ? ? ? recently , @xcite have suggested that convection zones will be present near the surface of rotating massive stars . this situation is the same as in the surface regions of the sun where convective motions and meridional circulation are present . thus , we suggest that it is reasonable to expect that there are massive stars with solar - type differential rotation . our study of protodiscs was motivated in large part by the claim that discs will not form around magnetic rotator stars because of centrifugal breakout found by @xcite in numerical simulations of uniformly rotating stars . @xcite had previously pointed out that centrifugal outflow of protodisc material due to increased density could diffuse into keplerian , especially when viscosity was present . in studying the evolutionary properties of protodiscs , we have focussed on centrifugal outflow and the onset of mri . it is possible that there are other mhd instabilities that may also play a role in the evolution of protodiscs to quasi - steady discs . we have been able to show that under certain highly plausible conditions at the stellar surface , mri can occur in protodiscs before they shrink significantly through centrifugal outflow . protodiscs in which mri occurs in an inner part before the outer edge shrinks to that part are of interest because viscosity arising from the onset of mri may be able to redistribute the angular velocity of the disc so that material may diffuse into a quasi - steady disc region with slow outflow . although , constructing detailed models of quasi - steady discs is beyond the scope our present work , we simply point out that there are such models that are plausible . in the first model , the disc magnetic field remains connected to the stellar field and a quasi - steady disc forms with inflow from the wind zone and slow radially outward flow with field lines stretching out with the flow ( e.g. , * ? ? ? another possible quasi - steady model is one in which a protodisc evolves after the onset of mri into a magnetic rotator decretion disc ( analogous to the model of * ? ? ? * ) with inflow of super - keplerian wind material . therefore , the centrifugal breakouts found in the numerical simulations of @xcite in uniformly rotating models does not mean that quasi - steady discs can not form around differentially rotating magnetic rotator stars . we have not addressed the question of how protodiscs may be detected observationally . because the duration of the protodisc phase in most cases is rather short , being of the order of a few hours to many days , special observational efforts will be required during disc formation phases to infer their presence . in some special cases , protodiscs can survive for a few years in the form of rings . we gratefully acknowledge the following grants : ( mm ) uwmc summer research grant ; ( jpc ) the nsf center for magnetic self organization in laboratory and astrophysical plasmas grant . we are grateful to john brown for valuable discussions and comments on the paper . we thank an anonymous referee for useful comments on the paper . balbus , s. a. 2003 , ara&a , 41 , 555 balbus , s. a. , & hawley , j. f. 1991 , apj , 376 , 214 bjorkman , j. e. , & cassinelli , j. p. 1993 , apj , 409 , 429 brown , j. , & cassinelli , j. p. 2005 , in asp conf . 337 , the nature and evolution of disks around hot stars , ed . r. ignace & k. gayley ( san francisco : asp ) brown , j. , cassinelli , j. p. , & maheswaran , m. 2008 , apj , 688 , 1320 brown , j. , telfer , d. , li q. , hanuschik r. , cassinelli , j. p. , & kholtygin , a. 2004 , mnras , 352 , 1061 cassinelli , j. p. , brown , j. c. , maheswaran , m. , miller , n. a. , & telfer , d. c. 2002 , apj , 578 , 951 cohen , d. h. , cassinelli , j. p & macfarlane , j. j. 1997 , apj , 487 , 867 cox , d. h. in allen s astrophysical quantities , 2000 , ed a. n. cox ( new york : aip press : springer ) donati , j .- f . , howarth , i. d. , jardine , m. m. , petit , p .. catala , c. , landstreet , j. d. , bouret , j .- c . ; , alecian , e. , barnes , j. r. , forveille , t. , paletou , f. & manset , n. 2006 , mnras , 370 , 629 hawley , j. f. , gammie , c. f. & balbus , s. a. , 1995 , apj , 440 , 742 hawley , j. f. , gammie , c. f. & balbus , s. a. , 1996 , apj , 464 , 690 lamers , h. j. g. l. m. , & cassinelli , j. p. 1999 , introduction to stellar winds , ( new york : cambridge univ . press ) lee , u. , saio , h. & osaki , y. 1991 , mnras , 250 , 432 maeder , a. , georgy , c. & meynet , g. 2008 , a&a , 479 , l37 maheswaran , m. 1968 , mnras , 140 , 93 maheswaran , m. 2003 , apj , 592 , 1156 maheswaran , m. 2005 , in asp conf . 337 , the nature and evolution of disks around hot stars , ed . r. ignace & k. gayley ( san francisco : asp ) mestel , l. 1968 , mnras , 138 , 359 neiner , c. 2007 in asp conference series , vol . 361 p.91 , active ob - stars : laboratories for stellar and circumstellar physics , ed . s. stefl , s. p. owocki & a. t. okazaki . ( san francisco : asp ) okazaki , a. t. 2001 , pasj , 53 , 119 porter , j. m. & rivinius , t. 2003 , pasp , 115 , 1153 pringle , j. e. 1981 , ara&a , 19 , 137 schnerr , r. s. , rygl , k. l. j. ; , van der horst , a. j. , oosterloo , t. a. , miller - jones , j. c. a. , henrichs , h. f. , spoelstra , t. a. th . ; & foley , a. r. 2007 , a&a , 470 , 1105s shakura , n. i. & sunyaev , r. a. 1973 , a&a , 24 , 337 strittmatter , p. a. 1970 , ara&a , 7 , 665s tassoul , j. l. , 2000 , stellar rotation , ( new york : cambridge univ . press ) ud doula , a. , townsend , r. h. d. & owocki , s. p. 2006 , apj , 640 , l191 ud doula , a. , owocki , s. p. & townsend , r. h. d. 2008 , mnras , 385 , 97

the study of traversable wormholes is very hot topic for the past 30 years . one of the best possible way to make traversable wormhole is using the thin - shells to cut and paste two spacetime which has tunnel from one region of space - time to another , through which a traveler might freely pass in wormhole throat . these geometries need an exotic matter which involves a stress - energy tensor that violates the null energy condition . however , this method can be used to minimize the amount of the exotic matter . the goal of this thesis study is to study on thin - shell and thin - shell wormholes in general relativity in 2 + 1 and 3 + 1 dimensions . we also investigate the stability of such objects . * keywords * : wormhole , thin - shells , junction conditions , black holes . solucan delikleri bilim ve bilim kurgu dnyasinin en popler konularindan biridir . 30 sene boyunca popleritesini daha da artirdi . olasi solucan delii yapabilmek iin en kullanilive kararli yntemlerden biri einsteinin yerekimi kurami ierisinde ince - kabuklu uzay solucan delii yapmaktir . bu kuramlarda nemli olani geii yapacak olanin , solucan deliinin boazindan serbeste geiine olanak vermesi ve belirli artlari salamasidir , ve egzotik madde miktarini en dk seviyeye ekebilmektir . bu alimada kara deliklerin etrafinda oluabilecek ince - kabuklu zari , ve bunlari kullanarak kararli yapida solucan delii oluturmaya alitik . * anahtar kelimeler * : solucan delikleri , kara delikler , `` imagination is more important than knowledge . '' - albert einstein i would like to express my deep gratitude to prof . dr . mustafa halilsoy , my supervisor , for his patient guidance , enthusiastic encouragement and useful critiques of this research work . i also thank to my co - supervisor assoc . prof . dr . habib mazharimousavi for willingness to spend his time discussing to me and for help on some difficult calculations . i would also like to thank assoc . prof . dr . izzet sakalli , for his advice and assistance in keeping my progress on schedule . my grateful thanks are also extended to prof . dr . zay grtu . i would like to thank my friends in the department of physics and chemistry , the gravity and general relativity group for their support and for all the fun we have had during this great time . i wish to thank also other my friends for their support and encouragement throughout my study . finally , i would like to thank my family . * this ph.d thesis is based on the following 13 sci and 1 sci - expanded papers :* 1 . thin - shell wormholes from the regular hayward black hole , m. halilsoy , * a. ovgun * and s. h. mazharimousavi , _ eur . phys . j. c 74 , 2796 ( 2014 ) . _ 2 . tunnelling of vector particles from lorentzian wormholes in 3 + 1 dimensions , i. sakalli and * a. ovgun * , _ eur . phys . j. plus 130 , no . 6 , 110 ( 2015 ) . _ 3 . on a particular thin - shell wormhole , * a. ovgun * and i. sakalli , _ arxiv:1507.03949 ( accepted for publication in theoretical and mathematical physics ) . _ + * other papers by the author : * 4 . existence of wormholes in the spherical stellar systems , * a. ovgun * and m. halilsoy , _ astrophys space sci 361 , 214 ( 2016 ) . _ 5 . gravitinos tunneling from traversable lorentzian wormholes , i. sakalli and * a. ovgun * , _ astrophys . space sci . 359 , 32 ( 2015 ) . _ 6 . gravitational lensing effect on the hawking radiation of dyonic black holes , i. sakalli , * a. ovgun * and s. f. mirekhtiary . _ int . j. geom . meth . mod . phys . 11 , no . 08 , 1450074 ( 2014 ) . _ 7 . uninformed hawking radiation , i. sakalli and * a. ovgun * , _ europhys . lett . 110 , no . 1 , 10008 ( 2015 ) . _ 8 . hawking radiation of spin-1 particles from three dimensional rotating hairy black hole , i. sakalli and * a. ovgun * , _ j. exp.theor . phys . 121 , no . 3 , 404 ( 2015 ) . _ 9 . quantum tunneling of massive spin-1 particles from non - stationary metrics , i. sakalli and * a. ovgun . * , _ gen . rel . grav . 48 , no . 1 , 1 ( 2016 ) . _ 10 . entangled particles tunneling from a schwarzschild black hole immersed in an electromagnetic universe with gup , * a. ovgun * , _ int . j. theor . phys . 55 , 6 , 2919 ( 2016 ) . _ 11 . hawking radiation of mass generating particles from dyonic reissner nordstrom black hole , i. sakalli and * a. ovgun * , _ arxiv:1601.04040 ( accepted for publication in journal of astrophysics and astronomy ) . _ 12 . tunneling of massive vector particles from rotating charged black strings , k. jusufi and * a. ovgun * , _ astrophys space sci 361 , 207 ( 2016 ) . _ 13 . massive vector particles tunneling from noncommutative charged black holes and its gup - corrected thermodynamics , * a. ovgun * and k. jusufi , _ eur . phys . j. plus 131 , 177 ( 2016 ) . _ 14 . black hole radiation of massive spin-2 particles in ( 3 + 1 ) dimensions , i. sakalli , * a. ovgun * , _ eur . phys . j. plus 131 , 184 ( 2016 ) .

1610.08118

100 years ago , albert einstein presented to the world his theory of general relativity ( gr ) which is said that space and time are not absolute , but can be distorted or warped by matter . einstein s genius was in his willingness to confront the contradictions between different branches of physics by questioning assumptions so basic that nobody else saw them as assumptions . one prediction of the gr is that in particular , a cataclysmic event such as the collapse of a star could send shock waves through space , gravitational waves ( gws),discovered and announced in february 2016 by ligo ( laser interferometer gravitational - wave observatory ) @xcite . the gws the other proposal that matter can warp space and time leads to many other predictions , notably that light passing a massive body will appear bent to a distant observer , whereas time will appear stretched . each of these phenomena has been observed ; the bending of distant starlight by the sun was first observed in 1919 , and the synchronization of clocks in gps satellites with earthbound clocks has to take account of the fact that clocks on earth are in a strong gravitational field . another prediction of gr is that if enough matter is concentrated in a small volume , the space in its vicinity will be so warped that it will curve in on itself to the extent that even light can not escape@xcite . astronomical observations suggest that each galaxy has a super massive black hole ( bh ) at its centre@xcite . mind - blowing concepts wormholes ( whs ) can be considered his last prediction , which is also known as einstein rosen bridge@xcite . it is a short - cut connecting two separate points in spacetime . moreover , gr is used as a basis of nowadays most prominent cosmological models@xcite . einstein s equations start to break down in the singularities of bhs . before going to study deeply bhs and whs , lets shortly review the gr . the gr is the einstein s theory of gravity and it is a set of non - linear partial differential equations ( pdes ) . the einstein field equations are @xcite @xmath0 where @xmath1 is the newton constant , @xmath2 is a einstein tensor and @xmath3 is a energy momentum tensor , which includes both energy and momentum densities as well as stress ( that is , pressure and shear ) , which must satisfy the relation of @xmath4 . this relation can be called as the equation of motion for the matter fields . furthermore , the einstein tensor @xmath2 is also divergence free @xmath5 noted that the einstein tensor @xmath2 is @xmath6where @xmath7 is called the ricci tensor @xcite @xmath8the ricci scalar of curvature scalar @xmath9 is the contraction of the ricci tensor.@xmath10where the christoffel symbols @xmath11 are defined by @xmath12 in which @xmath13 is the inverse of the metric function and it is the main ingredient of the einstein field equations . gr is based on two important postulates . the first one is the principle of special relativity . the second one is the equivalence principle . einstein s ` newton s apple ' which in sighted to gravitation was the equivalence principle . as an example , consider two elevators one is at rest on the earth and the other is accelerating in space . inside the elevator suppose that there is no windows so it is impossible to realize the difference between gravity and acceleration . thus , it gives the same results as observed in uniform motion unaffected by gravity . in addition , gravity bends light in which a photon crossing the elevator accelerating into space , the photon appears to fall downward . a bh is an object that is so compact that its gravitational force is strong enough to prevent light or anything else from escaping . it has a singularity where all the matter in it is squeezed into a region of infinitely small volume . there is an event horizon which is an imaginary sphere that measures how close to the singularity you can safely get@xcite . by far the most important solution in this case is that discovered by karl schwarzschild , which describes spherically symmetric vacuum spacetimes . the fact that the schwarzschild metric is not just a good solution , but is the unique spherically symmetric vacuum solution , is known as birkhoff s theorem . it only has a mass , but no electric charge , and no spin . karl schwarzschild discovered this bh geometry at the close of 1915@xcite , within weeks of einstein presenting his final theory of gr . real bhs are likely to be more complicated than the schwarzschild geometry : real bhs probably spin , and the ones that astronomers see are not isolated , but are feasting on material from their surroundings . the no - hair theorem states that the geometry outside ( but not inside ! ) the horizon of an isolated bh is characterized by just three quantities : mass , electric charge , spin . bhs are thus among the simplest of all nature s creations . when a bh first forms from the collapse of the core of a massive star , it is not at all a no - hair bh . rather , the newly collapsed bh wobbles about , radiating gws . the gws carry away energy , settling the bh towards a state where it can no longer radiate . this is the no - hair state . in this thesis , we will use some of the exact solution of the einstein field equations to construct whs and thin - shells . in gr , birkhoff s theorem states that any spherically symmetric solution of the vacuum field equations must be static and asymptotically flat . this means that the exterior solutions ( i.e. the spacetime outside of a spherical , non - rotating , gravitating body ) must be given by the schwarzschild metric . @xmath14 noted that it is singular at @xmath15to see that this is a true singularity one must look at quantities that are independent of the choice of coordinates . one such important quantity is the kretschmann scalar , which is given by @xmath16 note that schwarzschild solution has those properties : spherically symmetric , static , coordinates adapted to the time - like killing vector field @xmath17 , time - symmetric and time translation invariant , a hyper - surface - orthogonal time - like killing vector field @xmath18 , asymptotically flat and geometric mass @xmath19 . wormholes ( whs ) is a hypothetical connection between widely separated regions of space time @xcite . although flamm s work on the wh physics dates back to 1916 , in connection with the newly found schwarzschild solution @xcite , wh solutions were firstly considered from physics standpoint by einstein and rosen ( er ) in 1935 , which is known today as er bridges connecting two identical sheets @xcite . then in 1955 wheeler used `` geons '' ( self - gravitating bundles of electromagnetic fields ) by giving the first diagram of a doubly - connected - space @xcite . wheeler added the term `` wormhole '' to the physics literature , however he defined it at the quantum scale . after that , first traversable wh was proposed by morris - thorne in 1988 morris . then morris , thorne , and yurtsever investigated the requirements of the energy condition for whs @xcite . after while , visser constructed a technical way to make thin - shell whs which thoroughly surveyed the research landscape as of 1988 @xcite . after this , there are many papers written to support this idea @xcite . whs are existed in the theory of gr , which is our best description of the universe . but experimentally there is no evidence and no one has any idea how they would be created . [ fig : wormhole ] a wh is any compact region on the space time without any singularity , however with a mouth to allow entrance@xcite . firstly one considers an interesting exotic spacetime metric and solves the einstein field equation , then finds the exotic matter needed as a source responsible for the respective geometry . it is needed the exotic matter which violates the null energy condition,@xmath20 , where @xmath21 is a null vector . it also violates the causality by allowing closed time - like curves . furthermore , the other interesting outcome is that time travel is possible without excess the speed of light . those outcomes are based on theoretical solutions and useful for gedanken - experiments . as it is well known the casimir effect has similar features that violates this condition in nature@xcite . now , lets give an example of a traversable wh metric which is given by @xmath22 the first defined traversable wh is morris thorne wh@xcite @xmath23 where @xmath24 is the red - shift function ( the lapse function ) that change in frequency of electromagnetic radiation in gravitational field and @xmath25 is the shape function . at @xmath26 and @xmath27 , the 2-curved surface is embedded into 3-dimensional euclidean space @xmath28to be a solution of a wh , one needs to impose that the geometrical throat flares out conditions , the minimality of the wormhole throat , which are given by @xcite @xmath29 with new radial coordinate @xmath30 ( proper distance ) , while @xmath31@xmath32although there is a coordinate singularity where the metric coefficient @xmath33 diverges at the throat , the proper radial distance which runs from @xmath34 to @xmath35can be redefined as @xmath36 * properties of the metric : * -no event horizons ( @xmath37 @xmath38 ) , -@xmath24 must be finite everywhere , - spherically symmetric and static : static means that the metric does not change over time and irrotate , on the other way , stationary spacetime means that the metric does not change in time , however it can rotate . hence , the kerr metric is a stationary spacetime and not not static , the schwarzschild solution is an example of static spacetime . -solutions of the einstein field equations , -physically reasonable stress energy tensor , -radial coordinate @xmath39 such that circumference of circle centered around throat given by @xmath40 -@xmath39 decreases from @xmath41 to @xmath42 ( minimum radius ) at throat , then increases from @xmath43 to @xmath44 -at throat exists coordinate singularity where @xmath39 component diverges , -reasonable transit times , -proper radial distance @xmath45 runs from @xmath46 to @xmath41 and vice versa , -throat connecting two asymptotically flat regions of spacetime , -bearable tidal gravitational forces , -stable against perturbations , -physically reasonable construction materials . the einstein field equations are @xcite @xmath47@xmath48@xmath49 , \ ] ] the only non - zero components of @xmath50 are @xmath51@xmath52 @xmath53 @xmath54 \ , , \label{tauwh } \\ p_{t } & = & \frac{1}{8\pi } \left ( 1-\frac{b}{r}\right ) \left [ f^{\prime \prime } + f^{\prime 2}-\frac{b^{\prime } r - b}{2r^{2}(1-b / r)}f^{\prime } -\frac{% b^{\prime } r - b}{2r^{3}(1-b / r)}+\frac{f^{\prime } } { r}\right ] \ , . \label{pressurewh}\end{aligned}\]]at the throat they reduce to the simplest form @xmath55 note that the sign of @xmath56 and energy density @xmath57 must be same to minimize the exotic matter , since the condition of @xmath58 must be satisfied . moreover , the next condition is the flare outward of the embedding surface ( @xmath59 ) at / near the throat . therefore it shows that @xmath60 on this regime , in which the radial pressure is named by @xmath61 . so , the exotic matter which supports this wh spacetime violates the null energy condition . the four velocity is calculated by @xmath62 and four acceleration is @xmath63 , so one obtains @xmath64 in which the prime @xmath65 is `` @xmath66 '' . furthermore , the geodesics equation is used for a radial moving particle to obtain the equation of motion as following @xmath67 in addition , static observers are geodesic for @xmath68 . wh has an attractive feature if @xmath69 and repulsive feature if @xmath70 . the sign of @xmath71 is important for the behaviour of the particle geodesics . the shape function is obtained by @xmath72this is used to find the effective mass of the interior of the wh that gives@xmath73 also at the limit of infinity , we have @xmath74 whs are supported by exotic matter , and the suitable energy condition is defined as diagonal @xcite , @xmath75 in which @xmath76 and @xmath77 denotes to the mass density and the three principal pressures , respectively . perfect fluid of the stress - energy tensor is obtained if @xmath78 . it is believed that the normal matters obey these energy conditions , however , it violates the energy conditions and needs certain quantum fields ( casimir effect ) or dark energy . the nec is @xmath79 where @xmath21 is null vector . using the eq.([diagonalt ] ) , we obtain @xmath80 the wec is @xmath81where the timelike vector is given by @xmath82 . eq.([eq : wec ] ) is for the measured energy density by moving with four - velovity @xmath82 of any timelike observer . it must be positive and the geometric defination refer to the einstein field equations eq.([efe ] ) @xmath83 . it can be written as @xmath84the wec involves the nec . the sec asserts that @xmath85 in which @xmath86 is the trace of the stress energy tensor . because @xmath87 , according to einstein field equations eq . ( [ efe ] ) the sec is a statement about the ricci tensor . then by using the diagonal stress energy tensor given in eq . ( diagonalt ) , the sec reads @xmath88the sec involves the nec , but not necessarily the wec . the dec is @xmath89the energy density must be positive . moreover , the energy flux should be timelike or null . the dec involves the wec , and automatically the nec , but not necessarily the sec . it becomes @xmath90 it can be verified that whs violate all the energy conditions . therefore using the eq.s ( [ rhowh])-([tauwh ] ) with @xmath91 we obtain @xmath92 . \label{necthroat}\]]thanks to the flaring out condition of the throat from eq . ( [ flareout ] ) : @xmath93 , one shows that @xmath94 at the throat and because of the finiteness of @xmath24 , from eq . ( [ necthroat ] ) we have @xmath95 . hence all the energy conditions are violated and this matter is named as the exotic matter . since einstein , hawking s significant addition to understanding the universe is called the most significant@xcite . hawking showed that the bhs are not black but grey which emit radiation . this was the discovery of hawking radiation , which allows a bh to leak energy and gradually fade away to nothing . however , the question of what happens to the information is remains . all the particles should fall into bh and we do not know what happened to them . the particles that come out of a bh seem to be completely random and bear no relation to what fell in . it appears that the information about what fell in is lost , apart from the total amount of mass and the amount of rotation . if that information is truly lost that strikes at the heart of our understanding of science . to understand whether that information is in fact lost , or whether it can be recovered , hawking and colleagues , including andrew strominger , from harvard , are currently working to understand supertranslations to explain the mechanism by which information is returned from a bh and encoded on the hole s event horizon@xcite . in the literature , there exist several derivations of the hawking radiationwilczek1,damour , mann , review , ao3,ao4,ao5,ao6,ao7,ao8,ao9,ao10,ao14,ao15,ao16 . the transmission probability @xmath96 is defined by @xmath97 where @xmath98 is the action of the classically forbidden trajectory . hence , we have found the hawking temperature from the tunneling rate of the emitted particles . for studying the hr of traversable whs , we consider a general spherically symmetric and dynamic wh with a past outer trapping horizon . the traversable wh metric can be transformed into the generalized retarded eddington - finkelstein coordinates as following@xcite@xmath99 where @xmath100 @xmath101 and @xmath102 . the gravitational energy is @xmath103 which is also known as a misner - sharp energy . @xmath104it reduces to @xmath105 on a trapping horizon@xcite . furthermore , there is a past marginal surface at the @xmath106 ( at horizon : @xmath107 ) for the retarded coordinates @xcite . firstly , we give the equation of motion for the vector particles which is known as the proca equation in a curved space - time @xcite:@xmath108 in which the wave functions are defined as @xmath109 by the help of the method of wkb approximation , the following hj anstz is substituted into eq . ( proca ) @xmath110 with the real constants @xmath111 . furthermore , we define the action @xmath112 as following @xmath113 because of the ( [ met2 ] ) is symmetric , the killing vectors are @xmath114 and @xmath115 . then one can use the separation of variables method to the action @xmath116 : @xmath117it is noted that @xmath118 and @xmath119 are energy and real angular constants , respectively . after inserting eqs . ( [ 4psi ] ) , ( [ 5psi ] ) , and ( 6psi ) into eq . ( [ proca ] ) , a matrix equation @xmath120 ( to the leading order in @xmath121 ) is obtained@xmath122 which has the following non - zero components : @xmath123 ^{2}r^{2},\ \notag \\ \delta _ { 12 } & = & \delta _ { 21}=2m^{2}r^{2}b+2b\partial _ { r}w(r)er^{2}+2bj^{2}+2k^{2 } , \notag \\ \delta _ { 13 } & = & -\frac{2\delta _ { 31}}{r^{2}}=-2bj\partial _ { r}w(r),\ \notag \\ \delta _ { 14 } & = & \frac{\delta _ { 41}}{br^{2}}=-2k\partial _ { r}w(r ) , \notag \\ \delta _ { 22 } & = & -2bcm^{2}r^{2}+2e^{2}r^{2}b-2j^{2}bc-2k^{2}c,\ \label{7n } \\ \delta _ { 23 } & = & \frac{-2\delta _ { 32}}{r^{2}}=2jbc\partial _ { r}w(r)+2ejb , \notag \\ \delta _ { 24 } & = & \frac{\delta _ { 42}}{br^{2}}=2kc\partial _ { r}w(r)+2ke , \notag \\ \delta _ { 33 } & = & m^{2}r^{2}b+2ber^{2}\partial _ { r}w(r)+r^{2}bc\left [ \partial _ { r}w(r)\right ] ^{2}+k^{2 } , \notag \\ \delta _ { 34 } & = & \frac{-\delta _ { 43}}{2b}=-kj , \notag \\ \delta _ { 44 } & = & -2r^{2}bc\left [ \partial _ { r}w(r)\right ] ^{2}-4ber^{2}% \partial _ { r}w(r)-2b(m^{2}r^{2}+j^{2 } ) . \notag\end{aligned}\ ] ] the determinant of the @xmath124-matrix ( @xmath125 ) is used to get@xmath126 ^{2}+ber^{2}\partial _ { r}w(r)+\frac{b}{2}\left ( m^{2}r^{2}+j^{2}\right ) + \frac{k^{2}}{2}\right\ } ^{3}=0 . \label{8n}\ ] ] then the eq . ( [ 8n ] ) is solved for @xmath127 @xmath128the above integral near the horizon ( @xmath129 ) reduces to @xmath130 as shown in the eq . ( [ ab1 ] ) , the probability rate of the ingoing / outgoing particles only depend on the imaginary part of the action . eq . ( [ 10int ] ) has a pole at @xmath106 on the horizon . using the contour integration in the upper @xmath39 half - plane , one obtains @xmath131 from which @xmath132 that the @xmath133 is the surface gravity . note that the @xmath134 is positive quantity because the throat is an outer trapping horizon @xcite . when we define the probability of incoming particles @xmath135 to @xmath136 such as @xmath137 . consequently @xmath138 stands for the outgoing particles . then we calculate the tunneling rate of the vector particles as mann , k2 @xmath139 the boltzmann factor @xmath140 @xmath141 where @xmath142 is the inverse temperature is compared with the eq . ( [ 13tn ] ) to obtain the hawking temperature @xmath143 of the traversable wh as @xmath144however @xmath143 is negative , as also shown by @xcite . the main reason of this negativeness is the phantom energy @xcite , which is located at the throat of wh . moreover , as a result of the phantom energy , the ordinary matter can travel backward in time because in qft particles and anti - particles are defined via sign of time . surprisingly , we derive the the negative @xmath143 that past outer trapping horizon of the traversable wh radiate thermal phantom energy ( i.e. dark energy ) . additionally , the radiation of phantom energy has an effect of reduction of the size of the wh s throat and its entropy . nonetheless , this does not create a trouble . the total entropy of universe always increases , hence it prevents the violation of the second law of thermodynamics . moreover , in our different work , we show that the gravitino also tunnels through wh and we calculate the tunneling rate of the emitted gravitino particles from traversable wh . the procedure of dealing with a given surface of discontinuity is well known since the newtonian theory of gravity@xcite . firstly , the continuity of the gravitational potential should be checked and then from the surface mass the discontinuity of the gravitational field might be occured . those boundary conditions are derived from the field equation . notwithstanding , for the case of gr there is different problem because of the nonlinearity of the field equations as well as the principle of general covariance . to solve this headache , on a hypersurface splitting spacetimes one must introduce specific boundary conditions to the induced metric tensor and the extrinsic curvature . this method is called the darmois - israel formalism or the thin - shell formalism@xcite . there are many different using areas of this method such as dynamic thin - layers , connecting branes , quantum fields in thin - shell spacetimes , whs , collapsing shells and radiating sphereslake , counter , eid , es2,sotiriou , israeljunction , mv1,mv2,mv3,mv4,mv5,mv6plot , loboreview0,loboreview1,loboreview2,simeo , bhar . it is well known that this method and the searching for distributional solutions to einstein s equations are same . this method is also used for the stars that are expected to display interfacial layers much smaller than their characteristic sizes with nontrivial quantities , such as surface tensions and surface energy densities . another example is compact stars with interfaces separating their cores and their crusts , ( i.e. strange quark stars and neutron stars)@xcite . note that the darmois - israel formalism gives the nontrivial properties of transitional layers fully taking into account gr , but just under the macroscopic point of view . furthermore , the thin - shell formalism would be the proper formalism for approaching any gravitational system that presents discontinuous behaviors in their physical parameters . on the other hand , this method can be used to screen the bh s hairs against the outside observer at infinity@xcite . one should show that the thin - shell is a stable under the perturbations . our main aim is to apply such a formalism to the general charge carrying rotating bh solution in 2 + 1 dimensions . the bh has a thin - shell which has a radius greater than the event horizon@xcite . once this situation is occurred , we check the stability analysis . the metric for the general rotating bh solution in 2 + 1 dimensions is given as @xcite@xmath145 ^{2}\ ] ] by using the famous cut - and - paste method introduced by help of darmois - israel junction conditions , the thin - shell wh is constructed . firstly we take two copies of the bulk @xmath146with the line elements given above . then we paste them at an identical hypersurface @xmath147for convenience we move to a comoving frame to eliminate cross terms in the induced metrics by introducing @xmath148 then for interior and exterior of wh , it becomes @xmath149 ^{2}.\ ] ] the geodesically complete manifold is satisfied at the hypersurface @xmath150 which we shall call the throat . we define the throat for the line element by@xmath151first of all the throat must satisfy the israel junction conditions so @xmath152and it is found that@xmath153and @xmath154 in which a @xmath155 stands for the derivative with respect to the proper time @xmath156 second step is the satisfaction of the einstein s equations in the form of israel junction conditions on the hypersurface which are @xmath157 in which @xmath158 @xmath159 and the extrinsic curvature with embedding coordinate @xmath160 @xmath161 the parametric equation of the hypersurface @xmath150 is given by @xmath162 [ fig : thinshel ] and the normal unit vectors to @xmath163 defined by @xmath164where @xmath165@xmath166@xmath167so they are found as follows @xmath168@xmath169 the normal unit vector must be satisfied @xmath170 non - zero normal unit vectors are calculated as @xmath171 @xmath172 @xmath173it reduces to@xmath174 before calculating the components of the extrinsic curvature tensor , we redefine the metric of the bulk in 2 + 1 dimensions given as @xmath175 ^{2},\ ] ] where @xmath176 it becomes @xmath177 dt^{2}+\frac{dr^{2}}{u(r)}% + r^{2}d\psi ^{2}+2r^{2}\omega ( r)dtd\psi . \label{metr}\ ] ] note that the line element on the throat is @xmath151 and the corresponding levi - civita connections which is defined as @xmath178for the metric given in eq.([metr ] ) these are calculated as @xmath179@xmath180 , \]]@xmath181@xmath182@xmath183@xmath184@xmath185and@xmath186 one finds the extrinsic curvature tensor components using the definitiongiven in eq.([k ] ) @xmath187@xmath188note that on the hyperplane i.e @xmath189 , @xmath190 . the extrinsic curvature for tau tau is @xmath191after substituting all the variables it becomes @xmath192also it is found that the psi - psi component of the extrinsic curvature is @xmath193@xmath194@xmath195lastly the tau - psi component of the extrinsic curvature is also found as @xmath196 @xmath197@xmath198@xmath199 we can write them also in the following form @xmath200@xmath201@xmath202and @xmath203 for a thin - shell with different inner and outer spacetime , they become @xmath204@xmath205@xmath206and @xmath207 as a result one obtains @xmath208@xmath209@xmath210 in this section , we briefly review the darmois - israel junction conditions @xcite . the action of gravity is @xmath211where the first term is einstein - hilbert action and second term is gibbons - hawking boundary action term . @xmath212 the variation of this action is @xmath213@xmath214 @xmath215\]]where@xmath216 @xmath217@xmath218\]]@xmath219 we have found that @xmath220@xmath221where @xmath222 @xmath223and @xmath207 @xmath224@xmath225-[k]\ ] ] in which @xmath226=a^{+}-a^{-}.$ ] also @xmath104@xmath227 @xmath228 @xmath229@xmath230@xmath231then for other components@xmath232@xmath233@xmath234\]]and the last component is @xmath235 the special condition of @xmath236 @xmath237 so @xmath238therefore it implies that the upper - shell and the lower - shell are corotating . the surface stress - energy tensor is @xmath239 where@xmath240and @xmath241\ ] ] the case of the static is obtained by assuming @xmath242 and @xmath243 , @xmath244and @xmath245 .\ ] ] the darmois - israel junction condition for connecting a hypersurface @xmath246 with a hypersurface @xmath247 can be written as @xmath248 = 0 \label{eqn - jumpg}\]]and @xmath249 = 0 \label{eqn - jumpk}\ ] ] the boundary surface @xmath150 is defined when both ( [ eqn - jumpg ] ) and ( [ eqn - jumpk ] ) are satisfied . if only ( [ eqn - jumpg ] ) is satisfied then we refer to @xmath150 as a thin - shell . + conditions ( [ eqn - jumpg ] ) and ( [ eqn - jumpk ] ) require a common coordinate system on @xmath250 and this is easily done if one can set @xmath251 . failing this , establishing ( [ eqn - jumpg ] ) requires a solution to the three dimensional metric equivalence problem . after the signs of normal vector is choosen , there is no ambiguity in ( eqn - jumpk ) and ( [ eqn - jumpg ] ) and ( [ eqn - jumpk ] ) are used in conjunction with the einstein tensor @xmath252 to calculate the identities @xmath253 = 0 \label{eqn - gnn}\]]and @xmath254 = 0 . \label{eqn - gfn}\ ] ] this shows that for timelike @xmath150 the flux through @xmath150 ( as measured comoving with @xmath150 ) is continuous . + the israel formulation of thin shells follows from the lanczos equation @xmath255 - g_{ij } \left[{k_i}^i % \right])\ ] ] and we refer to @xmath256 as the surface stress - energy tensor of @xmath250 . the `` adm '' constraint @xmath257 along with einstein s equations then gives the conservation _ identity _ @xmath258.\ ] ] the `` hamiltonian '' constraint @xmath259 gives the evolution _ identity _ @xmath260.\ ] ] the dynamics of the thin - shell are not understood from the identities ( eqn - claw ) and ( [ eqn - elaw ] ) . the evolution of the thin - shell is obtained by the lanczos equation([eqn - lanc ] ) @xcite . the @xmath261 and @xmath262 are used to satisfy the energy condition @xmath263 where @xmath264 } { \sqrt{u_{+}+\dot{a}^{2}}% \sqrt{u_{-}+\dot{a}^{2}}}\ ] ] note that `` @xmath265 '' prime stands for the derivative respect to @xmath266 and the energy on the shell is not conserved . another relation between the energy density and pressure which is much helpful is the energy conservation relation which is given by@xmath267this relation must be satisfied by @xmath262 and @xmath261 even after the perturbation which gives an inside to the problem . the idea is to perturb the shell while it is at the equilibrium point @xmath268 . by using the eq.([sag ] ) , we apply the perturbation and find the equation of motion of the shell ( note that @xmath269 ) is@xmath270 where @xmath271 this one dimensional equation describes the nature of the equilibrium point whether it is a stable equilibrium or an unstable one . to see that we expand @xmath272 about @xmath268 and keep the first non - zero term which is @xmath273one can easily show that @xmath274 and therefore everything depends on the sign of @xmath275 let s introduce @xmath276 and write the equation of motion again@xmath277which after a derivative with respect to time it reduces to@xmath278this equation explicitly implies that if @xmath279 the @xmath280 will be an oscillating function about @xmath281 with the angular frequency @xmath282 but otherwise i.e. , @xmath283 the motion will be exponentially toward the initial perturbation . therefore our task is to find @xmath284 and show that under what condition it may be positive for the stability and negative for the instability of the shell . can naturally formed absorber thin - shells , in cosmology to hide the reality from our telescopes ? this can be revise the ideas of no - hair black hole theorem . thin - shell can be used to find exhibiting a remarkable property of qcd - like charge confinement to cover the dirac and yang - mills fields . the charged interior spacetime is completely con- fined within the finite - spacial - size analog of qcd quark confinement . naturally this takes us away from classical physics into the realm of gravity coupled qcd . let s set the lapse function of @xmath285 for the inner shell which is de sitter spacetime with mass @xmath286 and outer shells which is a btz bh with mass @xmath287and charge @xmath288 to be as follows@xcite @xmath289 and @xmath290 where s and l are constants . after some calculations , we obtain that the energy density and the pressures can be recast as@xmath291and @xmath292 for a static configuration of radius a , we obtain ( assuming @xmath242 and @xmath293 ) @xmath294and @xmath295 to obtain the stability criterion , one starts by rearranging eq . ( [ stabsig0 ] ) in order to obtain the equation of motion @xmath296where the @xmath297 is the potential as @xmath298 now we impose the energy conservation condition which must be satisfied after the perturbation and try to find out weather the motion of the shell is oscillatory or not . this openly means a relation between @xmath261 and @xmath299 finally in order to have the thin - shell stable against radial perturbation , @xmath300 at the equilibrium point i.e. , @xmath268 where @xmath301to keep our study as general as possible we assume @xmath261 to be an arbitrary function of @xmath142 and @xmath262 i.e. , @xmath302where @xmath303 in fig.2.2 we plot @xmath304 for the specific value of @xmath305 and @xmath306 as one observes in the region with @xmath307 the thin - shell is stable while otherwise it occurs for @xmath308 [ fig : thinshell1btz ] the generalized spacetime metric is given as follows@xmath309 ^{2}.\ ] ] we define the throat for the line element by@xmath151the throat must satisfy the israel junction conditions so @xmath310and it is found that@xmath311and @xmath312 in which a @xmath155 stands for the derivative with respect to the proper time @xmath156 for convenience we move to a comoving frame to eliminate cross term in the induced metrics by introducing @xmath313 then for interior and exterior of wh , it becomes @xmath314 ^{2},\ ] ] where @xmath176 the parametric equation of the hypersurface @xmath150 is given by @xmath162and the normal unit vectors to @xmath163 defined by @xmath164where @xmath315 @xmath166@xmath316 in which @xmath317@xmath318 @xmath319 so @xmath320 the normal unit vector must satisfy @xmath170 so that non - zero normal unit vectors are calculated as @xmath321 @xmath322 so corrresponding unit vectors are @xmath323 now we will calculate the extrinsic curvatures @xmath187@xmath324 where the levi - civita connections calculated @xmath325 ; @xmath326@xmath327@xmath328 @xmath329 @xmath330 @xmath331 and@xmath332 so @xmath333 @xmath334@xmath335 @xmath336 @xmath337\ ] ] also it is found that the phi , phi component of the extrinsic curvature is @xmath193@xmath338 and lastly the tau , phi component of the extrinsic curvature is also found as @xmath339 @xmath340@xmath341@xmath342 finally @xmath337\]]@xmath343@xmath344 in which @xmath345 ^{\prime } = h(r)^{\prime } .$ ] they become @xmath346 , \]]@xmath347@xmath348and @xmath349 if we take @xmath350 with @xmath351 one finds same as part 1 . now we define the junction conditions by using the exterior curvatures as follows@xmath352 , \]]@xmath353@xmath354and @xmath355 @xmath356 + \frac{1}{ag_{\pm } } \sqrt{\theta } \]]@xmath225-[k]\delta^{j}_{i}\ ] ] where @xmath357 $ ] is the trace of @xmath358 $ ] and @xmath359 is the surface stress - energy tensor on @xmath262 , and @xmath226=a^{+}-a^{-}$ ] . also @xmath360 @xmath228 @xmath229@xmath230@xmath361then for other components@xmath232@xmath233@xmath362@xmath363\ ] ] and the last component is @xmath364 @xmath365 the special condition of @xmath236 @xmath237 so @xmath238therefore it implies that the upper - shell and the lower - shell are co - rotating . the surface stress - energy tensor is @xmath366 one calculates the charge density and the surface pressure as follows@xmath367and @xmath368@xmath363\ ] ] thin - shell whs is constructed with the exotic matter which is located on a hypersurface so that it can be minimized . constructing whs with non - exotic source is a difficult issue in gr . on this purpose , firstly , visser use the thin - shell method to construct whs for minimizing the exotic matter on the throat of the whs . we need to introduce some conditions on the energy - momentum tensor such as@xcite [ fig : thin - shell wormhole ] -weak energy condition this energy condition states that energy density of any matter distribution must be non - negative , i.e. , @xmath369 and @xmath370 . - null energy condition this condition implies that @xmath370 . - dominant energy condition this condition holds if @xmath369 and @xmath371 . - strong energy condition this condition demands @xmath370 and @xmath372 in this section @xcite , we introduce a shbh ( shbh ) investigated by mazharimousavi and halilsoy recently @xcite . the following action describes the einstein - maxwell gravity that is minimally coupled to a scalar field @xmath373 @xmath374where @xmath9 denotes the ricci scalar , @xmath375 is the maxwell invariant , and @xmath376 stands for the scalar ( @xmath373 ) potential . from the action above , the shbh solution is obtained as @xmath377which @xmath378here @xmath379 and @xmath30 are constants , and event horizon of the bh is located at @xmath380 . it is clear that this bh possesses a non - asymptotically flat geometry . metric ( [ eq2 ] ) can be written in the form of @xmath381it is noted that the singularity located at @xmath382 , which is also seen from the ricci and kretschmann scalars : @xmath383@xmath384moreover , one obtains the scalar field and potential respectively as follows @xmath385 @xmath386 in which @xmath387 are constants . the corresponding hawking temperature is calculated as @xmath388which is constant . having a radiation with constant temperature is the well - known isothermal process . it is worth noting that hawking radiation of the linear dilaton bhs exhibits similar isothermal behavior . in this section , we take two identical copies of the shbhs with @xcite @xmath389 : @xmath390and the manifolds are bounded by hypersurfaces @xmath391 and @xmath392 , to get the single manifold @xmath393 , we glue them together at the surface of the junction @xmath394where the boundaries @xmath150 are given . the spacetime on the shell is@xmath395 where @xmath396 represents the proper time . setting coordinates @xmath397 , the extrinsic curvature formula connecting the two sides of the shell is simply given by @xmath398where the unit normals ( @xmath399 are @xmath400with @xmath401 . the non zero components of @xmath402 are calculated as @xmath403@xmath404where the dot over a quantity denotes the derivative with respect to @xmath396 . then , the non - zero extrinsic curvature components yield @xmath405@xmath406 since @xmath407 is not continuous around the shell , we use the lanczos equation : @xmath408-[k]g_{ij}\right ) . \label{ee}\ ] ] where @xmath409 is the trace of @xmath407 , @xmath410=k_{ij}^{+}-k_{ij}^{-}$ ] . firstly , @xmath411 $ ] while @xmath410=0.$ ] for the conservation of the surface stress energy @xmath412 and @xmath256 is stress energy - momentum tensor at the junction which is given in general by @xmath413with the surface pressure @xmath261 and the surface energy density @xmath262 . due to the circular symmetry , we have @xmath414.\]]thus , from eq.s ( [ enerji ] ) and ( [ ee ] ) one obtains the surface pressure and surface energy density . using the cut and paste technique , we can demount the interior regions @xmath415 of the geometry , and links its exterior parts . the energy density and pressure are @xmath416@xmath417 then for the static case ( @xmath268 ) , the energy and pressure quantities reduce to @xmath418@xmath419 once @xmath420 and @xmath421 hold , then wec is satisfied . besides , @xmath421 is the condition of nec . furthermore , sec is conditional on @xmath421 and @xmath422 . it is obvious from eq . ( 24 ) that negative energy density violates the wec , and consequently we are in need of the exotic matter for constructing thin - shell wh . we note that the total matter supporting the wh is given by @xmath423\right\vert _ { r = a_{0}}d\phi = 2\pi a_{0}\sigma ( a_{0})=-\frac{1}{4a_{0}^{\frac{1}{2}}|l|}% \sqrt{-l^{2}u+a_{0}}.\ ] ] stability of the wh is investigated using the linear perturbation so that the eos is @xmath424where @xmath425 is an arbitrary function of @xmath262 . furthermore , the energy conservation equation is introduced as follows @xmath426where @xmath427 is the bulk energy - momentum tensor . it can be written in terms of the pressure and energy density : @xmath428from above equation , one reads @xmath429and its second derivative yields@xmath430 where prime and tilde symbols denote derivative with respect to @xmath266 and @xmath431 , respectively . the equation of motion for the shell is in general given by @xmath432where the effective potential @xmath433 is found from eq . ( [ ed ] as @xmath434 in fact , eq . ( [ eqmt ] ) is nothing but the equation of the oscillatory motion in which the stability around the equilibrium point @xmath268 is conditional on @xmath435 . we finally obtain @xmath436 \right\vert _ { a = a_{0}},\]]or equivalently , @xmath437 -u\}\right\vert _ { a = a_{0}}.\ ] ] the equation of motion of the throat , for a small perturbation becomes @xmath438 note that for the condition of @xmath435 , tsw is stable where the motion of the throat is oscillatory with angular frequency @xmath439 . in this section , we use particular gas models ( linear barotropic gas ( lbg ) , chaplygin gas ( cg ) , generalized chaplygin gas ( gcg ) and logarithmic gas ( logg ) ) to explore the stability of tsw . the equation of state of lbg is given by @xmath440and hence @xmath441where @xmath442 is a constant parameter . by changing the values of @xmath443 and @xmath379 in eq . ( 35 ) , we illustrate the stability regions for tsw , in terms of @xmath442 and @xmath444 , as depicted in fig.3.2 . [ fig : stabilitylbg ] the equation of state of cg that we considered is given by @xmath445 and one naturally finds @xmath446 after inserting eq . ( 39 ) into eq . ( 35 ) , the stability regions for thin - shell wh supported by cg is plotted in fig.3.3 . [ fig : stabilitycg ] by using the equation of state of gcg @xmath447 and whence @xmath448 substituting eq . ( 41 ) in eq . ( 35 ) , one can illustrate the stability regions of thin - shell wh supported by gcg as seen in fig.3.4 . [ fig : stabilitygcg ] in our final example , the equation of state for logg is selected as follows ( @xmath449 @xmath450 are constants ) @xmath451which leads to @xmath452 after inserting the above expression into eq . ( 35 ) , we show the stability regions of thin - shell wh supported by logg in fig.3.5 . [ fig : stabilitylogg ] in summary , we have constructed thin - shell wh by gluing two copies of shbh via the cut and paste procedure . to this end , we have used the fact that the radius of throat must be greater than the event horizon of the metric given : ( @xmath453 ) . we have used lbg , cg , gcg , and logg eos to the exotic matter . then , the stability analysis ( @xmath435 ) is plotted . we show the stability regions in terms @xmath444 and@xmath442 . the problem of the angular perturbation is out of scope for the present paper . that s why we have only worked on the linear perturbation . however , angular perturbation is in our agenda for the extension of this study . this is going to be studied in the near future . the metric of the hayward bh is given by @xcite @xmath454with the metric function @xmath455and @xmath456it is noted that @xmath457 and @xmath30 are free parameters . at large @xmath458 the metric function behaves@xmath459whereas at small @xmath39 @xmath460one observes that for small @xmath39 the hayward bh becomes a de sitter bh and for large @xmath39 it is a schwarzschild spacetime . the event horizon of the hayward bh is calculated by using@xmath461and changing @xmath462 and @xmath463 , it turns to@xmath464note that for @xmath465 there is no horizon , for @xmath466 single horizon which is called a extremal bh and for @xmath467 double horizons . hence the ratio @xmath468 is important parameter where the critical ratio is at @xmath469 . set @xmath470 where @xmath471 for the case of @xmath472 the event horizon is given by@xmath473{\delta } + \frac{4}{\sqrt[3]{\delta } } % + 2\right)\]]with @xmath474 the extremal bh case occurs at @xmath475 and the single horizon occurs at @xmath476 when @xmath477 , the temperature of hawking is given by@xmath478which clearly for @xmath475 vanishes and for @xmath479 is positive so note that @xmath480 . entropy for the bh is obtained by @xmath481 with @xmath482 to find the heat capacity of the bh @xmath483and it is obtained as@xmath484it is clearly positive.when the heat capacity of the bh is positive @xmath485 , it shows the bh is stable according to thermodynamical laws . to find the source of the hayward bh , the action is considered as @xmath486where @xmath9 is the ricci scalar and the nonlinear magnetic field lagrangian density is @xmath487 ^{2}}=-\frac{6}{l^{2}\left [ 1+\left ( \frac{% \beta } { f}\right ) ^{3/4}\right ] ^{2 } } \label{lhay}\]]with the maxwell invariant@xmath488with two constant positive parameters @xmath30 and @xmath142 . the analyses of the stability depends on the fixing the @xmath142 . moreover , the magnetic field is@xmath489where the charge of the magnetic monopole is @xmath490 . it implies @xmath491with the line element given in eq.([haywardline ] ) . the einstein - nonlinear electrodynamics field equations are ( @xmath492 ) @xmath493 in which @xmath494with @xmath495 after using the nonlinear magnetic field lagrangian @xmath496 inside the einstein equations , one finds @xmath497 for the hayward regular bh@xmath498 the limit of the weak field of the @xmath499 is found by expanding it around @xmath500,@xmath501 note that at the limit of the weak field , the lagrangian of the ned does not reduce to the lagrangian of the linear maxwell @xmath502hence , this spacetime has not any reissner nordstrm limit at weak field . we use the cut and past technique to constructe a thin - shell wh from the hayward bhs . we firstly take a thin - shell at @xmath189 where the throat is outside of the horizon ( @xmath503 ) . then we paste two copies of it at the point of @xmath189 . for this reason the thin - shell metric is taken as@xmath504where @xmath396 is the proper time on the shell . the einstein equations on the shell are @xmath505 -\left [ k\right ] \delta _ { i}^{j}=-s_{i}^{j}\]]where @xmath506 = x_{2}-x_{1},$ ] . it is noted that the extrinsic curvature tensor is @xmath507 . moreover , @xmath409 stands for its trace . the surface stresses , i.e. , surface energy density @xmath262 and surface pressures @xmath508 , are determined by the surface stress - energy tensor @xmath359 . the energy and pressure densities are obtained as @xmath509 @xmath510then they reduce to simple form in a static configuration ( @xmath268)@xmath511and @xmath512stability of such a wh is investigated by applying a linear perturbation with the following eos @xmath513moreover the energy conservation is@xmath514which in closed form it equals to@xmath515after the line element in eq.([elem ] ) is used , it opens to @xmath516the 1-d equation of motion is @xmath517 in which @xmath518 is the potential , @xmath519the equilibrium point at @xmath268 means @xmath520 and @xmath521 then it is considered that @xmath522 , one finds @xmath523 to obtain @xmath524 we use the given @xmath525 and it is found as follows@xmath526and @xmath527where @xmath528 after we use @xmath529 finally it is found that@xmath530\ ] ] in this section , we consider some specific models of matter such as linear gas ( lg ) , chaplygin gas ( cg ) , generalized chaplygin gas ( gcg ) , modified generalized chaplygin gas ( mgcg ) and logarithmic gas ( logg ) to analyze the effect of the parameter of hayward in the stability of the constructed thin - shell wh . for a lg , eos is choosen as@xmath531 in which @xmath532 is a constant and @xmath533 fig.[fig.lg ] shows the stability regions in terms of @xmath534 and @xmath444 with different hayward s parameter . it is noted that the @xmath98 shows the stable regions . for cg , we choose the eos as follows@xmath535where @xmath532 is a constant and @xmath536 in fig.[fig.cg ] , the stability regions are shown in terms of @xmath532 and @xmath444 for different values of @xmath537 the effect of hayward s constant is to increase the stability of the thin - shell wh . the eos of the gcg is taken as@xmath538where @xmath539 and @xmath532 are constants . we check the effect of parameter @xmath539 in the stability and @xmath540 becomes @xmath541we find @xmath542 . in fig.[fig.gcg ] , the stability regions are shown in terms of @xmath539 and @xmath543 with various values of @xmath537 in this case , the mgcg is @xmath544 in which @xmath545 , @xmath532 and @xmath539 are free parameters . therefore,@xmath546to go further we set @xmath547 and @xmath548 . in fig.[fig.mgcg ] , the stability regions are plotted in terms of @xmath532 and @xmath444 with various values of @xmath549 . the effect of hayward s constant is to increase the stability of the thin - shell wh . lastly logg is choosen by follows@xmath550 in which @xmath532 is a constant . for logg , we find that @xmath551 in fig.[fig.logg ] , the stability regions are plotted to show the effect of hayward s parameter clearly . the effect of hayward s constant is to increase the stability of the thin - shell wh . now we check the stability of the hayward wh using the small velocity perturbations ( @xmath268 ) . the eos is calculated by @xmath552 and @xmath553@xmath554one finds the 1-d motion @xmath555after integrating both sides,@xmath556then it is obtained by taking the second integral@xmath557it is noted that @xmath558 is the equilibrium point , however we consider that there is an initial small velocity after perturbation , which is called @xmath559 we firstly consider the schwarzschild bh as an example . for the schwarzschild spacetime , @xmath560 yields@xmath561it is non - oscillatory so it means that the throat isnt stable . now we give the another example of the hayward thin - shell wh . for this case , using the hayward metric and the relation given in eq.([lastint]),expanding to the second order of @xmath562 , admits @xmath563here again we find that there is not stability of the throat against the small velocity perturbation . therefore this motion is not also oscillatory . the throat s acceleration is @xmath564 . note that for the schwarzschild and hayward spacetimes we show that it is positive . therefore , the corresponding thin - shell wh is not stable . in this section we constructe thin - shell whs from the hayward bh . firstly it is showed that a magnetic monopole field in the ned is the source of the hayward bh . on the thin - shell we use the different type of eos with the form @xmath525 and plot possible stable regions . we show the stable and unstable regions on the plots . stability simply depends on the condition of @xmath565 . we show that the parameter @xmath549 , which is known as hayward parameter has a important role . moreover , for higher @xmath549 value the stable regions are increased . it is checked the small velocity perturbations for the throat . it is found that throat of the thin - shell wh is not stable against such kind of perturbations . hence , energy density of the wh is found negative so that we need exotic matter . for constructing the thin - shell wh , we study 2 copies of a 2 + 1-dimensional background spacetime given . these copies are used at a radius @xmath566 , where the throat of the wh will be located larger than the event horizon . all steps for thin - shell by using the isreal junction conditions are also valid for the constructing of wh , only difference appears when calculating the energt density and pressure of wh . then we have the jump if the extrinsic curvature components at the surface @xmath189 , furthermore , we take the @xmath567 and also @xmath568 for the wh case , and calculate associated linear energy density and pressure for wh . [ fig : thin - shell wh ] once more we add that all functions are evaluated at @xmath569therefore it gives@xmath570@xmath571and@xmath572to finalize this section we add that at the equilibrium point where @xmath573 one finds@xmath574@xmath575and@xmath576 in which all functions are evaluated at @xmath577 furthermore , the equation of motion for the thin - shell is @xmath270with the effective potential ( @xmath269)@xmath578it is noted that stability of wh solution depends upon the conditions of @xmath579 and @xmath580 @xmath273let s introduce @xmath581 and write the equation of motion again@xmath277which after a derivative with respect to time it reduces to@xmath278hence our main aim is to discover the behaviour of @xmath582 and point out the conditions of its positive value for the stability by also the help of the eq.([energy ] ) . it is now interesting to reduce our general results to some specific examples . such as rotating ads - btz thinshell wh and rotating btz thinshell wh . our last wh example is on very recent work where they introduce a counter rotating thin shell whs as follows @xcite . in @xmath415 and @xmath583 the spacetimes are given by rotating btz whose line elements are written as with @xmath584 which are the rotating btz . here @xmath585 and @xmath586 are the integration constants which respectively correspond to the mass and angular momentum , and @xmath587 is the charge carried by the bh . the exterior curvatures are derived as follows @xmath588@xmath589and@xmath590then it is found that@xmath591 } { r } \\ -\frac{\left [ n_{+}^{\varphi } \left ( r\right ) + n_{-}^{\varphi } \left ( r\right ) \right ] } { r } & \frac{2\sqrt{\dot{r}^{2}+f}}{r}% \end{array}% \right ) .\]]note that @xmath359 is diagonal so that @xmath592 which consequently admits @xmath593 it means that between the upper and lower shells there is a counterrotating . after some calculations , the energy density and the pressures is calculated as@xmath594and @xmath595 the static case where @xmath242 and @xmath243 ) @xmath596and @xmath597 we have already imposed a generic potential equation in eq.([veffwh ] ) . now , eq . ( [ veffwh ] ) can be written for the rotating btz thinshell wh as @xmath598 with @xmath272 given by @xmath599 using the more general form of the pressure as a function of @xmath266 and @xmath600 i.e. , @xmath601 next , at equilibrium point @xmath268 it is obtained that @xmath602 one can show that @xmath603 . the eos of a lg is given by @xmath604 in which @xmath142 is a constant parameter . one observes that for the cases @xmath605and@xmath606@xmath607 and the equilibrium is stable . in fig . 3.12 we plot the stability region with respect to @xmath142 and @xmath608 in the same figure the result for different @xmath586 are compared . as one can see , increasing the value of @xmath586 increases the region of stability fig : stabilityrotating in this thesis , we have studied the stability of thin - shells around the bhs and the stability of the thin - shell whs . firstly , we have introduced generally the bhs , whs and their properties in the theory of the gr . furthermore , we have calculated the hawking radiation from the traversable whs . in chapter 2 , we have studied the rotating thin - shells methods and gave some examples . especially , the bardeen bh with a charged thin - shell is used to produce thin - shell matching inside and outside . we have created the thin - shell by using the israel junction conditions which we derived earlier . then as a result of the thin - shell , the effect of the charge is disappeared to outside . moreover , we have checked the stability regions of the thin - shell and we have plotted the areas of the stability . experimentally , there is a possibility that thin - shells can cancel the charge s effects for the outside observers . our example is in the 2 + 1 dimensions but it can be found similar effects for the higher dimensions . in chapter 3 , we have defined the methods of the constructing thin - shell whs and we have studied two example of thin - shell whs in 2 + 1 dimensions and 3 + 1 dimensions by using same method which is gluing two copies of bulks via the cut and paste procedure . to this end , we have used the fact that the radius of throat must be greater than the event horizon of the metric given : ( @xmath609 ) . we have adopted lbg , cg , gcg , and logg gas equation of states to the exotic matter locating at the throat . then , the stability analysis has become the study of checking positivity of the second derivative of an effective potential at the throat radius @xmath444 : @xmath610 . in all cases , we have managed to find the stability regions in terms of the throat radius @xmath444 and constant parameter @xmath611 , which are associated with the eos employed . hence , in our studies the exotic matter is needed to support construction of the throat of the thin - shell . in last section , we have studied the rotating thin - shell whs in 2 + 1-dimensions . this time we use the counterrotating shells around the throat which has no effect on the gluing procedure and lanczos energy conditions . it can be thought that the upper and lower pressures cancel each others and leaving no cross terms of the pressure component . stability of a counterrotating tsw for a gas of linear equation of state turns out to make the tsw more stable . for very fast rotation the stability region grows much larger in the parameter space . for a velocity dependent perturbation , however , it is shown that tsw is no more stable . that is , perturbation of the throat radius ( @xmath612 ) that depends on initial speed ( @xmath613 ) , no matter how small , does nt return to the equilibrium radius @xmath612 again . although our work is confined to the simple @xmath614dimensional spacetime it is our belief that similar behaviours are exhibited also by the higher dimensional tsws . however , even for the @xmath615dimensional spacetime cases it is not easy task to check the stability because there are more degrees of freedom in @xmath615 dimensional spacetime . ligo scientific , & virgo collaborations ( abbott , b.p.,(caltech ) et al . ) gw150914 : the advanced ligo detectors in the era of first discoveries . _ , 116 , no.13 , 131103 . doi : 10.1103/physrevlett.116.131103 straub , o.,vincent , f. h. , abramowicz , m. a. , gourgoulhon , e. , & paumard , t. ( 2012 ) . modelling the black hole silhouette in sgr a * with ion tori . _ , 543 , a83 . doi : 10.1051/0004 - 6361/201219209 garcia , n. m. , lobo , f. s. n. , & visser , m. ( 2012 ) . generic spherically symmetric dynamic thin - shell traversable wormholes in standard general relativity . , 86 , 044026 . doi : 10.1103/physrevd.86.044026 lemos , j. p. s. , lopes , f. j. , minamitsuji , m. ( 2015 ) . rotating thin shells in ( 2 + 1)-dimensional asymptotically ads spacetimes : mechanical properties , machian effects , and energy conditions . , 24 , 1542022 . doi : 10.1142/s0218271815420225 mazharimousavi s. h. , halilsoy , m. , & amirabi , z. ( 2011 ) . black hole and thin - shell wormhole solutions in einstein - hoffman - born - infeld theory , _ phys . a _ , 375 , 3649 . doi : 10.1016/j.physleta.2011.08.036 ovgun , a. , & jusufi , k. ( 2016 ) , massive vector particles tunneling from noncommutative charged black holes and their gup - corrected thermodynamics . j. plus _ , 131 , no.5 , 177 . doi : 10.1140/epjp / i2016 - 16177 - 4 sakalli , i. , ovgun , a. , & mirekhtiary , s. f. ( 2014 ) . gravitational lensing effect on the hawking radiation of dyonic black holes . _ , 11 , 8 , 1450074 . doi : 10.1142/s0219887814500741

we present far - ultraviolet ( 905 1182 ) , time - series spectroscopy of the eclipsing , novalike cataclysmic variable , ux uma , acquired with fuse . the time - averaged spectrum is complex and is dominated by overlapping spectral features . the most prominent features are emission lines of , , , and . they are broad ( fwhm 1800 km s ) and double - peaked with a central absorption at zero velocity . during eclipse , the spectrum is simpler : the emission lines remain bright , but the absorption components of the lines and the weaker features between the emission lines disappear entirely , leaving a flat continuum . this behavior is also evident in ghrs ( 1149 1660 ) spectra that we retrieved from the hst archive . the fuv spectra show flickering on time scales of several minutes . the flickering is seen primarily in the continuum and/or the weaker lines rather than in the prominent emission lines . the orbital light curve has a dip in the fuv flux between orbital phases 0.45 0.65 , similar to a pre - eclipse dip detected in hst observations . the ews of the line absorption features decrease during the dip . we have detected a systematic wavelength shift of spectral features on the orbital period , but with a phase lag of , a phenomenon that has been reported at optical wavelengths . we discuss the implications of our results in the context of models of an accretion disk with a chromosphere between the disk and the extended wind . finally , we note that the observed fuv flux is too low to be consistent with the temperature and radius of the wd derived by baptista et al . ( 1995 ) , suggesting that their remaining binary parameters , including a mass ratio of 1 , ought to be viewed with skepticism .

astro-ph0210398

cataclysmic variables ( cvs ) are interacting binary systems in which a late - type donor star overflows its roche lobe and transfers mass to a white dwarf ( wd ) . in the non - magnetic cvs , mass is accreted through a disk around the wd . novalikes ( nls ) are the non - eruptive class of cvs ; that is , they do not show dwarf novae outbursts . they are believed to have high mass - transfer rates , which keeps their disks in a permanent outburst state . the accretion disks in nls are typically well described by optically thick , steady - state disk models with accretion rates @xmath3 yr@xmath1 ( e.g. , horne 1993 ) . because their underlying accretion behavior is fairly well understood , novalike cvs are excellent laboratories to study the more complicated processes associated with disk accretion . whether in the form of collimated jets or winds , outflows are closely linked to the disk accretion process . wind signatures broad , blueshifted absorption and/or p cygni profiles in resonance lines such as @xmath41548,1552 , @xmath41393,1402 , and @xmath41232,1238 are seen in most high accretion rate , disk - fed cvs , including dwarf novae in outburst and novalikes ( e.g. , heap et al . 1978 ; holm , panek & schiffer 1982 ; crdova & mason 1982 ) . typical terminal velocities in the wind lines range from 3000 5000 km s@xmath1 , near the escape velocity of the primary @xcite . in eclipsing systems , the resonance lines seen in emission are only partially eclipsed or completely unocculted , indicating that the winds are vertically extended ( e.g. , mauche et al.1994 ) . much of the phenomenology of disk accretion and outflow in cvs has been revealed by study of the novalike cv , ux uma . ux uma has an orbital period of 4.72 hr and is eclipsing ( it is , in fact , the brightest eclipsing nl ) . the binary parameters for ux uma are typically taken from @xcite , and are based on their identification of wd eclipse features in the uv light curve , a standard mass - radius relation , and previously published limits on the @xmath5 amplitude of the wd @xcite . @xcite obtained an inclination of @xmath6 and , surprisingly , a mass ratio @xmath7 . they also determined a wd flux and a temperature 52,000 k , or 70,000 k if the lower half of the wd is obscured by the disk that would indicate a strong wd contribution in the fuv . @xcite modeled the profile changes in the @xmath41548,1552 doublet of ux uma through eclipse . they showed that modeling the line with a biconical disk wind model provides a good fit to the line profile in and out of eclipse . the line has narrow absorption reversals , or dips , that disappear at mid - eclipse . @xcite showed that the orbital behavior of the dips could be explained by the presence of a slow , dense transition region ( between the disk and the fast , extended wind ) that is occulted during eclipse . this transition region , or disk chromosphere , may correspond to the chaotic , slowly - outflowing wind regions that have appeared in some radiative transfer models of cv disk winds @xcite . the structure and behavior of winds and chromospheres in ux uma and other high state cvs is closely tied to unresolved questions concerning the structure of the underlying accretion disk . @xcite modeled optical and uv spectra of ux uma and found that steady - state accretion disk models provided poor fits to the shape of the continuum at blue wavelengths and at the balmer jump ( see also baptista et al . the discrepancy between disk models and the observed spectrum is ameliorated if significant amounts of optically thin continuum emission are present in the system . an accretion disk chromosphere is a plausible , though currently ad hoc , source of such emission . other possibilities have been suggested to resolve the mismatch between disk models and the observed spectrum in ux uma and other nls , such as a disruption of the inner accretion disk that removes the bluest emitting regions @xcite . to date , virtually all knowledge of the properties of the wind in ux uma and its relationship with the accretion disk are based on observation and modeling of the doublet . this is not ideal , since we do not know whether the behavior of is representative of other wind - dominated lines . the strength and profile of has been observed to change over a single orbit and over longer timescales @xcite . this variability may be due to asymmetries in the wind which cause orbital variations and/or to secular fluctuations in the mass loss rate ; alternatively , it may indicate that there are components of the line that do not originate in the wind at all . there is a wealth of further information that can be gathered by extending high spectral and time resolution observations of ux uma to shorter wavelengths . the fuv is an excellent window in which to probe the structure of disks and winds in cvs . for typical maximum temperatures in high state accretion disks ( @xmath8 k ) , the disk sed peaks in the fuv , so the fuv spectrum is a sensitive tracer of changes in the temperature and structure of the innermost regions of the disk . the fuv is also richly populated with spectral lines of a variety of elements and ionization states which , in principle , can be used to determine the thermal and ionization structure of the wind . consequently , we have pursued fuv spectroscopy of ux uma with the far ultraviolet spectroscopic explorer ( fuse ) , which covers 905 1182 at spectral resolutions ( through its large aperture ) near 20,000 @xcite . we have also examined archival hst uv spectroscopy to obtain a broader picture of the uv spectrum in ux uma . in this manuscript , we present observations ( [ sec_obs ] ) and analysis ( [ sec_analysis ] ) of the fuv spectrum of ux uma and discuss what the observations reveal about the structure of disks and outflows in disk accreting systems ( [ sec_discussion ] ) . we present our conclusions in [ sec_conc ] . we observed ux uma with fuse on 2001 march 23 25 . a summary of the observations is given in table [ tab_obs ] . the observations were spread over six binary orbits and covered each orbital phase at least twice . the fuse telescope collects data through four optical channels which cover the wavelength range of the instrument ( 905 1182 ) with some redundancy . ( for a summary of the fuse optical design and its on - orbit performance , see sahnow et al . 2000 or the fuse observer s guide . ) the four channels were co - aligned during our observations , so no correction to the relative wavelength solutions or fluxes of the channels was necessary . all of the data were acquired through the lwrs aperture ( @xmath9 ) in time - tag , or photon counting , mode , in which each photon event is recorded to an accuracy of 1 s. we re - reduced the data using the latest version of the calfuse pipeline ( v. 2.0.5 ) , extracting spectra in 200 s intervals . for the orbital period of ux uma @xcite , this corresponds to an orbital phase resolution of 0.012 cycles per spectrum . we used the 200 s spectra as our base set of data , which we binned up in time and/or spectral resolution for specific tasks . when combining spectra from the four optical channels , we specified a linear dispersion scale for the output spectrum and added each input datum to the appropriate output bin , weighting the bin average by the sensitivity of the input pixels . regions of poor flux calibration , most notably the `` worm '' in channel lif1 , were excluded . we also created high time resolution light curves of the observations using the raw time - tag photon event files . to construct the light curves , we defined rectangular apertures for each optical channel and summed all of the events occurring in each channel each second . in defining the apertures , we excluded ly@xmath10 airglow and the `` worm '' . we also screened photon events using the default pulse height limits adopted by the calfuse pipeline . the detector background , summed over astigmatic spatial height and dispersion direction ( see the fuse observer s guide ) , is @xmath110.5 counts / s / segment at night and @xmath110.8 counts / s / segment during the day . the background is not negligible compared to the fuv count rate of ux uma ( @xmath12 counts / s ) , particularly during eclipse , so we subtracted background emission from the counts in each 1 s time interval . the background was estimated by defining apertures away from the target spectrum of the same size as the target apertures and summing the counts therein . the time - averaged spectrum of the complete data set is shown in figure [ fig_uxuma ] . the spectrum has been binned to a dispersion of 0.1 , which is the typical resultant resolution for a lwrs spectrum when the channels are combined ( see froning et al . the spectrum is dominated by strong , broad emission lines from a range of elements and ionization levels . the strongest lines are resonance lines and low - excitation energy excited state transitions of @xmath13977 and the blend at @xmath131175 ; @xmath13990 ; the blend at @xmath13923 ; and @xmath41032,1038 . in between the strong emission lines , the spectrum is populated by a host of weaker spectral features . the line spectrum is quite complex , and a comparison of the time - averaged spectrum with the spectrum at mid - eclipse ( see figure [ fig_eclipse ] ) shows that many lines overlap and are blended with each other . some of the other lines that can be identified in the time - averaged and/or the mid - eclipse spectrum are @xmath4933,944 ; @xmath13951 ; @xmath131013 ; @xmath131063 and @xmath131073 ; and/or @xmath131085 ; @xmath41108 ,1109 , and 1113 ; and @xmath41122,1128 . there is a rise in flux at the shortest wavelengths , @xmath14965 . some lines that contribute to the increase are identifiable , but they do not account for all of the emission observed . excess emission near the lyman limit is a common feature in the fuv spectra of cvs : a similar bump is seen in the outburst spectrum of the dwarf nova , wz sge , and in the quiescent spectrum of ss cyg ( long et al.2002 , 2002b , in preparation ) . all of these spectra also show fuv lines in emission , which suggests that the blue excess is a blend of line features , dominated by emission from the upper order lyman series lines . neutral hydrogen emission is also evident in the presence of ly@xmath15 and ly@xmath16 emission features in the spectrum of ux uma . the profiles of the strongest lines in the fuv spectrum of ux uma are entangled with those of the weaker features , but a rough examination of line shapes can be made . the emission lines are broad , with fwhm @xmath17 km s@xmath1 ( where the instrumental resolution is @xmath18 km s@xmath1 ) . they have double - peaked profiles with a central absorption reversal at the rest wavelength of each line . the line peaks are separated by 1000 1500 km s@xmath1 and are roughly equidistant from the line center . in the strongest lines @xmath13923 , @xmath13977 , @xmath13990 , and @xmath41032,1038 the red peak is enhanced relative to the blue peak . the fwhm of the absorption components of the lines are @xmath19600 km s@xmath1 . both members of the and doublets are present . we fit the ly@xmath15 and doublet blend using the iraf task `` specfit '' and assuming gaussian profiles for all three lines . the equivalent widths of the two components of are equal , indicating that the lines are optically thick . the fwhm of the lines ( fixed to a 1:1 ratio ) are 3800 km s@xmath1 . in contrast to , the red component of the doublet appears to be weaker than the blue component , suggesting that these lines are not fully optically thick . finally , the time - averaged spectrum also shows narrow interstellar absorption lines from and metal transitions . identified interstellar lines and their ews are given in table [ tab_ism ] . we pursued a curve of growth analysis to determine the interstellar column density along the line to ux uma , resulting in an estimate of the column density of @xmath20 . the curve of growth analysis is presented in appendix [ sec_cog ] . the out of eclipse and mid - eclipse spectra of ux uma are compared in figure [ fig_eclipse ] . also shown is the difference between the two spectra ; this is the spectrum of the eclipsed light . the out of eclipse spectrum was created by binning the time series 200 sec spectra into 80 phase bins ( 0.0125 cycles orbital phase resolution ) and averaging the 18 bins preceding and following the eclipse ( i.e. , from orbital phases 0.85 0.95 and 0.05 0.15 ) . it is qualitatively similar to , but slightly brighter than , the time averaged spectrum ; see [ sec_lc ] for a discussion of orbital variations in the fuv spectrum away from the eclipse . the mid - eclipse spectrum is the spectrum of the phase bin at the light curve minimum . the mid - eclipse spectrum and the difference spectrum are binned to a resolution of 0.6 . the eclipse spectrum is less complex than the out of eclipse spectrum . at mid - eclipse , the fuv spectrum is characterized by broad emission lines separated by regions of continuum emission . the average flux of the continuum regions of the mid - eclipse spectrum is and the continuum is relatively constant over the fuv . many of the emission lines present in the out of eclipse spectrum remain strong in eclipse , indicating that they arise from regions that are largely unocculted . the excess emission at wavelengths @xmath14955 is not eclipsed either . many features that were not readily identifiable in the time - averaged spectrum due to line blending can be identified at mid - eclipse . mid - eclipse emission lines include transitions of , , , , , , , and . there is an emission line at @xmath131085 from and/or and an unidentified feature at 1143 . another emission feature at @xmath13951 may be from , although @xmath13949 is also present in that region . because of the presence of clean , line - free regions in the eclipse spectrum , we were able to measure the parameters of most of the emission lines seen at mid - eclipse . the ews and line widths of these lines are given in table [ tab_eclipse ] . a comparison of the out of eclipse and mid - eclipse spectra indicates that most of the weaker , blended features are occulted in eclipse . the difference spectrum shows that the lines are a complex blend of components , predominately in absorption . the line absorption can be just as strong or stronger than the line emission , so that some lines , like the doublet , can only be identified at mid - eclipse , after their absorption components are occulted . along with the weaker features , the absorption reversals seen in the profiles of strong lines in the time - averaged spectrum are eclipsed . we searched for evidence of a rotational component in the lines by examining changes in the fuv spectrum through the eclipse and comparing the profiles of the strong emission lines in the out - of - eclipse and mid - eclipse spectra . if the line regions are rotating in keplerian motion , the blue wing and peak should be occulted earlier than the red peak and wing . this was seen in @xmath41548,1552 in ux uma @xcite . most of the lines in the fuv spectrum are too confused with other features to examine their eclipse behavior in different velocity bins , but we examined the eclipse of two relatively unblended lines , @xmath131175 and @xmath131073 . in both of these lines , the fuv eclipse minimum occurs at earlier phases at blue velocities relative to the red components of the line . there also appears to be some narrowing of the lines in eclipse : the fwhm of and @xmath131073 decreases from @xmath113000 km s@xmath1 to @xmath112000 km s@xmath1 during the eclipse . a narrower line profile during eclipse has been seen in in ux uma @xcite . kinematic models predict narrowing of line profiles in eclipse for a collimated disk wind viewed at high inclination , as the more rapidly - rotating parts of the wind near the disk plane are occulted ( knigge & drew 1997 and sources therein ) . figure [ fig_eclc ] shows phase - binned fuv light curves of the eclipse in ux uma . the upper panel shows the light curve over the full 910 1182 range , while the lower panel shows the mean light curve for four regions of the spectrum in which only continuum emission is present at mid - eclipse : 955 965 , 1045 1057 , 1090 1103 , and 1149 1164 . we created both light curves by phase - binning the 1 s count rate light curves into 5000 bins per orbit . we shifted both light curves earlier in phase by 0.006 cycles to bring the phases of full width at half depth of the lower , continuum - dominated , light curve symmetric about phase 0 . this phase shift is typical of small - scale variations in the phasing of the eclipse as indicated by the scatter in the ( o c ) diagram for the ephemeris @xcite . also shown in figure [ fig_eclc ] are the contact points for the eclipse of the wd as found by @xcite . figure [ fig_eclc ] indicates a narrow eclipse in the fuv . the full eclipse width from start to end is @xmath190.065 cycles in both light curves . the full width at half depth for the upper light curve is 0.043 cycles , but is 0.055 cycles for the lower light curve , suggesting that the emission lines are slightly more centrally concentrated than the continuum emission . the upper light curve is also more asymmetric than the lower curve : the minimum occurs at earlier phases in the former , and its egress is longer than its ingress . both eclipses are fairly smooth , although there is a hint of a break around phase 0.0175 that could be an egress feature . the fuv eclipse is not total , although it is much deeper in the regions between the resonance emission lines . the eclipse depth over the full fuv range is 40% of the pre - eclipse flux , but it increases to an average depth of 82% for the four regions dominated by continuum emission . give slightly shallower eclipses of 45% and 73% for the upper and lower curves , respectively . the discrepancy between fluxed and raw eclipse depths most likely indicates that our method for estimating the background when creating the high time resolution light curves underestimated the true background . ] the residual continuum emission at mid - eclipse probably originates from the back of the accretion disk , which is not occulted during eclipse . emission from disk annuli as close as 1114 r@xmath21 from the center of the primary is still visible at mid - eclipse ( the lower value assumes the system parameters of baptista 1995 , while the upper value is for the geometry of smak 1994 , and using m@xmath22 from nauenberg 1972 ) . past analysis of time and orbital variability outside of the eclipse has shed light on behavior of the inner disk and outflow in ux uma . for example , ux uma is known to occasionally exhibit low - amplitude 28-s oscillations in optical and uv light ( e.g. warner & nather 1972 ; nather & robinson 1974 ; knigge et al.1998b ) . the oscillations originate very near or on the wd surface , and may indicate an appreciable magnetic field that affects the structure of the inner disk . on longer time scales , @xcite found an orbital variability in the uv light curve of ux uma in the form of a decline in flux , or `` light curve dip '' , centered on orbital phase 0.6 . they attributed the phenomenon to asymmetric vertical extension in the outer accretion disk absorbing flux from the disk interior . they found a decrease in absorption line ews during the dip , which indicates that the narrow absorption components seen in the uv lines in ux uma occur not in the outermost accretion disk annuli but interior to the source of the orbital light curve dip . in an effort to search for the oscillations and orbital variability in ux uma in the fuv , we examined the high time resolution light curves created from the raw photon event files of the fuse observations and spectral variations in the 200 sec spectra . the top three panels of figure [ fig_contlc ] show the fuv light curve of ux uma in 2001 march , with the count rate in each 1 s bin plotted vs. cycle number . the bottom panel shows the same data binned into 250 phase bins per orbit . the 1 s light curve shows significant flickering in the fuv light on short time scales , with flickering peaks typically separated by 200 1000 s. in several epochs , the count rate changed by a factor of 2 or more within a 5 min time period ( @xmath190.02 cycles ) . we compared the spectrum of the flickers to the spectrum away from the flickers by averaging all of the 200 sec spectra that occurred at or near the peak of a flicker and all of the 200 sec spectra obtained between flickers ( excluding the eclipse ) . the `` flicker '' and `` non - flicker '' spectra are shown in figure [ fig_flare ] . also shown is the difference spectrum between the high and low states . the difference spectrum , which is the spectrum of the flickering light , is quite similar in morphology to the spectrum of the eclipsed light at mid - eclipse . that is to say , the source of the flickering is the same material , located near the accretion disk plane , that is occulted in eclipse . although the continuum shows strong flickering , the prominent emission lines do not flicker . to search for periodicity in the fuv flux variations , we carried out a power spectrum analysis of the 1 s overall fuv light curve . no sign of the 29 s oscillations was found . we also failed to detect an unambiguous periodicity to the flickering . by injecting sinusoidal signals of varying amplitude into the data stream , we found that we would certainly ( probably ) have detected the 29 s oscillations with an amplitude of 3% ( 1% ) in our data set . for comparison , the average amplitude of the oscillations in knigge et al.s ( 1998b ) hst data ( which covered 2000 8000 ) was roughly 0.5% . however , in their data set , the oscillation amplitude rose to well over 1% near 2000 , due to the extremely blue spectrum of the oscillations . it therefore seems likely that we would have been able to detect oscillations of the same strength and spectral shape as those seen by knigge et al . the absence of the oscillations from our data is not particular surprising , however . for example , knigge et al . saw the oscillations in only two of their four observing runs . the phase - binned fuv light curve in figure [ fig_contlc ] shows that in addition to the short time scale flickering , there is an orbital modulation in the form of a decline in flux centered on phase 0.55 . on average , the fuv flux is 18% lower between phases 0.45 0.65 compared to the rest of the orbit outside of eclipse , and it reaches a maximum depth of 25% . the depth of the eclipse over the full fuv is 40% , so the light curve dip is over half the eclipse depth in the fuv . the dip in the fuv light curve occurs at roughly the same orbital phases as the dip observed by @xcite in 1996 november hst / ghrs spectra of ux uma ( these spectra are discussed further in [ sec_hst ] ) . the 1996 observations did not have full coverage of the orbit , but on two epochs ( different binary orbits ) the flux between phases 0.6 0.7 was 45% and 60% lower than the flux at orbital phases 0.25 0.4 ( measured in the 1415 1433 region ) , a deeper modulation than seen in the fuv . @xcite also noted that the ews of the absorption features in @xmath41393,1402 decreased during the phases of the dip . we compared the fuv spectrum in the dip phases ( 0.45 0.65 ) to the spectrum outside of the dip ( 0.05 0.25 and 0.75 0.95 ) . as with the short time scale flickering , the spectral changes in the orbital dip are seen in the weak lines and the continuum and not in the prominent line emission . the ews of the absorption components of the fuv lines decrease during the light curve dip , consistent with the results of mason et al . that place the material responsible for the line absorption interior to the source of the broadband light curve dip . this conclusion is based on uv and fuv observations that only covered the binary orbit approximately twice each , however , and an observation covering multiple binary orbits is necessary to ensure that orbital effects are not contaminated by the spectral variations from flickering . finally , we searched for orbital radial velocity variations in the fuv spectra using the 200 s spectra . with the `` crosscor '' task in iraf , we cross - correlated the individual spectra with respect to the time - averaged spectrum of the observation in several different wavelength regions . we also measured the centroid of several of the narrow absorption features in the lines by fitting a gaussian to their profiles in each spectrum . a representative example of the results is shown in figure [ fig_vrad ] , which plots the wavelength shift of each 200 s time series spectrum as a function of orbital phase ( corrected by @xmath230.006 cycles ; see [ sec_eclipse ] ) for the 1040 1182 region ( this region is shown because this part of the spectrum is largely free of non - varying features from interstellar lines and airglow ) . the best - fit sinusoid , also shown in figure [ fig_vrad ] , has an amplitude of 0.26 ( or a total peak - to - peak shift of 0.52 ) and a phase shift of 0.057 cycles . the amplitude corresponds to a velocity of 70 km s@xmath1 . the phase shift gives a 21@xmath24 lag between the photometric phasing and crossing times in the line spectrum . in general , this amplitude and phase shift are descriptive of the orbital radial velocity variations of the fuv spectrum . one possible exception are the absorption reversals in the lines , which show much larger amplitudes , from 140 km s@xmath1 for the and absorption features to a peak amplitude of 200 km s@xmath1 for the absorption component of . the wavelength shift in the absorption lines shows the same phase offset as seen in the broader spectrum . we caution that , due to the complex line blending in the fuv spectrum , the apparent change in the velocities of the absorption component centroids could be an artifact of a profile change in the emission components of the lines . the amplitude of the orbital variation in the overall fuv spectrum is smaller than the value of @xmath25 km s@xmath1 found by @xcite from radial velocity variations in the emission wings of h@xmath10 ( although the fuv absorption lines may show a range of motion equal to or larger than the @xmath5 determination ) . @xcite performed a spectrophotometric study of the optical lines in ux uma , observing it twice separated by a year . although they did not observe a velocity amplitude as low as 70 km s@xmath1 , they found that the optical lines exhibited a range of radial velocity amplitudes ( 121 267 km s@xmath1 ) that varied from line to line and within components of the same line from one epoch to another . more interesting , schlegel et al.observed a 0.06 cycle phase shift , consistent with the lag we observe in the fuv , in the radial velocity curve of @xmath134686 in their 1981 observations . the phase lag was not present in 1982 . because of the presence of a pre - eclipse light curve hump in 1981 but not 1982 , schlegel et al . attributed the phase lag at the earlier epoch to a substantial bright spot component in . there were problems with this interpretation , however , namely that there was no s - wave component and no eclipse in the line in 1981 , both of which are expected for bright spot emission . as a result , they suggested that emission region may have had substantial vertical extent in that epoch . the phase shift in the fuv between the wavelength shifts and the photometric ephemeris likely derives from some asymmetry in the line formation region , but while the phasing is suggestive of an asymmetry tied to the bright spot , the fuv emission originates interior to the outside edge of the disk , so it is not obvious how to reproduce the observed amplitude and phase shifts . as noted earlier , ux uma has previously been observed at uv wavelengths with hst . of particular interest for its combination of moderate spectral resolution and broad wavelength coverage is the 1996 november observation acquired with the ghrs . this observation obtained 1149 1660 wavelength coverage at 0.6 spectral resolution . the spectra were obtained in rapid mode with 5 s integration time and sampled much of the binary orbit . orbital light curves and spectra of @xmath41393,1402 from this data set were presented by @xcite as part of an analysis of the pre - eclipse dips discussed in [ sec_lc ] , but the full uv spectrum was not published . as these spectra provide a longer - wavelength complement to our fuv observations , we present them here . figure [ fig_eclipse_hst ] shows the time - averaged , mid - eclipse , and difference spectra of the 1996 ghrs observation of ux uma . the mid - eclipse spectrum covers the central 0.0125 cycles of the eclipse . the hst time - averaged spectrum is similar in character to the fuse fuv spectrum : in both , continuum emission is everywhere overlaid by a rich spectrum of lines with emission and absorption components . strong resonance lines of , , and , as well as , are joined by a wealth of other lines , including transitions of , and . the strong emission lines have line profiles similar to those seen in the fuv . the lines are double - peaked , with central absorption at the rest wavelengths of the lines . at this epoch , the absorption reversal is relatively shallow in @xmath41548,1552 , @xmath41238,1242 , and @xmath131640 . the absorption dips are deeper in @xmath131175 , @xmath131334 , and @xmath41393,1402 . in and , the lines are primarily or entirely in absorption outside of eclipse . the fuse and ghrs spectra overlap from 1149 1182 , a region including the @xmath131175 blend . a comparison of the line at two epochs indicates that the profile of the line is variable . although the line is double - peaked in both spectra , the blue peak dominates in the 1996 spectrum , while the red peak is stronger in 2001 . spectral minima ( around 1150 , for example ) are at the same flux in both spectra , as is the blue peak in , but other spectral peaks , at 1158 , 1162 , and the red peak of , are depressed by 15% 25% in the 1996 observation relative to 2001 . the line also shows a strong peak blueward of the rest wavelengths of the doublet lines . this enhanced blue peak is not present in other high - resolution spectra of ( 1993 ghrs spectra , see mason et al . 1995 , and unpublished 1998 stis spectra ) . the mid - eclipse hst spectrum is also quite similar to the fuse fuv eclipse spectrum . the weaker line emission is gone , and the spectrum consists of strong emission lines separated by a flattish continuum . the mean continuum flux in the overlap region , 1149 1165 , is in the hst spectrum , compared to in the fuse observation . the uv flux ( and , by implication , the mass accretion rate ) fluctuates in ux uma ( see , e.g. , knigge et al . 1998 ) and the difference in continuum flux between the 1996 and 2001 observations is typical for the system . in @xmath131175 , the line profile is reversed in the hst spectrum compared to the profile in the fuse observation , with the blue peak dominant in the former and the red peak in the latter . the difference spectrum in the hst observation shows that the eclipsed component of the lines , as in the fuse observations , are mainly or entirely in absorption . ( note that here , because of phase coverage gaps , the difference spectrum is the difference between the time - averaged spectrum and the mid - eclipse spectrum , rather than a direct comparison of mid - eclipse to phases immediately preceding and following . ) the absorption components are narrow , with typical fwhm of 600900 km s@xmath1 , and very deep : several of the dips reach zero flux at their line centers , which did not occur in the fuse observations . the absorption dips are occulted in some lines , such as , , and , but remain at diminished ew in other lines , including and . the broad blend at @xmath131300 appears weakly in emission only after its broad , deep absorption component is eclipsed . figure [ fig_eclc_hst ] shows the light curve of the hst observations for four regions of the spectrum away from the prominent lines and dominated by continuum emission at mid - eclipse . also shown for reference is the fluxed eclipse light curve from the 2001 fuse observation ( at lower time resolution than the light curves in figure [ fig_eclc ] but created from flux - calibrated spectra ) . the pre - eclipse flux was higher than the post - eclipse flux in 1996 , while the opposite trend was true in 2001 . in other respects , the uv and fuv light curves are quite similar , however . the full width at half depth of the 1996 uv eclipse ( where the out of eclipse flux was taken as the mean of the pre- and post - eclipse fluxes ) is 0.052 cycles , consistent with the eclipse width in the fuv . moreover , the depth of the eclipse does not change significantly with wavelength : the mean depth in the continuum - dominated regions is 82% in the fuv and 77% in the uv . even in the longest wavelength range , 1565 1665 , the depth of the eclipse remains 70% of the out of eclipse flux . the fuv spectrum of ux uma is everywhere characterized by line features . the fuse and archival hst spectra show that from 910 1600 , the time - averaged spectrum is composed of prominent emission lines and a complex blend of weaker features . the emission lines arise from the usual transitions seen in cv spectra in the uv : resonance lines of neutral hydrogen and ionized metals and prominent excited state transitions . in eclipse , these lines are largely unocculted , while the weaker features are completely eclipsed . the spectrum of the eclipsed light suggests that the weaker line features are primarily in absorption . because of the severe line blending , it is difficult to determine where to place the continuum in the fuv and uv spectra , and some of the eclipsed line region may be in emission as well as in absorption . at mid - eclipse , the spectrum in between the prominent resonance emission lines is smooth and appears to consist only of continuum emission . as a result , the emission line component of the eclipsed light , if present , must have substantial asymmetry or originate very close to the wd to explain why it is not seen during eclipse , when much of the back of the accretion disk remains unocculted . a possible source for asymmetric emission is the mass accretion stream . @xcite created spatially resolved spectra of the accretion disk in ux uma from uv and optical hst / fos observations . they found that the quadrant of the disk containing the mass accretion stream is a source of uv emission , including line emission , down to 0.1r@xmath26 , which they attributed to overflow of the mass stream above the disk . the introduction of an asymmetry in the line emission by the mass stream is also consistent with the phase lag in the orbital radial velocity variations in the fuv spectrum with respect to the photometric ephemeris . the origin of the line absorption also has important implications for the structure of the accretion disk and the wind . @xcite modeled the @xmath41548,1552 line profile in ux uma in and out of eclipse with a kinematic model of a rotating , biconical disk wind . they found that to match the eclipse of the narrow absorption features at the rest velocity of the doublet , their model had to include a dense , slowly outflowing transition region between the fast wind and the disk , which they dubbed an accretion disk chromosphere . some hydrodynamical models of radiatively - driven disk winds have predicted regions of chaotic , low velocity motion that may be the source of the chromospheric features seen in ux uma @xcite . the results of knigge & drew were based on only one epoch of observation of ux uma and only one line , however , which left open the possibility that secular variability and/or non - wind contributions to could be responsible for the narrow absorption components of the line . an alternate explanation for the absorption components is absorption by low - velocity material in the outer disk ; narrow , low - velocity absorption lines from the outer disk are present in the fuv spectrum of the dwarf nova u gem in outburst , for example @xcite . @xcite showed that the narrow line absorption in ux uma also occurred in @xmath41393,1402 and that the absorption line ews decreased during a broadband light curve dip that they attributed to asymmetric vertical extension in the outer disk . the ew decline suggested that the line absorption components were interior to the outer edge of the disk , which made the disk wind / chromosphere model of @xcite more attractive . the fuse observations , obtained 5 years after the hst spectra , show that the line absorption components are present throughout the uv and fuv in a wide variety of species and ionization states and are persistent . the ews of the absorption lines in the fuv spectrum also decrease during the broadband light curve dip . if the line absorption occurred in the outer disk edge , the line ews would increase , not decrease , when the amount of absorbing material increased around phase 0.6 . the low velocities and narrow profiles of the absorption lines are consistent with origin in the dense , slowly outflowing transition region between the accretion disk and the fast wind . nevertheless , both the hst and fuse observations sample a small number of orbits , so it remains possible that the orbital spectral changes are being confused with variability caused by flickering ; additional broadband uv observations spread over the binary orbit of ux uma will be necessary to properly average out the flickering effects . another ongoing question in ux uma and in other cvs concerns the continuum source in the ultraviolet . in ux uma , as in other nls , the accretion disk is believed to be in a high mass - accretion rate steady - state , and we expect it to be the dominant continuum source in the optical and uv . previous studies have not been successful in modeling the continuum emission in ux uma with steady - state disk models , however . @xcite explored model disk fits to broadband hst spectra ( 1250 9000 ) of ux uma . their models were constructed from sums of appropriately weighted stellar atmosphere spectra at the temperature and gravity of each accretion disk annulus . they demonstrated that the model disk spectra provide a poor fit to the overall spectral shape . at optical wavelengths , the models overestimate the amplitude of the balmer jump , while in the uv , the models are too blue to match the observed spectrum . the discrepancy at blue wavelengths between model and observed spectra becomes worse when the fuv spectrum is also taken into account : the model disk spectrum of @xcite would exceed the observed 2001 march fuv flux by @xmath2750% . we examined the fuv fluxes predicted by accretion disk models for the geometry of ux uma ( testing the geometries of both baptista et al . 1995 and smak 1994 ) . optical and uv eclipse mapping analyses have given mass accretion rates for ux uma from to @xcite , where the range reflects a real variation in the observed flux and a likely variation in the mass flow rate through the disk . the distance to ux uma has been estimated to lie from 215 345 pc ( baptista et al . 1995 and sources therein ) , and the reddening is e(b v ) = 0 , or at most e(b v ) @xmath14 0.04 @xcite . we compared the fluxes of fuv and uv spectra of ux uma to model disk spectra constructed from summed , area - weighted , doppler - broadened spectra of stellar atmospheres set to the appropriate temperature and gravity for a given disk annulus , assuming the standard , steady - state temperature distribution in the disk ( see froning et al . 2001 for a full description of the models ) . there is a continued discrepancy between the models and the observed fluxes at blue wavelengths . if the accretion disk is assumed to be the dominant fuv continuum source which corresponds to scaling the model to something like the minima in the fuv spectrum away from the strong lines : at 965 and at 1095 the predicted mass accretion rate is , assuming the upper range to the distance ( increasing the reddening to the upper limit gives @xmath28 ) . although the 1996 november and 2001 march hst and fuse spectra have the same out of eclipse fluxes in their region of overlap ( 1149 1182 ) , the same accretion disk model will not fit both spectra . the @xmath28 model that scales to the flux in the fuv only provides half of the flux observed in 1996 november at 1250 and one third of the observed flux at 1400 . increasing the reddening to e(b v ) = 0.04 has a negligible effect on the slope of the spectrum in the uv . the unknown degree of line blanketing affecting the fuv and uv limits our ability to assess the discrepancy between the observed continuum and the disk models , but it is clear that a single accretion disk model can not fit the continuum over the full fuv+uv range . although the discrepancy between the disk models and observed spectra in high state cvs could result from some problem with the physics of the models , a more likely cause is an incomplete description of the three - dimensional structure of the accretion disk . the presence of substantial vertical extent in the disk , as indicated by the absorption components in the spectral lines and the orbital dip at phase 0.6 , points to a more complex accretion disk structure than current disk models assume . @xcite discussed several ways in which to bring model continuum spectra in closer agreement with the observed spectrum in ux uma . one possibility , first discussed by @xcite with respect to discrepant disk model fits to the spectrum of the nl , ix vel , is that the hottest , inner annuli of the disk are not present , causing the disk spectrum to be less blue than predicted by a standard disk model . the presence of 29 s oscillations in the optical and uv light in ux uma is an inner disk phenomenon and may be indicative of disrupted accretion at the innermost annuli @xcite , but the oscillations are not always present and we do not detect them our observations . another way to bring model spectra into agreement with observed spectra is to posit a second continuum source : @xcite showed that the addition of optically thin recombination emission can flatten both the slope of the continuum from red to blue and fill in the balmer jump . the optically thin continuum emission may originate from the accretion disk chromosphere ; further modeling will be needed to determine if this picture is consistent with theoretical understanding of the properties of the disk - wind transition region . the wd may also be a significant source of uv continuum emission . if so , its presence will worsen the discrepancy between the models and the observed continuum in ux uma @xcite . @xcite concluded from their analysis of uv light curves that the wd is seen in the uv . they identified a signal in the derivative of the eclipse light curve from which they determined contact points for the wd eclipse ingress and egress . they also determined the flux of the wd at 1600 , 4.8 mjy ( ) , which was 25% of the continuum flux at the time of their observation . from their identification of the wd in the uv eclipse and a distance estimate of 345 pc obtained from their optical and uv eclipse maps , @xcite determined the wd diameter and a temperature : 52,000 k if the full wd is visible , or 70,000 k if the lower hemisphere of the wd is obscured by the disk . they combined their measure of the duration of wd eclipse with an empirical zams mass - radius relation and limits on @xmath5 to determine the binary parameters for ux uma , including the mass ratio , inclination , binary separation , and masses and radial velocities of the primary and the secondary . their derived parameters have become the standard adopted values for ux uma . a simple extension of the spectrum of a wd with the flux and temperature found by @xcite to fuv wavelengths indicates a problem with their analysis , however . this is illustrated in figure [ fig_wd ] , in which we compare model wd spectra to the 2001 fuv and ( for reference ) the 1996 november uv spectra of ux uma . the da wd models shown were created using tlusty and synspec @xcite , and have @xmath29 . a wd with a 1600 flux of 4.8 mjy ( ) and a temperature of 52,000 k exceeds the observed flux at nearly every wavelength in the fuv ; only in regions of strong fuv line emission does the model even fall below the observed flux . at its peak near 1000 , the model wd spectrum exceeds the observed flux by 60% . a 70,000 k wd ( the baptista et al . temperature if half the wd is obscured by the disk ) will deviate even more severely from the observed fuv spectrum . we examined the possibility that the wd flux found by @xcite is correct but that their temperature is too high . the second wd model in figure [ fig_wd ] shows that a cooler wd with a 4.8 mjy flux at 1600 is equally unlikely , however . the model shown is for a 20,000 k wd , which is the hottest wd of that uv flux consistent with the observed fuv spectrum . the problem with a cool wd with substantial uv flux is evident : its spectral shape is too variable with wavelength to be consistent with the observed spectrum . moreover , such a wd would need to have a radius of ( for the 216 345 pc distance range to ux uma ) to emit the flux shown , substantially larger than the radius found by baptista et al . we conclude , therefore , that the wd in ux uma is unlikely to have the high flux value determined by baptista et al . a strong change in the wd flux over time is also unlikely . the 1600 flux in 1993 ( the time of the baptista et al . observations ) is roughly the same as the flux in the 1996 hst spectrum . a comparison of the baptista et al . wd model to that of the 1996 spectrum in figure [ fig_wd ] shows that the model begins to overshoot the observed flux around 1200 , even before an accretion disk contribution to the fuv flux is taken into account . this inconsistency between the derived wd parameters and observed fuv spectra calls into question whether the features identified by baptista et al . in the uv eclipse were actually the ingress and egress of the wd . the s / n of the uv eclipse light curve used by baptista et al . to identify the wd was not high , and we consider their identification of the wd contact points in the derivative light curve as insecure . as all subsequent binary parameters found by @xcite depend on the accurate identification of the wd in the eclipse , we also consider their determination of the wd temperature , wd radius , primary and secondary masses , and inclination unreliable . the derived parameters themselves indicate a problem : a mass ratio of 1 , as found by baptista et al . , is theoretically inconsistent with stable mass transfer , for which there is ample and unambiguous evidence in ux uma . although the wd does not have the large flux contribution found by baptista et al . , it may still be an important source of flux in the fuv . we do not see obvious wd eclipse features in the fuv high time resolution light curve , however . figure [ fig_eclc ] shows a slight notch in eclipse egress near phase 0.02 , but that feature is only present in one of the two eclipses we observed ( the individual eclipses in broadband fuv light can be seen in figure [ fig_contlc ] ) . we can set limits on the wd contribution based on the spectrum of the eclipsed light , shown in the bottom panel of figure [ fig_eclipse ] . since the wd is fully eclipsed , the flux in the eclipsed light spectrum gives an upper limit to the wd contribution . if we assume a standard wd mass @xmath30 , a corresponding radius of order @xmath31 cm ( using the mass - radius relation from nauenberg 1972 ) , and a maximum distance of 345 pc , we find that the maximum temperature of the wd can not exceed 40,000 k if the wd flux is to fall below the observed fuv eclipsed flux . at this temperature and projected area , the wd would be responsible for all of the eclipsed flux , which is unlikely , considering that most of the inner accretion disk and the chromosphere are also eclipsed , and no clear - cut wd eclipse features can be seen in the fuv light curve . if we assume therefore that the wd is only responsible for 50% of the fuv eclipsed flux , its temperature would drop to 25,000 k , and a 20% wd contribution would predict a cool wd of 20,000 k. theoretically , half of the energy of mass accretion is emitted in the boundary layer at or near the wd surface ; for the typical accretion rates of ux uma , this indicates that the surface of the wd ( or some part thereof ) should be quite hot ( @xmath11100,000 k ) . if the wd is hot , it must have a small emitting area to be consistent with the eclipsed fuv flux . if fully visible , a 50,000 k wd must have a radius equal to or smaller than , assuming the minimum distance to ux uma of 216 pc . a 70,000 k wd would have to have a radius @xmath32 cm for its flux to be smaller than the observed fuv eclipsed flux . both of these values correspond to a massive wd ( m@xmath33 ) if a standard wd mass - radius relation holds for for ux uma . observationally , a hot bl / wd has not been observed in ux uma , whose x - ray luminosity ( @xmath34 ergs s@xmath1 for a 340 pc distance ) is too low to be consistent with the theoretically predicted bl emission @xcite . this may indicate that we are not seeing the wd in ux uma at all due to shrouding or occultation by the outer disk . occultation of the wd has been demonstrated in the the nl dw uma ( @xmath35 ) when a low state in the disk exposed the normally hidden wd @xcite . a clear detection of the eclipse of the wd would rule out the occultation hypothesis . ideally , a search for the contact points of the wd would be conducted using multiple ( at least 10 ) uv observations ( with fuse or hst ) of the eclipse to decrease the effects of flickering and variability in the shape of the eclipse . in this manuscript , we presented the observational characteristics of the first high resolution time series fuv spectroscopy of the eclipsing nl , ux uma . our main results are as follows : 1 . the time - averaged fuv spectrum of ux uma is dominated by line emission from numerous , overlapping transitions of , , , , , , , , and , and additional unidentified , blended features . there is no region in the time - averaged spectrum in which the continuum alone can be discerned . 2 . the emission lines have double - peaked profiles , with a narrow ( fwhm @xmath19 600 km s@xmath1 ) central absorption reversal at the rest wavelength of each line . in the strongest lines @xmath13977 , @xmath13992 , the blend at 920924 , and the doublet lines of @xmath41032,1038 the red emission peak is enhanced relative to the blue peak . 3 . the mid - eclipse spectrum is much less complex than the time - averaged spectrum . most of the weaker spectral features are occulted in eclipse , exposing a flat continuum with the prominent emission lines superposed . the central absorption components of these lines are eclipsed . the spectrum of the eclipsed light is a mix of continuum and line emission , with the line features primarily in absorption . several lines appear in emission only at mid - eclipse , when the strong absorption components of the lines are occulted . archival hst spectra of ux uma show similar morphology and behavior during eclipse . 4 . over the full fuv range ( 9101182 ) , the mid - eclipse flux is 40% below the uneclipsed level . for regions of the spectrum away from the prominent emission lines , the mid - eclipse flux is on average 82% below the uneclipsed level . the eclipse depth does not change significantly at longer uv wavelengths : the mid - eclipse flux is 77% of the mean out of eclipse flux in the continuum - dominated regions of the hst ( 1149 1660 ) spectra . the full width at half depth of the eclipse away from the prominent emission lines is roughly the same over the fuv and uv as well . the high time resolution ( 1 s ) light curve of the fuv observation shows strong flickering on time scales of several minutes . we do not detect a periodicity associated with the flickers and we do not detect the 29 s oscillations seen in some previous optical and uv observations . spectra of the flickering peaks and minima show that the flickering occurs in regions near the plane of the disk ( i.e. , the regions that are eclipsed ) ; the continuum fluctuates but the prominent emission lines do not flicker . there is an orbital variation in the fuv light curve : the phase - binned flux around orbital phase 0.55 dips to a minimum 25% below the flux at other phases outside of eclipse . the light curve dip occurs at roughly the same orbital phase as but is shallower than the dip seen in hst observations of ux uma . spectra in the light curve dip , averaged over two orbits , show lower ews for the line absorption features than spectra taken outside the dip . the same behavior was seen in the 1996 hst data . we have detected a wavelength shift in the fuv spectrum phased on the orbital period . the amplitude of the shift is 70 km s@xmath1 over broad regions of the fuv but may range as high as 140 200 km s@xmath1 for the absorption components of the lines . the wavelength shifts lag the phasing of the photometric ephemeris by @xmath36 . the wd parameters found by @xcite from hst uv observations requires a fuv flux from the wd in excess of the observed fuv flux . a wd with a radius ( corresponding to a mass range 0.7 0.47 @xmath37 ) must be cooler than 40,000 k to remain below the observed fuv flux . alternatively , the wd can be hotter but smaller in radius . . a curve of growth analysis of interstellar absorption lines in the fuv spectrum of ux uma ( see appendix [ sec_cog ] ) gives a range of @xmath20 for the column density along the line of sight to ux uma . the time - averaged spectrum of ux uma shows narrow interstellar ( is ) absorption lines . identified lines and their ews are given in table [ tab_ism ] . the ews of the lines are measured averages from time - averaged spectra of each of the four optical channels , giving two or four measurements of each line , depending on the wavelength region . the is lines are from and metals ; there is no molecular hydrogen absorption in the spectrum . the lack of h@xmath38 absorption is somewhat unusual since most fuse spectra , especially for targets outside the local ism , show interstellar h@xmath38 , and distance estimates place ux uma at least 216 pc away , and probably closer to 345 pc away ( see baptista et al . 1995 and sources therein ) . non - local sight lines without molecular hydrogen absorption do exist , however ( see shull et al.2000 , for example ) , and ux uma lies on a sight line directed out of the galactic plane ( @xmath39 , @xmath40 ) . at least two distinct velocity components contribute to interstellar absorption along the line of sight to ux uma . the is lines have fwhm @xmath110.36 , about twice that seen in the spectra of nearby cvs observed by fuse ( froning et al . 2001 ; long et al . 2002 , in preparation ) . the strong , ionized metal lines @xmath13977 the blend of @xmath13989.8 and @xmath13989.9 , @xmath131036 , and @xmath41 have two components , one near zero velocity and a second at a velocity of -60 km s@xmath1 ( the resolution element in the individual channel spectra is @xmath1125 km s@xmath1 ) . we examined archival stis high - resolution ( r = 0.075 ) spectra in the vicinity of @xmath41548,1552 ( the spectra are unpublished ) . the interstellar component of the nearby @xmath131527.71 line also shows two absorption components separated by 60 km s@xmath1 . the is ews given in table [ tab_ism ] for the fuse spectra are integrated over the full is absorption profile of each line . a measure of the absorbing column is of general interest for analysis of ux uma , particularly for x - ray and euv observations , so we undertook a determination of the hydrogen column density along the line of sight to ux uma using a curve of growth analysis . the higher order lyman series lines of can not be used to directly determine @xmath42 because they are all saturated . instead , we used measurements of unblended and is lines and assumed standard is abundances relative to hydrogen for these species ( see mauche , raymond & crdova 1988 and froning et al . 2001 for other applications of this method ) . the gas - phase interstellar abundances of oxygen and nitrogen have been well determined @xcite and are reliable tracers of hydrogen within several hundred parsecs of the sun ( but see moos et al . 2002 concerning nitrogen , which may be less reliable than oxygen as a tracer of hydrogen in the lism ) . unfortunately , none of the measured metal lines are completely optically thin , so a direct measurement of the column density independent of the broadening parameter governing the line widths is not possible . the metals also cluster in a small range of optical depths and do not tightly constrain the column density . the best fit to the and line ews gives @xmath43 @xmath44 and @xmath45 = 4.5 km s@xmath1 for the broadening parameter ( @xmath46 with 12 degrees of freedom ) . the best fit for alone gives the same result . fits within the 95% confidence interval for @xmath42 ( assuming gaussian statistics ) can be obtained for @xmath42 down to @xmath47 @xmath44 ( @xmath48 km s@xmath1 ) . the fit deviates more rapidly from the data points for larger column densities as the growth curve becomes too flat to match the slope in the observed points , leading to an upper estimate of the column density ( for a 95% confidence interval ) of @xmath49 @xmath44 ( @xmath50 km s@xmath1 ) . the resulting range for the column density for the line of sight to ux uma is @xmath51 . the best fit and the upper and lower statistical limits are shown in figure [ fig_cog ] . the curve of growth analysis assumes a single cloud with a single broadening parameter along the line of sight . the broad widths and large ews of the is lines show that multiple clouds must contribute to absorption in neutral hydrogen . if placed on the curve of growth , the lines are described by a broadening parameter of @xmath52 24 km s@xmath1 . this value is implausibly large if dominated by thermal broadening , @xmath45 = 24 km s@xmath1 implies a thermal temperature of 34,600 k ! the growth curve over - estimates the broadening parameter in with a single cloud ; two or more clouds with offset velocities will have smaller , more physical , broadening parameters in each cloud and be consistent with observed ews . the more pertinent question concerns the reliability of the column density derived above , based on and line measurements . in the absence of high resolution spectroscopy or complicated modeling of the sight line to ux uma beyond the scope of this manuscript , we can not know the number and parameters of is absorption components along the sight line , and any number of scenarios could be devised to make the true column density larger or smaller than the range indicated by the curve of growth . there is reason to believe that the column density range given above is a reasonable estimation for the line of sight to ux uma , however . @xcite demonstrated that the curve of growth method can give column densities close to the true value even when multiple components with well - separated or partially overlapping velocities contribute to the absorption . this is true even when individual components are very optically thick , as long as the distribution of cloud parameters is smooth and no single or few highly saturated components dominate the absorption . there is no evidence from the is spectrum of ux uma that any divergent component dominates the absorption , so we conclude that , given the current data , the range of @xmath20 is the best estimate for the column density along the line of sight to ux uma . knigge , c. , drake , n. , long , k. s. , wade , r. a. , horne , k. , & baptista , r. 1998b , , 499 , 429 knigge , c. , long , k. s. , wade , r. a. , baptista , r. , horne , k. , hubeny , i. , & rutten , r. g. m. 1998 , , 499 , 414 lccc h & 915.329 & 0.000385 & 0.3908 + h & 915.824 & 0.000468 & 0.3901 + h & 916.429 & 0.000577 & 0.4184 + h & 917.181 & 0.0007226 & 0.4226 + h & 918.129 & 0.0009213 & 0.4346 + h & 919.351 & 0.00120 & 0.4215 + h & 920.963 & 0.001605 & 0.4290 + h & 923.150 & 0.002216 & 0.4125 + o & 924.950 & 0.001540 & 0.0473 + h & 926.226 & 0.003183 & 0.426 + o & 929.517 & 0.002295 & 0.0544 + h & 930.748 & 0.004816 & 0.4600 + o & 936.630 & 0.003650 & 0.0765 + h & 937.803 & 0.007804 & 0.4451 + o & 948.686 & 0.005420 & 0.0606 + h & 949.743 & 0.01394 & 0.4349 + o & 950.885 & 0.001570 & 0.0592 + n & 953.415 & 0.01314 & 0.0619 + n & 953.655 & 0.02492 & 0.0608 + n & 953.970 & 0.03479 & 0.1231 + n & 963.990 & 0.0148 & 0.0534 + n & 964.626 & 0.0094 & 0.0607 + n & 965.041 & 0.00675 & 0.0676 + o & 971.738 & 0.01367 & 0.0729 + h & 972.537 & 0.0290 & 0.3131 + o & 976.448 & 0.003310 & 0.0700 + c & 977.020 & 0.7620 & 0.2482 + o & 988.578 & 0.000553 & 0.2021 + o & 988.655 & 0.0083 & + o & 988.773 & 0.04650 & + n & 989.799 & 0.1066 & 0.1892 + si & 989.873 & 0.1330 & + h & 1025.722 & 0.0791 & + o & 1025.762 & 0.01705 & + c & 1036.337 & 0.1231 & 0.2602 + o & 1039.230 & 0.009200 & 0.0760 + ar & 1048.220 & 0.26275 & 0.0404 + ar & 1066.660 & 0.06652 & 0.0236 + n & 1083.990 & 0.1031 & 0.2357 + n & 1134.165 & 0.01342 & 0.0692 + n & 1134.415 & 0.02683 & 0.085 + n & 1134.980 & 0.04023 & 0.1102 + cccccc n & 923.2 & 924.5 & & & + p & 950.7 & 950.0 & -50 & 2100 & 2900 + c & 977.0 & 976.7 & -90 & 2600 & 4900 + n & 989.8 & 992.1 & -69 & 2550 & 5300 + s & 1012.5 & 1012.8 & -30 & 3400 & 5300 + h & 1025.7 & 1025.0 & -25 & 1600 & 4100 + o & 1031.9,1037.62 & 1034.4 & -160 & 3300 & 5800 + s & 1062.7 & 1062.9 & -58 & 2050 & 4900 + s & 1073.0 & 1073.2 & -58 & 2200 & 4750 + n /he & 1085.0 & 1085.1 & -19 & 1750 & 3700 + si & 1108.4 & 1110.9 & -113 & 2700 & 5500 + si & 1122.5,1128.3 & 1124.8 & & 3600 & 6000 + ? ? & & 1142.7 & -61.2 & 2300 & 5200 + c & 1175.3 & 1176.0 & -88.0 & 1950 & 4500 +

we perform a linear analysis of the stability of a magnetized relativistic non - rotating cylindrical flow in the aproximation of zero thermal pressure , considering only the mode . we find that there are two modes of instability : kelvin - helmholtz and current driven . the kelvin - helmholtz mode is found at low magnetizations and its growth rate depends very weakly on the pitch parameter . the current driven modes are found at high magnetizations and the value of the growth rate and the wavenumber of the maximum increase as we decrease the pitch parameter . in the relativistic regime the current driven mode is splitted in two branches , the branch at high wavenumbers is characterized by the eigenfunction concentrated in the jet core , the branch at low wavenumbers is instead characterized by the eigenfunction that extends outside the jet velocity shear region . [ firstpage ] galaxies : jets , mhd , instabilities , relativistic processes

1307.6388

the formation and propagation of astrophysical jets are strongly affected by plasma instabilities , whose study is therefore of fundamental importance for understanding their dynamics and their associated phenomenology . in jets there are several possible different kinds of instability , among them the most studied are the kelvin - helmholtz instability ( khi ) driven by the velocity shear between the jet and the ambient medium and the current driven instability ( cdi ) associated with a longitudinal current and therefore with the toroidal component of magnetic field . since the most promising models for the acceleration and collimation of jets involve the presence of a magnetic field with footpoints anchored to a rotating object ( an accretion disk or a spinning star or black hole ) , the presence of a toroidal field component is a natural consequence and cdi may play an important role in the jet propagation . among cdi , the @xmath1 kink mode is the most effective , leading to an helical displacement of the whole jet body , and for this reason is the only one we will consider in this paper . recent numerical simulations have shown the development of such instabilities both in newtonian @xcite and in relativistic @xcite jets . an important step towards a better understanding of simulation results is a linear analysis of the instabilities , that is still largely missing for the relativistic magnetohydrodynamic ( mhd ) regime . khi have been extensively studied in several different configurations in the newtonian ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and relativistic ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) cases , but very few linear analyses have been presented for a relativistic magnetized jet @xcite . similarly , cdi have been widely studied in the newtonian limit ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , however , for the relativistic mhd case , only the force - free limit has been considered @xcite . @xcite have considered the case in which the longitudinal magnetic field is constant , showing that in this case the jet is stable , on the contrary @xcite showed that jets with longitudinal magnetic field decreasing outward can be unstable . @xcite finally derived a general necessary condition for instability . the force - free limit is valid when the energy density of electromagnetic fields is much larger than the energy density of matter , i.e. when the jet energy flux is mainly in the form of poynting flux . acceleration models predict that jets start being poynting dominated and progressively undergo a transition to a matter dominated state , but how fast the transition occurs is uncertain . from the observational point of view , the evidence is that , above the parsec scale , jets can not be poynting dominated @xcite . on the other hand , it has been suggested @xcite that jet instabilities may play a role in this conversion and that blazar activity could be linked to the development of such instabilities either through shock formation or through direct magnetic energy dissipation processes . these considerations suggest that it is very important to go beyond the force - free limit in the instability analysis and this is exactly the aim of the present paper in which we will analyze the linear stability properties of a cylindrical magnetized relativistic flow , taking into account the effects of matter inertia , but still neglecting thermal pressure . since the equilibrium configurations of the jet may be quite complex , with many possible sources of instabilities , like the longitudinal jet velocity , the toroidal field and the jet rotation , we will not start from a full configuration where all the above elements are present , but from a simpler case where one of the above elements , namely rotation , is absent . the additional effects introduced by rotation will then be examined in a following paper . in section 2 we will describe the physical problem , the relevant equations , the general equilibrium configuration and the relevant parameters , while in section 3 we will derive the linearized equations and describe the procedure for finding the normal modes . in section 4 we will present the results of our analysis for the non rotating case , first for a static configuration and then for a moving jet . finally in section 5 we will summarize our findings . we study the stability of a cold ( zero pressure ) relativistic magnetized cylindrical flow . the relevant equations are continuity and momentum coupled with maxwell equations : @xmath2 @xmath3 @xmath4 @xmath5 where @xmath6 is the proper density , @xmath7 is the lorentz factor , and @xmath8 , @xmath9 , @xmath10 , @xmath11 are respectively the velocity , magnetic field , electric field and current 3-vectors . the units are chosen so that the speed of light is @xmath12 , we also remark that in the following a factor of @xmath13 will be reabsorbed in the definitions of @xmath10 and @xmath9 . the first step in the stability analysis is to define an equilibrium state satisfying the stationary form of eqs . ( [ eq : drho / dt]- [ eq : de / dt ] ) and this will be done in the next subsection . we adopt a cylindrical system of coordinates @xmath14 ( with versors @xmath15 ) and seek for axisymmetric steady - state solutions for a relativistic magnetized jet , i.e. , @xmath16 . we assume that the jet propagates in the vertical ( @xmath17 ) direction and the magnetic field configuration consists of a vertical ( poloidal ) component @xmath18 and a toroidal component @xmath19 and can be expressed as @xmath20 the magnetic field configuration can be characterized by the pitch parameter @xmath21 from the stationarity condition , the continuity equation and the independence of @xmath22 on @xmath17 and @xmath23 , we get @xmath24 and from the condition @xmath25 we obtain that the velocity can then be written as @xmath26 where @xmath27 is the fluid toroidal velocity and @xmath28 is the angular velocity of field lines and they are related by @xmath29 the electric field is always directed radially and can be expressed as @xmath30 the only remaining non - trivial equation is given by the radial component of the momentum equation ( [ eq : dm / dt ] ) which simplifies to @xmath31 where @xmath32 . in the nonrelativistic limit @xmath33 reduces to @xmath19 and the equilibrium condition acquire the classical newtonian form @xmath34 eqs . ( [ eq : radial_eq ] ) or ( [ eq : radial_eqmhd ] ) leave the freedom of choosing the radial profiles of all flow variables but one and then solve for the remaining profile . we begin by prescribing the profiles of the proper density and lorentz factor that well describe a jet configuration , with the velocity and density variations concentrated inside the jet radius @xmath35 @xmath36 where @xmath37 is the ambient / jet density contrast , @xmath38 is the lorentz factor relative to the @xmath17 component of the velocity only , while @xmath39 is the lorentz factor on the axis where the vertical flow velocity is @xmath40 . from now on , we will use the subscript @xmath41 to denote values at @xmath42 . we note that , in the newtonian case , the presence of a longitudinal velocity has no effect on the radial equilibrium eq . ( [ eq : radial_eqmhd ] ) , while it changes the relativistic eq . ( [ eq : radial_eq ] ) modifying the centrifugal term . in the newtonian limit , it is then customary to prescribe the profile of the azimuthal field @xmath19 and this choice is more arbitrary since we have no direct information about the magnetic configuration in astrophysical jets . the choice of the @xmath19 distribution is equivalent to a choice of the distribution of the longitudinal component of the current and also determines the behavior of the pitch parameter @xmath43 , that is important for the stability properties . in principle , one can then have several equilibria characterized by different forms of the current distribution , that can be more or less concentrated , can peak on the axis or at the jet boundary , and can close in different ways ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in this limit , we start by considering the azimuthal field profile @xmath44\,,\ ] ] where @xmath45 is the magnetization radius and @xmath46 determines the maximum field strength . this profile corresponds to a current distribution peaked on the jet axis and , choosing @xmath47 , concentrated inside the jet . in addition , one can assume that the current closes at very large distances from the jet . in the relativistic limit the most natural generalization is to prescribe in a similar way the behavior of @xmath33 @xmath48\,.\ ] ] we observe that , in the absence of rotation , @xmath33 represents the azimuthal field strength measured in the proper frame . prescribing the profile of @xmath33 instead of @xmath19 modifies the current profile in the laboratory frame , introducing a return current in the region of the velocity shear and this has consequences for the stability properties as it will be discussed below . furthermore , the equilibrium configuration may be modified by the presence of rotations to different degrees : in one extreme case the gradient of @xmath49 in eq . ( [ eq : radial_eq ] ) ( or the gradient of @xmath50 in eq . ( [ eq : radial_eqmhd ] ) ) is exactly balanced by the centrifugal force and @xmath18 is constant , in the other extreme it is balanced by the gradient of @xmath51 . this suggests to introduce , more generally , a parameter @xmath52 $ ] so that the equilibrium poloidal magnetic field is given by @xmath53 where @xmath54 is the error function , @xmath55 corresponds to the absence of rotation , while @xmath56 corresponds to maximum rotation . introducing the expression for @xmath18 given by eq . ( [ eq : bz_prof ] ) in eq . ( [ eq : radial_eq ] ) we can get the azimuthal velocity from @xmath57 finally the azimuthal field is obtained from the definition of @xmath58 using @xmath59 . this yields a quadratic equation in @xmath19 with solution @xmath60 here we consider the negative branch because it guarantees that @xmath19 and @xmath27 have opposite signs , as suggested by acceleration models . thus , in our model the radial profile of the pitch parameter , eq . ( [ eq : pitch ] ) , is always negative . we choose to control the magnetic field configuration by specifying the value of the pitch on the axis @xmath61 and the ratio between the energy density of the matter and the magnetic energy density @xmath62 , where @xmath63 and @xmath64 represents the average across the beam : @xmath65 we note that @xmath66 by construction although the radial profile of the pitch parameter ( eq . [ eq : pitch ] ) using the negative branch of eq ( [ eq : bphi ] ) in the definition of @xmath19 is always negative . the constants @xmath67 and @xmath68 appearing in the above equations can be found in terms of @xmath61 and @xmath69 from the simultaneous solution of the two expressions given in eq . ( [ eq : p_cnd_ma ] ) , together with ( [ eq : bav ] ) . in particular , from the definition of the pitch parameter , after some algebra we find ( in the @xmath70 limit ) @xmath71 where @xmath72 . in practice , eq ( [ eq : bz0 ] ) is used to compute @xmath67 for trial values of @xmath68 until the prescribed value of magnetic energy is satisfied . summarizing , our equilibrium configuration depends on the 5 parameters @xmath37 , @xmath73 , @xmath74 , @xmath61 and @xmath75 specifying , respectively , the jet density contrast , bulk flow velocity , strength of centrifugal force , magnetic pitch and the ratio between the energy density of the matter and the magnetic energy density . let us consider small perturbations @xmath76 to the equilibrium state described above , which hereafter will be identified by a zero subscript . the linearized continuity , momentum , induction equations and the ideal mhd conditions are @xmath77 @xmath78 @xmath79 @xmath80 assuming now the perturbations to be of the form @xmath81 , after lengthy algebraic manipulations described in the appendix [ ap : linear ] , we arrive at a system of two first order differential equations in the radial coordinate for the two basic variables the radial displacement @xmath82 and the perturbed electromagnetic pressure @xmath83 defined as @xmath84 and @xmath85 the system of equations can then be written as @xmath86 @xmath87\xi_{1r}+ \\+ \frac{1}{r}\left(\rho_0\gamma_0 ^ 2v_{0\varphi}^2c_3 - 2d + \frac{c_6}{r } \right)+ c_7 \pi_1 \label{eq : dp / dr}\end{gathered}\ ] ] where @xmath88\,,\ ] ] @xmath89 the quantities @xmath90 depend on the chosen profiles of the equilibrium solution and are given in the appendix [ ap : coefficients ] . these two equations together with the appropriate boundary conditions represent an eigenvalue problem , where @xmath91 is the eigenvalue ( we observe that we here adopt a temporal approach to the stability analysis ) . we have instability when @xmath91 has a negative imaginary part . the domain of integration for eqs . ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) cover the interval from @xmath92 to @xmath93 , so we have to specify the boundary conditions at @xmath42 and for @xmath94 . on the axis at @xmath42 the equations are singular but the solutions have to be regular while at infinity the solutions have to decay and no incoming wave is allowed ( sommerfeld condition ) . for finding the eigenvalue we use a shooting method with a complex secant root finder . the numerical integration can not start at @xmath42 ( because of the singularity ) , so we start at a small distance from the origin where the solution is obtained through a series expansion of the equations described in the appendix [ ap : small_r ] . similarly , we start a backward integration from a sufficiently large radius , where the asymptotic solution is obtained as described in the appendix [ ap : large_r ] and then we match the two numerical solutions at an intermediate radius . furthermore , we have to consider that eqs . ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) may have singular points , that arise in the following cases : 1 . when the doppler shifted frequency , @xmath95 , appearing in the denominators in @xmath96 and @xmath97 , becomes zero at some @xmath98 , @xmath99 . this is a well - known corotation singularity , when the phase speed of a wave perturbation coincides with the basic flow velocity at @xmath98 and corresponds to resonant interaction between waves and the background jet flow . this corotation singularity is a physical one . 2 . when the determinant becomes zero at some @xmath100 , @xmath101 . the latter condition is a quadratic equation with respect to @xmath95 that gives @xmath102 from which we derive a phase speed equal to that of a relativistic alfvn wave ( see keppens & meliani 2008 , eq . 28 , istomin & pariev 1996 ) @xmath103 where @xmath104 is the versor of the wave vector and @xmath105 is the total pressure . in other words , the above determinant is zero and therefore equations ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) have singular , or resonant points at @xmath100 , where the phase speed of the wave perturbation coincides with that of relativistic alfvn waves . like the corotation singularity , this singularity is also physical , and results from the resonant interaction of perturbations with relativistic alfvn waves . + @xcite have discussed the proper way to handle these singularities , however , since singular points are found only for real values of @xmath91 and since we are interested in unstable modes and therefore complex values of @xmath91 , we can assume that our integration will always avoid singular points . before discussing in detail our results , however , we recall in section [ subsec : parameters ] the units we will use , the full set of parameters defining our problem and which ones will be investigated in more detail in the present work . we then preliminarly examine the static case in subsection [ subsec : static ] , the full case in subsection [ subsec : moving ] and finally a more detailed analysis of the cdi is performed in subsection [ subsec : cd ] . as discussed in section [ sec : equilibrium ] , the basic equilibrium state is determined by five parameters : the jet density contrast @xmath37 , the bulk flow velocity which is derived from @xmath73 , the rotation parameter @xmath74 , the pitch on the axis @xmath61 and the ratio @xmath75 between the energy density of the matter and the magnetic energy density . we notice that our parameter @xmath106 is the inverse of the magnetization parameter often used in the literature . in the present paper , as already discussed , we consider only the case without rotation , so @xmath74 is kept fixed to @xmath92 , and we also consider jets with density equal to that of the ambient medium , so @xmath37 is kept fixed to @xmath107 . we focus our analysis on the dependence on the three parameters @xmath73 , @xmath61 and @xmath69 . in the following discussion , instead of @xmath69 we will make use of the quantity @xmath108 which , in the newtonian limit , corresponds to the ratio between the jet speed and the alfvn speed , while in the relativistic limit reduces to @xmath69 . the unit of velocity is the light speed @xmath41 and we choose as unit of length the jet radius @xmath35 , so , unless otherwise specified , the growth rate is expressed in units of @xmath109 . an additional parameter is represented by the width @xmath45 of the current distribution , this parameter , unless otherwise specified , is kept fixed to @xmath110 . of course , in the static case , the jet radius has no significance and is an arbitrary measure , however we keep it as the unit of length for consistency with the general case . static columns in the newtonian case have been already studied by many authors ( see e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and differ one another in many respects , but mainly for the parameter range considered and for the equilibrium magnetic field configuration . in particular the studies by @xcite consider cases with subthermal magnetic field strength . in the present work , conversely , we neglect the pressure term and our regime of investigation addresses the case of suprathermal field strengths and can be compared with that of @xcite , that is performed in the same regime . in the study of cdi , a very important role is played by resonant surfaces , i.e. the surfaces where the condition @xmath112 is satisfied . on these surfaces , in fact , the stabilizing effect of magnetic tension is absent and these regions are therefore more prone to instability . the resonant condition can be rewritten as @xmath113 this condition can be verified only for positive @xmath114 and , for a constant pitch distribution like that used by @xcite , marks the stability boundary : wavenumbers larger than this critical limit are stable , while smaller wavenumbers are unstable . . the values of @xmath61 for the different curves are reported in the legend and , for comparison , in the top panel , we plot also the case of a configuration with constant pitch . in the bottom panel we show how well the scaling given by eq . ( [ eq : scaling ] ) reproduces our results , a significant deviation can be observed only for @xmath115 . , title="fig:",width=340 ] . the values of @xmath61 for the different curves are reported in the legend and , for comparison , in the top panel , we plot also the case of a configuration with constant pitch . in the bottom panel we show how well the scaling given by eq . ( [ eq : scaling ] ) reproduces our results , a significant deviation can be observed only for @xmath115 . , title="fig:",width=340 ] this behavior can be observed in the top panel of fig . [ fig : staticmhd ] , where we plot the growth rate as a function of the wavenumber for the case with @xmath116 for the constant pitch configuration ( black curve , top panel ) and for our equilibrium model ( colored lines ) . in the top panel we measure the wavenumber in units of @xmath117 and the growth rate in units of @xmath118 . in these units , the constant pitch solutions of @xcite are all represented by the single black curve independently of the pitch value . instability is present for all wavenumbers @xmath119 ( i.e. for wavenumbers lower than the threshold given by the resonance condition eq . [ eq : resonance ] ) and , for @xmath120 , the growth rate decreases as a power law . we note that , while we employ radiation condition at infinity , @xcite in most of their calculations made use of boundary conditions consistent with a rigid wall set at a finite radial distance . this condition introduces a new scale and restricts the unstable range in wavenumbers , with a small wavenumber cutoff . . the different curves refer to different values of the parameter @xmath121 , whose values are reported in the legend.,width=340 ] our equilibria are characterized by two length scales : the pitch value on the axis @xmath61 and the width of the current distribution @xmath45 and therefore we have the additional parameter @xmath121 . the pitch profile @xmath122 plays a fundamental role for the stability properties and , for our equilibrium configurations , it is shown in fig . [ fig : pitch ] for different values of @xmath121 . it is always characterized by an increase at large radii , owing to the confinment of the current inside @xmath123 , while in the central region we can distinguish two regimes : for @xmath124 the pitch is constant up to @xmath125 while for @xmath126 it decreases to a minimum value at @xmath127 immediately after the inner flat region . furthermore , there exists a critical value of @xmath128 below which equilibrium configuration are not possible . below this critical value , the equilibrium condition would require a negative value of @xmath51 meaning that the longitudinal field pressure gradient is no longer able to balance the inward force of the azimuthal magnetic field . the behavior of the growth rate in the first regime ( @xmath124 ) has a form similar to the constant pitch situation , but its value decreases as the parameter @xmath121 increases , as it is shown in the top panel of fig . [ fig : staticmhd ] . in the same limit the growth rate takes the asymptotic form @xmath129 where the function @xmath130 is independent from @xmath121 . the validity of the previous scaling law for any value of @xmath121 is demonstrated in the bottom panel of fig . [ fig : staticmhd ] , where we plot @xmath131 as a function of @xmath132 . the curves in the figures , therefore , represent the function @xmath130 and should be independent from @xmath121 . in fact , the purple ( @xmath133 , @xmath134 ) , blue ( @xmath135 , @xmath134 ) and orange ( @xmath136 , @xmath137 ) curves are overimposed and almost coincident with the green curve corresponding to @xmath138 , @xmath110 . a significant deviation from the above scaling is observed only for @xmath115 , as shown by the red curve . in this case , as discussed above , we have a region of decreasing pitch whose effect is to widen the instability range to larger values of @xmath132 thus increasing the growth rate above the value predicted by the scaling law and moving the maximum towards somewhat larger values of @xmath132 . for a static configuration , relativistic effects are introduced only by increasing the magnetic field strength and the previous results shown in fig . [ fig : staticmhd ] remain the same provided the alfvn velocity is replaced by the correct relativistic expression @xmath139 where @xmath140 is the gas enthalpy . for the case with @xmath142 and @xmath143 . as discussed in the text , the results for the two cases @xmath116 and @xmath144 coincide . in the left panel we have the kelvin - helmholtz ordinary mode , while in the right panel we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate.,width=566 ] for the case with @xmath142 and @xmath146 . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . in the left panels , for high values of @xmath108 , we have the ordinary mode of the kelvin - herlmholtz instability , for low values of @xmath108 the case @xmath147 ( top ) shows the current driven instability , while the case @xmath148 ( bottom ) is stable . in the right panels we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] for the case with @xmath142 and @xmath146 . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . in the left panels , for high values of @xmath108 , we have the ordinary mode of the kelvin - herlmholtz instability , for low values of @xmath108 the case @xmath147 ( top ) shows the current driven instability , while the case @xmath148 ( bottom ) is stable . in the right panels we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] for the case with @xmath142 and @xmath149 . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . in the left panels , for high values of @xmath108 , we have the ordinary mode of the kelvin - herlmholtz instability , for low values of @xmath108 the case @xmath147 ( top ) shows the current driven instability , while the case @xmath148 ( bottom ) is stable . in the right panels we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] for the case with @xmath142 and @xmath149 . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . in the left panels , for high values of @xmath108 , we have the ordinary mode of the kelvin - herlmholtz instability , for low values of @xmath108 the case @xmath147 ( top ) shows the current driven instability , while the case @xmath148 ( bottom ) is stable . in the right panels we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] we start the discussion of the dynamic case considering a flow moving at @xmath142 , corresponding to a non - relativistic @xmath150 . in this case the results obtained in the newtonian limit coincide almost exactly with those obtained with the full relativistic treatment . an overview of the mode structure can be gained by looking at figs . [ fig : p10000mhd ] , [ fig : p10mhd ] , [ fig : p1mhd ] , where we show the behavior of the growth rate as a function of the wavenumber @xmath151 and of @xmath108 for @xmath143 , @xmath146 and @xmath149 , respectively . figs . [ fig : p10mhd ] and [ fig : p1mhd ] present the results for @xmath152 in the upper panels and for @xmath148 in the lower panels , while in fig . [ fig : p10000mhd ] the system has no way to distinguish between @xmath152 and @xmath148 and the results for the two modes are coincident . as mentioned , we now expect the appearence of two types of instability , namely , kelvin - helmholtz instabilities ( khi ) and current driven instabilities ( cdi ) . moreover , in the case of a purely longitudinal field , cdi are absent and only khi may be effective . this is the case of fig . [ fig : p10000mhd ] , where we show the results for @xmath143 , i.e. for a magnetic field almost exactly longitudinal . the two panels in the figure refer to the ordinary mode and to the first reflected mode ( for a discussion on reflected modes , see e.g. * ? ? ? from the figure , we can see that the jet is stable below @xmath153 , the maximum growth rate for the ordinary mode is @xmath154 and it is found at @xmath155 and @xmath156 . moving towards higher values of @xmath157 , the relative maximum shifts towards smaller values of @xmath151 and our results show that both the wavenumber of the maximum and the maximum itself scale as @xmath158 . the first reflected mode ( right panel ) has a smaller growth rate , @xmath159 , and remains unstable at larger values of @xmath151 as expected . [ fig : p10mhd ] refers to @xmath146 with the upper and lower panels showing , respectively , the growth rates computed for the @xmath152 and @xmath148 modes . clearly , the upper left panel shows a region of instability for small values of @xmath108 which is completely absent in the lower panel . this instability region corresponds to onset of cdi modes that become stable for @xmath148 since the resonance condition ( eq . [ eq : resonance ] ) can not be satisfied . for small values of @xmath108 , the instability behavior is very similar to the static case discussed in the previous subsection and its growth rate is two order of magnitude smaller than that of the khi . the mode becomes stable for @xmath160 , which corresponds to the stability limit given by the condition @xmath161 . for larger values of @xmath108 the cd mode merge with the kh mode and shows essentially no difference with respect to the previous case . the results for @xmath149 are shown in fig . [ fig : p1mhd ] . as expected , the cdi moves towards higher values of @xmath151 as the stability limit @xmath161 gives a limiting value of @xmath162 . the growth rate of the cd mode also increases and scales as @xmath163 as in the static case eventually becoming dominant over the khi . for @xmath144 ( lower panels ) the cdi is absent and the khi presents only slight differences with the cases at larger values of @xmath61 . for the case with @xmath164 and @xmath165 ( correponding to @xmath143 in the rest frame of the jet ) . as discussed in the text , the results for the two cases @xmath116 and @xmath144 coincide . in the left panel we have the kelvin - helmholtz ordinary mode , while in the right panel we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , width=566 ] for the case with @xmath164 and @xmath149 ( correponding to @xmath146 in the rest frame of the jet ) . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . in the left panels , for high values of @xmath108 , we have the ordinary mode of the kelvin - herlmholtz instability , for low values of @xmath108 the case @xmath147 ( top ) shows the current driven instability which is splitted in two branches ( note that the region between @xmath166 and @xmath167 is stable ) , while the case @xmath148 ( bottom ) is stable . in the right panels we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] for the case with @xmath164 and @xmath149 ( correponding to @xmath146 in the rest frame of the jet ) . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . in the left panels , for high values of @xmath108 , we have the ordinary mode of the kelvin - herlmholtz instability , for low values of @xmath108 the case @xmath147 ( top ) shows the current driven instability which is splitted in two branches ( note that the region between @xmath166 and @xmath167 is stable ) , while the case @xmath148 ( bottom ) is stable . in the right panels we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] for the case with @xmath164 and @xmath168 ( correponding to @xmath149 in the rest frame of the jet ) . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . as discussed in the text , for @xmath116 ( top ) , the kelvin - helmholtz ordinary mode and the current driven mode merged . in the right panel we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] for the case with @xmath164 and @xmath168 ( correponding to @xmath149 in the rest frame of the jet ) . the top panels refer to @xmath116 while the bottom panels refer to @xmath144 . as discussed in the text , for @xmath116 ( top ) , the kelvin - helmholtz ordinary mode and the current driven mode merged . in the right panel we have the first reflected kelvin - helmholtz mode . the levels are equispaced in logarithmic scale from @xmath145 to the maximum value of the growth rate . , title="fig:",width=566 ] increasing the flow velocity up to lorentz @xmath164 , we show the growth rate behavior as a function of @xmath151 and @xmath108 in figs [ fig : p10000rmhd ] , [ fig : p10rmhd ] , [ fig : p1rmhd ] corresponding , respectively , to @xmath165 , @xmath149 and @xmath168 . we remark that the pitch in the comoving frame is obtained by multiplying the pitch in the lab frame by @xmath7 and therefore the chosen values of @xmath61 in the comoving frame are exactly the same ones used for the the classical case . the upper and lower panels in figs . [ fig : p10rmhd ] and [ fig : p1rmhd ] refer respectively , to @xmath152 and @xmath148 . similarly to the classical case , we do not have any difference between the @xmath152 and @xmath148 modes for @xmath169 , see fig . [ fig : p10000rmhd ] . the right panels in each figure show , as before , the reflected kh mode . the general instability behavior is quite similar to the classical case and we can easily recognize the khi region and the cdi region . focusing on the khi , we see that the stability boundary and , consequently , the position of the maximum growth rate has moved towards larger values of @xmath69 ( see figs . [ fig : p10000rmhd ] , [ fig : p10rmhd ] left and right panels ) as it is expected to happen for relativistic flows @xcite . decreasing @xmath61 , the kh instability boundary moves towards smaller values of @xmath108 , as it is evident in the lower panels of fig . [ fig : p1rmhd ] corresponding to @xmath148 and in the right upper panel of the same figure ( reflected mode @xmath152 ) . indeed , the stabilizing longitudinal component of magnetic field decreases with decreasing @xmath61 . in the upper left panel of fig . [ fig : p1rmhd ] we see that , for @xmath152 , the khi and cdi have merged . moreover , for each value of @xmath61 , an increase of @xmath108 leads to a shift of the relative maximum of the growth rate towards smaller values of @xmath151 ( see figs . [ fig : p10000rmhd ] , [ fig : p10rmhd ] , [ fig : p1rmhd ] ) and both the wavenumber of the maximum and the maximum itself scale as @xmath170 precisely as in the newtonian limit . for @xmath149 and @xmath116.,width=340 ] in fig . [ fig : maxkh ] we plot the value of the maximum growth rate of the kh mode as a function of @xmath171 for @xmath149 and @xmath116 . this curve can be considered representative for any value of the pitch since , as we have seen , the growth rate of the khi is only weakly dependent on this parameter . in the newtonian limit , for small values of @xmath171 , the growth rate becomes essentially proportional to @xmath172 , reaches a maximum at @xmath173 and then progressively decreases . this result is in agreement with the work of @xcite that found that relativistic motion plays a stabilizing role on the growth of the khi . given in the legends and for @xmath174 . the three panels refer to three different values of @xmath73 , more precisely @xmath142 for the top panel , @xmath175 for the middle panel and @xmath176 for the bottom panel . the values of @xmath61 are chosen so that in the rest frame we have always @xmath177 . in the relativistic case we observe the splitting of the mode in two branches as described in the text . , title="fig:",width=340 ] given in the legends and for @xmath174 . the three panels refer to three different values of @xmath73 , more precisely @xmath142 for the top panel , @xmath175 for the middle panel and @xmath176 for the bottom panel . the values of @xmath61 are chosen so that in the rest frame we have always @xmath177 . in the relativistic case we observe the splitting of the mode in two branches as described in the text . , title="fig:",width=340 ] given in the legends and for @xmath174 . the three panels refer to three different values of @xmath73 , more precisely @xmath142 for the top panel , @xmath175 for the middle panel and @xmath176 for the bottom panel . the values of @xmath61 are chosen so that in the rest frame we have always @xmath177 . in the relativistic case we observe the splitting of the mode in two branches as described in the text . , title="fig:",width=340 ] the cd mode is present only for @xmath152 ( as in the newtonian case ) and , by lowering @xmath61 , it progressively shifts towards larger values of the wavenumber and increases also its growth rate . moreover , for @xmath149 , we observe a splitting of the cdi in two unstable regions ( see the lower half in the top left panel of fig . [ fig : p10rmhd ] ) . this mode splitting can be better understood by inspecting fig . [ fig : cd ] where we plot the normalized growth rate as a function of the wavenumber @xmath151 for @xmath178 and different values of @xmath61 . since for @xmath179 the cdi becomes essentially independent of @xmath69 , the curves plotted in fig . [ fig : cd ] are representative of the instability behavior for small values of @xmath69 . the three panels refer to different values of @xmath73 , namely , @xmath180 in the upper panel , @xmath181 in the middle panel and @xmath182 in the lower panel . in the upper panel we see that the behavior of cdi at small non - relativistic velocities is essentially the same as in the static case ( see fig . [ fig : staticmhd ] for comparison ) . , in the middle panel we plot the pitch profile @xmath122 and in the bottom panel we plot the eigenfunctions for the inner ( orange ) and outer ( blue ) modes , for @xmath181 , @xmath149 . the orange curve is for @xmath183 ( inner mode ) and the blue curve is for @xmath184 ( outer mode).,width=340 ] when @xmath73 is increased to @xmath185 ( middle panel ) , we observe that the mode splits into two . in order to comprehend the reason behind the observed mode splitting , we plot in fig . [ fig : eigen ] the radial profiles of the lorentz factor ( top panel ) , pitch ( middle panel ) and the eigenfunctions relative to the electromagnetic pressure perturbation for two different values of the wavenumber , in the case with @xmath181 and @xmath149 ( bottom panel ) . the red shading marks the shear region where the jet velocity decreases and pitch profile becomes the steepest . in plotting the eigenfunctions , we have chosen the wavenumbers @xmath186 ( blue curve ) and @xmath187 ( orange curve ) which correspond to the maximum growth rate of each of the two branches shown by the green curves in the middle panel of fig . [ fig : cd ] . the modes present a resonant behavior and the peak positions are located at the radii where the condition @xmath188 is fulfilled . looking at the form of the eigenfunctions in the bottom panel of fig . [ fig : eigen ] , we see that the mode with @xmath183 peaks inside the jet core in the flat part of the pitch profile . at the opposite , the mode with @xmath189 reaches a maximum outside the velocity shear region . for this reason , we denote the right branch at large wavenumbers as the `` inner mode '' and the left branch at small wavenumbers as the `` outer mode '' . given that the absolute value of the pitch is monotonically increasing with radius , we note that the resonant position , expressed by the condition @xmath190 , has to shift to larger radii as @xmath151 is decreased . thus , if we focus on the inner mode , the resonant position will move from @xmath191 ( at large @xmath151 ) toward the exterior until , for a lower value of @xmath151 , it will fall inside the shear region where the pitch profile is steeper thereby becoming stabilized . a further decrease in the wavenumber leads the resonance point outside the velocity shear region , where the pitch slope decreases , thus giving rise to the outer branch of the instability . the outer mode , for the considered parameters , has a growth rate that is always somewhat larger than that of the inner mode and it is the one visible in fig . [ fig : p1rmhd ] ( upper left panel ) . for smaller values of @xmath61 , [ fig : cd ] shows that the deviation of the growth rate from the @xmath192 scaling ( eq . [ eq : scaling ] ) becomes larger and the outer mode stretches while moving towards larger values of @xmath151 . at the same time , the inner mode widens while moving towards lower values of @xmath151 , until a region of superposition between the two modes is formed . this behavior can be attributed to the non - monotonic trend of the pitch for small values of @xmath61 , as already shown in see fig . [ fig : pitch ] . the inner mode , being confined inside the jet core , does not feel the effect of the velocity shear and its properties may be derived by simply applying the appropriate lorentz transformations to the results obtained in the static case . we can in fact relate the pitch , growth rate and wavenumber in the jet frame to those measured in the laboratory frame by the following relations @xmath193 where the primed quantities are measured in the jet frame , while the unprimed quantities are measured in the laboratory frame . from the scaling given by eq . ( [ eq : scaling ] ) , we obtain @xmath194 the scaling of the growth rate with @xmath195 is demonstrated by fig . [ fig : cd_scaling ] , where we plot @xmath196 as a function of the wavenumber for three different values of @xmath73 . the scaling is excellent around the maximum of the growth rate , where the eigenfunction is more concentrated in the jet core . as we move towards smaller values of @xmath151 , the radial extension of the eigenfunction increases , the effect of the velocity shear becomes more important and the three curves deviates from one another . besides , from eq . ( [ eq : pc_scaling ] ) , we have that @xmath197 and the stabilization condition in the laboratory frame can be also written as @xmath161 . , as indicated in the legend . the values of @xmath61 are chosen so that in the jet rest frame the pitch on the axis is equal to 2 . the figure demonstrates the validity of the scaling of the growth rate with @xmath195 close the maximum . for lower values of the wavenumber the curves deviate from the above scaling.,width=340 ] given in the legends and for @xmath174 and @xmath181.,width=340 ] all the results presented so far have been obtained for @xmath198 . we recall that the value of the parameter @xmath45 determines the width of the current distribution and we can then ask how the results depend on it . the width of the current distribution is related to the extent of the flat part in the pitch profile ( see fig . [ fig : pitch ] ) : thus , if we decrease @xmath45 , the pitch grows to larger values inside the jet , the inner mode becomes stabilized at lower wavenumbers and the outer mode is also found at lower wavenumbers . the position of the stable region between the inner and outer branches is therefore a function of the value of @xmath45 and moves towards smaller wavenumbers ( when @xmath45 is decreased ) and to larger wavenumbers ( when @xmath45 is increased ) . in addition , the growth rate of the outer branch , being found at small values of @xmath151 , decreases with @xmath45 , while the growth rate of the inner branch is only determined by @xmath61 and does not depend on @xmath45 . we then expect that , by decreasing @xmath45 , the inner mode will become dominant . as an example , we show in fig . [ fig : a03 ] how the results obtained for @xmath199 confirm these expectations . the figure refers to @xmath181 and can be directly compared to the middle panel of fig . [ fig : cd ] . we can now summarize the results obtained for the cdi in the relativistic regime : we have two branches of instability , an inner mode concentrated inside the jet and an outer mode in which the perturbation is concentrated outside the jet . the relative importance of the two branches depends both on the current concentration determined by the parameter @xmath45 and on the lorentz factor @xmath73 . in fact , as we decrease @xmath45 , the growth rate of the outer branch also decreases and the mode shifts towards smaller wavenumbers , while the growth rate of the inner branch does not change . conversely , as we increase @xmath73 , the growth rate of the inner branch decreases while the growth rate of the outer branch does not change . we have examined the stability properties of a relativistic magnetized cylindrical flow in the approximation of zero thermal pressure , neglecting also the effects of rotation and focusing only on the @xmath0 mode . in this configuration we have two kinds of instability that may be present : kelvin - helmholtz and current driven . the instability behavior depends of course on the chosen equilibrium configuration and this is somewhat arbitrary since we have no direct information on the magnetic field structure , although some indications are provided by the acceleration models ( see e.g. * ? ? ? nevertheless , the general outcome and the properties of the solutions obtained for a particular configuration , such as the one we adopted here , can be considered valid for a more general class of equilibria . our results can then be considered representative of an equilibrium configuration characterized by a distribution of current concentrated in the jet , with the return current assumed to be mainly found at very large distances . we can summarize our results by considering the behavior of the system for different values of the ratio between matter and magnetic energy densities . for matter dominated flows , the dominant instability is kh and the wavenumber corresponding to the maximum growth rate as well as the growth rate itself scale both as @xmath170 . somewhat above equipartion khi reaches its maximum growth rate and then it becomes rapidly stabilized . below this stabilization limit the dominant instability becomes cdi the dependence of the khi on the value of the pitch is relatively weak and only for the smallest value of the axial pitch ( @xmath61 ) , in the relativistic case , we observe a displacement of the stability limit towards lower values of @xmath108 and a merging with cdi . cdi are therefore prevailing for flows in equipartition or magnetically dominated and the dependence on @xmath108 in these regimes is quite weak . the wavenumber corresponding to the maximum growth rate scale as @xmath117 , the growth rate itself increases with decreasing @xmath61 . and the modes have a resonant character with a peak in the eigenfunction at the radial position where @xmath188 . at low jet velocity , our equilibrium has no return current inside the domain , while at relativistic velocities we have a small portion of the return current corresponding to the velocity shear region . the corresponding steepening of the pitch profiles induces a stabilization of the modes for which the resonance condition corresponds to radial positions where the return current is found . we then observe a splitting of the cdi in two branches , one at high wavenumbers ( the inner mode ) characterized by an eigenfunction with a resonant peak in the inner radial part of the flow , and one at smaller wavenumbers ( the outer mode ) for which the resonant peak is outside the jet region . which of the two branches is dominant depends on the width of the current distribution and on the lorentz factor of the flow . an increase in the current concentration ( small @xmath45 ) favors the growth of the inner mode , while an increase in @xmath73 enhances the development of the outer mode . the different behavior in the explored parameter ranges may have crucial implications for the nonlinear stages as distinct types of instability may evolve differently . this study is therefore an essential first step for the interpretation of the results of numerical simulations that will be presented in a following paper and for their comparison with astrophysical data . s. , 1996 , , 314 , 995 s. , camenzind m. , 1992 , , 256 , 354 s. , lery t. , baty h. , 2000 , , 355 , 818 h. , keppens r. , 2002 , , 580 , 800 m. c. , 1998 , , 493 , 291 m. , 1991 , the stability of jets . pp 278+ g. , rosner r. , ferrari a. , knobloch e. , 1989 , , 341 , 631 g. , rosner r. , ferrari a. , knobloch e. , 1996 , , 470 , 797 a. , urpin v. , 2008 , , 488 , 1 a. , urpin v. , 2011 , , 84 , 056310 a. , ghisellini g. , 2008 , , 385 , 283 a. , trussoni e. , zaninetti l. , 1978 , , 64 , 43 p. e. , 1979 , , 234 , 47 p. e. , 2006 , in p. a. hughes & j. n. bregman ed . , relativistic jets : the common physics of agn , microquasars , and gamma - ray bursts vol . 856 of american institute of physics conference series , agn jets : a review of stability and structure . pp 5777 p. e. , cooper m. a. , norman m. l. , stone j. m. , 1992 , , 399 , 478 y. n. , pariev v. i. , 1994 , , 267 , 629 y. n. , pariev v. i. , 1996 , , 281 , 1 s. s. , barkov m. v. , vlahakis n. , knigl a. , 2007 , , 380 , 51 y. e. , 1999 , , 308 , 1006 j. c. , blandford r. d. , 2009 , , 394 , l126 a. , rossi p. , bodo g. , ferrari a. , massaglia s. , 2010 , , 402 , 7 y. , hardee p. , nishikawa k .- i . , 2007 , , 662 , 835 r. , spruit h. c. , obergaulinger m. , 2008 , , 492 , 621 m. , meier d. l. , 2004 , , 617 , 123 r. , li j. , tchekhovskoy a. , 2009 , , 697 , 1681 z. , mignone a. , massaglia s. , bodo g. , ferrari a. , 2008 , , 490 , 493 m. , hanasz m. , mart j. m. , sol h. , 2004 , , 427 , 415 m. , mart j. m. , cela j. m. , hanasz m. , de la cruz r. , rubio f. , 2010 , , 519 , a41 + m. , begelman m. c. , madejski g. m. , lasota j .- , 2005 , , 625 , 72 a. , matsuoka t. , takahashi m. , 2001 , , 64 , 123003 v. , 2002 , , 385 , 14 we start from the linearized system ( [ eq : lincont ] - [ eq : linmhd ] ) and , assuming the perturbations to be of the form @xmath81 , rewriting the vectorial equations in components , substituting the condition @xmath200 to the @xmath17 component of the induction equation , we obtain the following mixed system of @xmath201 differential and algebraic equations in the @xmath201 unknowns @xmath202 . , where we introduced the total electromagnetic pressure perturbation @xmath83 : @xmath203 @xmath204 @xmath205 @xmath206 @xmath207 @xmath208 @xmath209 @xmath210 @xmath211 @xmath212 @xmath213 where we defined @xmath214 furthermore , following @xcite , it is convenient to introduce the radial displacement of fluid elements , @xmath215 and use it instead of @xmath216 . we will now try to express all the variable in terms of @xmath82 and @xmath83 and substitute them in eqs . ( [ eq : app_lin_momr ] ) and ( [ eq : app_lin_divb1 ] obtaining a system of two first order differential equations in the two unkowns @xmath82 and @xmath83 . from equations ( [ eq : app_lin_indr ] ) , ( [ eq : app_lin_mhdphi ] ) and ( [ eq : app_lin_mhdz ] ) , we solve for @xmath217 , @xmath218 and @xmath219 in terms of @xmath82 @xmath220 @xmath221 @xmath222 for further use it is convenient to define the velocity components parallel and transversal to the background magnetic field @xmath223 @xmath224 where @xmath225 . substituting @xmath219 from equation ( [ eq : app_e1z ] ) and @xmath226 from ( [ eq : app_lin_mhdr ] ) into equation ( [ eq : app_lin_indphi ] ) and eliminating the radial derivative @xmath227 from equation ( [ eq : app_lin_divb1 ] ) , we get @xmath228 where @xmath229 a second relation comes from the definition of the electromagnetic pressure , @xmath83 , eq . ( [ eq : app_lin_p1 ] ) , if we substitute into it @xmath230 and @xmath226 from equation ( [ eq : app_lin_mhdr ] ) : @xmath231 now expressing @xmath232 and @xmath233 through @xmath234 and @xmath235 from equations ( [ eq : app_u1par ] ) and ( [ eq : app_u1perp ] ) , substituting into equations ( [ eq : app_lin_momphi ] ) and ( [ eq : app_lin_momz ] ) and then eliminating @xmath234 , we get the third relation : @xmath236 where @xmath237 @xmath238 @xmath239.\end{gathered}\ ] ] equations ( [ eq : app_alg1 ] ) , ( [ eq : app_alg2 ] ) and ( [ eq : app_alg3 ] ) form a system of three linear equations @xmath240 @xmath241 @xmath242 from which one can solve for @xmath243 and @xmath244 in terms of @xmath82 and @xmath83 . the solubility of this system depends on the determinant of the left hand side @xmath245 which after simplification reduces to @xmath246 . \label{eq : determinant}\ ] ] the possible singularities deriving from this determinant are discussed in the main text in section [ sec : linequations ] . at this point we are able to express all the variables in terms of @xmath247 ( in particular for expressing @xmath248 we make use of eq . ( [ eq : app_lin_cont1 ] ) ) , and substituting them into equations ( [ eq : app_lin_momr ] ) and ( [ eq : app_lin_divb1 ] ) after a long but straightforward algebra , we arrive at the system of two first order differential equations in the radial coordinate for the two basic variables the radial displacement and the perturbed electromagnetic pressure : @xmath249 @xmath250\xi_{1r}+ \\+ \frac{1}{r}\left(\rho_0\gamma_0 ^ 2v_{0\varphi}^2c_3 - 2d\right)\pi_1+\frac{2b_{0z}}{r}d b_{1z}+a_2d u_{1\bot}\end{gathered}\ ] ] where @xmath251 and the long expressions for @xmath252 and for @xmath253 through @xmath247 are given in the appendix [ ap : coefficients ] . @xmath254 @xmath255,\\\ ] ] @xmath256,\\ c_3&\equiv \sigma \left(\omega^2-\frac{m^2}{r^2}-k^2\right)+\frac{\tilde{\omega}^2b_0 ^ 2}{b_0 ^ 2-e_0 ^ 2}\end{aligned}\ ] ] and @xmath235 and @xmath244 expressed through @xmath82 and @xmath83 are @xmath257\xi_{1r } + \\ + \sigma \tilde{\omega}\left(kb_{0\varphi}-\frac{m}{r}b_{0z}+\omega\omega r b_{0z}\right)\pi_1,\end{gathered}\ ] ] @xmath258\xi_{1r}+ \\ + \left[\sigma ( \omega\bar{\omega}b_{0z}-kk_b)+\frac{\tilde{\omega}^2b_0 ^ 2b_{0z}}{b_0 ^ 2-e_0 ^ 2}\right]\pi_1,\end{gathered}\ ] ] where @xmath259 is the determinant ( [ eq : determinant ] ) . the second equation a25 , after expressing @xmath260 and @xmath261 through @xmath82 and @xmath83 takes the form @xmath262\xi_{1r}+ \\+ \left[\frac{1}{r}\left(\rho_0\gamma_0 ^ 2v_{0\varphi}^2c_3 - 2d+c_6\right)+c_7\right]\pi_1\end{gathered}\ ] ] , where the coefficients @xmath263 are given by @xmath264 @xmath265\ ] ] @xmath266 @xmath267 to find solution of equations ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) at small radii , we calculate the coefficients entering these equations at @xmath268 taking into account that in this limit the equilibrium quantities @xmath269 , while @xmath270 and @xmath271 tend to constant values . thus we have ( primes everywhere denote radial derivative ) @xmath272 @xmath273 @xmath274 @xmath275 @xmath276 \left(\sigma \overline{\omega}+\tilde{\omega}\right)\ ] ] @xmath277 @xmath278 @xmath279\pi_1.\end{gathered}\ ] ] substituting these coefficients into equations ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) , to leading order , we obtain @xmath280\xi_{1r}+\frac{m^2\sigma } { r^2d}\pi_1 , \label{eq : ap_small_r1}\ ] ] @xmath281\xi_{1r}. \label{eq : ap_small_r2}\end{gathered}\ ] ] we look for solutions in the form @xmath282 . after substitution of this form into equations ( [ eq : ap_small_r1 ] ) and ( [ eq : ap_small_r2 ] ) we get @xmath283 but because a solution must be regular at @xmath42 we take only @xmath284 , and after that the ratio @xmath285_{|r=0}\ ] ] this equation together with the choice @xmath286 serves as our boundary condition at small radii . to find the asymptotic limit of eqs . ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) and their corresponding solutions at large radii , we notice that the equilibrium azimuthal velocity , @xmath287 , the vertical velocity @xmath270 and @xmath28 decay very quickly ( exponentially ) with radius according to eqs . ( [ eq : omega ] ) , ( [ eq : vz_prof ] ) , ( [ eq : h2_prof ] ) and ( [ eq : vphi_prof ] ) ) , so we can put them effectively zero , @xmath288 at large radii and hence @xmath289 . at large @xmath290 , the equilibrium density and vertical magnetic field are constant , while the azimuthal magnetic field falls off as @xmath291 , as follows from eqs . ( [ eq : rho_prof ] ) , ( [ eq : h2_prof ] ) , ( [ eq : bz_prof ] ) and ( [ eq : bphi ] ) . taking this into account , the asymptotic form of each coefficient entering eqs . ( [ eq : dxi / dr ] ) and ( [ eq : dp / dr ] ) was calculated at large @xmath292 . then , neglecting everywhere terms of the order of @xmath293 and higher , after a rather lengthy algebra , we arrive at the following second order differential equation for the total electromagnetic pressure perturbation @xmath83 @xmath294\pi_1=0,\ ] ] which is of the bessel s equation type . its solution corresponding to radially propagating waves that vanish at infinity is the hankel function of the first kind @xmath295 , where @xmath296 with the leading term of the asymptotic expansion at @xmath297 @xmath298}. \label{eq : p1_asympt}\ ] ] the complex parameter @xmath299 can have either positive or negative sign , @xmath300 requiring that the perturbations decay at large radii , we choose the root of @xmath301 in eq . ( [ eq : p1_asympt ] ) that has a positive imaginary part , @xmath302 . these perturbations are produced within the jet and hence at large radii should have the character of radially outgoing waves . this implies that the real parts of @xmath299 and @xmath91 should have opposite signs , @xmath303 ( sommerfeld condition ) , in order to give the phase velocity directed outwards from the jet . the asymptotic behaviour of the displacement @xmath82 can be readily obtained from @xmath83 again correct to @xmath293 @xmath304 the asymptotic solutions ( [ eq : p1_asympt ] ) and ( [ eq : xi1_asympt ] ) , together with the above requirements of outgoing waves with decaying amplitides at @xmath292 , are used as an initial condition at outer jet boundary in our numerical scheme for finding eigenvalues of @xmath91 , when doing integration backwards , from large to small radii .

we present on the fly " maps of the co(1 - 0 ) and co(2 - 1 ) emission covering a 10 10 region of the ngc 6946 . using our co maps and archival vla hi observations we create a total gas surface density map , , for ngc 6946 . the predominantly molecular inner gas disk transitions smoothly into an atomic outer gas disk , with equivalent atomic and molecular gas surface densities at r = 3.5 ( 6 kpc ) . we estimate that the total mass is , roughly 1/3 of the interstellar hydrogen gas mass , and about 2% of the dynamical mass of the galaxy at our assumed distance of 6 mpc . the value of the co(2 - 1)/co(1 - 0 ) line ratio ranges from 0.35 to 2 ; 50% of the map is covered by very high ratio , , gas . the very high ratios are predominantly from interarm regions and appear to indicate the presence of wide - spread optically thin gas . star formation tracers are better correlated with the total neutral gas disk than with the molecular gas by itself implying . using the 100 and 21 cm continuum from ngc 6946 as star formation tracers , we arrive at a gas consumption timescale of 2.8 gyr , which is relatively uniform across the disk . the high star formation rate at the nucleus appears to be due to a large accumulation of molecular gas rather than a large increase in the star formation efficiency . the mid - plane gas pressure in the outer ( r 10 kpc ) hi arms of ngc 6946 is close to the value at the radial limit ( 10 kpc ) of our observed co disk . if the mid - plane gas pressure is a factor for the formation of molecular clouds , these outer hi gas arms should contain molecular gas which we do not see because they are beyond our detection limit .

0707.2636

we have observed ngc 6946 as part of a program of deep mapping of extended , cold co in nearby spiral galaxies @xcite . large , fully - sampled , and deep images of co in nearby galaxies are relatively uncommon . interferometric images either do not contain zero spacing data ( e.g. , * ? ? ? * ) or if they do , the images typically do not go deep enough to detect cold , molecular disk gas . the detection of cold , extended molecular gas at large galactocentric radii in large spirals is a goal of this mapping project with the nrao 12 meter telescope . a gas - rich sc galaxy at a distance of 6 mpc @xcite , ngc 6946 is known for its bright and asymmetric optical spiral arms @xcite . the pronounced asymmetry may be caused by interactions with neighboring dwarf galaxies , ugc 11583 and l149 @xcite . ngc 6946 has a 9 optical diameter on the sky and an atomic hydrogen ( hi ) gas disk extending to 25 @xcite . like many spirals , ngc 6946 has a central @xmath5 diameter hole " in its hi disk . radio continuum , far infrared ( fir ) , optical line , and x - ray observations indicate vigorous star formation in the disk and an interstellar medium ( ism ) stirred by supernovae and stellar winds @xcite . the high level of star formation in the disk of ngc 6946 has been attributed both to its strong spiral density wave @xcite , and to stochastic , self - propagating , star formation @xcite . until recently , the co morphology beyond the inner 3 radius ( @xmath65 kpc ) was not well known . like many spiral galaxies , ngc 6946 has a bright nuclear co peak which falls off nearly exponentially with galactocentric radius @xcite . a 3 diameter map of @xcite revealed substantial co emission in the inner disk with several emission peaks . @xcite mapped two 2 circular fields in the disk of ngc 6946 in a study of arm and interarm regions . the inner 2 region has been mapped with higher resolution single dish and aperture synthesis telescopes in transitions of co and co isotopomers @xcite . a larger interferometer+single dish map of ngc 6946 has been made by the bima song team @xcite , although with lower sensitivity to cold extended emission than the maps we present here . @xcite presented 21 resolution fully - sampled single dish maps of co(1 - 0 ) . we compare our results to theirs . the on the fly " ( otf ) observing mode at the nrao was ideally suited to the imaging of extended gas in galaxies . here , we present 16 by 10 maps of co(1 - 0 ) and co(2 - 1 ) covering the optical disk of ngc 6946 made with the 12 meter telescope . the maps are deep enough to detect cold , extended interarm co to levels of @xmath7 , or @xmath8 . we use these primary tracers of the molecular gas phase and combine them with archival and published data in order to study the molecular gas disk , the total neutral gas surface density , and their relationship to star formation in the disk of ngc 6946 . the observations of co(1 - 0 ) at 115 ghz and co(2 - 1 ) at 230 ghz were made at the nrao 12 meter telescope at kitt peak , on two separate observing runs in december 1996 and march 1997 . an equivalent of @xmath614 hours on - source observing was accumulated for each co line . calibration was by the chopper wheel method @xcite . we convert the recorded values @xcite to main beam temperature , , = @xmath9 at 115 ghz and = @xmath10 at 230 ghz @xcite . we report @xmath11 throughout this paper . the filter bank spectrometer was configured for 2 mhz channel widths with 256 total channels , producing spectral channels of a 5.2 width at 115 ghz and 2.6 at 230 ghz . the on the fly " ( otf ) observing mode was used @xcite . scanning rates between 30 and 40 /sec and scan row spacings of 18 at 115 ghz and 8 at 230 ghz were selected to ensure sampling better than nyquist over the mapped 16 @xmath0 10 region . the beamsize ( fwhm ) at 115 ghz is 55 and 27 at 230 ghz . chopper wheel calibration and sky offs were measured every 2 scans . the rms noise level was reduced by averaging the 16 individual otf maps made at 115 ghz and 10 individual otf maps at 230 ghz . the otf data was gridded using the nrao aips package . a linear baseline was removed from each spectrum . at this stage of the data reduction the background in the channel maps is uneven from one scan row to the next . this spatial variation , due to sky brightness fluctuations between successive off - source sky measurements , appears as stripes " in the continuum level running in the scan direction . the maps were made large enough to give us emission free , 2.5 wide , regions at the east and west ends of the maps , verified by the examination of individual spectra from these regions . these regions were used to remove a linear baseline from each row of the channel maps in the scan direction , effectively giving us a uniform background in all the channel maps . the rms variation in the channel maps before and after de - striping " differed by 10 mk at 115 ghz and 30 mk at 230 ghz ; in both cases a 20% reduction in the channel map noise . the mean rms noise ( @xmath12 ) from fully reduced , line free channels is = 0.050 k and = 0.145 k for the co(1 - 0 ) and co(2 - 1 ) respectively . the channels were averaged to a 10.4 channel width ; the mean rms noise in the averaged channel maps is 0.035 k for co(1 - 0 ) and 0.085 k for co(2 - 1 ) . the error beam of the 12 m telescope , which has been measured at 345 ghz , is extrapolated to be 8 - 9 fwhm @xcite . the contributions from the error beam in the individual channel maps is estimated to be less than a few percent , which we neglect here ( see discussion in * ? ? ? * ) . in the 10.4 channel width cubes the rescaled uncertainty in due to sky brightness fluctuations is @xmath13 mk for co(1 - 0 ) and @xmath14 mk for co(2 - 1 ) . we are interested in bringing out as much of the extended low level co emission as possible in our integrated intensity maps . to do this we want to minimize the noise that would normally be added to an map when the channel maps are added together . to test the effects of the level of flux cut ( level in the individual channels below which the data is deemed to be noise and not added to the integrated map ) on the final integrated flux , we did the following test . we convolved the co cubes to twice the beam size and used the regions with emission @xmath15 to create mask cubes . the mask cubes allow us to define areas of emission using smoothed , hence more sensitive images , which is one way of discriminating between signal and noise . we then created maps from both the masked and unmasked cubes using flux cuts from 1.0 @xmath16 to 0 @xmath16 in 0.1 @xmath16 decrements . the total flux in the co(1 - 0 ) integrated intensity map made from the masked cube with a 0.5 @xmath16 flux cut agrees with flux from the unmasked cube with no flux cut , to within 1% . for the co(2 - 1 ) map , a more conservative 0.8 @xmath16 was used . we created maps of the per pixel uncertainty for our maps by combining the per channel sky brightness fluctuation and rms noise , counting the number of channels contributing to the integrated emission at each pixel , and then adding the uncertainty in using error analysis techniques for the sums and products of independent variables . we created maps of the fractional uncertainty in our integrated intensity maps by dividing the aforementioned uncertainty maps by the maps themselves . the uncertainty is dominated at the nucleus by the uncertainty in while at the outer edges of the maps the uncertainty is dominated by the rms noise . for the co(1 - 0 ) integrated intensity map the combined uncertainty is on the order of 7% at the nucleus , 12% along the prominent spiral arm pattern , and 25% for the outer disk emission . for the co(2 - 1 ) map the uncertainties are on the order of 11% at the nucleus , 20% along the spiral arm pattern , and 30% for the outer disk emission . to obtain @xmath17 column densities from we applied the standard conversion factor " , @xcite . is estimated to be accurate to a factor of two in the milky way disk ( * ? ? ? * and references therein ) , although it may overpredict in the central arcminute of large spirals @xcite . we assume that these uncertainties hold for ngc 6946 as well . we will adopt the standard conversion factor , @xmath18 of @xcite throughout the remainder of the paper . using this conversion factor , can be converted into ( no he or heavier elements ) surface densities using @xmath19 including a geometric correction for inclination of the galaxy ( 40 , table 1 ) to correct the surface densities to their face - on values . we obtained archival vla d array 21 cm observations of the hi line in ngc 6946 , which are described in @xcite . the beam size is 49 @xmath0 42 , with @xmath20 , and each channel has a velocity width of 10.3 . the integrated intensity map was constructed from signals greater than 1.7 ( 1.2 @xmath16 ) . absolute flux calibration error is on the order of the uncertainty in the flux of 3c 286 , @xmath21 . the single channel rms noise of 1.4 corresponds to an hi column density of @xmath22 . integrated intensities were converted to face - on hi surface densities ( no he or heavier elements ) using , @xmath23 from a comparison to the single dish observations of @xcite we estimate 40% of the hi flux of ngc 6946 is missing from the vla maps due to the lack of short baselines ( emission extended over 15 ) . if the missing flux exists as a screen across the disk of ngc 6946 it constitutes a missing uniform hi surface density of @xmath24 . the total observed hi emission in the vla maps amounts to @xmath25 , not including missing under - sampled emission . @xcite , using the same vla hi data , obtained @xmath26 , rescaled to d = 6 mpc . their mass is probably slightly lower than ours because they used a higher flux cutoff to produce their integrated intensity map . @xcite obtained @xmath27 ( rescaled to 6 mpc ) . mass includes single dish observations merged with the westerbork interferometer data to cover the short spacings and corrections for milky way absorption and emission . the difference between our mass and theirs , 33% , is in line with our estimate of the missing flux in the vla map . a brandt rotation curve was fit to the hi radial velocities , for radii @xmath28 . our resulting kinematic parameters , table 1 , are in good agreement with the rotation parameters found by @xcite . a total dynamical mass estimate based on the fit ( see table 1 ) is : @xmath29 . we produced a 21 cm continuum map by averaging 4 line free " channels in the ngc 6946 hi channel cube . the @xmath12 noise level is 1.2 . high resolution ( hires ) 60 and 100 iras maps were obtained from ipac . the 100 hires map was produced by 200 iterations of a maximum entropy method yielding a 69 @xmath0 65 , @xmath30 beam and a @xmath12 noise level of 2.0 . the 60 hires map was produced by 50 iterations of a maximum entropy method yielding a 45 @xmath0 41 , @xmath31 beam and a @xmath12 noise level of 0.6 . a relatively heavy molecule , co is easily excited at the temperatures of molecular clouds , with its first rotational level at an energy of e / k = 5.5 k. since it has a small dipole moment and is generally optically thick , the excitation of the co rotational levels is driven by collisions at densities above @xmath32 for the j=1 - 0 transition . the j = 2 - 1 transition , with an upper level energy of e / k = 16.6 k , has a higher critical density , although its higher optical depth somewhat compensates for that . for gas kinetic temperatures of @xmath67 - 10 k , we expect co(2 - 1 ) to be thermalized at critical densities above @xmath33 , slightly higher than co(1 - 0 ) values . one might therefore expect co(2 - 1 ) emission to be more tightly confined to the spiral arms than co(1 - 0 ) , since the arms probably have denser gas , and are also closer to hii regions that might warm the clouds ( e.g. , * ? ? ? * ) . however , as we shall see , the co(2 - 1 ) and co(1 - 0 ) properties in ngc 6946 are not precisely what one might expect based on galactic clouds . we do not present co kinematic data ( radial velocity maps , velocity dispersion maps , rotation curves ) . our kinematic results do not differ significantly from those of @xcite , and we refer the reader to their presentation of the kinematics of ngc 6946 derived from iram single dish data . the 24 co(1 - 0 ) channels containing line emission are displayed in figure [ co10.chnl ] . the channels have been smoothed to a @xmath34 channel width . the channel maps show the butterfly pattern characteristic of emission from an inclined rotating disk . high velocity nuclear gas shows up as emission at the location of the nucleus spread across nearly the entire bandwidth of the line emission . the disk of co emission in ngc 6946 extends 8 ( 14 kpc ) north - south by 10 ( 18 kpc ) east - west as shown in the integrated intensity , ( figure [ co10.m0tb]a ) , and the peak main beam brightness temperature , , ( figure [ co10.m0tb]b ) maps . @xcite also found co emission extended out to 10 in 22 resolution single dish iram observations . there are no significant differences between our and walsh s map that can not be attributed to the difference in resolution . the co(1 - 0 ) emission covers the entire optical disk of a digital sky survey ( dss ) image of ngc 6946 ( figure [ co10.m0tb]c ) . the total co(1 - 0 ) flux density measured over a velocity width of @xmath35 is @xmath36 , in agreement with the value published in the fcrao survey of @xmath37 @xcite , but higher than the @xmath38 detected in a 22 resolution single dish iram map @xcite . converting our total observed into a total molecular mass , we obtain @xmath39 for ngc 6946 ; the mass will be higher if there is outer disk co that we do not detect . this is consistent with the mass obtained by @xcite , @xmath40 ( rescaled to 6 mpc and our @xmath41 ) . the co integrated intensity ( ) and temperature maps ( ) both show the same features : a bright nucleus , a broad asymmetrical spiral arm pattern and the overall asymmetrical distribution of the disk emission . the nucleus has a co emission peak at ( j2000 ) , with @xmath42 , @xmath43 , or in terms of surface density , @xmath44 ( face - on value ) , averaged over our 1.6 kpc diameter beam . the total molecular mass within the central beam is @xmath45 . @xcite obtained @xmath46 ( adjusted to 6 mpc and our conversion factor ) in a 65 diameter region from interferometric observations that recovered 70% of the total flux . large scale co spiral structure is traced in a broad s " pattern , outlined by the 7.5 contour in the map and more clearly seen in the map , outlined by the 0.23 k contours . here , molecular gas surface density is @xmath47 . there is also extended , weaker co emission from the disk outside of the 7.5 contour in the map . large regions of molecular gas exist to the northwest and southeast of the nucleus , as well as patchy emission out to the edges of the maps . not all of this gas is associated with the weaker optical arm pattern . the face - on gas surface density for this extended region has a mean of @xmath48 corresponding to @xmath49 or @xmath50 . this extended portion of the gas disk contributes 1/4 of the total co emission we observe . the total mass of this extended gas disk is @xmath51 . the 24 channels containing co(2 - 1 ) emission are displayed in figure [ co21.chnl ] . the rms of these 10.4 channels is three times as high as for co(1 - 0 ) , so the maps are less sensitive to faint emission . the total co(2 - 1 ) flux density measured over a velocity width of @xmath35 is @xmath52 . the butterfly patterns for the co(2 - 1 ) and co(1 - 0 ) are generally the same , with the differing resolutions , 27 and 55 respectively , accounting for most of the differences . as with co(1 - 0 ) the bright nucleus and asymmetric distribution of arm and disk emission are also apparent in the integrated intensity , , and peak main beam brightness temperature , , maps ( figures [ co21.m0tb]a and [ co21.m0tb]b ) . the co(2 - 1 ) emission in ngc 6946 is almost as extensive as that of the co(1 - 0 ) : the north - south and east - west dimensions are only 1 shorter . the observed co(2 - 1 ) emission covers the brighter portions of the dss optical image of ngc 6946 , figure [ co21.m0tb]c . the nuclear peak of the co(2 - 1 ) integrated intensity map is at ( j2000 ) with @xmath53 . using a galactic disk @xmath41 , this corresponds to a peak column density of @xmath54 or 450 averaged over the 790 pc beam . if other nearby galaxies are any indication , this may be an overestimate of the true molecular gas mass in the nucleus , by factors of @xmath55(@xmath563.1 . ) the largest value , 0.85 k , occurs at the same location as the peak and both peaks are consistent with the location of the peak . at the resolution presented of maps , inside a 2 ( 3.5 kpc ) radius both the co(2 - 1 ) and co(1 - 0 ) emission trace a broad 2 arm pattern . at a 2.5 radius ( 4.4 kpc ) the co(2 - 1 ) still traces a broad 2 arm pattern while the co(1 - 0 ) emission from a larger beam has clearly bifurcated into a 4 arm spiral ( figure [ coaz]a ) . at a 3.5 radius ( 6.1 kpc ) 5 arms are discernible in azimuthal emission plots of both co emission lines ( figure [ coaz]b ) . along the optical arm pattern the co(2 - 1 ) emission is uneven , ( figure [ co21.m0tb]c ) with typical variations of 15 and 5 separated by 40 or less . this same clumpy pattern of co arm emission can be seen in the 22 co(1 - 0 ) and co(3 - 2 ) maps presented by @xcite . the emission on each arm at the same galacto - centric radius is also uneven as can be seen in figures [ coaz]a and b. at 27 resolution it becomes possible to distinguish between arm and interarm emission . the contrast between the brighter , on arm , clumps and obvious off - arm regions is @xmath57 . using the dss optical image to mask the outer ( r@xmath4 2.5 ) arm / interarm regions , we obtain a mean arm / interarm contrast of 2.4 . this is higher than both the 1.2 and 1.8 values found from co(1 - 0 ) and co(3 - 2 ) iram observations of @xcite who interpret the increase in contrast at the higher co transitions verses co(1 - 0 ) as evidence that the on - arm molecular gas is warmer . the presence of widespread interarm gas becomes plainly apparent in a comparison to an optical image , figure [ co21.m0tb]c . the fact that interarm co(2 - 1 ) is present and at emission levels comparable to co(1 - 0 ) suggests that the interarm gas is also not very cold and that it must be warmer than @xmath58 k. this widespread distribution of interarm gas is not seen in the @xcite co(2 - 1 ) survey of the milky way . a co ratio map , @xmath59 , is presented in figure [ rco]a . the co(2 - 1 ) channel cube was convolved to the beam size of the co(1 - 0 ) data and the @xmath60 emission from the beam matched cubes was used to produce the map . the spiral pattern is only weakly reflected in the map . outside of the inner few arcminute radius the pattern breaks up into a patchwork of high and low regions . the ratio uncertainty for the regions of strong emission is dominated by the error in the conversion to main beam temperature , while the uncertainty for the weaker emission regions is dominated by the rms noise in the contributing channels of emission . adding in quadrature the fractional uncertainties from the maps we obtain an error estimate on the order of 15% at the nucleus , 25% along the spiral arm pattern , and 40% for the outer disk ratios . we were concerned that the ratio might be biased towards higher values due to the uncertainty in the denominator . we performed a standard statistical analysis to correct a ratio for bias this bias . the bias corrected map was not significantly different from the map presented in figure [ rco]a and well within our reported error . despite the large uncertainty estimates , our ratios are consistent with those derived from other independent observations @xcite . in principle , , the ratio of co(2 - 1 ) to co(1 - 0 ) integrated intensities , gives information on the excitation of the co emission and depending on circumstances , the kinetic temperature , . the observed main beam temperature at frequency , @xmath61 , is related to a planck function , , with excitation temperature , , by the areal beam filling factor , , a conversion to the rayleigh - jeans temperature scale , and radiative transfer through a optical depth , : @xmath62including a correction for the cosmic microwave background contribution to the beam , , at temperature 2.73 k. standard assumptions are that is the same for both emission lines , that both lines are optically thick ( @xmath63 ) and that the emission can be characterized by a single @xcite . with these assumptions we can naively derive values for the emission : = 0.9 corresponds to = 20 k , while = 0.5 corresponds to = 3.5 k. higher ratio values are found in the nuclear regions of starburst galaxies where elevated star formation heats the molecular gas ; lower values are typical of cold or subthermally excited ( @xmath64 ) disk molecular clouds . values of @xmath65 indicate optically thin gas in this picture . within 1 of the nucleus , @xmath66 . this is consistent with other studies @xcite that indicate the presence of relatively dense molecular gas as temperatures of about @xmath67 k. this is comparable to the milky way mean of 0.74 @xcite . along the optical arms the mean is 0.83 , ranging from 0.44 to 1.4 . to within our uncertainties , this is the same as the 0.73 mean found for milky way molecular arms where a similar range of values is found @xcite . from a naive interpretation of the ratios , a mean of 12 k is implied for the arm co in ngc 6946 . @xcite concluded that the interarm does not differ significantly from that of the arms based on observations of two circular regions in the arms , one on the east side of ngc 6946 , the other on the west side . the two circular regions they observed have a mean of 0.8 in our ratio map , also not significantly different from that of the arm regions . however when larger regions of the co disk are sampled , a very different picture emerges . there are regions ( 5% of the ratio map ) where @xmath68 indicating cold ( @xmath69 5 k ) or sub - thermally excited co. there are also much larger regions where @xmath65 . very high ratio , @xmath65 , gas covers 50% of our ratio map ; and these are predominantly interarm regions , figure [ rco]b . we have eliminated the possibility that these very high ratios are due to contributions from the telescope error beam at 230 ghz : the error beam at the 12 meter telescope is a measured quantity and modeling of the error beam contribution in the individual channel maps shows that it contributes at most a few percent of the emission in the map . the high ratios appear to indicate the presence of widespread , warm , optically thin co - emitting gas in the interarm regions . while optically thin co emission is not typical of galactic molecular clouds , our large beam and unusual perspective from outside the disk of ngc 6946 may allow us to trace a component that is difficult to detect in the galaxy . we note that the photon - dominated regions ( pdr ) around cool stars can produce bright co emission with high temperatures @xcite since their softer radiation fields allow the co to be photoelectrically - warmed at low a@xmath70 . also , as pointed out by @xcite it does not take much optically thin gas to swing the emissivity ; the co(2 - 1 ) emission that we see could easily come from a small amount of optically thin , but co - loud " gas that is not the dominant cloud population . other less likely possibilities are discussed in @xcite . we have also seen a have a high overall @xmath71 in m83 @xcite ; confirming earlier reports of unusually high in these two galaxies @xcite . large scale maps of the milky way @xcite show high ratios in immediate proximity to pockets of star formation . the widespread regions of high seen in m83 and ngc 6946 are not seen in surveys of co in our galaxy . although co emission covers only 15% of the hi disk surface area , ngc 6946 is nevertheless a molecular gas - rich galaxy ; 1/3 of the hydrogen gas disk is in molecular form . this is consistent with the 36% value from @xcite once the flux missed in the interferometer hi maps is taken into account . in figure [ hico]a , we present our co(1 - 0 ) emission in contours superimposed on a grey scale image of the hi emission . this image shows that co emission fills the inner 6 of the hi disk . the hi disk has a central depression , which rises to peaks in a clumpy ring of emission . the hi ring is coincident with the outer boundary of co(1 - 0 ) . while spiral structure is readily apparent in the outer hi disk @xcite , it is insignificant in the inner disk , r@xmath72 ( 9 kpc ) , where the hi column densities decline and @xmath73 takes over . at 1 resolution there are no obvious systematic displacements of the hi relative to the co(1 - 0 ) spiral structure other than the strongest hi tends to be near the outermost ends of the co(1 - 0 ) arms . in fact , if anything , hi and co peaks tend to coincide , which is more apparent in the higher resolution co(2 - 1 ) maps . the co(2 - 1 ) image in contours is shown in figure [ hico]b , again with a grey scale image of hi . outside the inner 2 ( 3.5 kpc ) region of the hi depression , the peaks of co(2 - 1 ) emission tend to coincide with hi peaks along the ring . this could be evidence for photodissociated creating a higher hi column density at these locations . @xcite and @xcite suggest the same explanation for hi in the spiral arms of m51 , m83 and m33 . from these images , we have constructed plots of the azimuthally - averaged gas surface densities , , and , as a function of galactocentric radius in figure [ sdg_r ] ( corrected for inclination ) . this is similar to the plot of @xcite , which was based on a radial cross map , and those of @xcite and @xcite . dominates the neutral gas distribution in within the central 3 ( 5 kpc ) of ngc 6946 , rapidly fading into a predominantly hi disk beyond . a least squares fit of an exponential to over the entire range of radii has a scale length of @xmath74 ( 1.8 kpc ) . @xcite and @xcite obtained scale lengths of 1.5and 1.4 for the co disk from fcrao and iram single disk observations ; @xcite obtained 1.3 from the bima interferometer survey data combined with nrao 12 m observations . all these latter scale lengths exclude the nuclear contribution to the fit and are sensitive to the range of radii over which they computed . we obtain @xmath75 and @xmath76 fitting to the range of radii used by @xcite and @xcite , 40 - 280 and 60 - 260 respectively . we conclude that the co falls off with an exponential scale length of @xmath77 on arcminute scales , excluding the inner 1 nuclear region and about 1 if the nucleus is included . a total neutral gas surface density map , @xmath78 is presented in figure [ sdg]a . this map has been corrected for inclination but not for missing due to under - sampling in the vla image ( @xmath61 ) , a false color image is presented in figure [ sdg]b , with hi in red and co in green , so that the molecular contribution to the total gas disk can be distinguished . values for range from 240 at the nucleus , to 20 - 45 along the spiral arms within the optical part of the galaxy , to 5 - 10 for the outer arms . at first glance , and hi blend into a single global gas structure extending from the nucleus to the outer gas disk . a closer look reveals some subtleties . in the southern part of ngc 6946 , the neutral density gradient is smooth and gradual . in the north , the gradient is much steeper , with the gas density falling to a level of @xmath79 in 90 ( 2.6 kpc ) to the north , into an interarm void , as compared to a 180fall to the same mass density in the south ( this can also be seen in figures [ hico ] ) . this steeper falloff on the northern side of the central gas peak may be related to the more well - defined spiral arm to the north , and perhaps also to the prominent northeastern optical arm @xcite . the global properties of three late galaxies for which we have a similar data set , along with the milky way for reference are listed in table 2 . ic 342 , m83 and ngc 6946 all have gas disks that extend well beyond their optical diameters and all three show some evidence for a bar . the @xmath80 values for ic 342 , ngc 6946 and m83 are all significantly higher than that of the milky way ; a factor of four higher for ic 342 and ngc 6946 , a factor of seven higher for m83 . the ngc 6946 @xmath80 value is similar to that of ic 342 , while its @xmath81 intermediate between ic 342 and m83 . all four galaxies have similar relative proportions of and hi , global / values between 0.3 and 0.5 are also seen in the nearby galaxy , maffei 2 @xcite . m83 stands out as a particularly gas - rich object relative to its dynamical mass , with a factor of two higher m@xmath82/m@xmath83 than ic 342 and ngc 6946 , and four times that of the milky way . the and optical disk diameters of the three external galaxies are the same , d@xmath84/d@xmath85 . the dynamical masses of these three galaxies are an order of magnitude smaller than that of the milky way , although their total gas mass is similar . it is difficult to place too much emphasis on subtle variations that rely on distance estimates , since the uncertain distances to these nearby galaxies can easily affect their masses , both gaseous ( @xmath86 ) and dynamical ( @xmath87 ) . because stars form from molecular gas , we expect to see correlations between co and tracers of star formation . these tracers include : emission from ionized gas associated with massive young stars , fir emission from dust heated by young stars , and radio continuum emission from relativistic electrons created by supernovae . to calculate linear correlation coefficients all the maps were convolved to the largest beam size in our collection of maps , a 70 circular beam size , _ then clipped to use @xmath60 emission _ , and sampled on a 35 grid . emission @xcite and are compared in figure [ cosf]a . the nuclear bar traced by within the central 2is aligned with the elongation of the central contours . the arms are traced at higher resolution by the emission . in the outer disk , the brighter patches of co(2 - 1 ) emission are aligned with the brightest patches of which is most clearly seen along the northeastern spiral arms . this supports the conjecture that the ring of hi emission seen at these radii is the result of photodissociated . it is apparent from this map that ratio of emission to is not the same in the nucleus and outer arms ; due to substantial extinction particularly in the nucleus , is probably not the optimal tracer of star formation for this gas - rich galaxy . when is compared to or our primary molecular mass tracer , , we obtain a linear correlation coefficient of @xmath88 . the correlations do not improve when a 1 diameter region centered on the nucleus is excluded from the correlation . figure [ cosf]b shows that the 60 fir emission generally reproduces the morphology seen in ( grey scale ) , despite the noticeable artifacts ( the boxy " structure of the central contours ) produced by the hires maximum entropy algorithm . an isophot 60 map by @xcite confirms the overall morphological structure seen in the hires map . the 60 nuclear peak is at ( j2000 ) which is , within the uncertainties , coincident with the co peak . if co(2 - 1 ) along with 60 fir trace a warmer component both heated by star formation ; we expect these emission pairs to have similar feature in the disk . the 60 emission traces the overall co(2 - 1 ) arm structure better than the 100 emission shown in figure [ cosf]c . the 4 prominent peaks in the outer 60 emission disk are aligned with peaks in the outer arm structure . the 100 fir and 21 cm continuum maps have resolution comparable to our co map ( figures [ cosf]c and d ) . the 100 emission roughly mimics the co(1 - 0 ) emission disk , but the co arm structure east of the nucleus is not well traced by the 100 . the 100 nuclear peak is at ( j2000 ) ; northeast of the co emission peaks . the radio continuum does a better job reproducing the co morphology , with the exception of the 21 cm peak 4 northwest of the nucleus . this is consistent with the suggestion that cosmic rays may be more important for the heating of molecular gas on large scales such as observed here than the radiation from in situ massive star formation @xcite . the continuum peak to the north and west of the nucleus at is probably a background radio galaxy ( * ? ? ? * based on its spectral index , @xmath89 and flux . ) the 21 cm continuum nuclear peak at is consistent with the co(1 - 0 ) peak . much of the outer co arm morphology and prominent emission peaks are traced by non - thermal radio continuum . @xcite also found statistically significant correlations between co(1 - 0 ) emission and the total radio - continuum at 6 cm , and between co(1 - 0 ) and 20 cm non - thermal continuum in ngc 6946 . when fluxes are extrapolated to the expected 20 cm values ( * ? ? ? * using @xmath90 ) , we obtain a mean /t@xmath91 of @xmath92 from our 70 beam convolved maps for the disk of ngc 6946 . this is higher that the /t@xmath91 = 1.0 value determined by @xcite from sparely sampled data , but in line with the 1.3 mean value for all the galaxies in their sample . one might naively think that the fir has a more direct causal link to newly formed stars , while radio continuum emission , which arises from relativistic electrons produced in supernova remnants , would be more spatially removed . in reality the far - infrared cirrus and non - thermal disk emission on arcminute scales represent a less temporally localized reflection of star formation ; the close correspondence of radio continuum and co supports a model of cosmic ray confinement on 1 kpc scale lengths . and highlights the importance of gas surface density in containing the magnetic field required to extract synchrotron emission from relativistic electrons @xcite . table 3 lists least squares fits to the radial distribution of ratios of or to our star formation tracers , . with the exception of , all the star formation tracers are highly correlated with co emission . from the 70 beam maps we obtain linear correlation coefficients on the order of @xmath93 or better in comparisons between either of our co maps and any of the star formation tracers except for . the radial gradient is reduced by a factor of two when is used instead of indicating remains the best tracer of molecular gas associated with star formation on kpc size scales . while co is highly correlated with star formation tracers , the correlation of total gas , hi@xmath94co , is even better ( see table 3 ) . the radial gradient is reduced by a factor of ten when is used . if we assume that the 21 cm and fir emission trace star formation , this result implies that the star formation rate in ngc 6946 is more closely tied to than . restating this result : a simple schmidt law applies , sfe @xmath95 @xcite , at least on kpc size - scales . _ moreover , it appears that in the outer parts of ngc 6946 , star formation is likely to take place in regions of predominantly hi gas_. we obtained similar results for ic 342 and m83 @xcite . while the correlation coefficient for / in m83 is also high , 0.95 , for ic 342 the correlation coefficient is only 0.73 . the lower map correlation for ic 342 is related to the peculiar 1 armed appearance of the 21 cm continuum map that contrasts starkly with its grand design appearance in other tracers . from least squares fits to the / ratio we obtain slopes of @xmath96 ( dex ) for ic 342 and @xmath97 ( dex ) for m83 . for m83 , the slope is not as flat as in ic 342 and ngc 6946 . this gradient in m83 may indicate an age difference with larger / values resulting from younger star forming regions in its disk @xcite . the / comparison for ngc 6946 suffers none of these peculiarities . a complication in the interpretation of molecular gas mass could arise because of a radial gradient in metallicity and because the galactic co conversion factor over - predicts the mass in the nucleus of ngc 6946 @xcite . based on the metallicity gradient for ngc 6946 measured by @xcite and the metallicity adjustment to the co conversion factor used by @xcite for giant molecular clouds in m33 , could be a factor of two higher at the edge of the ngc 6946 co disk . this would increase the molecular surface density derived from co at the disk edge by a factor of two , and potentially reduce the radial gradient of molecular mass implied by with radius . however , this variation in does not appear to be universal in low metallicity systems @xcite , so we do not adopt the correction here . studies of the links between star formation and gas , or star formation efficiency , have recently focused on correlations as a function of position in the galaxy , as opposed to global measures . @xcite examined the relation between and the star formation rate , sfr , as traced by ; @xcite compare a radio continuum based sfr and ; @xcite compare a sfr from emission to . @xcite pointed out that the falloff of in ngc 6946 with radius was very similar to the falloff in b luminosity , implying a constant star formation rate per unit h@xmath98 . the most widely used formulation is the schmidt law , @xmath99 @xcite ; and the most widely accepted calibration was done by @xcite who finds @xmath100 and @xmath101 for in and in . how well does the schmidt law describe star formation in ngc 6946 ? we can estimate star formation rates from our fir and 21 cm continuum data . unlike emission , these tracers are relatively unaffected by absorption . although cirrus heated by an older population of stars may contaminate fir in some galaxies , this is probably not a big effect in scd type spirals @xcite . the 21 cm continuum appears to be a good star formation tracer on global scales , although it must break down at smaller scales . the maps for this analysis were convolved to the the largest circular beam in the set , 70 . we built a fir map from the 60 and 100 iras maps using : @xmath102 @xcite . we adopt the fir star formation rate : @xmath103 @xcite , where @xmath104 and @xmath105 is the distance to ngc 6946 . we derive the map from the @xmath106 map using : @xmath107 where @xmath108 gives us face - on surface values . for the non - thermal continuum at 21 cm we use @xmath109 to get the star formation rate @xcite and use : @xmath110 to build the map . the maps were azimuthally averaged in 30 radius bins . the results are shown in figure [ sfr]a . the sfrs derived from the fir and radio continuum differ by a factor of two . @xcite finds that the non - thermal continuum from low luminosity galaxies is substantially suppressed , and so calibrations of the sfr from radio continuum from surveys that do not take into account the overall luminosities of the constituent galaxies tend to underestimate the sfr . bell provides an alternate calibration : @xmath111 where the total galactic flux satisfies the condition , @xmath112 , @xmath113 . in the case of ngc 6946 , @xmath114 this calibration of is also shown in figure [ sfr]a and agrees with the fir derived rate . although the agreement with the result is clearly promising , this alternate calibration was established using total galaxy fluxes and has not been studied with respect to the varying non - thermal continuum fluxes within galaxies . figure [ sfr]a also shows the sfr predicted by the @xcite calibration of the schmidt law using our map ; in ngc 6946 , it overestimates the sfr relative the and values . another way to look at the schmidt law is in terms of the star formation efficiency , sfe . in figure [ sfr]b we express the sfe for each of our sfrs as , the time it would take to consume all the gas at the given star formation rate , @xmath115 . using the and bell s calibration for , the mean for ngc 6946 is @xmath116 gyr and @xmath3 . this is consistent with the mean of 2.6 gyr for undisturbed galaxies found by @xcite . the @xmath117 schmidt law is also shown on figure [ sfr]b ; the schmidt law over - predicts the sfe and its dependence on . @xcite found @xmath118 values as low as 1.1 for the galaxies in their sample . * see references within ) has noted that the schmidt law is inconsistent with observations that the sfe is roughly constant across range of galactic environments . this certainly is the case for ngc 6946 . on kpc scales , the sfe for the nucleus of ngc 6946 does not differ from that of the disk ; in other words , the strong central fir emission from the nucleus ngc 6946 results from having more molecular material from which to form stars rather than any increase in the star formation efficiency . this can be seen in figure [ rsf ] where we show the radial distribution of the ratio of to 21 cm continuum ; the ratio does not decline at the nucleus . the lowest values ( highest efficiencies ) are found between 2and 4 radii ( 3.5 to 7 kpc ) . the highest values ( lowest efficiencies ) are at the largest radii . in two other late type galaxies for which we have a similar data set , ic 342 and m83 , there is a noticeable drop in log(/ ) from the peak values in the disk to the nucleus ; an indication of a elevated star formation efficiency relative to the disk rates @xcite . using the definition of @xcite , ngc 6946 would not be a starburst galaxy because it does not have a gas cycling time shorter than @xmath119 years . other authors have come to a similar conclusion about the star formation efficiency in ngc 6946 : @xcite based on a determination of the nuclear molecular mass from high resolution observations of co isotopes and multiple line transitions , and @xcite by evaluating the sfr using cii maps . does the outer edge of our co(1 - 0 ) map represent a real threshold for the formation of molecular clouds ? clearly in ngc 6946 , there is not a steep or marked decline in the gas as we detected in m83 ; that galaxy has a real gas edge , in both co and hi , possibly due to tidal interaction @xcite . in ngc 6946 the exponential disk continues as far as we can detect it ( figure [ sdg_r ] ) . but there may be other subtle changes in the gas with radius and we examine these below . we use our galaxy as guide on what to expect for the radial falloff in gas and star formation in a spiral galaxy . the number of hii regions per kpc@xmath120 , , and at @xmath121 and 12 kpc ( rescaled to @xmath121 = 8 kpc ) for the milky way from @xcite are presented in table 4 . in the galaxy , between 8 and 12 kpc @xmath122/kpc@xmath123 has fallen off by a factor of 15 , by a factor of 20 , while stays relatively flat . star formation per unit remains relatively constant over this range of radii . @xcite found evidence for molecular outflows at distances as large as 16 kpc . @xcite detected galactic molecular gas in a search for co absorption lines against continuum sources ; their findings suggest cold molecular gas may be far more abundant at large radii than previously suspected ; they suggest the molecular gas mass may be 4 times the hi mass at 12 kpc . @xcite detect a sparse distribution of milky way molecular clouds at 13 kpc with a mean @xmath124 k and a mean radius of 20 pc . the edge of the observed co disk in ngc 6946 lies midway between 8 and 12 kpc for an assumed distance of 6 mpc . comparing ngc 6946 to the milky way ( table 4 ) : at 8 kpc ( 4.6 ) @xmath122/kpc@xmath123 is a factor of four less in ngc 6946 than the galactic value at @xmath121 , and could be due to extinction in . is the same for both galaxies ( within the errors ) , and is at least twice the galactic value . at 12 kpc , @xmath122/kpc@xmath123 is the same as galactic ( extrapolating 20 in radius past the radial limit of the @xcite data set ) , is no longer detected in ngc 6946 by our observations , and is at least as large as the value for the milky way . we conclude that given the similarity of @xmath122/kpc@xmath123 and between these two galaxies , we would expect that in ngc 6946 should be also be @xmath125 at 12 kpc . complicating these comparisons is the uncertainty in the ngc 6946 distance . to date , there are no cepheid - based distance estimates to ngc 6946 , which has a low galactic latitude . recent estimates are ( in mpc ) : @xmath126 based on type ii sn @xcite , @xmath127 based on the brightest stars @xcite , and @xmath128 based on blue supergiants @xcite . as with any comparison based on absolute scales , our comparison changes if the the 6 mpc distance estimate is incorrect by more than @xmath129 mpc . the mapping of the co disk presented here is sensitivity - limited ; our 3@xmath16 detection limit is 2 - 3 . a combination of low and a low filling factor for molecular clouds in the beam combine to create a detection limit rather than a real threshold . there is also the possibility that declining metallicity in the outer disk lowers the co emissivity relative to , although this effect has yet to be seen . there are probably clouds here , but more widely dispersed than they are in the inner parts of the ngc 6946 . but our suspicion that there is substantial molecular gas at large galactic radii is not based entirely on a comparison to the milky way . there are theoretical reasons to expect molecular gas at radii beyond our detected co edge @xcite . according to @xcite molecular clouds are not formed where the ism pressure , , becomes too low to facilitate the formation of cold , dense clouds . or conversely , molecular clouds form in regions where @xmath130 . the outer hi disk of ngc 6946 has considerable surface density structure that exists on kpc in - plane size scales and where may locally exceed this minimum . to calculate the mid - plane gas pressure we use : @xmath131 where we have ignored the stellar contribution to the pressure which can only act to increase the value @xcite . along the outer ( r@xmath132 , 10 kpc ) hi gas arms , @xmath133 which is close to values at the edge of the co disk . the outer gas arms are where one would expect to find molecular gas , if we had deeper maps . several of the hii regions identified by @xcite are located beyond the edge of our co disk , particularly to the south , and it is likely that they are associated with molecular clouds that are too beam - diluted and cold to show up in our images . @xcite have detected emission associated with the formation of massive stars , up to 3 beyond our co disk edge and precisely at the local maxima in the outer hi arms of ngc 6946 . these star forming outer arms would be good places to look for co with a smaller beam and higher sensitivity . based on theoretical arguments and the detection of in ngc 6946 at large galactic radii , the answer is : no , the edge " of the co disk at a galactocentric radius of 5 ( @xmath6 9 - 10 kpc ) in our images is not a real edge , it is a detection limit . there is no significant change in properties to indicate that the disk is undergoing a change at this radius . there could easily be molecular clouds that escape detection beyond this radius we have obtained deep co(1 - 0 ) and co(2 - 1 ) images of a 10 @xmath0 10 region centered on the sc galaxy , ngc 6946 , with 55 and 27 resolution using the nrao 12 meter telescope . we have combined these deep co images with vla hi images to obtain images of the neutral gas in this galaxy . to summarize our findings : + 1 ) the co(1 - 0 ) and co(2 - 1 ) disks shows many of the same features seen in the optical disk : a strong nuclear peak of emission , molecular gas arms spatially coincident with the asymmetric optical arm pattern , and a 10 diameter inner disk filled with co emission . the co disk is roughly exponential with a scale length of @xmath77 or @xmath134 kpc . + 2 ) we obtain @xmath135 for the molecular hydrogen gas mass of ngc 6946 . this can be compared to the total atomic hydrogen gas mass of @xmath136 ( which includes a vla missing flux estimate ) . molecular hydrogen constitutes 1/3 of the interstellar hydrogen gas mass and 2% of the dynamical mass of ngc 6946 . at the nucleus , = 170 , the molecular gas surface density is @xmath47 along the strong optical arm pattern , and @xmath137 regions beyond . + 3 ) the mean value for @xmath138 in the the nucleus and optical arm pattern is @xmath139 . this is consistent with optically thick gas , with an excitation temperature of 10 - 15 k. interarm co shows the largest variation in , with a range of 0.35 to 2 . only 4% of the co ratio map has @xmath68 which is likely to be cold or subthermally excited co ( @xmath140 k ) . fully half of the co disk of ngc 6946 has @xmath65 . the high appears to indicate the presence of widespread , optically - thin gas in between the spiral arms . + 4 ) molecular gas dominates the total gas surface density within a galactocentric radius of 3.5 ( 6 kpc ) . co emission fills the central depression in the hi disk . beyond the nuclear region , there is a good correlation between the co and hi gas , which both peak on the spiral arms . the inner disk arm pattern transitions smoothly from predominantly co arms to predominantly hi arms at the outskirts of the co disk . at the nucleus , = 240 , the total gas surface density is @xmath141 along the optical arm pattern , and @xmath142 along the outer gas arms ( @xmath143 ) . co(2 - 1 ) emission peaks are coincident with clumps of hi , , and fir emission outside the nuclear region , presumably an indication of warmer molecular gas and dissociated . + 5 ) we find strong correlations between star formation tracers ( except for ) and co emission in ngc 6946 . the correlation coefficients are @xmath144 . when is used instead of just the molecular gas tracer in a radial comparison , the dependence on galactic radius is reduced . star formation is more closely related to the total gas surface density than the molecular gas surface density alone , a schmidt law relationship . this also implies star formation is likely well beyond the radial limits of our observed co disk in regions of predominantly hi gas . surprisingly , is not as well correlated with co or as are the other star formation tracers . the correlation coefficient is significantly less , @xmath88 , and the ratio formed from and co or declines more rapidly with galactic radius . + 6 ) the star formation efficiency in the disk of ngc 6946 is relatively uniform ; 2.8 gyr when expressed in terms of the gas consumption time - scale . this uniformity implies @xmath3 . the star formation rate at the nucleus is due to the high molecular surface density rather than an elevated star formation efficiency . + 7 ) the edge of the observed co disk represents a detection limit rather than a threshold for molecular cloud formation . we suspect the exponential fall of the co disk continues smoothly past our detection limit . the mid - plane gas pressure in the outer hi arm structure , @xmath133 is close to the value at the limits of the observed co disk . these gas arms should support the formation of molecular clouds which could be detectable in higher resolution observations . the authors would like to thank the nrao 12 m telescope operators , duane clark , paul hart , victor gasho and harry stahl , for their help and company during the many observing sessions in which the data for ngc 6946 and other galaxies were acquired . we thank l. tacconi giving for us permission to retrieve and use the hi 21 cm line observations from the vla archive . and we appreciate the careful reading and critical suggestions made by the anonymous referee . this work was supported in part by nsf grants ast00 - 71276 and ast03 - 07950 . we made use of the nasa / ipac / iras hires data reduction facilities as well as stsci digital sky survey facilities . the digitized sky surveys were produced at the space telescope science institute under u.s . government grant nag w-2166 . adler , d. s. , allen , r. j. , & lo , k. y. 1991 , , 283 , 475 arp , h. 1966 , , 14 , 1 arsenault , r. 1989 , , 217 , 66 ball , r. , et al . 1985 , , 298 , l21 bell , e.f . 2003 , , 586 , 794 belley , j. & roy , j. 1992 , , 78 , 61 bicay , m.d . , helou , g. & condon , j.j . 1989 , , 338 , l53 bicay , m.d . & helou , g. 1990 , , 362 , 59 boulanger , f. & viallefond , f. 1992 , , 266 , 37 carignan , c. , et al . 1990 , , 234 , 43 casoli , f. , et al . 1990 , , 233 , 357 castets , a. , et al . 1990 , , 234 , 469 catalogued galaxies and quasars in the iras survey 1985 , prepared by lonsdale , c.j . ( jpl ) combes , f. 1991 , , 29 , 195 condon , j.j . 1992 , , 30 , 575 crosthwaite , l.p . , et al . 2001 , , 122 , 797 crosthwaite , l.p . , 2002 , , 123 , 1892 dahmen , g. , et al . 1998 , , 331 , 959 degioia - eastwood , k. , et al . 1984 , , 278 , 564 de vaucouleurs , g. , de vaucouleurs , a. & corwin , h. 1976 , _ second reference catalogue of bright galaxies _ , univ . of texas press dickman , r.l . , snell , r.l . & schloerb , f.p . 1986 , , 309 , 326 eastman , r.g . , et al . 1996 , , 466 , 911 elmegreen , b.g . 1989 , , 338 , 178 elmegreen , b.g . 1993 , , 411 , 170 elmegreen , b.g . 2002 , , 577 , 206 elmegreen , b.g . & parravano , a. 1994 , , 435 , l121 engargiola , g. 1991 , , 76 , 875 ferguson , a.m.n . , 1998 , , 506 , l19 gordon , k.j . , remage , n.h . & roberts , m.s . 1968 , , 154 , 845 helou , g. & bicay , m.d . 1993 , , 415 , 93 hodge , p.w . & kennicutt , r.c . 1983 , , 88 , 296 honma , m. , sofue , y. & arimoto , n. 1995 , , 304 , 1 hunter , s.d . , et al . 1997 , , 481 , 205 ishizuki , s. et al . 1990 , , 355 , 436 kamphuis , j. & sancisi , r. 1993 , , 273 , l31 karachentsev , i.d . , sharina , m.e . & hutchmeier , w.k . 2000 , , 362 , 544 kennicutt , r.c . 1998a , , 498 , 541 kennicutt , r.c . 1998b , , 36 , 189 kulkarni , s.r . , & heiles , c. 1987 , in _ interstellar processes _ , ed . hollenbach & thronson , p. 87 kutner , m.l . & ulich , b.l . 1981 , , 250 , 341 lacey , c. , duric , n. & goss , w.m . 1997 , , 109 , 417 lequeux , j. , allen , r.j . & guilloteau , s. 1993 , , 280 , l23 leroy , a . 2005 , , 625 , 763 madden , s.c . 1993 , , 407 , 579 maloney , p. & black , j.h . 1988 , , 325 , 389 mangum , j.g . user s manual for the nrao 12 m millimeter - wave telescope _ , nrao 12 m publication mangum , j.g . 1996 , _ on the fly observing at the 12 m _ , nrao 12 m publication mangum , j.g . 1997 , _ equipment and calibration status for the nrao 12 m telescope _ , nrao 12 m publication mason , a. m. , & wilson , c. d. 2004 , , 612 , 860 mead , k.n . & kutner , m.l . 1988 , , 330 , 399 meier , d.s . & turner , j.l . 2004 , , 127 , 2069 meier , d.s . & turner , j.l . 2001 , , 551 , 687 meier , d.s . , turner , j.l . , & beck , s. c. 2002 , , 124 , 877 morris , m. & lo , k.y . 1978 , , 223 , 803 murgia , m. , et al . 2002 , , 385 , 412 nikiforov , i.i . , petrovskaya , i.v . & ninkovic , s. 2000 , _ small galaxy groups , asp conference serier _ , 209 , 399 oka , t. , et al . 1996 , , 460 , 334 oka , t. , et al . 1998 , , 493 , 730 pisano , d.j . & wilcots , e.m . 2000 , , 319 , 821 regan , m.w . & vogel , s.n . 1995 , , 452 , l21 regan , m.w . 2001 , , 561 , 218 rogstad , d.h . , shostak , g.s . & rots , a.h . 1973 , , 22 , 111 rownd , b.k . & young , j.s . 1999 , , 118 , 670 saha , a , claver , j. & hoessel , j.g . 2002 , , 124 , 839 sakamoto , s. 1996 , , 462 , 215 sakamoto , s. , et al . 1995 , , 100 , 125 sakamoto , s. , et al . 1997 , , 486 , 290 sakamoto , s. , et al . 1999 , , 124 , 403 sauvage , m. & thuan , t.x . 1992 , , 396 , l69 schlegel , e. m. 1994 , , 434 , 523 schmidt , m. 1959 , , 129 , 243 scoville , n. z. & sanders , d.b . 1987 , in _ interstellar processes _ , hollenbach & thronson , p 21 scoville , n. z. , et al . 1987 , , 63 , 821 sharina , m.e . , karachentsev , i.d . & tikhonov , n.a . 1997 , astl , 23 , 373 sofue , y. , et al . 1988 , , 40 , 511 solomon , p.m. , et al . 1987 , , 319 , 730 spaans , m. , et al . 1994 , , 437 , 270 strong , a.w . 1988 , , 207 , 1 suchkov , a. , allen , r.j . & heckman , t.m . 1993 , , 412 , 542 tacconi , l.j . & young , j.s . 1986 , , 308 , 600 tacconi , l.j . & young , j.s . 1989 , , 71 , 455 tacconi , l.j . & young , j.s . 1990 , , 352 , 595 tilanus , r.p.j . & allen , r.j . 1989 , , 339 , l57 tilanus , r.p.j . & allen , r.j . 1993 , , 274 , 707 tuffs , r.j . 1996 , , 315 , l149 ulich , b.l . & hass , r.w . 1976 , , 30 , 247 van der kruit , p.c . , allen , r.j . & rots , a.h . 1977 , , 55 , 421 wall , w.f . 1993 , , 414 , 98 walsh , w. , et al . 2002 , , 388 , 7 weliachew , l. , casoli , f. & combes , f. 1988 , , 199 , 29 wiklind , t. , et al . 1990 , , 232 , l11 wilson , c.d . 1995 , , 448 , l97 wilson , c.d . & scoville , n. 1991 , , 370 , 184 wong , t. & blitz , l. 2002 , , 569 , 157 wouterloot , j.g.a . , brand , j. & henkel , c. 1988 , , 191 , 323 young , j.s . & scoville , n.z . 1982 , , 258 , 467 young , j.s . & scoville , n.z . 1991 , , 29 , 581 young , j.s . , 1995 , , 98 , 219 hubble type & scd + r.a . ( j2000 ) & @xmath145 + dec . ( j2000 ) & @xmath146 + @xmath147 & @xmath148 + distance & 6 mpc + @xmath149 & 50.5 @xmath150 + inclination & 40@xmath151 + position angle & 242@xmath152 + @xmath153 & 170 @xmath154 + @xmath155 & 5.1@xmath156 ( 9.0 kpc ) + @xmath157 & 1.07 @xmath158 + @xmath159 & @xmath160 + @xmath161 & @xmath162 + @xmath163 & @xmath164 + @xmath165 & @xmath166 + @xmath167 & @xmath168 + & hubble type & scd & scd & sbc & sbbc + distance ( mpc ) & 3.3 & 6 & 4 & ... + d@xmath84 ( arcminutes ) & 20 & 11 & 11 & ... + d@xmath84/d@xmath169 & 0.23 & 0.38 & 0.15 & ... + d@xmath84/d@xmath170 & 1.1 & 1.0 & 1.0 & ... + m@xmath171 ( @xmath172 ) & 1.8 & 1.9 & 0.7 & @xmath57 + @xmath173 ( @xmath174 ) & 16 & 33 & 23 & 25 + @xmath173 ( @xmath174 ) & 57 & 70 & 62 & 48 + / & 0.27 & 0.47 & 0.37 & 0.52 + m@xmath82/m@xmath83 & 0.055 & 0.074 & 0.16 & 0.03 + @xmath175 ( ) & 140 & 170 & 220 & 400 + @xmath176 ( ) & 11 & 13 & 23 & 3.4 + & 100 & -0.196 ( 0.015 ) + & 60 & -0.189 ( 0.015 ) + & 21 cm continuum & -0.170 ( 0.014 ) + & & -0.224 ( 0.015 ) + & & + & 100 & -0.102 ( 0.008 ) + & 60 & -0.095 ( 0.009 ) + & 21 cm continuum & -0.076 ( 0.009 ) + & & -0.130 ( 0.014 ) + & & + & 100 & -0.002 ( 0.006 ) + & 60 & 0.005 ( 0.007 ) + & 21 cm continuum & 0.024 ( 0.007 ) + & & -0.030 ( 0.013 ) +

we study a two - player zero - sum game in which one of the players is restricted to mixed strategies with limited randomness . more precisely , we consider the maximum payoff that the maximizer ( alice ) can secure with limited randomness . this problem finds an operational interpretation in the context of repeated games with non - ideal sources of randomness as shown by gossner and vieille , and neyman and okada . we begin by simplifying the proof of gossner and vieille and also generalize their result . then , we turn to the computational aspect of the problem , which has not received much attention in the game theory literature . we observe the equivalence of this problem with entropy minimization problems in other scientific contexts . next , we provide two explicit lower bounds on the entropy - payoff tradeoff curve . to do this , we provide and utilize new results for the set of distribution that guarantee a certain payoff for alice ( mixed strategies corresponding to a security level for alice ) . in particular , we study how this set of distribution shrinks as we increase the security level . while the use of total variation distance is common in game theory , our derivation indicates the suitability of utilizing the renyi - divergence of order two .

1702.05719

consider a two - player and zero - sum game between alice and bob . the game is played once . a nash equilibrium assigns a strategy to alice and bob such that no player has any incentive to unilaterally change his strategy @xcite . it is proven that if the players can randomize on their pure strategies set according to _ any _ probability distribution , then the game has at least one nash equilibrium in the randomized mixed strategies . however , assume that alice has restrictions on implementing probability distributions when choosing his mixed strategy . more specifically , assume that alice can only choose a mixed action with entropy at most a given number @xmath0 . for instance , if @xmath1 , alice is only allowed to play pure actions . we are interested in the maximum payoff that alice can secure with mixed actions of entropy at most @xmath0 ( regardless of the action of bob ) . let us denote this by @xmath2 . equivalently , we may ask : in order to guarantee average payoff of @xmath3 , how much randomness needs to be utilized by alice ? let us denote the minimum entropy of the randomness consumed by alice to guarantee payoff @xmath3 by @xmath4 . we may call @xmath4 , the _ min - entropy function_. observe that @xmath5 is the inverse function of @xmath6 . even though @xmath2 is defined in a one - shot setting , the ( upper concave envelope ) of @xmath2 finds an operational meaning in @xcite in the context of repeated games as the payoff that alice can secure in the long run . for instance , gossner and vieille @xcite study a repeated game in which alice can not randomize freely , but has access to an external i.i.d . source @xmath7 which are revealed symbol by symbol ( causally ) to alice as the game is played out . they show that the entropy of the source @xmath8 characterizes the payoff that alice can secure in the long run , and the answer is given in terms of @xmath9 . this is similar to shannon s compression formula @xmath8 which is defined on a single copy of the source ( _ single letter _ ) , but gives the ultimate compression limit when multiple copies of the source are observed . another motivation for @xmath2 is given in @xcite , arguing that limitation on random strategies of players stems from simplicity of humans ; that humans are known to be bad at generating random - like sequences " . to compute @xmath4 , one has to first consider the set of distributions on the action of alice that would secure a payoff @xmath3 for her . this set will be a polytope in the space of all probability distributions . then , one should solve an entropy minimization problem over this polytope in the space of probability distributions . in fact , minimizing and maximizing entropy arises in a wide range of contexts . computing maximum entropy under a set of linear constraints is a well - studied problem with a wide range of applications , _ see @xcite@xcite and the principle of maximum entropy . it is shown in @xcite that computing the minimum entropy can be also quite important . furthermore , many algorithms for clustering and pattern recognition are essentially solving entropy minimization problems @xcite . an important special case of the entropy minimization problem ( with its own applications ) is that of finding a joint probability of minimum entropy given its marginal distributions ( the marginal distribution is a linear constraint on the joint probability distribution ) @xcite . finally , the quantum version of the entropy minimization problem is of key theoretical importance.@xcite while the calculation of the function @xmath4 has received attention in the game theory literature , entropy minimization problem are known to be an np - hard non - convex optimization problem @xcite . since the entropy is a concave function over the probability simplex , its minimum occurs at a vertex of the feasible domain . as a result , computation of @xmath10 leads to a search problem over an exponentially large set . even though the algorithm of @xcite can be used to compute @xmath4 , but it has no guarantee of success . our contributions can be summarized as follows : * we study the properties of @xmath10 and give a number of easy - to - compute bounds on the value of @xmath10 . one of the bounds on @xmath10 is expressed in term of a linear program ; this linear program has a quite general form . we employ _ probabilistic tools _ to study this linear program to prove the following explicit lower bound on @xmath4 : @xmath11 where @xmath12 and @xmath13 are the minimum and maximum entries of the entries of the payoff table ; @xmath14 is the payoff that alice can guarantee with deterministic strategies ( pure - strategy security level ) and @xmath15 is the nash value of the game for alice ( payoff that alice can guarantee with no restriction on her action s entropy ) . because the linear program we consider has a quite general form ( and can be converted to other linear programs with a change of variables ) , our use of probabilistic tools for studying linear programs is novel . * we give and utilize new results on the set of mixed strategies that guarantee a certain security level . while the literature on game theory makes extensive use of total variation distance between distributions , we propose and illustrate the use of the @xmath16divergence ( or the tsallis divergence of order two ) in game theoretic contexts . the @xmath16divergence is defined as @xmath17 , where @xmath18 and @xmath19 are two probability distributions . to see the applicability of this divergence measure , consider a zero - sum game between alice and bob . fix an arbitrary ( mixed ) strategy for bob . let @xmath20 and @xmath21 be two probability distributions for the action of alice . let @xmath22 and @xmath23 denote the payoff of alice when she plays according to @xmath20 and @xmath21 respectively . then , @xmath24 $ ] and @xmath25 $ ] are the payoffs secured by @xmath20 and @xmath21 . also @xmath26 $ ] is a natural object , the risk associated to @xmath20 . then , as we see later in lemma [ lemmavar ] , these natural objects in a game come together as follows : @xmath27-\mathbb{e}[{w_2}])^2}{\mathsf{var}[w_1]}.\end{aligned}\ ] ] * the min - entropy function @xmath4 finds its operational interpretation in a repeated game ( e.g. , the result of gossner and vieille in @xcite ) . we simplify and generalize the proof of gossner and vieille in section [ sec:5op ] . the generalized setup considers the possibility of leackage of alice s random source sequence to bob in the zero - sum game . more specifically , we assume an i.i.d . sequence of pairs @xmath28 , @xmath29 distributed according to a given @xmath30 . the sequence @xmath31 is revealed symbol by symbol ( causally ) to alice as the game is played out , while the sequence @xmath32 is revealed symbol by symbol to bob . we can view @xmath33 as the side information that bob obtains about alice s observation . as before , alice can not randomize freely and can only use the randomness in the sequence @xmath31 . we show that the upper concave envelope of @xmath2 , at @xmath34 is the maximum payoff that alice can secure regardless of bob s actions ; furthermore , bob can ensure that alice does not get more than this payoff . this result is a generalization of the result of gossner and vieille since when @xmath35 is a constant random variable , the conditional entropy @xmath36 reduces to the unconditional entropy @xmath8 . the rest of this paper is organized as follows : in section [ s : prob_state ] we give a mathematical definition of the problem under study ; in section [ s : results ] we express the main results . finally , in section [ s : proofs ] we give the proofs . some of the details are left for the appendices . add notation of @xmath37 and @xmath38 . consider a zero - sum game between players alice ( @xmath39 ) and bob ( @xmath40 ) with respective pure strategy sets @xmath41 and @xmath42 where @xmath43 and @xmath44 are natural numbers . the payoff matrix is denoted by @xmath45 $ ] where @xmath46 is the real valued payoff that player @xmath39 gets from player @xmath40 when @xmath47 and @xmath48 are played . player @xmath39 ( @xmath40 ) wishes to maximize ( minimize ) the expected payoff . the set of all randomized strategies of players @xmath39 and @xmath40 are denoted by @xmath49 and @xmath50 respectively which are probability simplexes on sample spaces @xmath51 and @xmath52 respectively . thus , alice s strategy corresponds to a probability mass function ( pmf ) @xmath53 , which can be also illustrated as a column vector with non - negative entries that add up to one . assume that player @xmath39 uses randomized strategy @xmath37 . then the payoff of alice if bob plays @xmath54 is @xmath55 . we say that alice secures payoff @xmath3 with randomized strategy @xmath37 ( regardless of the action of player @xmath40 ) if @xmath56 for all @xmath54 . thus , the set of all distributions that guarantee payoff @xmath57 for player @xmath39 can be expressed as @xmath58 where @xmath59 is the transpose of the column vector @xmath37 , @xmath60 is a column vector of all ones and @xmath61 means any element of @xmath62 is greater than or equal to the corresponding element at @xmath63 . we define @xmath64 where @xmath65 is the shannon entropy of probability mass function ( pmf ) @xmath37 , defined as @xmath66 where all the logarithms in this paper are in base two . if the set @xmath67 is empty , we set @xmath68 . * a remark on notation : * two - player zero - sum games are completely characterized by payoff matrix . for the sake of simplicity , we call two - player zero - sum games with their payoff matrix . thus , game @xmath69 refers to a game with payoff table @xmath69 . [ d : parameters ] given a game @xmath70 $ ] , parameters @xmath71 , @xmath13 , @xmath72 and @xmath73 are defined as : * @xmath71 ( @xmath13 ) is the minimum ( maximum ) element of matrix @xmath69 : @xmath74 ( @xmath75 ) . * @xmath72 is the maximum payoff secured by pure actions ( _ pure - strategy security level _ ) : @xmath76 . * @xmath73 is the nash value of @xmath69 , which is the maximum guaranteed payoff with unlimited access to random sources : @xmath77 note that by definition , @xmath78 . according to definition [ d : parameters ] , @xmath72 is the payoff that is guaranteed without consumption of any randomness whereas @xmath73 is the maximum guaranteed payoff when unlimited randomness is available . thus , it is interesting to consider the min - entropy function @xmath79 in the domain @xmath80 . if @xmath81 , then @xmath82 ; if @xmath83 the feasible set of optimization problem in [ e : min_entropy2 ] is empty and @xmath84 . when @xmath80 , the function @xmath79 is not necessarily convex or concave as a function of @xmath3 : it is strictly increasing and piecewise concave @xcite .. ] the function @xmath86 is defined in using @xmath85 , which is the set of all distributions that guarantee a security level @xmath57 for player @xmath39 . observe that @xmath87 is a polytope defined via some linear constraints . as the matrix @xmath69 is completely arbitrary , with a change of variables , one can convert it to many different equivalent linear programs . we only need to study @xmath85 for @xmath88 . it is immediate from the definition @xmath85 that this set is decreasing in @xmath3 , _ i.e. , _ for any @xmath89 , @xmath90 ( 6,6 ) ( 0,.5 ) , @xmath91 and @xmath92,title="fig:",width=226 ] ( 2.7,0)@xmath92 ( 2.5,2.5)@xmath93 ( 4,1.2)@xmath91 we are interested to see if the inclusion in is strict , and if yes , quantify to what extent it is . to do this , we look at the distance between the set @xmath93 and the compliment of @xmath94 ( that is @xmath95 ) . the sets @xmath93 , @xmath91 and @xmath92 are illustrated in figure [ f : simplex ] . the distance between any two sets can be defined as @xmath96 where @xmath97 can be any arbitrary distance measure . the standard option is the total variational distance @xmath98 with this choice of the distance , we have [ thm : tvw1w2]for any @xmath99 satisfying @xmath100 , we have @xmath101 observe that the quantities @xmath12 and @xmath13 can be simply computed from @xmath102 . the idea of the proof of theorem [ thm : tvw1w2 ] is standard ( e.g. , see ( * ? ? ? ( 4 ) ) for a similar derivation ) , but is included in section [ proof : mainw1w2 ] for completeness . in this paper , we propose use of the rnyi divergence of order two between @xmath37 and @xmath103 to quantify the distance between two distributions : to be zero if @xmath104 , and infinity if @xmath105 while @xmath106 . we have that @xmath107 , and @xmath108 if and only if @xmath109 . ] our first result gives the following bound : [ thm : mainw1w2]for any @xmath99 satisfying @xmath100 , we have @xmath111 the proof can be found in section [ proof : mainw1w2 ] . the above result is derived by a probabilistic approach , which we believe is novel in the context of linear programming . note that the linear program given in is quite general , and given in terms of an arbitrary payoff matrix @xmath102 . with a proper change of variables , the above result can be applied to a large class of parametric linear programs . our second result is less crucial , but still useful . it gives a compact formula for the supporting hyperplanes of @xmath85 in terms of the nash equilibrium of a game . we need a definition : let @xmath112 denote the value of a two - player zero - sum game with payoff table @xmath45 $ ] ; thus @xmath112 is the maximum value that alice can guarantee using arbitrary mixed strategies . it is known that in a zero - sum game , while the game might have multiple nash equilibriums , the value of alice in all of the equilibriums is the same @xcite . given values @xmath113 , let us construct a new table whose @xmath114 entry is @xmath115 . in other words , we add @xmath116 to the entries in the @xmath117-th row of @xmath102 . the new table can be expressed as @xmath118 where @xmath119 is a column vector whose entries are @xmath120 and @xmath121 is a row vector of all ones . observe that the table @xmath122 can be intuitively understood as giving an additional incentive @xmath116 to alice for playing her @xmath117-th action ( it is actually a disincentive or tax " if @xmath123 ) . [ theorem - sec4 - 1]the set @xmath85 can be characterized as follows : @xmath124 note that @xmath125 is simply a linear program . equivalence of linear program and nash equilibriums are known in the literature @xcite . however , our construction of the game @xmath122 based on incentive or tax is new to best of our knowledge . in section [ proof - theorem - sec4 - 1 ] , we give the proof of theorem [ theorem - sec4 - 1 ] as well as a geometric picture of the nash equilibrium strategies of alice . we begin with a property of the min - entropy function . to state the property , we need the following definition : consider two games with payoff matrixes @xmath126 and @xmath127 . let @xmath128 be the direct - sum of @xmath126 and @xmath127 . then @xmath129 defines a new game in which players simultaneously play one instance of @xmath126 and one instance of @xmath127 and the resulting payoff is the sum of payoffs from @xmath126 and @xmath127 . [ p : convexity ] [ t : direct_sum ] then , we have @xmath130 . similarly , for every natural number @xmath131 , @xmath132 where @xmath133 is @xmath131 times direct sum of @xmath102 . an application of the above theorem is that given an expression @xmath134 that bounds @xmath86 from below for all @xmath3 and @xmath102 , we can conclude that @xmath135 . thus , @xmath136 is also a ( potentially better ) lower bound to @xmath86 . as a result , we expect that a good " lower ( or upper bound ) on @xmath86 should have the correct scaling behavior as we simultaneously play more and more copies of the game . the min - entropy function @xmath86 in is the minimum of a concave function on a polytope @xmath85 . this minimum occurs at a vertex of @xmath85 . this leads to a search in the exponentially large set of vertices of the polytope @xmath85 , which is computationally hard . we desire to find bounds on @xmath86 that are either explicit , or else can be computed in polynomial time . observe that @xmath137 where @xmath138 is the uniform distribution over @xmath139 , and @xmath140 is the kullback leibler ( kl ) divergence . thus , we are interested in finding the vertex of @xmath85 which has maximum distance from the uniform distribution ( with respect to kl divergence ) . * lower bound : * to prove lower bounds for @xmath141 one idea is to replace the entropy function with a smaller function and compute the maximum over @xmath67 . another idea is to also relax the set of distributions @xmath67 and replace it with a potentially bigger set . we proceed with the first idea , and comment on the second idea afterwards . using the fact that the rnyi entropy is decreasing in its order , we obtain that for any @xmath142 @xmath143 where @xmath144 is the rnyi entropy of order @xmath145 : @xmath146 the case of @xmath147 is related to the euclidean norm and results in an optimization problem similar to the one given in for the euclidean norm instead of the kl divergence , which is still not tractable . however , the case @xmath148 relates to the maximum norm and results in the following lower bound : @xmath149 for each @xmath117 , the problem of finding the maximum of @xmath38 over @xmath150 is a linear program . from theorem [ theorem - sec4 - 1 ] , we can find an upper bound on the value of this linear program , yielding @xmath151 where @xmath152 is a vector of length @xmath153 whose @xmath117-th coordinate is one , and all its other coordinates are zero . the lower bound @xmath154 , or its relaxed version in can be found in polynomial time , even though they are not in explicit forms . to obtain an explicit lower bound , observe that @xmath155 note that the vector @xmath152 is a probability vector associated to a _ random variable that chooses @xmath117 with probability one . take some @xmath156 . by definition , deterministic strategies can not secure a payoff of more than @xmath14 . hence @xmath157 . thus , @xmath158 where follows from theorem [ thm : mainw1w2 ] . letting @xmath159 , we obtain @xmath160 with a similar argument and using the fact that @xmath161 , we obtain the following lower bound utilizing theorem [ thm : tvw1w2 ] : @xmath162 observe that when @xmath163 , @xmath164 equals @xmath165 . when @xmath166 , a simple calculation shows that @xmath167 if and only if @xmath168 . [ ex : g2_g3 ] consider a game with payoff matrix : @xmath169\ ] ] from payoff matrix we have @xmath170 , @xmath171 , @xmath172 and @xmath173 . therefore for @xmath174 , @xmath164 gives a better lower bound than @xmath165 on @xmath86 . one can inspect that just like that the min - entropy function , the explicit lower bounds @xmath175 and @xmath176 satisfy @xmath177 have the correct scaling behavior . additionally , by replacing entropy with rnyi entropy in the proof of theorem [ t : direct_sum ] , one also obtains that @xmath178 the function @xmath175 has second derivative for all @xmath179 , while the second derivative of the piecewise concave function @xmath86 is defined everywhere except for a finite number of kink points . the second derivative of the function @xmath175 may be positive or negative , while @xmath86 is piecewise concave . on the other hand , the function @xmath180 is piecewise convex . the reason is that if @xmath181 is attained by @xmath182 and a particular vertex of @xmath67 for @xmath183 $ ] , in this interval @xmath184 varies linearly in @xmath3 . then , convexity of @xmath185 results in convexity of @xmath180 in the interval @xmath186 $ ] . inequality shows that @xmath187 which gives an upper bound for the linear programming of @xmath188 . note that @xmath67 is a very generic polytope , parameterized by a variable @xmath3 . by a change of variables ( scaling and shifting ) , one can convert @xmath188 to a wide class of linear programs ( with no immediate connection to the probability simplex ) , and then use the bound given in . let us now turn to the second idea to prove a lower bound for @xmath189 in , namely replacing the set of distributions @xmath67 with a potentially bigger set . as mention earlier , minimization of the entropy over the set @xmath190 can be difficult . however , it could be possible to solve it ( or find good lower bounds for it ) for special choices of the matrix @xmath69 . has only one non - zero element , the set of constraints will be on the individual coordinates of the vector @xmath37 and minimizing entropy for such constraints is tractable . ] we show how a result for an special case of @xmath69 can be utilized to find a bound ( computable in polynomial time ) for an arbitrary @xmath69 . in particular , let us assume that one has a way to minimize entropy over the set @xmath191 for some given matrix @xmath192 , and any arbitrary column vector @xmath193 . then , we are interested in a value for @xmath193 such that @xmath194 then , we can relax minimization of the entropy over the set @xmath85 by instead , computing its minimum over the set @xmath195 . note that an appropriate @xmath193 in can be found by solving a number of linear programs : the product @xmath196 consists of a number of linear equations on coordinates of @xmath37 , and the minimum of each linear equation over the set @xmath85 is a linear program ( see also theorem [ theorem - sec4 - 1 ] ) . * upper bound : * it is clear that any @xmath197 for any arbitrary choice of @xmath150 . the following theorem gives a number of upper bounds , each of which are obtained by identifying @xmath150 in different ways . [ t : bounds ] consider a game with payoff matrix @xmath102 and parameters @xmath72 , @xmath71 , @xmath13 and @xmath73 defined in definition [ d : parameters ] . let @xmath198 be the entropy of some nash strategy of player @xmath39 and define @xmath199 where @xmath200 and @xmath201 can be defined with any choice of @xmath202 from the argmax set ( if there are multiple possible choices ) . therefore , @xmath203 proof of theorem [ t : bounds ] can be found in section [ prooftbound ] . the second derivative of @xmath204 is negative for @xmath205 . thus @xmath204 is a concave function of @xmath3 and one can readily inspect that : @xmath206 as @xmath207 may contain multiple elements , @xmath208 is not a well defined function of @xmath3 . the function @xmath209 is not necessarily scalable for game @xmath210 . [ ex : game ] consider a game with payoff matrix : @xmath211\ ] ] figure [ f : bounds ] illustrates the behavior of the bounds for this example . in this example @xmath163 ; hence @xmath212 coincides with @xmath213 and is not depicted in the figure . ( 17,7 ) ( 0,.1 ) . the horizontal line depicts @xmath3 and vertical line depicts the value of bounds.,title="fig:",width=642 ] ( 8.3,0)@xmath3 ( 3.2,6.5)@xmath175 ( 3.2,6.1)@xmath154 ( 3.2,5.7)@xmath214 ( 3.2,5.3)@xmath215 ( 3.2,4.9)@xmath216 ( 3.2,4.5)@xmath86 consider a @xmath217 stage repeated zero - sum game between players alice(@xmath39 ) and bob(@xmath40 ) with respective finite action sets @xmath41 and @xmath42 where @xmath43 and @xmath44 are natural numbers . let @xmath218 and @xmath219 be two sequences of random variables drawn i.i.d . from respective sample spaces @xmath220 and @xmath221 with law @xmath222 . in every stage @xmath223 , alice and bob observe respective random sources @xmath224 and @xmath225 privately and choose actions @xmath226 and @xmath227 . then alice gets stage payoff @xmath228 from bob and both players observe the chosen actions @xmath229 and @xmath230 . in order to choose actions at stage @xmath231 , players make use of the history of their observations until stage @xmath231 which is denoted by @xmath232 for alice and @xmath233 for bob let @xmath234 and @xmath235 be the functions mapping the history of observations of alice and bob to actions at stage @xmath231 , thus @xmath236 and @xmath237 . note that alice and bob do not have access to any private sources of randomness , besides @xmath238 and @xmath239 , and have to use deterministic functions @xmath240 and @xmath241 . we call the @xmath217-tuples @xmath242 and @xmath243 the strategies of alice and bob respectively . the expected average payoff for alice up to stage @xmath217 induced by strategies @xmath244 and @xmath245 is denoted by @xmath246 which is : @xmath247\ ] ] where @xmath248 denotes expectation with respect to the distribution induced by i.i.d . repetitions of @xmath222 and strategies @xmath244 and @xmath245 . alice ( bob ) wishes to maximize ( minimize ) @xmath246 . the above repeated zero - sum game between players alice(@xmath39 ) and bob(@xmath40 ) reduces to the one considered by gossner and vieille in @xcite if we set @xmath33 to be constant random variables . we call @xmath249 an unsecure randomness source since it is partially leaked to bob through @xmath33 . let @xmath14 be arbitraty real value : * alice can secure @xmath14 if there exists a strategy @xmath250 for alice such that for all strategy @xmath245 of bob we have @xmath251 . * bob defends @xmath14 if given an arbitrary strategy @xmath244 for alice there exists a strategy @xmath252 for bob such that @xmath253 . * @xmath14 is the maxmin value of the repeated game if alice can secure @xmath14 and bob can defend @xmath14 . the maxmin value of the repeated game under study is characterized as follows : [ t : maxmin_val ] the maxmin value of the repeated game is @xmath254 where @xmath255 is the smallest concave function greater than or equal to @xmath256 . to prove theorem [ t : maxmin_val ] , in section [ s : secure ] we show that alice can secure @xmath254 and in section [ s : defend ] we show that bob can defend @xmath254 , therefore by definition @xmath254 is the maxmin value of the repeated game . [ 1 ] ( 15,4 ) ( 0,0 ) ( .7,0)time slots in which alice draws random action with law @xmath257 ( .7,.61)time slots in which alice draws random action with law @xmath258 ( .7,1.22)time slots of the first block in which alice plays pure action @xmath259 ( 1.5,2)@xmath260 ( 4.5,2)@xmath261 ( 7.5,2)@xmath262 ( 10.5,2)@xmath263 ( 13.5,2)@xmath264 ( .1,3.5)@xmath260 ( .4,3.5)@xmath261 ( 1.5,3.5)@xmath263 ( 2.75,3.5)@xmath265 ( 3.1,3.5)@xmath260 ( 3.5,3.5)@xmath263 ( 4,3.5)@xmath266 ( 5,3.5)@xmath263 ( 5.7,3.5)@xmath265 ( 6.1,3.5)@xmath260 ( 6.5,3.5)@xmath263 ( 7,3.5)@xmath266 ( 8,3.5)@xmath263 ( 8.7,3.5)@xmath265 ( 10.55,3.5)@xmath263 ( 12.1,3.5)@xmath260 ( 12.5,3.5)@xmath263 ( 13,3.5)@xmath266 ( 14,3.5)@xmath263 ( 14.7,3.5)@xmath265 to show that alice can secure any payoff less than @xmath267 , we extend and simplify the proof of @xcite . the upper concave envelope @xmath268 at @xmath34 can be expressed as the convex combination @xmath269 for some @xmath270 $ ] and input distributions @xmath258 and @xmath257 on @xmath51 satisfying @xmath271 as a result , it suffices to show that for any @xmath258 and @xmath257 satisfying @xmath272 alice can secure payoffs arbitrarily close to @xmath273 the idea is to utilize the so - called block - markov " proof technique . take some @xmath217 of the form @xmath274 , and divide the total @xmath217 stages into @xmath264 blocks of length @xmath265 . excluding the first block , we generate the action sequence of each block as a function of the random source observed during the previous block ( we ignore the payoff of the first block throughout the discussion , since by taking the number of blocks @xmath264 large enough , the contribution of the first block in alice s net average payoff becomes negligible ) . more specifically , excluding the first block , each block is further divided into two subblocks , with the first subblock taking up @xmath275 fraction of the block . alice aims to use its observed random source during the previous block to play almost i.i.d . according to @xmath258 during the first subblock , and almost i.i.d . according to @xmath257 during the second subblock . in addition , alice wants its action at any given stage to be also ( almost ) independent of bob s previous observation history in that stage . this is depicted in figure [ f : blocks ] . observe that if alice could produce actions that are perfectly i.i.d . according to @xmath258 and independent of bob s history , then in @xmath275 fraction of the games in a block , she secures the payoff of @xmath276 , and in the remaining @xmath277 fraction secures the payoff of @xmath278 . this gives alice a total payoff of @xmath279 . in reality , alice can not produce the exactly independent actions that she desires . thus , the proof of @xcite ( for the special case of a constant @xmath35 ) goes at length to ensure that approximate independence is enough . we diverge from the proof of @xcite at this point . note that by symmetry , we only need to consider the payoff that alice gets in one of the blocks . for notational simplicity , we denote the observations of alice and bob in the previous block by @xmath280 and @xmath281 respectively , and use @xmath282 and @xmath283 to denote their actions in the current block . the proof relies on the following proposition whose proof is given in section [ s : mapping ] : [ pro : mapping ] let @xmath284 be an arbitrary real number and @xmath258 and @xmath257 be arbitrary distributions on @xmath51 such that @xmath285 . then , for any @xmath156 , there exists a natural number @xmath265 and a mapping @xmath286 such that @xmath287 where @xmath288 , @xmath289 are i.i.d . according to @xmath30 and @xmath290 is the total variation distance . note that @xmath291 is the ideal independent actions by alice , whereas @xmath292 is the real joint distribution . let alice use the actions @xmath293 in the current block where @xmath288 . bob s action at time @xmath231 depends on this history and we may represent it by @xmath294 . then , utilizing the following property of total variation for random variables @xmath35 and @xmath295 @xmath296 we conclude from that @xmath297 utilizing the following property of total variation for random variables @xmath35 and @xmath295 @xmath298 we conclude that the distance between the actual actions @xmath299 and the ideal one is less than or equal to @xmath300 , _ i.e. , _ @xmath301 by relating the total variation distance to the payoff differences ( with a similar derivation as in ) , we obtain that the payoff under the actual distribution and the ideal one is at most @xmath300 times a constant @xmath302 , where @xmath13 and @xmath12 are the maximum and minimum entries for each repetition of the game . and @xmath303 are the maximum and minimum entries of the @xmath265 repetitions of the game . ] thus , the distance between the average payoff of the block differs by at most @xmath304 from the average payoff under the ideal distribution . this completes the proof . we constructed the strategy @xmath244 for the case @xmath274 . in general for @xmath305 where @xmath306 one can extend the first block to contain @xmath307 stages and choose @xmath264 large enough to diminish the effect of first block on average payoff . this is an extension of the proof given in @xcite . let @xmath244 be an arbitrary strategy for alice and generate strategy @xmath245 for bob as follows : given @xmath308 , an arbitrary history of observations of bob until stage @xmath231 , @xmath309 is the best choice of bob that minimizes the expected payoff at stage @xmath231 : @xmath310\ ] ] where @xmath311 denotes expectation with respect to the probability distribution induced by @xmath244 and @xmath312 . thus : @xmath313 \leq \mathcal{j}\left(h(a_t|h_2^t = h_2^t)\right)\leq \mathcal{j}_{\text{cav}}\left(h(a_t|h_2^t = h_2^t)\right),\ ] ] since conditional on the observations @xmath308 of bob , alice s action @xmath229 has entropy @xmath314 . therefore : @xmath315 & = \sum_{h_2^t\in ( \mathcal{a}\times\mathcal{b})^{t-1}\times \mathcal{x}_2^t } \pr\left[h_2^t = h_2^t\right]{\mathbb{e}}_{\sigma,\tau}\left[u_{a_t , b_t}|h_2^t = h_2^t\right ] \notag \\ & \leq \sum_{h_2^t\in ( \mathcal{a}\times\mathcal{b})^{t-1}\times \mathcal{x}_2^t } \pr\left[h_2^t = h_2^t\right]\mathcal{j}_{\text{cav}}\left(h(a_t|h_2^t = h_2^t)\right ) \notag \\ & \leq \mathcal{j}_{\text{cav}}\left(\sum_{h_2^t\in ( \mathcal{a}\times\mathcal{b})^{t-1}\times \mathcal{x}_2^t } \pr\left[h_2^t = h_2^t\right ] h(a_t|h_2^t = h_2^t)\right ) \notag \\ & = \mathcal{j}_{\text{cav}}\left(h(a_t|h_2^t)\right ) \label{e : jensen1 } .\end{aligned}\ ] ] the first inequality results from and by using jensen s inequality for concave function @xmath316 the second inequality follows . by definition of @xmath246 and using we have : @xmath317 where the second inequality follows by using jensen s inequality for concave function @xmath318 . note that given the strategy @xmath245 for bob , @xmath319 is a function of @xmath320 and @xmath321 thus @xmath322 and as @xmath229 and @xmath230 are causally generated from i.i.d . sequences @xmath323 and @xmath324 hence @xmath325 is independent of @xmath326 and @xmath327 follows . therefore the second equality is correct . to prove the final equality consider that given strategies of the players and the sequence @xmath324 , @xmath328 is a function of @xmath323 thus @xmath329 and as @xmath316 is an increasing function the final equality holds . thus for all strategy @xmath244 of alice there exists a strategy @xmath245 such that @xmath330 hence bob can defend @xmath331 . take arbitrary @xmath332 and @xmath333 . since @xmath332 , when alice plays according to @xmath37 , bob has a ( pure ) strategy @xmath54 that reduces alice s expected payoff to a number less than @xmath334 . we can define a random variable @xmath335 that represents the payoff of alice when alice plays @xmath37 and bob plays @xmath336 . then , the alphabet set of @xmath335 is @xmath337 , and its probability distribution over this set is @xmath37 . we must have @xmath338<w_2 $ ] . let us assume that alice adopts @xmath103 , but bob keeps playing the same @xmath54 . let @xmath339 be the random variable , describing the payoff of alice when she plays according to @xmath103 . because @xmath333 , @xmath340\geq w_1 $ ] . we can think of @xmath339 as taking @xmath46 with probability @xmath38 . theorem [ thm : tvw1w2 ] follows from the following chain of inequalities : @xmath341 to prove theorem [ thm : mainw1w2 ] , the key step is to relate the rnyi divergence to _ remember that @xmath342 . [ lemmavar ] take an arbitrary sequence of real numbers @xmath343 and two probability distributions @xmath18 and @xmath19 . let @xmath339 be a random variable that takes value @xmath344 with probability @xmath345 , and @xmath335 be a random variable that takes value @xmath344 with probability @xmath38 . then , we have @xmath346-\mathbb{e}[\tilde{w}])^2}{\mathsf{var}[w]}.\label{eqn : dtforml}\end{aligned}\ ] ] furthermore , the above inequality becomes an equality if we set @xmath347 . observe that the left hand side of depends only on the probability values @xmath38 and @xmath345 , while the right hand side depends not only on the probabilities , but also the values that @xmath339 and @xmath335 take . the proof of this lemma is given in appendix [ appndixa ] . using this lemma and the fact that @xmath338<w_2<w_1\leq \mathbb{e}[w]$ ] , we can conclude @xmath348-w_2)^2}{\mathsf{var}[w]}.\ ] ] observe that @xmath349 holds with probability one . the proof is finished by the following lemma . [ lemmavar2 ] for any @xmath350 , we have @xmath351-w_2)^2}{\mathsf{var}[w]}\geq \frac { ( w_1 -w_2)^2 } { ( w_1 -\underbar{m})(\overline{m}- w_1 ) } , \ ] ] provided that @xmath340\geq w_1 $ ] , and @xmath349 . the proof of this lemma is given in appendix [ appndixb ] . in this section we provide additional details and build a geometric picture . this picture implies theorem [ theorem - sec4 - 1 ] , but also gives a geometric interpretation of nash strategies . let @xmath352 observe that the table @xmath122 can be intuitively understood as giving an additional incentive @xmath116 to alice for playing her @xmath117-th action ( it is actually a disincentive if @xmath123 ) . also , since @xmath353 we only need to understand @xmath265 when the sum of the incentives @xmath116 is zero . we need the following definition : let @xmath354 be the payoff that alice can guarantee with playing distribution @xmath355 with table @xmath69 . we extend the definition of @xmath356 to arbitrary @xmath357 by setting @xmath358 when the tuple @xmath359 does not form a valid probability distribution , _ i.e. , _ when any of the @xmath38 s becomes negative , or @xmath360 . note that @xmath361 a full geometric picture of @xmath85 as well as nash strategies are provided in the following theorem : [ theorem - sec4 - 2]we have 1 . the function @xmath362 is the convex conjugate dual of @xmath363 in the following sense : @xmath364 , \qquad\forall \mathbf{a}\in\mathbb{r}^n.\end{aligned}\ ] ] the function @xmath362 is jointly convex in @xmath365 , while @xmath363 is jointly concave in @xmath359 . + furthermore , the supporting hyperplanes to the convex curve @xmath366 characterize alice s nash strategies as follows : for any arbitrary @xmath119 , @xmath37 is a nash strategy of alice for table @xmath367 if and only if @xmath37 is a subgradient of the function @xmath265 at @xmath119 . in other words , take some arbitrary vector @xmath119 . then , @xmath368 if and only if @xmath37 is a nash strategy of alice for the payoff table @xmath369 . 2 . given a probability vector @xmath37 , we have @xmath370 if and only if @xmath37 guarantees a payoff of at least @xmath3 for game @xmath371 . in particular , setting @xmath372 , @xmath373 if and only if @xmath37 guarantees a payoff of at least @xmath3 for game @xmath69 , _ i.e. , _ @xmath374 . thus , having a payoff @xmath3 , we look for hyperplanes of the form @xmath375 that pass through @xmath3 at @xmath376 , and lie below the curve of @xmath265 . observe that the second part of theorem [ theorem - sec4 - 2 ] is equivalent with theorem [ theorem - sec4 - 1 ] . we begin with the first part of the theorem . using the max - min formulation for the nash value of a game @xmath122 , we have @xmath377\nonumber \\&=\max_{{p}_i\geq 0 , \sum_i p_i=1}\left[k(p_1 , p_2 , \cdots , p_n)+\sum_{i}p_ia_i\right].\nonumber \\&=\max_{{p}_i\in\mathbb{r}}\left[k(p_1 , p_2 , \cdots , p_n)+\sum_{i}p_ia_i\right].\label{eqn : kneg}\end{aligned}\ ] ] where follows from the fact that @xmath378 is minus infinity when @xmath37 is not a probability distribution . this shows the duality of @xmath379 and @xmath356 . next , note that @xmath356 is a minimum of linear functions ; hence it is a concave function . convexity of @xmath379 can be directly seen from which implies that at least one supporting hyperplane to its curve exists at any given point ( since at least one nash strategy exists for any arbitrary game ) . thus , it remains to prove . without loss of generality , it suffices to prove for @xmath380 ( @xmath372 ) , and get the result for arbitrary @xmath381 by changing variable @xmath382 . the inequality @xmath383 can be also expressed as @xmath384 from the duality relation and utilizing the fenchel s duality theorem , the left hand side is @xmath363 . thus , the expression is equivalent with can be written as @xmath385 which is equivalent with @xmath37 being a nash strategy . to proof for the second part of the theorem is similar . as before @xmath386 can be set to zero . then , we can express as @xmath387 from the duality relation and the fenchel s duality theorem , this is equivalent with @xmath388 or @xmath374 . let @xmath389 and @xmath35 be r.v.s representing the actions of player @xmath39 in game @xmath210 with distribution @xmath390 . note that if an arbitrary distribution @xmath391 secures payoff @xmath392 in game @xmath69 then @xmath393 secures payoff @xmath3 in game @xmath210 , where @xmath394 is indicator function . this is because the expected value of the sum payoff is the sum of the expected values of the payoffs of the two games . note that since @xmath395 and @xmath396 , the function @xmath397 is bounded above as follows : @xmath398 on the other hand , let @xmath399 be a distribution with minimum entropy that secures a payoff @xmath3 in game @xmath400 , _ i.e. , _ @xmath401 . note that such a distribution @xmath402 exist because @xmath403 is compact and closed . we have @xmath404 \label{e : j1j2 } \\&= \min_{j_1,j_2\in \mathcal{b } } \left({\mathbb{e}}_{\mathbf{p}^*_{x y } } [ u_{x , j_1}]+ { \mathbb{e}}_{\mathbf{p}^*_{x y}}[u_{y , j_2}]\right ) \\ & = \min_{j_1\in \mathcal{b } } { \mathbb{e}}_{\mathbf{p}^*_{x } } [ u_{x , j_1 } ] + \min_{j_2\in \mathcal{b } } { \mathbb{e}}_{\mathbf{p}^*_{y } } [ u_{y , j_2}]\end{aligned}\ ] ] as a result , there exists @xmath405 such that @xmath406 \geq\frac { w}{2}. \label{e : some_k}\end{aligned}\ ] ] thus , @xmath407 secures payoff @xmath392 in game @xmath69 . hence , @xmath408 where @xmath409 follows from properties of entropy . equations and conclude @xmath410 . the above line of proof can be extended for every natural number @xmath131 to prove that @xmath132 . let @xmath411 be a nash strategy for player a where @xmath412 , and @xmath413 be the pure strategy that guarantees payoff @xmath14 for player a. for @xmath414 define @xmath415 and @xmath416 where @xmath417 is a vertical vector of all zero elements except for its @xmath182th element which is @xmath260 . the pmf @xmath37 guarantees payoff @xmath3 since @xmath418 using properties of entropy one can inspect that : @xmath419 therefore @xmath420 . additionally , as @xmath198 secures payoff @xmath421 then @xmath422 . these two facts imply @xmath423 once we substitute the value of @xmath145 . from the definition of @xmath202 it follows that @xmath424 . therefore for every @xmath47 , @xmath425 and @xmath426 follows . for an arbitrary @xmath48 if @xmath427 then as @xmath428 also belongs to @xmath67 we have @xmath429 . then , for every @xmath48 : @xmath430 as the above inequality is correct for every @xmath48 , @xmath431 . to prove proposition [ pro : mapping ] we make use of the following lemmas . lemma [ l : yasayi ] is a direct consequence of ( * ? ? * theorem @xmath260 ) and lemma [ l : han ] is adopted from @xcite . [ l : yasayi ] consider the correlated random sequences @xmath432 and @xmath281 drawn i.i.d . from respective spaces @xmath220 and @xmath221 by joint probability distribution @xmath312 . let @xmath433 be a random variable independent of @xmath281 and uniformly distributed on @xmath434 where @xmath435 is a real number , then there exists a mapping @xmath436 such that @xmath437 [ l : han ] let @xmath432 and @xmath281 be arbitrary general sequences drawn from @xmath220 and @xmath221 respectively . there exists a mapping @xmath438 satisfying @xmath439 provided that @xmath440 , where @xmath441 and @xmath442 are defined as follows . @xmath443 @xmath444 while for a random sequence @xmath445 , @xmath446=0\right\}\ ] ] @xmath447=0\right\}.\ ] ] choose a real number @xmath448 such that @xmath449 and @xmath450 is a natural number ( for a sufficiently large @xmath265 ) . we define random variables @xmath433 and @xmath451 that are mutually independent of each other , and of @xmath281 with the following marginal distributions : let @xmath433 be a uniformly distributed random variable on @xmath434 , and @xmath451 be distributed as follows : @xmath452 where @xmath453 is the smallest integer greater than or equal to @xmath454 . note that @xmath435 ; thus according to lemma [ l : yasayi ] there exists a mapping @xmath455 satisfying @xmath456 on the other hand we have @xmath457 therefore according to lemma [ l : han ] there exists a mapping @xmath458 such that @xmath459 considering the fact that @xmath460 and @xmath451 both are independent of @xmath461 , the above equation results in @xmath462 furthermore note that @xmath463 where the first inequality follows from the triangular inequality for total variation and the second inequality holds because @xmath464 for any arbitrary deterministic function @xmath465on the sample space of r.v.s @xmath35 and @xmath295 . let @xmath466 . then by combining equations , and we have @xmath467 thus @xmath468 is the desired mapping . 99 j. f. nash , equilibrium points in n - person games , " proc . usa 36.1 ( 1950 ) : 48 - 49 . o. gossner , n. vieille , how to play with a biased coin ? " games and economic behavior , 41(2 ) , 206226 , 2002 . a. neyman and d. okada , repeated games with bounded entropy , " games and economic behavior , 30(2 ) , 228 - 247 , 2000 . a. neyman , d. okada , growth of strategy sets , entropy , and nonstationary bounded recall , " games and economic behavior , 66(1 ) , 404425 , 2009 . p. hubacek , m. naor , and j. ullman . when can limited randomness be used in repeated games ? " international symposium on algorithmic game theory . springer berlin heidelberg , 2015 , 259 - 271 . m. budinich and l. fortnow , repeated matching pennies with limited randomness , " proceedings of the 12th acm conference on electronic commerce . acm , 2011 , 111 - 118 . s. c. fang , j. r. rajasekera and h. s. j. tsao , entropy optimization and mathematical programming , " vol . 8 , springer science & business media , 2012 . t. m. cover and j. a. thomas , elements of information theory , " john wiley & sons , 2012 . j. n. kapur , g. baciu , and h. k. kesavan , the minmax information measure , " int . j. systems sci . 26 , pp . 112 , 1995 . s. watanabe , pattern recognition as a quest for minimum entropy , " pattern recognition , vol . 13 , pp . 381387 , 1981 . m. kocaoglu , a. g. dimakis , s. vishwanath and b. hassibi , entropic causal inference , in:31-st aaai conference on artificial intelligence(aaai-17 ) , to appear . also : arxiv:1611.04035 . f. cicalese , l. gargano , u. vaccaro , how to find a joint probability distribution of minimum entropy ( almost ) , given the marginals , " arxiv : 1701.05243 . m. kovacevic , i. stanojevic , and v. senk , on the entropy of couplings , information and computation , vol . 242 , ( 2015 ) 369382 . p. w. shor , equivalence of additivity questions in quantum information theory , " communications in mathematical physics , 246(3 ) , 453 - 472 , 2004 . m. kovacevic , i. stanojevic , v. senk , on the hardness of entropy minimization and related problems , " in information theory workshop ( itw ) , 512 - 516 , 2012 . m. srikanth , h. k. kesavan , p. roe , computation of the minmax measure , " in entropy measures , maximum entropy principle and emerging applications , springer berlin heidelberg , pp . 239251 , 2003 . y. narahari , game theory and mechanism design , " vol . 4 . world scientific , 2014 . g. b. dantzig , a proof of the equivalence of the programming problem and the game problem , " in : activity analysis of production and allocation , t c. koopmans ( eds ) . john wiley & sons , new york : 330335 . i. adler , the equivalence of linear programs and zero - sum games , " international journal of game theory , 42(1 ) , 165 - 177 , 2013 . yassaee , mohammad hossein , mohammad reza aref , and amin gohari . `` achievability proof via output statistics of random binning . '' ieee transactions on information theory 60.11 ( 2014 ) : 6760 - 6786 . information - spectrum methods in information theory . springer science & business media , 2013 . m. broniatowski , s. leorato , an estimation method for the neyman chi - square divergence with application to test of hypotheses , " journal of multivariate analysis , 97(6 ) , 1409 - 1436 , 2006 . we would like to prove that @xmath346-\mathbb{e}[\tilde{w}])^2}{\mathsf{var}[w]},\label{eqn : dtpq}\end{aligned}\ ] ] where @xmath469 and @xmath470 are defined on some arbitrary set @xmath220 . to the best of our knowledge , this inequality has not appeared in the game theory or information theory literature before . we had a direct proof of it , until we found out that it can be derived from a result of @xcite published in a math journal in 2006 . according to ( * ? ? ? * lemma 1 ) , we have @xmath471-\mathbb{e}[\tilde{w}]-\frac{1}{4}\mathbb{e}[{w}^2].\label{eqn:22dd}\end{aligned}\ ] ] since holds for all @xmath339 and @xmath335 , let us apply the transformation @xmath472)$ ] , and @xmath473)$ ] . this shows that for any real @xmath454 , we have @xmath474-\mathbb{e}[\tilde{w}]\right)-\frac{a^2}{4}{\mathsf{var}}[{w}].\label{eqn:22dd2}\end{aligned}\ ] ] the right hand side is a quadratic expression in terms of @xmath454 , and reaches its maximum at @xmath475-\mathbb{e}[\tilde{w}]\right)}{{\mathsf{var}}[{w}]}.\ ] ] substituting this value of @xmath454 into , we get the desired result . the second part of the theorem that becomes an equality for @xmath347 can be verified directly . let us minimize the expression @xmath476-w_2)^2/{\mathsf{var}}[x]$ ] over all r.v.s @xmath389 that satisfy @xmath477 and @xmath478\geq w_1 $ ] : @xmath479\geq w_1 } } \frac{\left ( { \mathbb{e}}[x]-w_2 \right)^2}{{\mathsf{var}}[x ] } & = \min_{w_1\leq \mu\leq \overline{m } } \left ( \min_{\substack{\underline{m}\leq x \leq \overline{m}\\ { \mathbb{e}}[x]=\mu } } \frac{\left ( { \mathbb{e}}[x]-w_2 \right)^2}{{\mathsf{var}}[x]}\right ) = \min_{w_1\leq \mu \leq \overline{m } } \left(\frac{(\mu -w_2)^2}{\underset{\substack{\underline{m}\leq x \leq \overline{m}\\ { \mathbb{e}}[x]=\mu } } { \max } { \mathsf{var}}[x]}\right)\nonumber \\&= \min_{w_1\leq \mu \leq \overline{m } } \left(\frac{(\mu -w_2)^2}{-\mu^2+\underset{\substack{\underline{m}\leq x \leq \overline{m}\\ { \mathbb{e}}[x]=\mu } } { \max } { \mathbb{e}}[x^2]}\right ) . \label{e : min_min}\end{aligned}\ ] ] we claim that @xmath480=\mu } } { \max } \mathbb{e}[x^2]=(\overline{m}-\mu ) ( \mu -\underline{m})+\mu^2.\ ] ] observe that if @xmath481 is the following binary random variable @xmath482=\frac{\overline{m}-\mu } { \overline{m}-\underline{m}},\quad \mathbb{p}[x^*=\overline{m}]=\frac{\mu -\underline{m}}{\overline{m}-\underline{m}}\end{aligned}\ ] ] we have @xmath483=\mu $ ] and @xmath484=(\overline{m}-\mu ) ( \mu -\underline{m})+\mu^2 $ ] . as a result , @xmath485=\mu } } { \max } \mathbb{e}[x^2]\geq ( \overline{m}-\mu ) ( \mu -\underline{m})+\mu^2.\ ] ] on the other hand , the function @xmath486 is convex and lies below the line that connects the two points @xmath487 and @xmath488 for any @xmath489 $ ] , _ i.e. , _ @xmath490.\ ] ] thus , @xmath491\leq \underline{m}^2+(\mathbb{e}[x]-\underline{m})(\overline{m}+\underline{m})=(\overline{m}-\mu ) ( \mu -\underline{m})+\mu^2.\ ] ] thus , equation holds . as a result , equation becomes @xmath492\geq w_1 } } \frac{\left ( { \mathbb{e}}[x]-w_2 \right)^2}{{\mathsf{var}}[x ] } = \min_{w_1\leq \mu \leq \overline{m } } g(\mu),\ ] ] where @xmath493 note that the function @xmath494 is increasing for any @xmath495 satisfying @xmath496 as @xmath497}{(\overline{m}-\mu ) ^2(\mu -\underline{m})^2 } \geq 0 . \end{aligned}\ ] ] since @xmath498 , this would then imply that the minimum on the right hand side of is obtained at @xmath499 . this will complete the proof .

we propose a new theoretical and practical framework for the study of light - matter interactions and the angular momentum of light . our proposal is based on helicity , total angular momentum , and the use of symmetries . we compare the new framework to the current treatment , which is based on separately considering spin angular momentum and orbital angular momentum and using the transfer between the two in physical explanations . in our proposal , the fundamental problem of spin and orbital angular momentum separability is avoided , predictions are made based on the symmetries of the systems , and the practical application of the concepts is straightforward . finally , the framework is used to show that the concept of spin to orbit transfer applied to focusing and scattering is masking two completely different physical phenomena related to the breaking of different fundamental symmetries : transverse translational symmetry in focusing and electromagnetic duality symmetry in scattering .

1206.5563

in the last decades , the angular momentum of light has received much attention from very diverse areas of physics . from experimental astrophysics proposals for the detection of exotic cosmic objects @xcite to the use of light beams to rotate atoms @xcite , through the exploitation of the infinite number of possible angular momentum values for increasing the capacity of optical communication networks @xcite or developing new concepts in quantum information @xcite . the list of applications is long indeed @xcite . the availability of appropriate theoretical tools for the study of light beams with angular momentum and their interactions with matter is crucial for the development of such a wide and promising range of applications . the current state - of - the - art theoretical framework is based on the separation of angular momentum @xmath0 in its orbital ( @xmath1 ) and spin ( @xmath2 ) components : @xmath3 . in the paraxial approximation the value of the total angular momentum along the optical axis can be split into a term which depends on the azimuthal spatial phase of the field and a term which depends on the polarization @xcite . several efforts have been undertaken to rigorously extend this approach to the non - paraxial regime , but have encountered fundamental difficulties @xcite . strongly non - paraxial tightly focused light fields are the bread and butter of applications where light is made to interact with nano - structures , molecules and atoms . the mechanism of spin to orbit angular momentum conversion ( sam to oam ) , also referred to as spin - orbit interaction , is the explanation of choice for numerically obtained observations in focusing @xcite and remarkable results in scattering experiments @xcite . for example , the presence of optical vortices in tightly focused fields and in scattered fields is explained by conversions between the two types of angular momenta . rigorously speaking , though , the separation between sam and oam can not be made on firm physical grounds . consequently , a conversion between the two quantities is not a fully satisfactory explanation for physical phenomena . the separate consideration of @xmath2 and @xmath1 is known to pose fundamental problems for the electromagnetic field @xcite and its quantum excitations @xcite , @xcite . from the point of view of quantum field theory only the total angular momentum operator is a valid observable property of the photon . even more generally , the strict non - separability is not restricted to photons . for example , it also applies to the electron since only in the non - relativistic limit can the orbital and spin parts of its angular momentum be separately considered @xcite . the fundamental reason for such non - separability is the geometry of rotations of vectors and spinors . such a restriction also applies to rotations of classical electromagnetic fields @xcite . in this article we put forward an alternative theoretical framework for the general and rigorous treatment of the angular momentum of light and its role in light - matter interactions . our approach solves the theoretical difficulties of the current framework , draws its predictive power from symmetry considerations , and can be simply applied in practice . by using it , we discover that the actual physical reason responsible for the presence of optical vortices in tightly focused beams is totally unrelated to the one responsible for the appearance of optical vortices in scattering experiments . in the current state - of - the - art framework , both cases are explained as sam to oam conversion . our proposal is based on total angular momentum and helicity . the role of helicity ( @xmath4 ) in light matter interactions has recently been considered @xcite . the macroscopic maxwell equations have been shown to be invariant under generalized electromagnetic duality transformations , and helicity has been identified as the generator of those transformations . by exploiting this connection , helicity can be used within the powerful formalism of symmetries and conserved quantities for the study of light matter interactions when the approximations implicit in the macroscopic maxwell equations hold @xcite . the use of symmetries and conserved quantities for the study of electromagnetic problems is the paradigm used in this article for the development of its theoretical concepts and their application to practical situations . in section [ sec : context ] we outline the paradigm , mathematical concepts and notation used throughout the paper . in section [ sec : lzsz ] we summarize the different aspects involved in the separation of sam and oam . in section [ sec : framework ] we outline our proposal . first , we discuss the concept of helicity and its associated symmetry and comment on a result from @xcite , which shows that , upon scattering , helicity transforms independently of the geometry of the scatterer . then we show that the combined use of angular momentum and helicity solves the problems associated with the separation of sam and oam in a way that is simpler than the existing theoretical solutions and comment on the practical applicability of our ideas . finally , we establish a relationship between helicity eigenstates and the transverse electric ( te ) and transverse magnetic ( tm ) components of the field . using this relationship we express the conservation law for helicity as a function of the partial scattering matrices in the te - tm basis . this expression becomes useful in the practical application of the framework . in section [ sec : spintoorbit ] we use the developed ideas to revisit the concept of sam to oam conversion in focusing and scattering . we are able to clearly identify the underlying reasons for the presence of optical vortices in focused and scattered fields , which happen to be two totally different physical phenomena connected to the breaking of two independent fundamental symmetries : transverse translational symmetry in focusing and electromagnetic duality in scattering . up to now , the two are explained by sam to oam conversion . during section [ sec : spintoorbit ] , we provide the analytical tools necessary for the practical application of our framework . section [ sec : conc ] contains our conclusions and discussion . the paradigm that we follow in our work is the use of symmetries and conserved quantities for the study of electromagnetic problems . in this article we consider classical maxwell fields . symmetry operations like rotations and translations are linear transformations that apply to the fields . similarly , we model the light - matter interactions as linear transformations of the free space fields . the fields themselves will hence always be transverse . these ideas are best formalized using the mathematical setting of hilbert spaces . therefore , in this article we make extensive use of the concepts and tools associated with a vector space endowed with an inner product , i.e. a hilbert space , and the linear operators acting within that vector space . the vector space we consider is the space of transverse solutions of maxwell s equations , or transverse maxwell fields , which we call @xmath5 . a linear operator within @xmath5 takes one of its vectors , a transverse maxwell field , and transforms it into another vector of the space , still a transverse solution of maxwell s equations . we are interested in symmetry transformations of the fields : time translation , spatial translations and rotations , etc . these transformations are operators acting within @xmath5 . such continuous symmetries are generated by hermitian operators , also acting within @xmath5 , which are associated with properties of the fields . for instance , energy generates time translations , the components of linear momentum generate spatial translations and the components of angular momentum generate spatial rotations . a hermitian operator @xmath6 generates a continuous symmetry transformation @xmath7 by means of @xmath8 see ( * ? ? ? * .21 ) for the detailed study of classical maxwell fields using angular momentum and its generated transformation , spatial rotations . the fact that @xmath5 has an inner product allows us to speak of hermitian operators . it also allows us to construct orthonormal basis into which any transverse maxwell field can be expanded . the basis vectors can be chosen to be transverse fields which are simultaneous eigenvectors of four commuting hermitian operators . the choice of the inner product is not necessarily unique , even though the one we have used , defined in section [ sec : scattering ] has a long tradition for vector fields ( * ? ? ? 13.1.21 ) . the consideration of @xmath5 , the hermitian operators associated with the fundamental quantities of the field and the transformations that these operators generate , allows to study electromagnetic problems using maxwell fields together with the powerful framework of symmetries and conserved quantities . when the electromagnetic equations of a given system are invariant under the transformations generated by a given hermitian operator , the property of the field associated with that operator is a constant of the motion . conversely , if the system does not possess that symmetry , we know that the associated property can , in general , change during evolution . the effect of the symmetry of the system is even stronger , as it must also preserve the eigenvectors and eigenvalues of the operator generating the symmetry , thus offering an adequate basis of vectors to solve the electromagnetic problem . as we will show , this is a simple yet insightful approach to electromagnetic problems . in this article , we want to exploit this approach for the study of the angular momentum of light . consequently , for our purposes , we may only use properties of the electromagnetic field which are associated with a hermitian operator in @xmath5 . only then can we consider their associated symmetry . this rules out the separate use of the components of @xmath1 and @xmath2 , since their associated operators transform a transverse maxwell field into a non - transverse field @xcite , ( * ? ? * b@xmath9 ) : they do not act within the required vector space . from this point of view , @xmath1 and @xmath2 are qualitatively different from @xmath0 , the linear momentum @xmath10 , the energy @xmath11 or , as we will discuss , helicity @xmath4 . in the most commonly used representation , the expressions corresponding to these operators are @xmath12 where @xmath13 , the @xmath14-th component of @xmath15 , is a matrix of indexes @xmath16 defined using the totally antisymmetric tensor @xmath17 with @xmath18 . as discussed above , the operators in equation ( [ eq : line1 ] ) transform a transverse maxwell field into a transverse maxwell field , while those in equation ( [ eq : line2 ] ) break the transversality of the fields . in our notation we use capital letters like @xmath19 and @xmath20 to denote operators , and lower case letters like @xmath21 ( or @xmath22 ) and @xmath23 to denote their eigenvalues . when we speak of a field having a `` sharp '' or `` well defined '' value for an operator , we mean that the field is an eigenvector of that operator with eigenvalue equal to its `` sharp '' value . in the text , names like `` helicity '' or `` third component of angular momentum '' refer to the operators unless it is clear from the context that this is not the case . also , all analytical calculations from section [ sec : scattering ] on , assume a time harmonic decomposition of the fields with an @xmath24 dependence . finally , we would like to mention that the main context of this work is that of classical maxwell fields . nevertheless , the approach here taken , which is based on the study of symmetries , is general and often used in quantum mechanics and other areas of modern physics . serious theoretical difficulties are encountered when attempting to separately consider @xmath1 and @xmath2 for the electromagnetic fields ( @xcite , @xcite ) or its quanta ( @xcite , ( * ? ? ? * b@xmath9 ) ) . we also know @xcite , ( * ? ? ? * b@xmath9 ) that the operators @xmath25 and @xmath15 break the transversality of the fields , taking a vector of @xmath5 out of that space . as far as we are concerned , this prevents the separate use of @xmath25 and @xmath15 for studying electromagnetic problems using symmetries and conserved quantities : @xmath25 and @xmath15 do not generate symmetry transformations for vectors in @xmath5 . on the other hand , the total integrated value of the angular momentum of the electromagnetic field , expressed here in convenient units as an integral over all space involving the electric and magnetic fields @xmath26 and @xmath27 @xmath28 can be separated into two gauge invariant integrals . to achieve such separation one needs to consider only the transverse parts of the field , which are always gauge independent . this restriction is justifiable because the degrees of freedom associated with the longitudinal part of the electric field can always be combined with the degrees of freedom of the sources ( * ? ? ? * i.b.5),(*???*,22 ) . then , in @xmath29 where index @xmath30 refers to the three spatial components , the two terms @xmath31 and @xmath32 are apparently gauge invariant . from now on , we will drop the @xmath33 underscripts since in this article we only consider transverse maxwell fields . the identification of the two parts of equation ( [ eq : oamsam ] ) with spin and orbital angular momenta is tempting due to the appearance of the operator @xmath34 and the relationship of the cross - product in @xmath35 with the spin-1 matrices representing @xmath15 . but , since @xmath25 and @xmath15 are not operators in @xmath5 , the question arises of which are the operators corresponding to the two parts of equation ( [ eq : oamsam ] ) . in his book on quantum mechanics , messiah ( * ch . xxi , problem 7 ) offers an expression which corresponds to the second quantization of the first part of equation ( [ eq : oamsam ] ) . in 1994 , van enk and nienhuis @xcite , in a more detailed study , arrived at the same result and also derived the expression for the second part of equation ( [ eq : oamsam ] ) . in that work , they showed that the two operators are not angular momenta because they do not satisfy the commutation relations which define angular momentum operators . these fock space operators have their corresponding operators in @xmath5 for classical fields ( see section [ sec : theorprac ] ) , and their third components commute @xcite . it is hence possible to split the total angular momentum @xmath36 into two operators @xmath37 and @xmath38 . consequently @xmath39 and @xmath40 in ( [ eq : oamsam ] ) are in reality @xmath41 and @xmath42 . unfortunately , in terms of the separation of the angular momentum operator , this approach is not fully satisfactory because the resulting operators are not angular momentum operators . this can be easily proved by checking that their vectorial components do not fulfill the commutation relations required for angular momentum operators . the consequence of this is that they can not separately generate meaningful rotations . in summary , @xmath36 can always be decomposed into two meaningful operators , @xmath43 , independently of the paraxial approximation . these two operators are never angular momenta since they never fulfill the correct commutation relationships . this last statement is also independent of the paraxial approximation . see section [ sec : theorprac ] for a more detailed discussion . following the paradigm discussed in section [ sec : context ] , we only consider properties of the field whose corresponding operators generate transformations in @xmath5 . as already mentioned several times , this disqualifies @xmath25 and @xmath15 . as substitutes , one may choose @xmath44 and @xmath45 . we prefer to disregard the question of the separation completely and use the total angular momentum @xmath36 and the helicity instead . our choice is based on the fact shown below ( section [ sec : theorprac ] ) that @xmath36 and helicity generate very simple symmetry transformations which lead to a simpler framework . finally , it is worth noting that expressions ( [ eq : jmean ] ) and ( [ eq : jtrans ] ) can be interpreted and computed as a weighted average . when the electromagnetic field is decomposed in a basis of eigenvectors of @xmath19 , the total integrated value @xmath46 is equal to a weighted average ( * ? ? ? the weights are the squares of the amplitudes in the linear decomposition of the field , and they multiply the different eigenvalues @xmath21 of each mode in the basis . this connection is not restricted to angular momentum . all the integrals that are used to compute total integrated values of properties of the electromagnetic field can be interpreted and computed as weighted averages involving the squares of the expansion coefficients and the eigenvalues of the operator related to each particular property . this connection is the key step used for obtaining a fock space operator from the classical spatial integral involving the fields . see for instance @xcite for the cases of @xmath11 and @xmath47 respectively . this connection relates the algebraic formalism introduced in sec . [ sec : context ] with the well known spatial integrals involving the fields , like ( [ eq : jmean ] ) for the case of angular momentum . helicity is defined as the operator which results from projecting the total angular momentum onto the linear momentum @xcite , i.e. @xmath48 . helicity commutes with all the generators of rotations @xmath0 and translations @xmath10 @xcite . in the case of the photon , the helicity can only take the values @xmath49 @xcite . for the electromagnetic field , @xmath4 has only two eigenvalues equal to @xmath50 . a useful interpretation of helicity is obtained by considering the expansion of the field as a superposition of plane waves . in such a representation , helicity is associated with the handedness of each plane wave . for a particle to have a well defined helicity , all the plane waves must be purely circularly polarized and have the same handedness with respect to its momentum vector . this is illustrated in fig . [ fig : helicity ] . given that a general light beam can always be expanded as a sum of plane waves and that helicity is associated with the polarization handedness of those plane waves , helicity seems a very suitable candidate for representing the polarization degrees of freedom of a general light beam . crucially , the action of helicity on a plane wave affects only its polarization vector : it leaves the plane wave momentum vector invariant . this is also true for other types of light beams like multipolar fields or bessel beams : the action of helicity does not change the quantities related to energy , linear or angular momenta which define those beams . helicity acts on separated degrees of freedom , which , in this article , we refer to as the degrees of freedom of the polarization . from its definition , it is readily checked that helicity is a hermitian operator . with respect to related transformations , helicity is the generator of the generalized electromagnetic duality transformation of the fields : @xmath51 where @xmath52 is a real angle . that helicity generates duality is a remarkable fact which can hardly be deduced from the mathematical definition of helicity containing the angular and linear momenta . generalized duality ( [ eq : dualitygeneralization ] ) is well known to be a symmetry of the source free microscopic maxwell equations . the results in @xcite allow to extend the relationship between helicity and duality to the macroscopic maxwell equations in material systems . that work shows that the helicity of the light interacting with a piecewise homogeneous isotropic medium will transform independently of the shapes of the different material domains , and that helicity will be preserved by the interaction if and only if the ratio between the relative electric and magnetic constants of all the involved materials @xmath53 $ ] is constant , that is : @xmath54 when condition ( [ eq : epsmu ] ) is met , the macroscopic maxwell s equations for the system are invariant under transformation ( [ eq : dualitygeneralization ] ) . the geometry independent character of the duality symmetry ( helicity conservation ) allows to separately consider the transformations of the polarization degrees of freedom from the geometry of the scattering problem . such notable simplification is very desirable in a framework for the study of light matter interactions . we propose the use of helicity , the generator of duality transformations , for treating the polarization degrees of freedom in electromagnetic problems . we will show that , when used in conjunction with the total angular momentum , the generator of rotations , we obtain a general framework for the study of problems involving the angular momentum of light and its interaction with matter using the language of symmetries and conserved quantities . it is important to recall that , in this article , we will remain within the approximations implicit in the macroscopic maxwell s equations @xcite . for completeness , a expression of the total integrated value of helicity in terms of electromagnetic fields can be found in @xcite . in this article , though , we are not concerned with the total integrated value of helicity . instead , we use modes with well defined helicity and consider their transformations in particular situations , which will depend on whether the system is invariant under electromagnetic duality . the fundamental problem of the separability of @xmath55 and @xmath56 is avoided by using @xmath4 and @xmath19 instead . both @xmath19 and @xmath4 are commuting hermitian operators in @xmath5 , and generate two simple and independent symmetry transformations of the electromagnetic field : @xmath19 generates rotations around the @xmath57 axis and @xmath4 generates the generalized duality transformation ( [ eq : dualitygeneralization ] ) . as previously mentioned , the two fock space operators obtained from the second quantization of expression ( [ eq : oamsam ] ) studied in @xcite ( and also later in @xcite ) have their corresponding operators for classical fields : in @xcite they are obtained in the momentum space . for instance , it can be seen that @xmath58 , the operator substituting @xmath55 is @xmath59 . this operator commutes with @xmath60 @xcite , whose expression is obviously @xmath61 . consequently , we could use @xmath58 and @xmath60 instead of @xmath19 and @xmath4 . considering the symmetries generated by each pair of operators we prefer to choose @xmath19 and @xmath4 for reasons of simplicity . while @xmath19 and @xmath4 are related to the two simple symmetries indicated above , the symmetries related with @xmath58 and @xmath60 are more complicated . an explicit mathematical expression for the transformation generated by @xmath45 can be found in @xcite . for once , @xmath58 involves a combination of duality and translational symmetries , while the symmetry related to @xmath60 has not been properly studied as such , up to our knowledge . this suggests that using @xmath19 and @xmath4 should , in most situations , result in a simpler analysis . for instance , it is explicitly seen in section [ sec : focusing ] that the conservation law associated with @xmath59 is not independent of the geometry of the problem . in the paraxial limit , when @xmath62 , it can be shown that @xmath63 and @xmath64 . even in this limit , neither of these operators generates physical rotations for the full electromagnetic field . for a paraxial beam , the helicity may be approximated by the real space circular polarization component perpendicular to the @xmath57 axis and , the value of @xmath65 coincides ( see equations ( [ eq : cparaxial])-([eq : dparaxial ] ) below ) with the azimuthal phase winding number of the dominant circular polarization . it is customary to use the paraxial correspondence between sam and circular polarization and oam and azimuthal phase . from the previous considerations , we think that it would be more insightful to use @xmath4 and @xmath65 , which have the advantage of retaining its meaning outside the paraxial approximation . from the experimental point of view , section [ sec : helvor ] shows that the preparation of general ( non - paraxial ) beams with well defined helicity can be done using simple optics in a straightforward fashion . the measurement of the helicity state of a general beam is also shown to be easily achieved using simple optics . in this article we focus in the combined use of @xmath4 and @xmath19 . however , it is worth mentioning that the fact that @xmath4 commutes with all the generators of rotations @xmath0 and translations @xmath10 , and that it generates a transformation which is independent of geometry , should allow its combined use with other degrees of freedom different from angular momentum when the particular problem requires it . in this section we arrive at a relationship between helicity conservation and the scattering of the transverse electric and transverse magnetic components of the field . this relationship is useful in practical problems , as will be seen in section [ sec : spintoorbit ] . let us consider the general scattering problem presented in figure ( [ fig_scatt ] ) . an incident electromagnetic field @xmath66 impinges onto a scatterer of arbitrary shape @xmath67 resulting in the scattered field @xmath68 . consider an orthonormal basis of transverse electromagnetic modes with well defined helicity @xmath69 @xmath70 @xmath71 , where the superindex denotes the sign of @xmath4 and the subindex @xmath71 is a composed index which contains three other commuting operators . for example @xmath72 $ ] for plane waves , with @xmath11 being the energy , and @xmath73 the first two components of linear momentum . both incident and scattered fields can be expanded using the @xmath69 basis , and the scattering process is fully characterized by the following transformation of the incident field : @xmath74\mathbf{e}_{in } , \end{split}\ ] ] where @xmath75 is the hermitian conjugate of @xmath76 . in ( [ eq : scatt ] ) , the transformation @xmath15 is represented by a doubly infinite weighted sum of projection operators of the type @xmath77 , whose action on the input field is @xmath78 and the inner product @xmath79 is assumed . transformation ( [ eq : scatt ] ) is specified by the infinite set of @xmath80 partial scattering matrices relating the coefficients of @xmath66 in @xmath81 to the coefficients of @xmath68 in @xmath82 for all @xmath71 , @xmath83 : @xmath84 let us impose helicity conservation upon scattering , forcing all the partial scattering matrices to be diagonal : @xmath85 for all @xmath71 , @xmath83 . consider the following linear combinations of modes with well defined helicity . @xmath86 which transform into each other by the action of @xmath4 @xmath87 in appendix [ appa ] we show that , for plane waves , bessel beams and multipoles , the sum and subtraction of modes with same @xmath71 and different helicity result in what are commonly known as te and tm modes . we will adopt the te - tm naming to denote general sum and subtractions of modes differing only by their sharp helicity eigenvalue . since @xmath69 is an orthonormal basis , so is @xmath88 . after changing ( [ eq : scatt ] ) to the @xmath88 basis , the condition for helicity preservation upon scattering , expressed in the te - tm basis reads : @xmath89 for all ( @xmath71 , @xmath83 ) . namely , that the scattering of te and tm components is on an equal footing for all ( @xmath71 , @xmath83 ) , as can be seen from the scattering matrix having the same values @xmath90 in the diagonal and @xmath91 in the off - diagonal . condition ( [ eq : tetmhel ] ) is clearly a restriction which will not be met in general . we conclude that , in general , a scatterer will partially convert the helicity of the incident field and that this partial helicity conversion is reflected in asymmetries of the scattering matrices with respect to the te and tm modes . since helicity conservation is equivalent to invariance under generalized duality transformations , having information on the te - tm scattering properties of a system can be used to assess its duality invariance and vice versa . in systems with a high degree of symmetry , like a sphere or a planar multilayer system , a wise choice of basis simplifies condition ( [ eq : tetmhel ] ) . the symmetries of those two systems make the te and tm multipoles @xcite and the te and tm plane waves ( definition contained in appendix [ appb ] ) the eigenmodes of the spherical and planar structures respectively . this means that @xmath92 unless @xmath93 and that @xmath94 for all @xmath71 , @xmath83 . all the partial scattering matrices are diagonal . the preservation of the te - tm components is related to the geometrical symmetries of the system . using mie scattering theory and fresnel s formulas , it is an easy and interesting exercise to analytically verify that , in these two cases , when the materials meet ( [ eq : epsmu ] ) , all the scattering matrices are proportional to the identity ( @xmath95 for all @xmath71 ) , hence preserving helicity as well . the conversion between spin and orbital angular momentum is widely used to explain phase singularities in numerical simulations of tightly focused fields @xcite , and in scattering experiments : @xcite . a detailed discussion of the sam to oam conversion can be found in @xcite . we will now use symmetries and conserved quantities arguments , particularly those related to @xmath19 and @xmath4 , to identify the actual physical reasons for those observations . we will demonstrate that the mechanism responsible for the presence of optical vortices in focused fields is totally different from the one responsible for the observation of optical vortices in scattering experiments . this , in our opinion , shows that the sam - oam formulation can be quite misleading : it assigns the same explanation to two distinct physical phenomena . the analytical tools and methodology employed in this section allow a simple application of the ideas developed in the previous section to practical electromagnetic problems . as already mentioned , @xmath19 and @xmath4 commute . for our analysis we will need a basis of transverse electromagnetic modes which are simultaneous eigenvectors of @xmath19 and @xmath4 . one such set of modes is a particular type of bessel beams . bessel beams have been thoroughly studied . see for instance the series of papers @xcite , @xcite , and @xcite . in appendix [ appb ] , we constructively derive a complete orthonormal basis for transverse electromagnetic fields consisting of bessel beams with well defined third component of angular momentum @xmath19 and helicity @xmath4 . these modes appear in @xcite , although their relationship with helicity is not considered in that paper . in @xcite , these type of electromagnetic modes are obtained in a different way as linear combinations of other type of more commonly used bessel beams , the transverse electric ( te ) and transverse magnetic ( tm ) modes . the constructive derivation in appendix [ appb ] shows clearly that the modes are eigenstates of @xmath4 . from appendix [ appb ] , equations ( [ eq : cd ] ) are the real space expressions in cylindrical coordinates @xmath96 $ ] . an implicitly harmonic @xmath97 dependence has been assumed . @xmath98 where : * @xmath99 , @xmath100 * @xmath101 , @xmath102 . by construction , the two types of vector wave functions @xmath103 and @xmath104 have a sharp value of the @xmath57 component of angular momentum @xmath19 equal to @xmath22 and a sharp value of helicity @xmath4 equal to @xmath105 and @xmath106 , respectively . additionally , they have well defined values for the energy @xmath11 and the @xmath57 component of the linear momentum @xmath107 proportional to @xmath14 and @xmath108 respectively . for a given value of @xmath14 , a well defined value of @xmath107 implies a well defined value of the transverse momentum @xmath109 proportional to @xmath110 . modes ( [ eq : cd ] ) form a complete orthonormal basis of transverse electromagnetic fields when @xmath111,\ \ { k_{\rho}}\in [ 0 , \ \infty),\text{and } \lambda = \pm 1,\ ] ] and both signs of @xmath108 in @xmath112 are considered . in the following , the consideration of both signs of @xmath108 is implicitly made . in order to study why optical vortices seem to appear in numerical studies of focalization of apparently vortex free beams @xcite , we analyze the aplanatic lens model @xcite , which is routinely used to study the effects of high numerical aperture lenses . as we show in appendix [ appc ] and has been discussed before @xcite , the action of an aplanatic lens preserves @xmath19 and @xmath4 . the cylindrical symmetry of the model can be reasonably expected , but its invariance under duality transformations is `` hidden '' in the assumption that the lens transmission coefficients for the two polarization components , te and tm , are identical and that there is no crosstalk between input and output te and tm components . that this assumption implies duality symmetry ( helicity conservation ) is obvious from condition ( [ eq : tetmhel ] ) and the discussion at the end of section [ sec : scattering ] . the preservation of @xmath4 by an aplanatic lens has been discussed in @xcite without using its relationship to electromagnetic duality . in appendix [ appc ] we explicitly analyze the conservation of @xmath19 , highlight the model s key assumption on te and tm scattering , and show that the model conditions for the conservation of @xmath4 and @xmath19 are , as expected , independent of each other . even though both cylindrical and generalized duality symmetries are preserved by the model , the focalized beam is quite different from the input beam : some other symmetry must be broken . the most obvious candidate is the lack of translational symmetry on the plane perpendicular to the optical axis of the lens . we know that the transverse momentum components @xmath20 and @xmath113 are the generators of that symmetry . below we show that the differences between the input and focalized beams are due to changes in @xmath114 . we will prove this point using the basis introduced in ( [ eq : cd ] ) . let us take a collimated right circularly ( rc ) polarized gaussian beam going through an aplanatic lens of high numerical aperture . the linear momentum components of a collimated beam are all almost totally aligned with the propagation direction @xmath57 : @xmath115 . consequently , a collimated beam only has components with small transverse momentum value @xmath110 . with respect to ( [ eq : cd ] ) , in the limit when @xmath115 ( @xmath116 ) , both @xmath103 and @xmath104 approach pure rc and lc polarized modes respectively . this can be easily seen by setting a @xmath117 in ( [ eq : cd ] ) : @xmath118 the other polarization components , the opposite circular and the longitudinal @xmath119 component , are strongly attenuated in this regime . importantly though , they are actually present : without them , the modes are not solutions of maxwell s equations , and its transformation properties can not be consistently analyzed as general electromagnetic fields . from ( [ eq : cparaxial ] ) , and since the collimated input rc gaussian beam does not have a phase singularity in its @xmath120 dominant polarization , we can conclude that mostly @xmath121 type modes with @xmath122 will exist in its expansion in the ( [ eq : cd ] ) basis : @xmath123 where @xmath124 is only significant when @xmath116 . equation ( [ eq : last ] ) defines a beam with a sharp value of @xmath19 , @xmath122 , and a sharp value of @xmath4 , @xmath125 . as per the above discussion regarding symmetries , the output beam must also have sharp @xmath19 and @xmath4 values of @xmath122 and @xmath125 . focusing can hence only change the relative weight of @xmath110 components . intuitively , modes with higher transverse momentum are needed to expand the field after focusing . @xmath126 which is in line the non - preservation of @xmath114 due to broken transverse translational symmetry . the fact that the change is limited to @xmath99 could have been foreseen : it stems from the cylindrical symmetry of the model . now , let s go back to equation ( [ eq : cd ] ) and check the spatial shape of @xmath103 modes when @xmath122 and the condition @xmath116 is not met : @xmath127 the @xmath128 modes are almost purely right polarized , but when @xmath110 increases , the terms multiplying @xmath129 and @xmath119 become significant . as it can be seen in ( [ eq : c ] ) , for @xmath122 these newly enhanced terms have phase singularities of orders minus two and minus one respectively . fig . [ fig : smallbig ] shows the radial spatial profiles of the three polarization components for two @xmath130 modes , one with @xmath131 and the other with @xmath132 . in the small @xmath110 case ( fig . [ figsmall ] ) , the dominant polarization component @xmath120 is much stronger than the @xmath129 and @xmath119 components ( which are nonetheless present ) . in the large @xmath110 case ( fig . [ figbig ] ) , the relative weight between the intensity of the different polarizations has shifted significantly . the vortices of charge @xmath105 in @xmath119 and charge @xmath133 in @xmath129 ( see ( [ eq : c ] ) ) become relatively more important . we argue that these are the optical vortices appearing in numerical simulations of focused beams , and that the correct explanation is not sam to oam conversion but just the inherent spatial properties of light modes with definite energy , @xmath107 , @xmath19 and @xmath4 propagating through a system that conserves energy , @xmath4 and @xmath19 while breaking transverse translational invariance . the lens shifts the weight distribution towards modes with larger @xmath110 values and optical vortices already existing in the initially strongly attenuated polarization components of the input beam gain relative importance in the focalized beam . for the theoretical study of optical vortices in focused beams , @xcite and @xcite use solutions of the paraxial equation as the input modes , instead of using solutions of the full maxwell equations as we have done . since paraxial input modes do not have the attenuated phase singularities in the other polarization components because only a single polarization component is non zero , the appearance of optical vortices upon focusing was , contrary to this paper s explanation , attributed to sam to oam conversion . it is interesting to note that the property associated with @xmath59 is not preserved upon focusing . helicity is conserved but , due to the lack of of translational symmetry in the transverse plane , @xmath109 changes , which implies that @xmath107 changes . the fact that the lack of a `` geometrical '' symmetry breaks the conservation law corresponding to @xmath59 indicates that the symmetry transformation generated by such `` spin operator '' is , in general , not independent of the geometry of the problem . the experimental observation of optical vortices in scattered fields has been reported in the literature @xcite . in these papers , the observations are explained by means of spin to orbit angular momentum conversion during the interaction with the scatterer . recently , similar observations have been analyzed in @xcite using symmetries and conserved quantities , and reaching a very different conclusion . in line with the discussion of the last paper , we will show in this section that the reason for all these experimental observations is not sam to oam transfer , but a partial helicity change during the light - matter interaction due to the breaking of electromagnetic duality symmetry in the system . we will also argue that the helicity change is enhanced by physical effects which strongly break duality , beyond the inherent duality asymmetry of general planar multilayer structures . we have already mentioned several times that helicity transforms independently of the geometry of the scatterer . in particular , it transforms independently of whether the considered system has cylindrical symmetry or not . nevertheless , as seen below , a change in helicity is very clearly identifiable in the spatial patterns of the scattered fields when the system has cylindrical symmetry . several of the experimental setups and input beams in the articles cited in this section have cylindrical symmetry and are similar to the system in fig . [ fig : telescope ] , which we will now analyze . in fig . [ fig : telescope ] , we distinguish the preparation and measurement apparatus from the focusing and scattering part of the setup . in the preparation stage , a collimated gaussian beam goes through a linear polarizer ( lp ) and quarter wave plate ( qwp ) , which results in a beam with well defined values of @xmath19 and @xmath4 ( see more details below ) . in the central part , the beam is focused onto a cylindrically symmetric target and the resulting scattered light is collected by another lens . the measurement part of the setup uses another qwp and lp to project light with different helicity depending on the setting of the lp ( see more details below ) . at first sight , the central part is cylindrically symmetric but lacks translational symmetry in the transverse plane . we should hence expect conservation of @xmath19 and , as in section ( [ sec : focusing ] ) , non - conservation of @xmath109 . since the lenses preserve helicity , the behavior of the system with respect to @xmath4 depends on whether the target sample meets condition ( [ eq : epsmu ] ) . let us assume that it does not meet such condition . let us imagine that , after the first lp and qwp , the input to the first lens is a predominantly rc polarized gaussian beam . assuming perfect optical elements , and using the same arguments as in section [ sec : focusing ] , we can see that the input beam can be expanded using basis ( [ eq : cd ] ) into modes with @xmath116 and sharp values of @xmath19 , @xmath122 , and of @xmath4 , @xmath125 . the first lens will focus the beam into the sample without altering either @xmath19 or @xmath4 , then , since it is assumed that the sample does not meet ( [ eq : epsmu ] ) , part of the light will change helicity upon interacting with the cylindrically symmetric target and , after collimation with the second lens , there will be two types of collimated modes , both with @xmath122 , but differing in the value of helicity . schematically representing the actions of the lenses and the cylindrically symmetric scatterer as transformations of fields expanded in the basis ( [ eq : cd ] ) by coordinates @xmath134 and @xmath135 , we may summarize the whole sequence as : @xmath136 the corresponding modal expressions for the components of the two types of output modes are : @xmath137 that is , a predominantly rc beam similar to the input , without any phase singularity in @xmath120 and a predominantly lc vortex beam with a singularity of charge two in @xmath129 . if we set the last lp to project the lc component of the output collimated beam , this vortex will appear in the ccd camera . we argue that these are the vortices observed in the cylindrically symmetric scattering experiments of ( * ? ? ? 4 ) , ( * ? ? ? 2(c)-(d ) ) and @xcite , and that the underlying reason for their appearance is not sam to oam transfer , but that the electromagnetic duality symmetry is broken in those systems . the samples used in @xcite and @xcite consist of nano - apertures on metallic thin films . the breaking of duality in a general planar multilayer structure , and in particular when it includes a metal , is clear since the relative electric and magnetic constants of the system will not meet condition ( [ eq : epsmu ] ) . nevertheless , as shown in @xcite , the effect due to the multilayer alone is typically small in terms of helicity conversion . the reason why the helicity conversion is enhanced making it easily detectable , is that the nano - apertures allow for light to couple to the natural modes of the multilayer system , where duality is strongly broken . in a multilayer system , the natural modes are either te or tm resonances . since a mode with well defined helicity has equal contributions from te and tm components ( [ eq : mnaa ] ) , a te only or tm only resonance implies a strong breaking of electromagnetic duality . the interfaces between a metal and a dielectric allow the existence of surface plasmon polaritons of tm only character . as shown in @xcite , the influence of these modes in the transmitted light is responsible for the experimentally detectable helicity change . in @xcite , optical vortices are observed upon propagation of light through a planar semiconductor microcavity , still a cylindrically symmetric system as noted in that work . in this case , duality is strongly broken in the multilayer itself by the energy splitting between te and tm modes in the structure . this splitting is identified in that work as the enabler for sam to oam conversion . references @xcite and @xcite contain also results for non - cylindrically symmetric setups . even though their study using @xmath19 is not as simple , we are confident that the generality of the methodology that we propose can rid their analysis from the artificial concept of sam to oam conversion , possibly leading to further physical insights . for instance , for the squared nano - apertures studied in @xcite , the same argumentation about the nano - aperture mediated coupling onto strong duality breaking multilayer natural modes applies . we postulate that any instance where sam to oam conversion is invoked can be properly explained using our framework . using the helicity of light fields to represent the polarization degrees of freedom , we have introduced a general framework for the study of light beams with angular momentum and their interactions with matter . in particular , the framework does not depend on the applicability of the paraxial approximation . our proposal is based on two hermitian operators in the hilbert space of transverse electromagnetic fields : the helicity @xmath4 and a component of angular momentum @xmath19 . these operators are the generators of simple transformations of the fields : generalized electromagnetic duality and rotations along the @xmath57 axis respectively . since the generalized duality transformation is independent from rotations or translations @xcite , the use of @xmath4 allows to consider the changes in polarization independently from other degrees of freedom like angular momenta ( @xmath36 ) and linear momenta ( @xmath47 ) . this simplification is crucial for the practical applicability of the framework . we recall that the approximations implicit in the macroscopic maxwell s equations are assumed in this paper . we propose this new framework as a substitute of the current state of the art treatment , which is based on the separation of spin and orbital angular momenta for the description of the angular momentum of light , and on the mechanism of sam to oam transfer in light - matter interactions and other situations . we have shown that it avoids the fundamental problems associated with the separation of sam and oam in a simpler fashion than the current theoretical solutions . we have also demonstrated its predictive power based on analyzing the broken and unbroken symmetries of the system to predict which properties of the light will change and which ones will be preserved . the current framework lacks predictive power and , at this point , can only be qualified as a descriptive theory . this is illustrated by the fact that , using our methodology , we have shown that , phenomena commonly explained as sam to oam transfer in focusing and scattering are actually due to two completely independent physical reasons , showing the inconsistency of the current framework . the results of this paper suggest that such inconsistency may be related with the use of quantities like sam and oam , which are not valid operators for transverse maxwell fields and hence can not be related to meaningful transformations of the fields . we are confident that the use of helicity and its associated transformation will become a powerful theoretical and practical tool which will improve the understanding and control of light - matter interactions at the nano - scale . applications of the ideas presented in this paper can be foreseen in nano - optics , for the control of the shape and polarization of electromagnetic fields , in metrology , for probing the equivalence of the electric and magnetic responses of a system , and in quantum science and technology , where the identification of two truly independently observable degrees of freedom of the field like @xmath4 and @xmath19 should allow to better understand the potential benefits of the use of angular momentum in quantum applications . the association of helicity and duality with other degrees of freedom and their corresponding transformations constitutes a general and robust methodology to study practical light - matter interaction problems by using fundamental concepts . this work was funded by the australian research council discovery project dp110103697 and the center of excellence in engineered quantum systems ( equs ) . t is also funded by the future fellowship ff110100924 . we set out to proof a general relationship between electromagnetic states of well defined helicity and the transverse electric and transverse magnetic modes . we start by making use of an elegant method for finding solutions of the monochromatic maxwell equations in a source free , isotropic and homogeneous medium @xcite , @xcite . under some suitable conditions of the coordinate system @xmath138 , two transverse independent solutions of the vectorial helmholtz equation of the medium can be obtained from a separable solution @xmath139 of the corresponding scalar helmholtz equation . with @xmath140 a unit vector perpendicular to the surface of constant coordinate @xmath141 , the two vector solutions are obtained as : @xmath142 there are six different coordinate systems for which a complete basis for transverse electromagnetic fields can be built in this way @xcite . plane waves , multipoles and bessel beams result from using cartesian , spherical and cylindrical coordinates , respectively @xcite . besides being eigenvectors of the hamiltonian due to its monochromatic character , each coordinate system produces electromagnetic modes which are eigenvectors of a different set of operators : * plane waves with @xmath143 : eigenvectors the of transverse momenta @xmath23 and @xmath144 . * multipoles with @xmath145 : eigenvectors of the squared angular momentum norm @xmath146 and a component of angular momentum , for instance @xmath19 . * bessel beams with @xmath143 : eigenvectors of the third components of the linear and angular momenta @xmath107 and @xmath19 . from now on , we will lump the energy @xmath11 , which is proportional to @xmath14 , and these other degrees of freedom into a collective index @xmath71 and use @xmath147 and @xmath148 . in all three reference systems , the @xmath147 and @xmath148 modes are commonly referred to as te and tm modes , respectively . using @xmath149 ( * ? ? ? * xiii.93 ) , it can be proved that the helicity operator @xmath4 can be written in real space as @xmath150 . in such formulation , it is also true that @xmath151 @xcite . then , together with ( [ eq : mn ] ) , we see that : @xmath152 namely , the te and tm modes are transformed into each other by the application of @xmath4 . we can hence obtain electromagnetic fields with well defined helicity as : @xmath153 equations ( [ eq : apam ] ) already make it obvious that helicity conservation will only happen for equivalent scattering of the te and tm components of the field , which is needed in order to preserve their linear combinations . it is clear that , since @xmath88 are an orthonormal basis for transverse fields when all values of @xmath71 are considered , so is @xmath69 . this derivation of the te , tm modes is valid for the systems of coordinates mentioned above . on the other hand , the helicity operator is well defined for any basis . thus , we think that it is more natural to define the te and tm modes as @xmath154 , and as such we use them in the main text . in this appendix we present a constructive derivation of electromagnetic modes with well defined energy @xmath11 , third component of angular and linear momenta @xmath19 and @xmath107 , and helicity @xmath4 . we start by considering te and tm modes of well defined energy and linear momentum . they are plane waves derived following the constructive procedure given in @xcite-@xcite which was already used in section [ sec : scattering ] . their explicit expressions are , in the cartesian @xmath155 $ ] basis : @xmath156 { \exp(i \textbf{k } \cdot \textbf{r})},\end{aligned}\ ] ] where @xmath99 , and @xmath157 $ ] and @xmath14 are proportional to the linear momentum vector and energy of the plane waves . as per equation ( [ eq : apam ] ) , sum and subtraction of te and tm modes result in states of well defined helicity : @xmath158 interestingly , any plane wave of well defined helicity and momentum vector proportional to @xmath157 $ ] can also be obtained @xcite by rotating a plane wave of the same helicity and momentum @xmath159 $ ] : @xmath160 where @xmath101 , @xmath102 , @xmath161 and @xmath162 . equation ( [ eq : rothel ] ) exploits the fact that helicity does not change under spatial rotations . a rotation operation of a vectorial field @xmath163 is , explicitly : @xmath164 where @xmath165 is the rotation matrix : @xmath166 where @xmath167 and @xmath168 are rotations around the @xmath119 and @xmath169 axis respectively . we now start our construction of modes with well defined @xmath11 , @xmath107 , @xmath19 and @xmath4 by expressing the most general monochromatic forward propagating transverse electromagnetic field as a combination of plane waves with well defined momentum : @xmath170.\ ] ] a backward propagating beam can be obtained by using @xmath171 and @xmath172 as the initial plane waves instead . having restricted ( [ eq : a ] ) to a single wavenumber @xmath14 assures that the resulting mode has a definite energy @xmath11 proportional to @xmath14 . with respect to @xmath4 , it is clear from ( [ eq : a ] ) that , if we desire a field @xmath76 with well defined helicity , we need only to set either @xmath173 or @xmath174 equal to zero for all @xmath175 , so that only plane waves of the same helicity type are present in ( [ eq : a ] ) . with respect to having a well defined @xmath107 , and because , in a plane wave , its value is proportional to @xmath176 , we must include only a single value of @xmath177 in ( [ eq : a ] ) . the conditions needed in ( [ eq : a ] ) to specify a beam with well defined @xmath19 are not as apparent as in the case of @xmath11 , @xmath107 and @xmath4 . the solution can be reached by applying the operator @xmath19 to the general mode @xmath76 . in order to do that , we use the following definition of @xmath19 as an infinitesimal rotation operation : @xmath178 where @xmath179 is the identity operator . applying ( [ eq : jz ] ) to ( [ eq : a ] ) and making use of the properties of rotation operators we obtain : @xmath180 = \lim_{d{\phi_k}\rightarrow 0 } \int_0^{\frac{\pi}{2 } } \sin { \theta_k}d { \theta_k}\int_0^{2\pi } d{\phi_k}{\alpha({\theta_k } , { \phi_k})}\frac{{\mathbf{r } ( { \theta_k } , { \phi_k})}- \mathbf{r}({\theta_k},{\phi_k}+ d{\phi_k } ) } { id{\phi_k } } \left(-{\mathbf{\hat{l}}}\exp(ikz)\right ) + \\ { \beta({\theta_k } , { \phi_k})}\frac{{\mathbf{r } ( { \theta_k } , { \phi_k})}- \mathbf{r}({\theta_k},{\phi_k}+ d{\phi_k } ) } { id{\phi_k } } \left({\mathbf{\hat{r}}}\exp(ikz)\right ) , \end{split}\ ] ] which , after integrating by parts can be reduced to : @xmath181 = \int_0^{\frac{\pi}{2 } } \sin { \theta_k}d { \theta_k}\int_0^{2\pi } d{\phi_k}\frac{\partial { \alpha({\theta_k } , { \phi_k})}}{i \partial { \phi_k } } { \mathbf{r } ( { \theta_k } , { \phi_k})}\left(-{\mathbf{\hat{l}}}\exp(ikz)\right)+\frac{\partial { \beta({\theta_k } , { \phi_k})}}{i \partial { \phi_k } } { \mathbf{r } ( { \theta_k } , { \phi_k})}\left ( { \mathbf{\hat{r}}}\exp(ikz)\right).\ ] ] so , in this representation , the operator @xmath19 acts by taking the partial derivative of the coordinates @xmath182 with respect to @xmath183 and dividing by @xmath30 . gathering together all the above , we can assert that @xmath184 specifies a mode with well defined energy @xmath11 proportional to @xmath14 , well defined @xmath4 with value @xmath185 , well defined @xmath19 with value @xmath22 and well defined @xmath107 proportional to @xmath186 , and that @xmath187 does the same for the opposite helicity . after inserting the specified coordinates into equation ( [ eq : a ] ) , substituting the rotated plane wave by their explicit expressions as linear combinations of ( [ eq : s ] ) and ( [ eq : p ] ) , changing basis from @xmath155 $ ] to @xmath188 $ ] and using @xmath189 before performing the integral in @xmath190 , we finally obtain the real space expressions of the modes in cylindrical coordinates that have been extensively used in the main text : @xmath191 note that , due to the integration limits of @xmath192 in ( [ eq : a ] ) , the derivation is restricted to propagating modes . in reality , values of @xmath193 in equations ( [ eq : cd2 ] ) are possible , specifying non - propagating modes with well defined @xmath11 , @xmath107 , @xmath19 and @xmath4 . including non - propagating modes from the start of the derivation can be done by using imaginary values of @xmath192 in order to obtain values of @xmath194 bigger than one . the aplanatic or spherical lens model allows to relate the real space field profile of the collimated input beam to the angular spectrum of the focused beam . originally developed by richards and wolf in @xcite , we can find an explanation of the model and its formulas in @xcite , which we reproduce here using a slightly different notation using that @xmath195 : * @xmath197 is the focused field in real space expressed in cylindrical coordinates @xmath198 $ ] . * @xmath199 is the focal distance of the lens , @xmath200 , where @xmath201 is the numerical aperture of the lens , and @xmath14 the wavenumber of the field . * @xmath202 $ ] , with @xmath203 and @xmath204 are both the real space cylindrical coordinate system for the input beam @xmath205 and the spherical coordinates in momentum space of @xmath206 , the angular spectrum of the focalized output beam , with @xmath207 . this dual role of the coordinates is the essence of the model . * @xmath208 are the lenses te and tm transmission coefficients and @xmath209 the indexes of refraction of the input and output media . we now analyze the properties of the model with respect to conservation of @xmath19 and @xmath4 . we start by using some of the ideas and techniques of appendix [ appb ] in order to calculate the angular momentum of the focused beam when the input beam has a definite value of @xmath19 . using ( [ eq : a ] ) , expression ( [ eq : aplanatic_us ] ) can also be written as : @xmath214,\ ] ] where @xmath215 as seen in appendix [ appb ] , in order to apply the operator @xmath19 to the output focused beam in ( [ eq : sum ] ) , we need to take the partial derivative of its coordinates @xmath216 with respect to @xmath183 and divide by @xmath30 . equation ( [ eq : gsgp ] ) relates @xmath216 to the input beam . note that the effect of the lens , which is given by the transmittivities @xmath217 , does not add any azimuthal dependence to @xmath218 and @xmath219 . thus , with this model , the lens will not affect the azimuthal dependence of the input beam , hence keeping the angular momentum constant . more explicitely , since vectors @xmath220 are proportional to the azimuthal and radial polarization vectors defined as : @xmath221 the focalized field angular spectrum coordinates at point @xmath222 , @xmath223 , are the real space azimuthal @xmath224 and radial @xmath225 components of the input field at point @xmath226 . in order to compute the angular momentum of the focused field we need to take the partial derivatives of the radial and azimuthal components of the input field with respect to @xmath227 and divide by @xmath30 . let us now find explicit expressions for @xmath224 and @xmath225 . the collimated input beam can be expanded in the plane @xmath228 into modes of the type ( [ eq : cparaxial ] ) and ( [ eq : dparaxial ] ) with @xmath116 , which we rewrite here after the change of basis @xmath229 , @xmath230 : note that in order to avoid confusion with @xmath192 , the letter for the input beam coordinate angle @xmath232 is now @xmath233 , instead of the original letter @xmath52 in expressions ( [ eq : cparaxial ] ) and ( [ eq : dparaxial ] ) . the sum and subtraction of the above modes result in approximately pure radially and azimuthally polarized modes , which separates the radial and azimuthal components of the field : @xmath234\exp(i\phi n)\frac{1}{\sqrt{2}}{\mathbf{\hat{\rho}}}=a^{\mathbf{\hat{\rho}}}_{n{k_{\rho}}}(\rho,\phi , z_0 ) { \mathbf{\hat{\rho}}},\\ { \mathbf{b}_{n{k_{\rho}}}}(\rho,\phi , z = z_0)&\approx-\sqrt{\frac{2{k_{\rho}}}{\pi}}i^{n+2}\exp(i{k_z}z_0)\left[j_{n+1}({k_{\rho}}\rho)-j_{n+1}({k_{\rho}}\rho)\right]\exp(i\phi n)\frac{1}{\sqrt{2}}{\mathbf{\hat{\phi}}}=b^{\mathbf{\hat{\phi}}}_{n{k_{\rho}}}(\rho,\phi , z_0 ) { \mathbf{\hat{\phi}}},\\ \end{split}\ ] ] which means that , ( [ eq : gsgp ] ) is explicitly @xmath235 then , we need to take the partial derivative of the azimuthal and radial input field components @xmath236 and @xmath237 with respect to @xmath183 and divide by @xmath30 . let us now assume that the input is a field with definite angular momentum @xmath238 . inspection of ( [ eq : ab ] ) reveals that such operation will leave those components invariant except for a multiplicative factor equal to @xmath22 . so , the output field is also a field with definite angular momentum equal to @xmath22 . we have just proved that the aplanatic lens model transfers the @xmath19 value of the input beam to the output beam without changing it . we now study the behavior of the model with respect to helicity . from ( [ eq : gsgp ] ) , it becomes clear that the te and tm content of the output field are independently determined by the real azimuthal and radial components ( [ eq : ab ] ) of the input field respectively . having taken linear combinations of modes with defined helicity , and using the results in appendix [ appa ] , it follows that the modes in ( [ eq : ab ] ) are also pure te and tm modes respectively . using the results from section [ sec : scattering ] about helicity preservation in the te tm basis , we now see that , since there is no cross - talk between the te and tm components of the input and focalized fields , the key condition for helicity preservation in the aplanatic lens model is : let us assume for a moment a more general dependence of the transmission coefficients @xmath240 . the new @xmath183 dependence could destroy the @xmath19 preservation since now more terms will be involved in the partial derivatives of @xmath218 and @xmath219 . on the other hand , as long as @xmath241 for all @xmath242 , helicity will be preserved . this is another example of the independence of the conservation laws of @xmath19 and @xmath4 . m. f. andersen , c. ryu , p. clad , v. natarajan , a. vaziri , k. helmerson , and w. d. phillips , `` quantized rotation of atoms from photons with orbital angular momentum , '' phys . * 97 , * 170406 ( 2006 ) . l. t. vuong , a. j. l. adam , j. m. brok , p. c. m. planken , and h. p. urbach , `` electromagnetic spin - orbit interactions via scattering of subwavelength apertures , '' phys . * 104 , * 083903 ( 2010 ) . f. manni , k. g. lagoudakis , t. k. paraso , r. cerna , y. lger , t. c. h. liew , i. a. shelykh , a. v. kavokin , f. morier - genoud , and b. deveaud - pldran , `` spin - to - orbital angular momentum conversion in semiconductor microcavities , '' phys . b * 83 , * 241307 ( 2011 ) . i. fernandez - corbaton , x. zambrana - puyalto , n. tischler , a. minovich , x. vidal , m. l. juan , and g. molina - terriza , `` experimental demonstration of electromagnetic duality symmetry breaking , '' arxiv:1206.0868 ( 2012 ) . k. y. bliokh , e. a. ostrovskaya , m. a. alonso , o. g. rodrguez - herrera , d. lara , and c. dainty , `` spin - to - orbital angular momentum conversion in focusing , scattering , and imaging systems , '' opt . express * 19 , * 2613226149 ( 2011 ) .

we determine the asymptotics of the joint eigenfunctions of the torus action on a toric khler variety . such varieties are models of completely integrable systems in complex geometry . we first determine the pointwise asymptotics of the eigenfunctions , which show that they behave like gaussians centered at the corresponding classical torus . we then show that there is a universal gaussian scaling limit of the distribution function near its center . we also determine the limit distribution for the tails of the eigenfunctions on large length scales . these are not universal but depend on the global geometry of the toric variety and in particular on the details of the exponential decay of the eigenfunctions away from the classically allowed set .

math0306189

a problem of considerable interest in both mathematics and physics is to determine the asymptotics of the distribution functions @xmath0 of an orthonormal basis @xmath1 of eigenfunctions of a laplacian ( or similar hamiltonian ) on a compact manifold ( see @xcite ) . in general , it is hopelessly difficult to obtain more than crude bounds on such distribution functions , which of course control the @xmath2-norms of the eigenfunctions . numerical analyses and heuristics from quantum chaos and disordered systems suggest however a rich picture in which the asymptotics of @xmath3 is related to the classical dynamics underlying the eigenvalue problem . ( some references will be discussed at the end of the introduction . ) the purpose of this paper is to give a rather complete analysis of the limit distribution of eigenfunctions in one of the few settings where such a detailed analysis is possible , namely where the phase space is a toric khler variety @xmath4 . let us recall the definitions ( see 1 for details ) . toric varieties are complex manifolds on which the complex torus @xmath5 acts with an open dense orbit . by a toric khler variety we mean a toric variety equipped with a khler form @xmath6 that is invariant under the underlying real torus @xmath7 . the action of @xmath8 is hamiltonian with respect to @xmath9 , and thus toric varieties are models of completely integrable systems . they are of a very special type because integrable systems usually generate an @xmath10 action rather than a @xmath8 action . although there are rigidity theorems limiting the class of such examples in the world of real riemannian manifolds @xcite , toric varieties provide a plentiful collection in the world of complex manifolds . the torus action can be ` quantized ' or linearized on the hilbert space completion of the coordinate ring @xmath11 where @xmath12 is a holomorphic line bundle with @xmath13 and where @xmath14 denotes the space of holomorphic sections of its @xmath15-th tensor power . this quantization is generally known as the holomorphic ( bargmann - fock ) representation in the physics literature . the space @xmath16 is spanned by joint eigenfunctions of the linearized @xmath5 action , which we refer to as ` monomials . ' in the fundamental case of @xmath17 , the joint eigenfunctions are the monomials given in an affine chart by @xmath18 the monomials lift ( by homogenization ) to homogeneous monomials on @xmath19 . we consider the case where @xmath20 is a smooth projective toric variety ; i.e. , @xmath21 , @xmath22 , where @xmath23 is an integral delzant polytope ( see [ background ] ) . then the linearized @xmath7 action is generated by @xmath24 commuting operators @xmath25 , @xmath26 on @xmath27 which preserve @xmath28 , and the joint spectrum of the eigenvalue problem @xmath29 consists of lattice points @xmath30 . our main results concern the asymptotics of their distribution functions @xmath31 with @xmath32 as @xmath33 . the function @xmath34 of @xmath35 is the pull - back of the fubini - study norm under a monomial embedding of the form @xmath36 : ( { { \mathbb c}}^*)^m \to { { { \mathbb c}}{{\mathbb p}}}^{d}\;,\qquad p\cap{{\mathbb z}}^m=\{{\alpha}(1),\dots,{\alpha}(d+1)\}\;,\ ] ] for a choice of constants @xmath37 . the volume in @xmath27 and the hermitian norm @xmath38 on @xmath39 are by definition the pull - backs of the fubini - study metric and form under this monomial embedding , and the @xmath40 norm on the space @xmath41 is in turn induced from the volume form and the hermitian pointwise norm of @xmath38 . ( see [ background ] for details . ) as figure [ monfig ] illustrates , the monomial @xmath42 is something like a gaussian bump centered on the real torus @xmath43 , where @xmath44 is the moment map for the classical hamiltonian @xmath8-action on @xmath27 ( see 1 ) . [ monfig ] times the monomial @xmath45 for @xmath46 ( left ) and @xmath47 ( right ) for @xmath48 and @xmath49 , where @xmath50 is the standard simplex . the variable @xmath51 is chosen as @xmath52 . see proposition [ localp ] in section [ massasym ] . , title="fig : " ] times the monomial @xmath45 for @xmath46 ( left ) and @xmath47 ( right ) for @xmath48 and @xmath49 , where @xmath50 is the standard simplex . the variable @xmath51 is chosen as @xmath52 . see proposition [ localp ] in section [ massasym ] . , title="fig : " ] since we would like to determine the properties of eigenfunctions @xmath53 when @xmath54 is large , but not necessarily a multiple of a lattice point in @xmath23 , we shall consider a sequence of approximate multiples , as in the following definition : let @xmath55 be a sequence of lattice points , and let @xmath56 . we say that @xmath57 is a sequence of approximate multiples of @xmath58 if @xmath59 our first result gives the pointwise behavior of the eigenfunctions : let @xmath58 be a point in the interior @xmath60 of the polytope @xmath23 , which is not necessarily a lattice point . then there exists a non - negative function @xmath61 such that @xmath62 if and only if @xmath63 , and for every sequence @xmath64 of approximate multiples of @xmath58 , we have @xmath65\ ] ] uniformly on @xmath66 , where @xmath67 . [ pointww ] the constant @xmath68 is defined in and ( and also appears in theorems [ unrescal ] and [ mldist ] ) . when we consider a sequence of the lattice points of the form @xmath69 with a lattice point @xmath70 , the assumption that @xmath71 is not necessary . in fact , we will give the pointwise asymptotics for @xmath69 with any lattice point @xmath72 in section [ massasym ] ( see propositions [ localss ] and [ localv ] ) . in the above theorem , the function @xmath73 ( @xmath74 ) on @xmath66 is defined by @xmath75 where @xmath76 is the vector given by the equation @xmath77 here @xmath78 is the moment map for the @xmath79 action on @xmath80 ( see for the definition ) , and we write @xmath81 we can express ( as in ( * ? ? ? * ( 17)(18 ) ) ) the function @xmath82 in the more intuitive form as follows : we introduce the real power ` monomials ' @xmath83 and define @xmath84 where ( cf . ) @xmath85 ( the normalized monomial @xmath86 has sup - norm @xmath87 , attained on the torus @xmath88 . ) then is equivalent to : @xmath89 since the sequence of monomials flattens out exponentially quickly away from the peak set @xmath88 , the distribution function is clearly tending to zero . the rate of decay of the distribution function is given by the following result : * let @xmath55 be a sequence of lattice points which are approximate multiples of @xmath74 ( see ) . then , for @xmath90 , we have @xmath91 * let @xmath92 . then , for @xmath90 , we have @xmath93 where @xmath94 , @xmath95 being the face of @xmath23 containing @xmath54 . [ unrescal ] here @xmath96 means the ratio of the left and right hand sides tends to @xmath87 , and the constants @xmath97 are given by and . we recall that if @xmath58 is a point in a face @xmath98 of codimension @xmath99 , then @xmath100 . hence @xmath101 . the exponentially localized behavior of the monomials suggests studying the distribution function on various length scales . first , we show that the @xmath102 have a universal scaling limit on a small length scale : [ mldist ] * let @xmath64 be a sequence of lattice points satisfying the condition with some point @xmath74 . then , for @xmath103 , we have @xmath104 * let @xmath105 . then @xmath106 for @xmath107 , where @xmath108 . a sample graph of the scaling limit distribution function for @xmath109 is given in figure [ limitd ] . with @xmath48 , @xmath110 , @xmath111 as in theorem [ mldist ] , the limit of the rescaled distributions has a universal form , _ i.e. _ it does not depend on the geometry of the manifold @xmath80 , and is given by a logarithmic power of the form @xmath112 with some constant @xmath113 . the logarithmic power appears because the @xmath114-normalized monomials are close to gaussian around the peak set @xmath115 on a vector space of dimension @xmath116 ( or dimension @xmath117 for the boundary lattice cases ) . more precisely , theorem [ pointww ] ( and propositions [ localss ] and [ localv ] ) shows that the function @xmath82 has a positive definite hessian at the peak point . we then observe that the distribution of a gaussian function on @xmath118 , @xmath119 ( where @xmath120 is a real positive @xmath121-matrix ) is given by the logarithmic power law @xmath122 c=\frac{1}{\sqrt{\det a}},\quad 0 < t \leq c,\end{gathered}\ ] ] relative to the normalized lebesgue measure @xmath123 thus , the rescaled distribution of an @xmath114-normalized monomial at the ` center ' of its localized bump has a universal gaussian form . to analyze the ` tails ' of the eigenfunctions , we next use an exponential rescaling of the distribution function so that the global distribution law has a non - zero limit as @xmath33 . as may be expected , it is no longer universal but depends on the geometry of @xmath124 . let @xmath64 satisfy the condition with a point @xmath74 . then @xmath125 where @xmath126 is the hessian matrix of @xmath127 a more general scaling limit law is given in theorem [ tnonrescaled ] . in the above theorem , the assumption that @xmath128 is not necessary . in fact , we will give similar result for @xmath69 with any lattice point @xmath129 ( theorem [ thnonrescaled ] ) in section [ invmprob ] . our strategy for proving theorems [ unrescal ] and [ tnonrescaledintro ] on the distribution functions of monomials is based on their pointwise asymptotics ( theorem [ pointww ] , and also propositions [ localss ] and [ localv ] in section [ massasym ] ) . pointwise asymptotics of monomials are more or less equivalent to asymptotics of their @xmath130 norms . since the latter are of independent interest , we state these asymptotics explicitly in : [ ml2knorm ] * let @xmath64 satisfy the condition for a point @xmath74 . let @xmath131 denote the @xmath132-norm of the @xmath114-normalized monomial @xmath133 with the weight @xmath134 . then we have @xmath135 where @xmath136 depends on @xmath137 . * let @xmath72 with @xmath108 . then we have @xmath138 we close the introduction with some general remarks and references . as mentioned above , our results pertain to phase space distribution of eigenfunctions ( husimi distributions ) rather than to their configuration space distribution . to our knowledge , the only prior example in dimension @xmath139 for which the limit distribution of eigenfunctions has been determined is the case of certain ( so - called ) hecke eigenfunctions of discrete quantum cat maps , due to kurlberg - rudnick @xcite . they work in a simpler discrete model rather than the holomorphic model . the main result of kurlberg - rudnick @xcite is that the distribution functions of ( un - scaled ) eigenfunctions tend to the semi - circle law . their method was to relate the eigenfunctions to exponential sums studied by katz @xcite and to apply the value distribution of exponential sums . since value distribution depends on the representation , it is not clear that the same semi - circle law would hold for hecke eigenfunctions in the holomorphic ( bargmann - fock ) representation , and this appears to be a challenging and interesting problem . part of the motivation for this paper was to set a baseline for eigenfunction distribution problems by studying a class of explicitly solvable examples . it would also be interesting to study the limit distribution of real eigenfunctions in the schrdinger representation , i.e. on the configuration space rather than the phase space . to our knowledge , the physics results mainly pertain to these configuration space results . the cases most studied and speculated about are those of chaotic or disordered systems . when eigenfunctions are delocalized , their spatial distribution is conjectured to be gaussian ( see e.g. @xcite ) . in the opposite regime where the eigenfunctions of disordered systems are exponentially localized , the expected distribution in the low amplitude ( tail ) region is given by a power of a logarithm @xcite , precisely the one we obtained in the high amplitude ( center ) region . the reason is that the distribution law is universal and gaussian in the tail region for exponentially localized eigenfunctions of disordered systems @xcite , while it is universal and gaussian in the center for our problem ( theorem [ pointww ] ) . the only studies we have located which are related to distribution laws of integrable eigenfunctions are those of one of the authors with j. a. toth ( cf . @xcite ) and that of berry - hannay - ozorio de almeida @xcite , which describes the asymptotic expansions of @xmath140 norms ( moment intensities ) of real oscillatory integrals of several stable types . such oscillatory integrals define quasimodes for a quantum integrable system , and in generic cases one can express eigenfunctions as oscillatory integrals of various kinds @xcite . in general , many possible kinds of oscillatory integrals could arise , including ones associated to singular lagrangean tori . hence , one can only expect complete results in special cases . to take the simplest example , our methods could be adapted to find the scaling limit distributions of squares of the standard spherical harmonics @xmath141 on @xmath142 ( or @xmath143 ) . these eigenfunctions are also joint eigenfunctions of commuting operators which generate a quantum torus action . to our knowledge , the distribution laws of even such simple eigenfunctions are unknown at this time . * acknowledgments : * this paper was written during a stay of the second author at johns hopkins university on a jsps fellowship . he would like to express his special thanks to the faculty in the department of mathematics of johns hopkins university . we summarize here some basic facts and terminology on toric varieties from @xcite . recall that a _ toric variety _ is a complex algebraic variety @xmath20 containing the complex torus @xmath144 as a zariski - dense open set such that the group action of @xmath5 on itself extends to @xmath20 . we consider here smooth projective toric varieties ; they can be given the structure of a symplectic manifold such that the restriction of the action to the underlying real torus @xmath145 is a hamiltonian action ( see [ s - torus ] ) . these toric varieties can be constructed from delzant polytopes either by symplectic reduction ( see @xcite ) or by gluing affine toric varieties described by the normal fan of the polytope ( see @xcite ) . however , for our analysis , it is more convenient to define the toric variety @xmath80 associated to the delzant polytope @xmath23 through a _ monomial embedding _ as follows ( see ( * ? ? ? * chapter 5 ) ) . suppose that @xmath23 is a delzant polytope and let @xmath146 we shall write @xmath147 . ( recall that a _ delzant polytope _ is a convex integral polytope in @xmath10 with the property that each vertex is incident to exactly @xmath24 edges and the primitive vectors in @xmath148 parallel to these edges generate @xmath148 . ) to define the monomial embedding , we fix an arbitrary @xmath149 . then , we define @xmath150 : ( { { \mathbb c}}^*)^m \to { { { \mathbb c}}{{\mathbb p}}}^{d}\ , ; \label{phpc}\ ] ] i.e. , @xmath151 , \quad z \in ( { { \mathbb c}}^*)^m.\ ] ] the toric variety @xmath152 is defined as the zariski - closure of the image @xmath153 of the monomial embedding @xmath154 in the complex projective space @xmath155 . since our polytope @xmath23 is assumed to be delzant , @xmath154 is an embedding and the variety @xmath80 is smooth . the symplectic ( or khler ) form on @xmath27 is given by @xmath156 where @xmath157 denotes the fubini - study khler form on @xmath158 with homogeneous coordinates @xmath159 . on @xmath5 , we have latexmath:[\[{\omega}_p^c=\frac{\sqrt{-1}}{2\pi}\partial \bar{\partial } \log \sum_{\alpha \in p } |c_{\alpha}|^2 @xmath27 is given by @xmath161 . we define the line bundle @xmath164 by @xmath165 , where @xmath166 denotes the hyperplane section bundle on @xmath158 . recall that the space of holomorphic sections @xmath167 consists of the linear functions @xmath168 , and that the _ fubini - study metric _ on @xmath166 is given by @xmath169)=\frac{|{\lambda}(\zeta)| } { \|\zeta\|}\qquad ( \zeta\in{{\mathbb c}}^{d+1})\;,\ ] ] which has curvature form @xmath170 . we endow @xmath162 with the hermitian metric @xmath171 , which has curvature @xmath172 . each monomial @xmath173 with @xmath174 corresponds to a section of @xmath175 and vice versa . to explicitly define this correspondence , we make the identifications : @xmath176 more generally , @xmath177 and a basis for @xmath178 is given by the sections @xmath179 corresponding to the monomials @xmath180 . these sections are given by @xmath181 where @xmath182 such that @xmath183 ( see@xcite ) . so far , we have not specified the constants @xmath184 . for studying our phenomena , the choice of constants defining the toric variety @xmath27 is not important . however , when our polytope @xmath23 is the full simplex @xmath185 , we shall use the special choice @xmath186 , where @xmath187 is the multinomial coefficient ( see [ s - proj ] ) . the associated principal @xmath188-bundle @xmath189 of the line bundle @xmath190 is defined by @xmath191 where @xmath192 denotes the norm of @xmath193 with respect to the hermitian metric on @xmath194 induced by @xmath195 . we identify sections @xmath196 of @xmath197 with equivariant functions @xmath198 on @xmath199 by the rule @xmath200 clearly , @xmath201 if @xmath202 . it should be noted that for each @xmath203 , we have @xmath204 where @xmath205 is in the fiber over @xmath206 , @xmath207 denotes the norm with respect to the hermitian metric on @xmath208 induced by the metric @xmath195 . in particular , for @xmath209 , the ` monomial ' @xmath210 given by ( [ identify ] ) lifts to an equivariant function @xmath211 on the circle bundle @xmath212 , and we write @xmath213 where @xmath214 is the lift of the embedding @xmath215 ( @xmath216 ) , that is , @xmath217 is the restriction to @xmath218 of the natural inclusion @xmath219 . ( of course , @xmath220 depends on @xmath221 , which we omit to simplify notation . ) we also consider the monomials @xmath222 so that @xmath220 is the equivariant lift of @xmath223 to @xmath163 . in terms of local coordinates @xmath224 on @xmath225 , we have @xmath226 since its absolute value is independent of @xmath227 , we can write @xmath228 we give @xmath229 the inner product @xmath230 and the @xmath40 norm @xmath231 . we note that the sections @xmath232 are orthogonal but not normalized . we normalize them to obtain an orthonormal basis for @xmath229 consisting of the sections @xmath233 their equivariant lifts @xmath234 form an orthonormal set of monomials on @xmath163 . we include below a table of notation to help the reader keep track of the various monomials : [ cols="^,^,^,^",options="header " , ] the group @xmath5 acts on @xmath235 and the subgroup @xmath79 acts in a hamiltonian fashion . let us recall the formula for its moment map @xmath236 : on the open orbit @xmath5 , we have @xmath237 for any @xmath221 , the image of @xmath238 under @xmath239 equals @xmath23 . the moment map @xmath78 is invariant under the @xmath240-action on @xmath80 . by the identification @xmath241 , the moment map @xmath78 defines a map from @xmath242 to @xmath23 , which is a diffeomorphism between @xmath242 and the interior @xmath60 of @xmath23 . the action of the real torus @xmath8 lifts from @xmath235 to @xmath163 and combines with the @xmath243 action to define a @xmath244 action on @xmath163 . recall that under the monomial embedding @xmath245\;,\ ] ] the @xmath79 action on @xmath246 is given by @xmath247= \big[e^{i\langle{\alpha}(1),{\varphi}\rangle}\zeta_1,\dots , e^{i\langle{\alpha}(d+1),{\varphi}\rangle}\zeta_{d+1}\big]\;.\ ] ] the action ( [ actionm ] ) lifts to an action on @xmath248 : @xmath249 since the circle bundle @xmath250 is invariant under this action , ( [ actionx ] ) also gives a lift of the action ( [ actionm ] ) to @xmath163 . we also have the standard circle action on @xmath163 : @xmath251 which commutes with the @xmath79-action ( [ actionx ] ) . combining ( [ actionx ] ) and ( [ circle ] ) , we then obtain a @xmath252-action on @xmath163 : @xmath253 in this section , we shall explain an aspect of fourier analysis on toric varieties which describe the complete integrability of the system . the hardy space @xmath254 is the hilbert space spanned by the equivariant lifts of sections to functions on @xmath163 with the inner product @xmath255 where @xmath256 is a contact 1-form defined by the hermitian connection on @xmath257 such that @xmath258 . under the identification @xmath259 , the inner product is the same as the inner product on @xmath260 given by ( [ inner ] ) . alternately , @xmath254 consists of the functions @xmath261 satisfying @xmath262 ( see e.g. , @xcite ) . under the @xmath243 action , the hardy space then has the orthogonal decomposition @xmath263 where @xmath264 consists of elements @xmath265 such that @xmath266 . we recall that the ( equivariant ) ` szeg projectors ' @xmath267 are the orthogonal projection onto @xmath268 . if @xmath269 denotes an orthonormal basis of @xmath270 , and @xmath271 denote their lifts to @xmath199 , then the projector @xmath267 is given by the kernel function @xmath272 we now describe how one can combine the eigenvalue problems given by for varying @xmath15 into a homogeneous scalar eigenvalue problem on @xmath199 . to do this , we define for each @xmath273 , the ` homogenization ' @xmath274 of the lattice point in the polytope @xmath275 to be the set of all lattice point @xmath276 of the form @xmath277 where @xmath278 . we also define the cone @xmath279 . it is well known that rays @xmath280 in this cone define a semiclassical limit . in this section , we use the more precise notation @xmath281 for the @xmath40-normalized monomial @xmath282 ( since @xmath15 is not specified in the latter ) , for @xmath283 . the torus action on @xmath163 can be quantized to define an action of the torus as unitary operators on @xmath284 specifically , we let @xmath285 denote the differential operators on @xmath163 generated by the @xmath79 action : @xmath286 we recall the following observation from @xcite : for @xmath287 , * @xmath288 ; * the lifted monomials @xmath289 satisfy @xmath290 ( @xmath291 ) . furthermore , we note that @xmath292 thus , the monomials @xmath293 are the solutions of the joint eigenvalue problem @xmath294 with the commuting operators : @xmath295 the joint eigenvalues are the lattice points @xmath283 . in the case of @xmath296 , the @xmath297-norms of the monomials @xmath298 can be evaluated explicitly in an elementary way . we give the details in this section . when the polytope @xmath23 is the unit simplex @xmath299 , we have @xmath300 ; furthermore @xmath301 is the hyperplane bundle @xmath166 . we can identify @xmath302 with @xmath303 with the origin blown up , and the circle bundle @xmath304 is identified with the unit sphere @xmath305 . the equivariant lifts to @xmath306 of sections of @xmath307 consist of homogeneous polynomials @xmath308 in @xmath309 variables . the induced fubini - study metric on @xmath310 is given by @xmath311 ( here the constants @xmath184 are taken to be @xmath87 . ) identifying @xmath98 with the polynomial @xmath312 , the norm can be written @xmath313 and the inner product is given by : @xmath314 we are interested here in the case @xmath315 . then @xmath316 , and the line bundle @xmath317 is identified with the @xmath318-th tensor power @xmath319 of the hyperplane section bundle . the circle bundle @xmath320 is the lens space @xmath321 . by lifting equivariant functions from the lens space @xmath321 to the sphere @xmath322 , we see that @xmath323 . hence , we shall replace the @xmath15-th hardy space @xmath324 by @xmath325 below . then the equivariant lift @xmath326 of @xmath327 is given by the homogenization : @xmath328 in this case , we shall use the special choice of the coefficients of the monomial embedding : @xmath329 so that by , we have @xmath330 furthermore , @xmath331 where the last equality follows by differentiating the identity @xmath332 . note that this choice gives us the scaling formula @xmath333 as before , for each @xmath24-dimensional multi - index @xmath334 with @xmath335 , we define an @xmath336-dimensional multi - index @xmath337 by @xmath338 recall that @xmath339 with @xmath340 , @xmath341 . the @xmath342-norms of the monomial @xmath298 can be evaluated explicitly as follows . for @xmath343 , we have the precise formula : @xmath344^{q/2 } \frac{\prod_{j=0}^m\gamma(nq{\widehat}{{\alpha}}_j/2 + 1)}{p^{m(q/2 - 1)}\gamma(npq/2+m+1)}.\ ] ] [ exact ] we write @xmath345 , which we consider as a function in @xmath346 . for @xmath347 , we set @xmath348 where @xmath349 denotes lebesgue measure on @xmath303 . we shall compute the integral @xmath350 in two ways . first we use polar coordinates on @xmath303 . the measure @xmath351 is expressed as @xmath352 then we have @xmath353 to relate the volume form @xmath354 on @xmath322 to that on @xmath355 , we recall that @xmath356 and therefore the volume form on @xmath355 is @xmath357 times the fubini - study volume . recalling our convention that @xmath358 , we then have @xmath359 , and hence @xmath360 on the other hand , if we use polar coordinates for each component @xmath361 of @xmath362 , we have @xmath363 hence we obtain @xmath364 and therefore @xmath365 since @xmath366 , the identity follows from . the following proposition is a direct consequence of stirling s formula and proposition [ exact ] . [ stirling ] let @xmath367 , and let @xmath368 , where @xmath369 . set @xmath370 . then we have @xmath371 @xmath372^{n}.\ ] ] in particular , again by stirling s formula , we have @xmath373 in the next section , we will obtain similar formulas for monomials on general toric varieties . we now consider the case of general toric varieties . our first purpose is to find the pointwise asymptotics of the monomials and to prove theorem [ pointww ] . we then asymptotically determine their @xmath132-norms . first of all , we shall consider a sequence @xmath374 of lattice points in @xmath275 . we assume that there exists a point @xmath74 such that @xmath375 under the assumption , the point @xmath376 is in the interior @xmath60 of the polytope @xmath23 for sufficiently large @xmath15 . thus the analysis is performed on the open orbit @xmath377 , and hence the coordinate @xmath378 will be useful . the moment map @xmath78 is also invariant under the hamiltonian @xmath240-action , and it is well - known ( @xcite ) that it induces a diffeomorphism : @xmath379 in these coordinates , the function @xmath82 defined by can be written simply as @xmath380 where the function @xmath381 is the ` polytope character ' @xmath382 in is given by @xmath383 we also define the symmetric real @xmath384 matrix @xmath385 the real symmetric matrix @xmath386 is positive definite , for all @xmath387 . [ posmat ] we must show that @xmath388 for @xmath389 . consider the vectors @xmath390 given by @xmath391 , @xmath392 . since @xmath393 and @xmath394 , we have @xmath395 since @xmath396 for all @xmath397 and @xmath398 is not constant on @xmath399 , the cauchy - schwartz inequality in ( [ s ] ) is strict . in particular , the matrix @xmath400 is positive definite . for a point @xmath74 , we now define the constant @xmath401 in the coordinates @xmath402 on @xmath66 , the volume form on @xmath27 is given by @xmath403 [ volume ] we note that @xmath404 where @xmath405 . the conclusion follows from ( [ hess1 ] ) , recalling that @xmath406 it should be noted that @xmath407 grows as @xmath408 , as stated in the following simple lemma . let @xmath409 be a compact set . then there exists positive constants @xmath410 , @xmath411 such that @xmath412 for @xmath413 , @xmath414 . [ lexpfunc2 ] for any @xmath415 , we define @xmath416 if @xmath74 , then the polytope @xmath417 contains the origin in its interior . thus , clearly we have @xmath418 for any @xmath419 . next , we note that the function @xmath420 is continuous . to see this , let @xmath421 be a sequence such that @xmath422 . then , for any @xmath423 , @xmath424 by using this inequality , we can show that @xmath425 now , for a compact set @xmath409 , we set @xmath426 we set @xmath427 . since @xmath399 is a finite set , there exists @xmath428 such that @xmath429 for @xmath430 with @xmath74 and @xmath431 . thus , for @xmath432 and @xmath433 , we have @xmath434 for @xmath435 and @xmath433 , which completes that proof . _ completion of the proof of theorem [ pointww ] : _ recalling that @xmath436 , we write @xmath437 instead of @xmath438 . by the definition of the hermitian metric on @xmath439 , we have @xmath440 where , in the right hand side , @xmath441 denotes the usual norm in @xmath442 . by the definition of the monomial embedding @xmath154 , we have @xmath443 hence , we have @xmath444 where the function @xmath445 for @xmath74 is defined in . we note that , since @xmath446 , we have @xmath447 . by lemma [ volume ] , the @xmath114-norm of the un - normalized monomial @xmath448 is given by @xmath449 here it should be noted that @xmath450 is a positive integrable function on @xmath451 . by lemma [ lexpfunc2 ] , we can choose @xmath410 , @xmath411 such that @xmath452 and @xmath453 for any @xmath414 and @xmath15 . thus , by choosing a cut - off function @xmath454 suitably , we may write @xmath455 recall that @xmath456 if and only if @xmath457 , and that @xmath458 is the unique critical point of @xmath82 . the hessian of @xmath82 at @xmath458 is the positive definite symmetric matrix @xmath459 . thus , by the morse lemma , there exists a change of coordinates @xmath460 from a neighborhood of the origin to a neighborhood of @xmath461 such that @xmath462 and that @xmath463 by choosing the cut - off function @xmath464 suitably , we get @xmath465 where @xmath466 is a compactly supported function in @xmath467 such that @xmath468 . by , the derivatives of @xmath466 with respect to @xmath467 are all bounded uniformly in @xmath15 . therefore , by using the plancherel formula and a formula for the fourier transform of the gaussian functions , we obtain @xmath469 which completes the proof . since @xmath448 is a monomial , the asymptotics of @xmath470 is essentially the same calculation as the asymptotics of the @xmath130 norm of another monomial . thus , determining the pointwise asymptotics of monomials is equivalent to determining the asymptotics of their @xmath130 norms . next , we consider the ray @xmath471 for a lattice point @xmath472 , which is allowed to lie in the boundary @xmath473 . in such a case , we need to work with other coordinates than the usual coordinates on the open orbit @xmath377 ( @xcite ) , since the open orbit @xmath377 does not cover the set @xmath474 . in the following , we mean that the faces are disjoint , and the facet is a face of codimension one . thus we call the closed face ( or facet ) @xmath475 the closure of @xmath98 in the minimal affine subspace containing @xmath98 . to describe the coordinates , let @xmath476 be a vertex of @xmath23 . since our polytope @xmath23 is delzant , we can choose lattice points @xmath477 in @xmath23 such that each @xmath478 is in an edge incident to the vertex @xmath476 , and the vectors @xmath479 form a basis of @xmath480 . we choose ( open ) facets @xmath481 , @xmath482 incident at @xmath476 so that @xmath483 . [ divisor ] let @xmath72 , and @xmath484 . then , @xmath485 if and only if @xmath486 if @xmath487 then automatically we have @xmath488 . thus , we may assume that @xmath489 . first , assume that @xmath490 for some closed facet @xmath475 which does not contain @xmath70 . the last formula in for the moment map is globally defined , for the function @xmath491 is globally defined . since @xmath490 , the coefficients in @xmath492 of the lattice points @xmath493 must vanish . thus , we have @xmath485 . conversely , assume that @xmath494 is not in the set described in . in our convention , the faces are disjoint and the boundary @xmath473 is the disjoint union of faces . thus , that @xmath494 is not in the set described in is equivalent to say that there exists an open face @xmath495 such that @xmath496 and @xmath497 . let @xmath498 be the set of vertex of @xmath499 . since @xmath500 and @xmath70 are lattice points , there exists a positive integer @xmath501 such that @xmath502 with @xmath503 integer such that @xmath504 . thus we have @xmath505 since @xmath492 is in the interior @xmath495 of the face ( polytope ) @xmath499 , each @xmath506 can not vanish , and hence @xmath507 . using lemma [ divisor ] , we set @xmath508 which covers @xmath80 as @xmath476 varies over all vertices . we define @xmath509 the map @xmath510 is a diffeomorphism and the inverse is given by @xmath511 where @xmath512 is the standard basis for @xmath451 ( or for @xmath513 over @xmath514 ) , and @xmath515 is an @xmath384-matrix with @xmath516 and integer coefficients defined by @xmath517 by definition , we have the obvious formula : @xmath518 by lemma [ divisor ] , @xmath519 if @xmath520 , @xmath521 . since @xmath522 , we have @xmath523 the set @xmath524 is the union of the sets @xmath525 , and hence the map @xmath510 extends a homeomorphism : @xmath526 by this homeomorphism , the set @xmath525 corresponds to the set @xmath527 . this coordinate @xmath528 is useful to explain toric subvarieties corresponding to faces . namely , let @xmath475 be a closed face with @xmath529 which contains @xmath476 . since @xmath530 , we can choose @xmath531 such that @xmath532 . then the subvariety @xmath533 corresponding @xmath475 is expressed , in the coordinate neighborhood @xmath534 , by @xmath535 now , we fix a lattice point @xmath70 in a ( relatively open ) face @xmath98 of dimension @xmath529 such that @xmath530 . without loss of generality , we may assume that @xmath536 . to state a result for the lattice point @xmath70 in the boundary corresponding to theorem [ pointww ] , we need to find a function corresponding to the function @xmath537 . since our coordinate @xmath510 is based on the lattice points @xmath538 , it is reasonable to introduce a new polytope defined by the affine linear transformation @xmath539 which maps @xmath480 bijectively onto itself . we set @xmath540 . then @xmath541 is contained in the positive orthant @xmath542 , and we have @xmath543 . the face of @xmath541 corresponding to @xmath98 is then given by @xmath544 we denote a point in @xmath545 as @xmath546 . in this expression , @xmath547 is a local coordinate of the submanifold @xmath533 . the modulus square of the monomials @xmath548 with @xmath72 is , in this coordinate , given by @xmath549 we then introduce the ` moment map ' corresponding to the face @xmath98 by : @xmath550 where @xmath551 , and the function @xmath552 is given by @xmath553 as mentioned above , the submanifold @xmath533 has the coordinate @xmath554 , and the torus @xmath555 acts on it . thus , it is natural to use the coordinate @xmath556 then , we write @xmath557 for @xmath558 . since we have assumed @xmath559 , we may write @xmath560 we also define @xmath561 [ localss ] in the coordinate @xmath562 as above , we write @xmath563 for the modulus square of the monomial @xmath564 . we also write @xmath565 , which is in the interior of the polytope @xmath566 . then we have @xmath567 where the function @xmath568 is given by @xmath569 and @xmath570 positive definite matrix @xmath571 is given by @xmath572 we shall prove proposition [ localss ] in the rest of this subsection . first of all , we need the following simple lemma : [ volume22 ] in the coordinate @xmath573 on @xmath574 , the volume form @xmath575 is given by @xmath576 where @xmath577 denotes the lebesgue measure on @xmath578 , and the function @xmath579 is given by the determinant of the following @xmath384 matrix : @xmath580 where @xmath581 is the function defined in . for the function @xmath582 on @xmath583 independent of the variable @xmath227 in @xmath584 , then the derivatives @xmath585 , @xmath586 is given , respectively , by @xmath587 we write @xmath562 . then , by this relation , we have @xmath588 where @xmath579 is given by . but , we have @xmath589 where @xmath590 denotes the lebesgue measure on @xmath591 . combining this with , we obtain the assertion . the following lemma can be shown by the same argument as in the proof of lemma [ posmat ] , and we shall omit the proof . [ posmatf ] the @xmath570 matrix defined by @xmath592 is positive definite for every @xmath593 . note that , the map @xmath594 is a diffeomorphism , and the lattice point @xmath595 is in @xmath596 . thus , the vector @xmath597 is well - defined . hence , the @xmath570 matrix @xmath598 is positive definite . _ completion of proof of proposition [ localss]._the modulus square of the monomial @xmath564 in this coordinate is given by @xmath599 where the function @xmath600 is defined by . thus , as in the proof of theorem [ pointww ] , what we need to analyze is the @xmath114-norm @xmath601 where we have used lemma [ volume22 ] . we note that the function @xmath602 defined in is of the form @xmath603 where @xmath604 are the standard basis for @xmath605 . the function @xmath606 is of order @xmath607 in @xmath467 , and hence its derivative up to the second order vanish at @xmath608 . thus , the function @xmath600 has only one critical point @xmath609 . it is not hard to show that the hessian @xmath610 of the function at the critical point @xmath611 is given by @xmath612 where the @xmath570-matrix @xmath571 is given by , and we have used the coordinate @xmath613 with @xmath614 . here , it should be noted that the polytope @xmath615 contains the standard basis in @xmath616 . thus the lattice points @xmath617 with the standard basis @xmath618 in @xmath605 is in the polytope @xmath541 , and hence the functions @xmath619 are all positive . this combined with lemma [ posmatf ] shows that the hessian @xmath610 is positive definite . by the same argument as in the proof of lemma [ lexpfunc2 ] , the function @xmath620 tends to @xmath621 as @xmath622 . therefore , by the standard laplace method as in the proof of theorem [ pointww ] , we have @xmath623 a direct computation will show that @xmath624 and hence , the asymptotics can be written in the form : @xmath625 dividing @xmath626 by the above , we conclude the assertion . when our fixed lattice point @xmath70 is a vertex , say @xmath627 in the description of the coordinate @xmath510 , the matrix @xmath571 is not defined suitably . however , clearly the similar asymptotics can be deduce by the same method . [ localv ] suppose that @xmath70 is a vertex of the polytope @xmath23 . then we have @xmath628 where the function @xmath629 on @xmath513 is given by . we also have @xmath630 where @xmath631 is the fixed point for the hamiltonian @xmath240-action such that @xmath632 . in the description of the coordinate @xmath510 , we put @xmath633 . then , clearly we have @xmath634 , where the fixed point @xmath631 corresponds to the origin @xmath635 and the function @xmath629 is defined in . in this case , we just use the coordinate @xmath510 itself without change of variable . in this coordinate , the monomial @xmath548 is given by @xmath636 the volume measure @xmath575 is of the form : @xmath637 with the lebesgue measure @xmath638 on @xmath513 . in is straight forward to see that the critical point of the function @xmath639 is the origin , and the determinant of the hessian at the origin is given by @xmath640 by using these facts with the laplace method , we obtain the assertion . as an application of the pointwise estimates , one can prove that eigenfunctions ` localize on tori ' . we also determine @xmath132 norm of the monomials . the following is easily shown by using theorem [ pointww ] , propositions [ localss ] and [ localv ] , and the argument is the same as in their proofs , and hence we shall omit the proof ( see also the proof of theorem [ ml2knormp ] ) . [ lt ] * let @xmath641 be a sequence of lattice points satisfying for some point @xmath74 . then , the measure @xmath642 weak@xmath643-converges to the normalized haar measure on the @xmath24-dimensional torus @xmath115 , i.e. , @xmath644 for @xmath645 . * let @xmath70 be a lattice point in @xmath23 with @xmath646 . then , we have @xmath647 where @xmath648 denotes the normalized haar measure on the @xmath649-dimensional torus @xmath650 . in the above proposition , if @xmath70 is a vertex , then the left hand side denotes the dirac measure at the fixed point @xmath651 of the hamiltonian @xmath240-action corresponding to @xmath70 . we note that @xmath652 is invariant under the @xmath8 action . we denote the hilbert space of @xmath8 invariant functions by @xmath653 . we can restate the conclusion as follows : if @xmath654 , then we can regard it as a function on the polytope @xmath23 . we can also regard @xmath655 as a function , say @xmath656 , on @xmath23 , equipped with action variables @xmath657 . we then have : for any @xmath72 , we have @xmath658 ; i.e. , @xmath659 , for @xmath660 . for @xmath661 satisfying with a point @xmath74 , we have @xmath662 . next , we determine the asymptotics of the @xmath132-norm of the @xmath114-normalized monomials . [ ml2knormp ] * let @xmath661 be a sequence of lattice points satisfying with a point @xmath74 . let @xmath131 denote the @xmath132-norm of the @xmath114-normalized monomial @xmath133 with the weight @xmath663 . then we have @xmath664 where the @xmath136 depends on @xmath137 . * let @xmath129 be a lattice point with @xmath646 . then , for @xmath665 , we have @xmath666 for a vertex @xmath70 , we have @xmath667 the proof of and is the same as that for , so we shall give a proof of only . by theorem [ pointww ] and proposition [ localss ] , we have @xmath668 where @xmath669 is given by . as in the proof of proposition [ localss ] , the critical point of @xmath669 is only the point @xmath611 with @xmath597 , @xmath565 , and which is non - degenerate . we have @xmath670 . thus , by the standard laplace method , we have @xmath671 therefore , the assertion follows from in the proof of proposition [ localss ] . by using proposition [ localss ] , we can determine the limit of the sup - norm @xmath672 . [ sup ] * let @xmath673 satisfy with a point @xmath74 . then we have @xmath674 * let @xmath70 be a lattice point with @xmath646 , @xmath665 . let @xmath675 be the point where @xmath564 attains its maximum . then , we have @xmath676 for a vertex @xmath129 , we have latexmath:[\[\label{limitv } \lim_{n\to \infty}n^{-m}\|\varphi_{n\alpha}^{p}\|_{\infty}^{2 } = |c_{\alpha}|^{-2m}= @xmath651 is the unique fixed point for the @xmath240 such that @xmath632 . we only give a proof of . the function @xmath568 defined in attains its minimum at the point @xmath611 , and it tends to @xmath621 as @xmath622 . thus , the function @xmath678 is of order @xmath679 outside a compact neighborhood of @xmath611 . on a compact neighborhood @xmath680 of @xmath611 , we have @xmath681 now follows from the above inequality . the same argument with proposition [ localss ] shows . the values of the @xmath132 norm in proposition [ exact ] and in the projective - space case of theorem [ ml2knorm ] may seem to be different . however , we can check that these two coincide by noting the following simple lemma . the determinant @xmath682 of the matrix @xmath683 is given by @xmath684 [ det ] by applying the differential operator @xmath685 twice to the formula @xmath686 for a vector @xmath687 , we have @xmath688 where @xmath689 is the kronecker s delta , and we have set @xmath690 . we put @xmath691 . then , by , we have @xmath692 and @xmath693 . from this the second equation follows . substituting @xmath694 for @xmath58 in the above formula , we obtain @xmath695 we set @xmath696 with @xmath697 . then a simple computation shows that @xmath698 thus the lemma follows by induction on the dimension @xmath24 . note that , for the simplex @xmath699 , the faces containing the origin is again a simplex in lower dimensional vector space . therefore , lemma [ det ] can be applied to compute @xmath700 . in the case of projective monomials , the function @xmath701 can be expressed as follows . for @xmath387 , @xmath702 where @xmath703 . this is a direct consequence of the formulas @xmath692 , @xmath691 and @xmath704 as mentioned before . the pointwise asymptotics of the @xmath114-normalized projective monomials is given in the following proposition . [ localp ] 1 . for @xmath387 , @xmath705 uniformly @xmath66 . 2 . for @xmath706 , we have @xmath707 uniformly for @xmath708 , where @xmath709 is the diagonal matrix with entries @xmath710 . the assertion @xmath711 in the above is a restatement of theorem [ pointww ] for the projective monomials . the assertion @xmath712 follows from @xmath711 and a taylor expansion of the function @xmath713 . in this section , we find asymptotics of rescaled and un - rescaled distribution functions . fix a lattice point @xmath70 in @xmath23 , and let @xmath99 denote the codimension of the face of @xmath23 ( possibly the open face @xmath714 ) containing @xmath54 . in analogy with , we define the constant @xmath715 by @xmath716 in the following discussion , we give the details for the case where @xmath69 with a lattice point @xmath717 . for general @xmath673 satisfying for a point @xmath74 , one needs only to put @xmath718 and replace @xmath719 and @xmath70 by @xmath374 and @xmath58 . we would like to understand the limit distributions of the measures @xmath720 , namely , the limit of its distribution function @xmath721 however , by theorem [ ml2knorm ] , the @xmath137-th moments of the measure @xmath720 tends to infinity as @xmath15 tends to infinity . therefore , we need to re - normalize the monomials . we write @xmath722 so that @xmath723 by propositions [ limits ] and [ limitv ] , we know that @xmath724 exists and we have @xmath725 furthermore , by theorem [ ml2knorm ] , we have @xmath726 we consider the limit distribution of the sequence of measures @xmath727 on the real line . the distribution function @xmath728 of the measure @xmath729 is given by @xmath730 the distribution function @xmath731 defined above can be expressed , in terms of the distribution function @xmath732 for the measure @xmath720 , as @xmath733 we should note that the total mass of the measure @xmath729 is @xmath734 , and hence it tends to infinity as @xmath15 goes to infinity . but its @xmath137-th moment satisfies , for each positive integer @xmath137 , @xmath735 furthermore , by , the measure @xmath737 is a finite measure on the real line whose support lies in a bounded interval in @xmath736 which is independent of @xmath15 . this implies that the sequence of the finite measures @xmath737 on the real line has a weak limit , say @xmath738 , and it must satisfy let @xmath221 be a positive constant , and let @xmath741 be a positive integer . let @xmath738 be a probability measure on the real line such that @xmath742 then @xmath738 must coincide with the measure @xmath743 where @xmath744 [ momentlemma ] we shall use the following formula : @xmath745 to compute the fourier transform @xmath746 of the measure @xmath738 satisfying the condition in the lemma . substituting @xmath747 , @xmath748 in the above formula , we get @xmath749 and hence @xmath750 . this completes the proof . _ proof of theorem [ mldist ] . _ the rescaled distribution function @xmath728 for @xmath90 of the measure @xmath756 is given by @xmath757 where @xmath758 is the characteristic function of the interval @xmath759 . now set @xmath760 and write @xmath761 by corollary [ cm ] , we know that the sequence of finite measures @xmath762 converges weakly to the probability measure @xmath763 . for fixed @xmath90 , by approximating the function @xmath764 by a sequence of continuous functions , and by using the lebesgue convergence theorem , one easily obtains that , for every @xmath765 , @xmath766 in the previous section , we derived the limit of the rescaled distribution functions @xmath728 . in the definition of the rescaled function @xmath728 , the parameter @xmath767 is rescaled by the factor @xmath768 so that it tends to @xmath769 as @xmath33 . then the corresponding volume @xmath770 is of order @xmath771 . this was the reason why we need to multiply the volume by the extra factor @xmath768 in the definition of the rescaled distribution functions . as a result , the limit distribution has a universal form ( theorem [ mldist ] ) . however , one may ask , of course , what the limit of the non - rescaled distributions @xmath772 is . but , for each fixed @xmath773 , @xmath774 as @xmath33 by theorem [ pointww ] and proposition [ localss ] . therefore , we need to replace @xmath772 by @xmath775 for an appropriate sequence @xmath776 of positive functions in @xmath90 which compensate for the flattening rate . our sequence will satisfy the conditions @xmath777 as @xmath33 and @xmath778 for a function @xmath779 . we then have the following limit distribution law : [ thnonrescaled ] let @xmath780 satisfy . then @xmath781 in particular , by taking the function @xmath782 as @xmath783 , we have @xmath784 if @xmath70 is in the interior @xmath60 of the polytope @xmath23 , we have @xmath785 [ tnonrescaled ] the proof of is the same as that for and , and follows from and lemma [ volume22 ] . thus , we give a proof only for . first of all , we note that the set @xmath534 introduced in section [ massasym ] is dense in @xmath80 , and hence we may consider the volume of the set @xmath786 . let @xmath787 . we write @xmath788 . then , by proposition [ localss ] , we have @xmath789 where @xmath790 is a constant , and @xmath791 is a function of order @xmath792 . thus , we obtain @xmath793 this combined with lemma [ volume22 ] shows the assertion . our final aim is to prove theorem [ unrescal ] , which gives the asymptotic limit of the distribution function @xmath794 itself without any rescaling . since , by theorem [ pointww ] and propositions [ localss ] and [ localv ] , the monomial @xmath795 decays exponentially away from the corresponding invariant torus @xmath796 , it is obvious that @xmath794 tends to zero as @xmath33 . so , a problem is to find the decay rate of @xmath794 for any ( but fixed ) @xmath773 . [ lest1 ] let @xmath773 . let @xmath129 be a lattice point with @xmath646 and lie in a face @xmath98 with @xmath529 . let @xmath799 satisfy @xmath800 . we write @xmath801 . then , we have @xmath802 locally uniformly in @xmath90 . for every @xmath411 , we set @xmath803 then , by the argument in the proof of proposition [ localss ] , we can find positive constants @xmath804 , @xmath805 such that @xmath806 therefore , by proposition [ localss ] , we have @xmath807 with some constant @xmath790 . therefore , we obtain @xmath808 which implies the assertion . thus , to obtain the asymptotic estimate of the distribution function @xmath794 , we need to find that of the monomial @xmath795 on the ball @xmath809 of radius @xmath810 around the point @xmath611 , where @xmath809 is defined in . let @xmath411 . we denote points @xmath809 as @xmath811 with @xmath812 . then we have @xmath813 where the hessian @xmath814 of the function @xmath669 on @xmath815 at the point @xmath611 is given by [ logest ] by a taylor expansion , we have @xmath816 for @xmath812 , where @xmath817 . in particular , we have @xmath818 > from this and the estimate in proposition [ localss ] , the assertion follows for @xmath819 . for @xmath820 , precisely the same argument replacing @xmath669 by @xmath537 or @xmath821 with theorem [ pointww ] and proposition [ localv ] will show the assertion . _ proof of theorem [ unrescal ] . _ for every @xmath90 , we set @xmath822 then , by lemma [ lest1 ] , there is a constant @xmath411 ( depending on a fixed @xmath773 ) such that @xmath823 . let @xmath824 . then , by lemma [ logest ] , @xmath825 if and only if @xmath826 where @xmath827 is a function of order @xmath828 uniformly in @xmath829 with @xmath830 , and @xmath715 is define in . this is equivalent to the following estimate : @xmath831 therefore , by lemma [ volume22 ] , we have @xmath832 where we set , for simplicity , @xmath833 . changing the variables @xmath834 to @xmath829 , we have @xmath835 a taylor expansion gives @xmath836 with the error term @xmath837 uniformly for @xmath812 . thus , we obtain @xmath838 note that @xmath839 while @xmath840 as @xmath33 . the function @xmath827 tends to zero uniformly in @xmath812 . therefore , we conclude , for a fixed @xmath90 , @xmath841 by and the well - known formula for the volume of the unit disk in @xmath842 , we conclude the assertion . d. a. hejhal , on eigenfunctions of the laplacian for hecke triangle groups , _ emerging applications of number theory ( minneapolis , mn , 1996 ) , _ , i m a vol . math.appl . 109 , springer - verlag , new york , 1999 , pp . 291315 . b. shiffman , t. tate and s. zelditch , harmonic analysis on toric varieties , to appear in the greene festschrift volume , contemporary mathematics , amer . soc , providence , ri ( preprint : arxiv.org/math.cv/0303337 ) . yau , open problems in geometry , _ differential geometry : partial differential equations on manifolds ( los angeles , ca , 1990 ) _ , proc.sympos . pure math . , 54 , part 1 , amer . soc . , providence , ri , 1993 , pp .

we present new lightcurve measurements of comet 10p / tempel 2 carried out with lowell observatory s discovery channel telescope in early 2013 when the comet was at aphelion . these data represent some of the first science obtained with this new 4.3-m facility . with tempel 2 having been observed to exhibit a small but ongoing spin - down in its rotation period for over two decades , our primary goals at this time were two - fold . first , to determine its current rotation period and compare it to that measured shortly after its most recent perihelion passage in 2010 , and second , to disentangle the spin - down from synodic effects due to the solar day and the earth s orbital motion and to determine the sense of rotation , i.e. prograde or retrograde . at our midpoint of 2013 feb 24 , the observed synodic period is 8.948.001 hr , exactly matching the predicted prograde rotation solution based on 2010 results , and yields a sidereal period of the identical value due to the solar and earth synodic components just canceling out during the interval of the 2013 observations . the retrograde solution is ruled out because the associated sidereal periods in 2010 and 2013 are quite different even though we know that extremely little outgassing , needed to produce torques , occurred in this interval . with a definitive sense of rotation , the specific amounts of spin - down to the sidereal period could be assessed . the nominal values imply that the rate of spin - down has decreased over time , consistent with the secular drop in water production since 1988 . our data also exhibited an unexpectedly small lightcurve amplitude which appears to be associated with viewing from a large , negative sub - earth latitude , and a lightcurve shape deviating from a simple sinusoid implying a highly irregularly shaped nucleus .

1309.2944

comet 10p / tempel 2 is the second largest jupiter - family comet in existence only 28p / neujmin 1 is known to be larger and is comparable in size to 1p / halley . unlike halley , however , tempel 2 has a very small fraction of its surface that is still active ( @xmath11% ) and is therefore one of the most highly evolved comets . these two properties have made it possible to study its nucleus at both larger heliocentric distances @xcite and at favorable perihelion passages @xcite . it also exhibits a strong asymmetry about perihelion in brightness @xcite and production rates due to the seasonal effects associated with the orientation of the rotation axis and a small source region near one pole that dominates the outgassing @xcite . of particular interest here is that tempel 2 is one of the first comets to exhibit not just a change in its rotational period @xcite but a progressive spin - down over multiple apparitions . our lightcurve measurements obtained at the 1999 apparition gave a period of [email protected] hr , about 32 s longer than the value of [email protected] hr two orbits earlier in 1988 based on a reanalysis of all available data @xcite . we followed this with further measurements in 2010 , finding a period of [email protected] hr , thereby strongly implying the spin - down is caused by progressive torquing associated with the outgassing from the near - polar source region @xcite . the situation , however , is somewhat more complicated than it first appears . while the measured spin - down was the same between 1988 and 1999 , and between 1999 and 2010 , the observations in the earlier apparitions were obtained several months prior to perihelion while the observations in 2010 were obtained between two and six months following perihelion . thus a somewhat larger change in period was expected in the second pairing if the amount of torque remained the same at every perihelion passage . while our 2010 result might be explained by a secular decrease in outgassing that we observed during the past quarter century , it remained unclear what period tempel 2 ultimately attained as it retreated from the sun in 2011 and activity dropped off . additionally , all of the reported values for its rotation period are synodic rather than sidereal . while some evidence has suggested that tempel 2 is in prograde rotation @xcite , a reanalysis of the same data gave conflicting results @xcite . since the spin - down of the nucleus adds a further complication to extracting the sense of rotation from varying synodic effects , we considered the direction of rotation as being unresolved , resulting in a pair of possible sidereal periods for each synodic result . for example , the 2010 synodic value of [email protected] hr corresponds to a prograde sidereal solution of [email protected] hr or a retrograde solution of [email protected] hr , where these offsets are asymmetric due to the earth s orbital motion . since the difference between synodic and sidereal varies with orbital position and observer geometries , this results in an added ambiguity to the specific amount of torque delivered each apparition . fortunately , the sense of rotation can be resolved with additional measurements of tempel 2 at a more distant location in its orbit when the synodic / sidereal difference is much smaller than near perihelion . with these motivations , and with access to our new 4-meter class discovery channel telescope ( dct ) , we decided to again measure tempel 2 s rotation period but this time near aphelion . the specific timing of the observations was based not only on the observing geometries but also required waiting for the comet to move sufficiently far away from the galactic plane . the initial night of planned observation also proved to be the first successful night of science operations for dct , still early in its commissioning phase . we present here lightcurve data obtained over a 3-month interval , and the resulting derived period along with the implications of this result . lowell observatory s discovery channel telescope saw first light in april 2012 , less than 7 years after ground - breaking . the telescope , with a 4.3-m primary , is located near happy jack , az , at an elevation of 2337 meters ( cf . site testing yielded a median seeing of 0.84 arcsec @xcite and delivered images already routinely exhibit fwhm of @xmath20.8 arcsec when atmospheric conditions are good . the first instrument designed for dct is the large monolithic imager ( lmi ) , which began service in september 2012 . containing a newly developed [email protected] ccd by e2v , this camera yields a 12.3 arcmin on a side field - of - view at the f/6.1 ritchey - chretien focus @xcite . our observations were obtained using 2@xmath32 on - chip binning , resulting in a pixel scale of 0.240 arcsec , and we used a kron - cousins r band filter throughout . successful and useful observations were acquired on two nights in january 2013 , one in march , and two in april ; planned observations in february were snowed out while limited observations in may were interrupted by technical problems ( including a regional power failure ) making the remaining data useless for this project . observational circumstances for each of the five useful nights are given in table [ table1 ] , including heliocentric distance ( @xmath4 ) , geocentric distance ( @xmath5 ) , and phase angle ( @xmath6 ) . the january and march runs each immediately followed large storms , resulting in quite poor and variable seeing ( @xmath21.02.4 arcsec ) , while conditions were better in april ( @xmath20.91.3 arcsec ) ; image quality additionally degraded at high airmasses as we attempted to maximize the temporal coverage each night . as expected during the very early science phase of a new telescope undergoing commissioning , a variety of small issues emerged and some affected the observations . in particular , pointing map problems in january required us to use the sidereal rate and manually guide on stars rather than track at tempel 2 s rate of motion . four - minute integrations were used , resulting in 1.6 arcsec of trailing of the comet . we were able to use our preferred mode of tracking at the comet s rate of motion on the other runs , resulting in trailed comparison stars , though no tracking was actually required nor performed in april because the comet was within days of its stationary point . with a shorter exposure of 150 s and the slower rate of motion of the comet as it approached its stationary point , stellar trailing was quite small , 0.6 arcsec , in march while the comet trailed by @xmath10.1 arcsec in april . the choice of the aperture size for photometric extractions was , as usual , a compromise between a variety of factors , including trailing and seeing as just described , along with the contrast of the object to the sky , and how crowded the field was . the faintness of the comet and the occasional close passage of field stars argued for a small aperture while the other factors pushed for larger sizes , especially in january . fortunately , as discussed below , no coma was detected and so no removal was necessary . we ultimately chose to minimize the resulting scatter in tempel 2 s lightcurve each night , which resulted in aperture radii of 11 pixels ( 2.6 arcsec ) in january , 8 pixels ( 1.9 arcsec ) in march , and 6 pixels ( 1.4 arcsec ) in april ; sky measurements were obtained from the median value in an annulus between 30 and 60 pixels ( 7.214.4 arcsec ) . the effects of changing seeing were directly compensated for by using identical apertures for on - frame comparison stars ( see below ) , while the differential effects associated with trailing were minimized with these relatively large extraction apertures . calibration was performed using standard techniques with the interactive data language software package for removing bias and applying median twilight flat fields . an extinction coefficient of 0.098 mag per unit airmass was determined on jan 13 , and was applied to all of each nights data as a first - order correction for airmass . a minimum of seven comparison stars were used each night ; the relatively slow motion of the comet at aphelion coupled with the large field of view of lmi meant these stars remained in the field throughout the night . remaining variations from frame to frame due to cirrus , changing seeing , and second - order airmass effects were compensated for by determining and applying the median value of the deviations of all the comparison stars . in practice , though the resulting standard deviations for the comparison stars were quite small , the scatter for the much fainter comet became unacceptable when this correction was larger than 0.5 mag , and we therefore excluded data above this threshold . we additionally excluded frames for which the comet was adversely affected by nearby stars , cosmic rays , etc . the uncertainties on each resulting measurement of tempel 2 were based on the photometric uncertainties combined with the standard deviations of the comparison star corrections . absolute flux calibrations were applied by using sdss r filter catalog values of each of our comparison stars . the average nightly uncertainty in the absolute flux calibrations was 0.06 mag , which encompasses both the uncertainty in the catalog magnitudes themselves and color effects in the extinction correction . the resulting nightly instrumental correction varied by as much as 0.3 mag because several nights had no interval when the sky was photometric . while landolt fields were observed on the few photometric occasions , these measurements were too sparse to significantly improve the absolute calibration but did confirm that our calibration was good to better than 0.1 mag . assuming a solar color , we applied an average offset of 0.22 mag to convert from sdss r to cousins r. final lightcurve measurements and uncertainties of all 422 useful data points ( as defined above ) are listed in table [ table2 ] , as well as the observed mid - times for each image . the reduced lightcurves are displayed in figure [ fig1 ] as a function of time for each of the five nights . immediately evident is the large decrease in brightness during the apparition that will be shown later to be simply due to the combination of increasing distance and increasing phase angle . gaps in the data on jan 12 and the apparent late start on jan 13 are due to clouds and/or the comet passing too close to a star . series of images having the best effective seeing from jan 13 and again from apr 6 were stacked to search for evidence of a coma ; none was detectable on either night . comparison of the wings of the comet s radial profile to stellar radial profiles on apr 4 and apr 6 ( when the non - sidereal motion was negligible ) suggests than any non - stellar component had an upper limit of 5% of the total cometary flux . therefore no coma removal was required ( or possible ) from these observations , unlike observations near perihelion . to remove the observed secular trend , we next performed a standard asteroidal normalization for these nuclear magnitudes to absolute values by correcting for heliocentric and geocentric distances and to zero phase angle , i.e. from the reduced magnitude , @xmath7 , to @xmath7(1,1,0 ) . our first attempts at normalization used the same linear phase function that we applied in our previous papers @xcite . this failed , however , to yield consistent brightnesses among the three observing runs no matter what phase coefficient was used . in retrospect , this was not surprising due to the much smaller phase angles involved in 2013 and the typical asteroidal non - linear opposition effect . rather than empirically determining the needed curvature , we decided to use a model phase curve that best matched c - type asteroids , those with physical characteristics such as albedo most similar to comet nuclei . fitting and interpolating the values from table iv of @xcite , we computed phase adjustments ( @xmath5@xmath8 ) for each night as given in table [ table1 ] . the resulting @xmath7(1,1,0 ) values at lightcurve maxima were in remarkable agreement , indicating that the phase function for tempel 2 s nucleus is indeed well - represented by c - type asteroids , and no further adjustments were applied . we continue to use our interactive period search routine for period determinations . as described in more detail in @xcite , phase plots are instantaneously updated as we systematically step through possible periods and examine by eye the tightness of resulting lightcurves . data are color - coded by date , making it easy to see small offsets in either brightness or phase throughout the phased curve . this is particularly useful when the intrinsic scatter due to photometric uncertainties is larger than usual , as we have here ; a slightly incorrect period will only have a very slight increase in scatter and most period search algorithms will not `` see '' that one day is offset from another . in fact , we also used the phase dispersion minimization method @xcite and , while this gave the same result , its uncertainty is much greater than what can be easily estimated from the color - coded phase plots , i.e. when clear shifts between observing runs is evident . note that standard routines can also misinterpret changes in the shape of the lightcurve , another factor here . our overall best observed period solution was [email protected] hr , and the corresponding phase plot is shown in the top panel of figure [ fig2 ] . since there is no obvious preferred zero point for phasing , we choose 2013 jan 0.0 as this placed zero phase away from interesting features and the @xmath5@xmath9 values correspond to the day of the year ; prior to phasing , the @xmath5@xmath9 values were first corrected for light travel time from the observed mid - times in table [ table1 ] . as with previous apparitions , we have a double - peaked light curve where the two maxima are nearly identical in brightness but the minima are quite different , with the deeper minimum nearly `` v''-shaped while the smaller minimum varied in depth and shape with changing viewing geometry . unfortunately , our phase coverage is incomplete in april and the shallower minimum is not sampled , so we can not determine if the change in depth from january to march continued . also evident is a @xmath20.02 mag shift of much ( but not all ) of the jan 12 lightcurve from that of a day later on january 13 , for which we have been unable to identify a cause . while this has no effect on the the period solution , it alters the perception of the shape for the first half of the lightcurve . we also phased subsets of the data to look for evidence of a change in the observed period associated with the changing viewing geometries . as expected , phasing jan / apr yielded the same result as the entire data set , while a minor change was evident for the other two pairings jan / mar gave a slightly smaller value ( [email protected] hr ) and mar / apr gave a slightly larger value ( [email protected] hr ) with the higher uncertainty in the latter value directly caused by the shorter interval . within the uncertainties , this increase matches the predicted change of @xmath100.001 hr , caused simply by the earth s orbital motion between our earliest midpoint pairing and the latest ; the solar component to the synodic period near aphelion is both small and unchanging . these calculations all use the pole orientation from @xcite of r.a . = 162@xmath11 and dec . = @xmath1058@xmath11 ( and the diametrically opposite solution for retrograde rotation ) , but similar pole positions , such as from @xcite , would yield the same results . as discussed in the introduction , one goal of these observations was to uniquely discriminate between prograde and retrograde rotation for tempel 2 s nucleus . the sense of rotation always determines the sign of the solar component of the synodic period while the earth s component is unaffected by the direction of rotation . at the midpoint of our ensemble of measurements for 2013 , the solar and earth components are coincidentally nearly identical and in the prograde case have opposite signs that cancel out , resulting in the sidereal period being the same as the synodic ( [email protected] hr ) , while in the retrograde case they compound yielding a sidereal value of [email protected] hr . to determine which solution is correct , we next re - examine the solutions from 2010 . our overall observed ( synodic ) value was [email protected] hr with a corresponding midpoint time of 2010 oct 25 or 112 days following perihelion @xcite . based on our water production rates throughout the apparition , over 95% of the total outgassing had taken place by this date . with an observed change in period of about @xmath100.004 hr per perihelion passage , we concluded that further torquing late in the 2010 apparition should be negligible and that the sidereal period in 2013 would therefore be unchanged from the late 2010 result . the corresponding sidereal values at the 2010 midpoint are 8.948 hr ( prograde ) and 8.955 hr ( retrograde ) . note that these values are slightly different from those listed in @xcite which erroneously only included the solar component but not the earth s component . with the expectation that the sidereal value should be the same in early 2013 as in late 2010 , and the possible sidereal values in 2013 being [email protected] hr ( prograde ) and [email protected] hr ( retrograde ) , we conclude that tempel 2 _ must _ be in prograde rotation , with a sidereal period in late 2010 and in early 2013 of 8.948 hr . as further evidence that tempel 2 is in a prograde rotation , we also examined the retrograde scenario . if it was the retrograde case , then as already stated the sidereal period in 2010 would have been 8.955 hr . since the difference between the retrograde synodic and sidereal periods in 2013 would be @xmath120.002 hr , we should have measured a period of 8.953 hr . phasing with this value is shown in the bottom panel of figure [ fig2 ] and it is clear that this solution does not yield a viable lightcurve , therefore the comet must instead be in prograde rotation . looking back at the earlier apparitions requires disentangling the synodic effects from the actual changes in period due to torquing . an early attempt to derive the sidereal period from the changing synodic period during 1988 @xcite suffered from several problems , including the claim of a smoothly decreasing period which could not be reproduced during a detailed reanalysis @xcite and our new determination that his synodic / sidereal modeling only included the earth s motion but neglected the dominant solar component . having now definitively determined the sense of rotation as prograde , we can reexamine prior apparitions . the computed sidereal periods for each epoch are given in table [ table3 ] ; for completeness and to illustrate the asymmetries of the pro- and retrograde solutions with respect to the observed synodic periods , we also tabulate the retrograde values . again , these values differ somewhat from the values given in @xcite since the earth component is now included along with the solar component , thereby causing the asymmetry . our conclusions regarding a decrease in the amount of spin - down remain unchanged from @xcite , with sidereal periods before perihelia in 1988 of 8.931 hr and in 1999 of 8.939 hr , for an average spin - down by 0.004 hr per apparition . the same rate of spin - down would have predicted a sidereal period of 8.951 hr in late 2010 , a larger change than we observe . we continue to think the most likely cause of this is the decrease in total water production from 1988 to 1999 and to 2010 , with lower production rates providing a smaller amount of torquing . however , within the uncertainties of the period determinations , the data are also consistent with no change in the rate of spin - down . the agreement between the late 1994 period with early 1999 also remains , with both yielding a sidereal period of 8.939 hr during an interval for which there was not an intervening perihelion passage . our 2013 lightcurves of tempel 2 revealed two unexpected findings . first , the amplitude within the lightcurve was smaller than we had assumed it would be and , second , the peak brightness was higher than predicted from earlier apparitions even after compensating for a non - linear phase function . after considering a variety of possibilities , such as removing too much or too little coma for differing apparitions or systematic effects when different filters were used , we think we have arrived at a self - consistent explanation . our expectations for 2013 were based on assuming that tempel 2 is a prolate ellipsoid with similar dimensions for the small and intermediate axes while the long axis is about 2.1@xmath3 greater . this is based on the 1988 thermal ir lightcurve from @xcite and that the sub - earth latitude then varied between @xmath121@xmath13 and @xmath129@xmath13 , i.e. the comet was viewed in 1988 nearly equator - on using our preferred pole solution . in early 2013 , the sub - earth latitude was @xmath1247@xmath13 , yielding a predicted amplitude of 0.4 mag , significantly greater than the observed value of @xmath20.2 mag . if , however , the intermediate axis is actually intermediate in length rather than the same as the small axis , then the total cross - section will appear to increase as one views from more pole - on and the amplitude will become even smaller than the simple function of the cosine of the sub - earth latitude ( the intermediate axis begins to dominate over the small axis ) . note that in this tri - axial case , it is the ratio of the long axis to the intermediate axis that determines the amplitude for the equator - on view . therefore , to explain the 1988 amplitude , the intermediate axis remains the same as originally assumed for the prolate case , and it is the short axis which must be even shorter . for our observed amplitude in 2013 , this requires the short axis to be less than one - third that of long axis , while the intermediate axis remains at about one - half of the long axis . this implies an even more elongated nucleus than the tri - axial solution with ratios of @xmath14 proposed by @xcite . in addition to explaining the observed amplitude , our tri - axial solution also naturally explains the increased brightness at lightcurve maxima in 2013 as being due to the increased total cross - section relative to the equator - on view . we strongly suspect , however , that the shape of the nucleus significantly departs from a tri - axial ellipsoid for several reasons . the lightcurve shape is not sinusoidal , but rather has one v-shaped minimum and one rounded minimum . also , the shapes and depths of the minima change rapidly in 2013 with only a small change in viewing geometries even though the sub - earth and sub - solar latitudes did not vary . finally , as discussed below , the amplitudes are quite different when the comet was viewed from high positive latitudes as compared to high negative latitudes . thus , we conclude that the nucleus must have large - scale protuberances . to first - order , the tri - axial ellipsoid explains most of the brightness and amplitudes measured at the other apparitions . most similar to 2013 were the circumstance in early 1987 with a sub - earth latitude of @xmath1252@xmath13 , and where jewitt & meech ( 1988 ) measured an amplitude of @xmath20.3 mag and a peak magnitude essentially identical to our 2013 value once adjustments are made for filter and phase angle . the 1999 apparition was nearly identical to that of 1988 , and the lightcurve characteristics are also essentially the same . the peak brightness in 1994 @xcite is also consistent with 1988 and 1999 but the amplitude is somewhat smaller ( by @xmath20.10.2 mag ) with no clear explanation since the sub - earth latitude was close to the equator . most different were our results from 2010 @xcite , where the peak brightness is @xmath20.20.3 mag fainter while the amplitude is larger than expected for a sub - earth latitude near @xmath1040@xmath13 . while an over removal of coma might partially explain both aspects , our methodology was the same used for the 1999 apparition which has no such issues , leading us to suspect that tempel 2 s actual shape characteristics are another significant factor ; this is supported by the fact that only in 2010 did the sub - earth latitude have a high positive value ( @xmath1040@xmath13 ) while all other apparitions were either near the equator ( 1988 , 1994 , 1999 ) or at high negative values ( 1987 , 2013 ) . the observations reported here represent some of the first science collected with lowell observatory s new discovery channel telescope . despite a few problems associated with a facility still in the early stages of commissioning , we were able to easily fulfill our major science objectives regarding comet tempel 2 s rotational state , in spite of a smaller lightcurve amplitude than expected . we obtained a precise measurement of its current rotation period that , when combined with our 2010 measurements , yielded a definitive determination of its sense of rotation . this result in turn allowed us to determine the correct sidereal period associated with observed synodic values at each prior apparition . for lowell observatory and its partners , a new era has begun in which projects requiring time - intensive or long - duration observations can be accomplished for targets much fainter than have ever been possible . in the specific case of comet nuclei , we can now measure lightcurves of comet nuclei far from perihelion . the measured periodicity in a nucleus lightcurve depends on the changing cross - section both as illuminated by the sun and as seen from the earth , and in turn depends on the solar day as well as the earth s motion with respect to the comet and the sun . disentangling these effects from the change in the physical rotation due to torquing had been problematic at best . however , the relative importance of the solar component and earth s motion component are quite different far from the sun as compared to near perihelion . because of this , we planned and executed these new observations . the resulting synodic period , [email protected] hr , matched that expected for the prograde scenario from 2010 while the nominal period associated with the retrograde solution is clearly ruled out . we therefore conclude that tempel 2 s rotation is prograde with respect to both the ecliptic and to its orbital plane . the sidereal period , [email protected] hr , is coincidentally identical to the synodic value as the solar and earth components of the synodic period just cancel out during the interval of our 2013 observations . the new data also confirm that no additional spin - down took place following our late 2010 measurements , consistent with the water production curve showing a near - cessation of activity by that time . this lack of a change in the sidereal period very late in the 2010 apparition thereby implies that the smaller change in the period per perihelion passage between 1999 and 2010 as compared to 1988 to 1999 is real ( although a constant rate of spin - down can not be excluded within the uncertainties ) . note that the inferred changes match those given by @xcite for the prograde case . thus our hypothesis that a decrease in torquing is associated with the secular decrease in water production from 1988 to 2010 is probably correct . we note , however , that there is no strong evidence for a long - term decrease in activity over the past century , and the secular drop over the past two decades may reverse as the surface is eroded and varying proportions of ice are exposed from successive perihelion passages . similar to prior apparitions , the light curve is double - peaked with near - equal maxima but quite differently shaped minima . the shape and depth of one of the minima also changed over the three - month interval , another trait observed in the past . both characteristics directly indicate that the shape of tempel 2 s nucleus is not a simple prolate ellipsoid or even a triaxial ellipsoid but must instead have large - scale protuberances to cause such clear changes in the lightcurve from such a small change in viewing geometries ( @xmath120@xmath13 ) and essentially no change in sub - earth and sub - solar latitudes . the unexpectedly small amplitude we measured in the lightcurve ( @xmath20.2 mag ) appears to also be due to peculiarities of the nucleus shape our value is in good agreement with the only other data obtained at a similar sub - earth latitude @xcite , while data taken from near - equator on or from the other hemisphere all exhibit a much larger amplitude . the data further suggest that the length of the short axis is substantially shorter than that of the intermediate axis , though the shape is probably not as extreme as that of comet 103p / hartley 2 as imaged from epoxi @xcite . thus a more complete story of the properties of tempel 2 s nucleus is beginning to emerge . while determining the detailed shape must continue to await an upclose view tempel 2 was the proposed target of the planned comet rendezvous asteroid flyby mission more than two decades ago additional numerical modeling and new investigations from afar can continue to reveal clues to its physical structure and evolving behavior . we thank t. farnham for calculations of the offsets between the prograde and retrograde synodic periods with respect to a given sidereal period . we gratefully acknowledge the assistance of m. fendrock with the first observing run and preliminary analyses , and a. venetiou , m. sweaton , j. sanborn , r. winner , and s. strosahl for their successful operations during the early commissioning phase of the dct , thereby making these observations possible . we also thank r. millis , w. l. putnam , and j. hendricks for their vision and support which allowed the dct to become a reality , and all members of the dct and lmi teams . these results made use of lowell observatory s discovery channel telescope , supported by lowell , discovery communications , boston university , the university of maryland , and the university of toledo . m.m.k . is grateful for office space provided by the university of maryland department of astronomy and johns hopkins university applied physics laboratory while working on this project . the lmi instrument was funded by the national science foundation via grant ast-1005313 . this research has been supported by nasa s planetary astronomy program ( grant nnx09ab51 g ) . , t. a. , dunham , e. w. , bright , l. p. , & corson , c. 2004 , in ground - based telescopes . j. m. oschmann , jr . , proc . of spie , 5489 , 196 , t. a. , dunham , e. w. , nye , r. a. , chylek , t. , & oliver , r. c. 2012 , in ground - based and airborne telescopes iv . , eds . l. m. stepp , r. gilmozzi , h. j. hall , proc . of spie , 8444 , 844451 , s. e. , bida , t. a. , chylek , t. , et al . 2012 , in ground - based and airborne telescopes iv . , l. m. stepp , r. gilmozzi , h. j. hall , proc . of spie , 8444 , 844419 , p. , dunham , e. w. , bida , t. a. , et al . 2013 , aas meeting # 221 , poster # 345.02 , b. e. a. , & ferrin , i. 1996 , icarus , 123 , 463 lcccccccccccccc jan 12&3:3413:29&@xmath1012.4&4.697&3.770&4.5&0.42&203.8&309.3&clouds + jan 13&3:3513:27&@xmath1013.4&4.698&3.765&4.3&0.41&204.1&309.4&intermittent clouds + mar 11&2:5110:38&@xmath1070.3&4.709&3.970&8.8&0.64&225.4&313.1&cirrus + apr 4&3:018:52&@xmath1094.2&4.707&4.293&11.6&0.76&234.6&314.6&cirrus + apr 6&2:568:45&@xmath1096.2&4.706&4.323&11.8&0.76&235.5&314.8&clouds + [ table1 ] jan 12&4.188&20.025&0.039&&&jan 13&7.857&20.109&0.013&&&mar 11&2.983&20.433&0.022&&&mar 11&7.064&20.482&0.023 + jan 12&6.337&20.216&0.031&&&jan 13&7.931&20.110&0.013&&&mar 11&3.040&20.456&0.023&&&mar 11&7.112&20.458&0.023 + jan 12&6.486&20.209&0.027&&&jan 13&8.005&20.126&0.012&&&mar 11&3.090&20.434&0.021&&&mar 11&7.161&20.465&0.022 + jan 12&6.635&20.230&0.027&&&jan 13&8.079&20.136&0.013&&&mar 11&3.199&20.436&0.022&&&mar 11&7.209&20.446&0.021 + jan 12&6.710&20.224&0.023&&&jan 13&8.153&20.114&0.013&&&mar 11&3.250&20.434&0.020&&&mar 11&7.258&20.455&0.021 + jan 12&6.786&20.247&0.026&&&jan 13&8.226&20.119&0.013&&&mar 11&3.300&20.392&0.020&&&mar 11&7.307&20.444&0.022 + jan 12&6.861&20.208&0.030&&&jan 13&8.301&20.144&0.013&&&mar 11&3.350&20.386&0.020&&&mar 11&7.355&20.449&0.023 + jan 12&6.936&20.237&0.029&&&jan 13&8.374&20.152&0.012&&&mar 11&3.413&20.407&0.021&&&mar 11&7.403&20.478&0.025 + jan 12&7.011&20.259&0.032&&&jan 13&8.448&20.177&0.013&&&mar 11&3.464&20.387&0.020&&&mar 11&7.451&20.455&0.023 + jan 12&9.159&20.054&0.016&&&jan 13&8.522&20.170&0.013&&&mar 11&3.514&20.398&0.021&&&mar 11&7.503&20.446&0.021 + jan 12&9.235&20.112&0.015&&&jan 13&8.597&20.173&0.013&&&mar 11&3.576&20.396&0.022&&&mar 11&7.552&20.459&0.021 + jan 12&9.309&20.077&0.014&&&jan 13&8.676&20.166&0.012&&&mar 11&3.630&20.403&0.023&&&mar 11&7.600&20.431&0.020 + jan 12&9.384&20.079&0.014&&&jan 13&8.750&20.180&0.012&&&mar 11&3.681&20.389&0.023&&&mar 11&7.648&20.417&0.019 + jan 12&9.458&20.064&0.013&&&jan 13&8.824&20.169&0.013&&&mar 11&3.736&20.393&0.022&&&mar 11&7.696&20.421&0.020 + jan 12&9.533&20.034&0.013&&&jan 13&8.898&20.182&0.013&&&mar 11&3.793&20.395&0.022&&&mar 11&7.746&20.429&0.022 + jan 12&9.607&20.030&0.013&&&jan 13&9.195&20.213&0.013&&&mar 11&3.846&20.413&0.023&&&mar 11&7.795&20.435&0.022 + jan 12&9.681&20.067&0.013&&&jan 13&9.268&20.220&0.013&&&mar 11&3.895&20.394&0.022&&&mar 11&7.843&20.406&0.022 + jan 12&9.756&20.076&0.014&&&jan 13&9.342&20.224&0.014&&&mar 11&3.945&20.411&0.022&&&mar 11&7.891&20.461&0.023 + jan 12&9.830&20.075&0.013&&&jan 13&9.416&20.213&0.013&&&mar 11&3.994&20.396&0.020&&&mar 11&7.939&20.445&0.023 + jan 12&9.905&20.089&0.015&&&jan 13&9.490&20.216&0.013&&&mar 11&4.045&20.387&0.021&&&mar 11&7.990&20.418&0.022 + jan 12&9.979&20.107&0.015&&&jan 13&9.564&20.235&0.013&&&mar 11&4.098&20.390&0.021&&&mar 11&8.039&20.443&0.021 + jan 12&10.055&20.116&0.014&&&jan 13&9.638&20.193&0.013&&&mar 11&4.147&20.415&0.022&&&mar 11&8.087&20.395&0.022 + jan 12&10.130&20.116&0.016&&&jan 13&9.934&20.176&0.013&&&mar 11&4.196&20.424&0.021&&&mar 11&8.135&20.431&0.020 + jan 12&10.428&20.114&0.022&&&jan 13&10.008&20.181&0.012&&&mar 11&4.244&20.423&0.022&&&mar 11&8.184&20.418&0.020 + jan 12&10.576&20.136&0.023&&&jan 13&10.082&20.167&0.013&&&mar 11&4.294&20.451&0.022&&&mar 11&8.237&20.430&0.021 + jan 12&10.651&20.132&0.016&&&jan 13&10.156&20.172&0.012&&&mar 11&4.751&20.434&0.024&&&mar 11&8.285&20.391&0.021 + jan 12&10.725&20.128&0.017&&&jan 13&10.230&20.126&0.012&&&mar 11&4.801&20.451&0.025&&&mar 11&8.334&20.443&0.022 + jan 12&10.800&20.139&0.016&&&jan 13&10.304&20.132&0.012&&&mar 11&4.852&20.473&0.025&&&mar 11&8.382&20.400&0.022 + jan 12&10.873&20.110&0.017&&&jan 13&10.380&20.133&0.012&&&mar 11&4.907&20.478&0.027&&&mar 11&8.430&20.422&0.024 + jan 12&10.948&20.106&0.016&&&jan 13&10.454&20.136&0.012&&&mar 11&4.956&20.494&0.027&&&mar 11&8.496&20.394&0.023 + jan 12&11.022&20.136&0.015&&&jan 13&10.528&20.150&0.012&&&mar 11&5.005&20.504&0.026&&&mar 11&8.545&20.407&0.023 + jan 12&11.096&20.159&0.015&&&jan 13&10.601&20.119&0.012&&&mar 11&5.054&20.491&0.027&&&mar 11&8.593&20.440&0.024 + jan 12&11.170&20.131&0.014&&&jan 13&10.675&20.104&0.012&&&mar 11&5.102&20.511&0.027&&&mar 11&8.641&20.419&0.024 + jan 12&11.245&20.114&0.014&&&jan 13&10.750&20.103&0.012&&&mar 11&5.151&20.485&0.025&&&mar 11&8.689&20.464&0.025 + jan 12&11.319&20.106&0.014&&&jan 13&10.824&20.088&0.012&&&mar 11&5.199&20.504&0.025&&&mar 11&8.837&20.473&0.026 + jan 12&11.394&20.158&0.015&&&jan 13&10.898&20.067&0.012&&&mar 11&5.248&20.535&0.024&&&mar 11&8.886&20.468&0.026 + jan 12&11.544&20.145&0.023&&&jan 13&10.972&20.070&0.012&&&mar 11&5.296&20.509&0.025&&&mar 11&8.935&20.472&0.027 + jan 12&11.916&20.149&0.023&&&jan 13&11.046&20.081&0.012&&&mar 11&5.345&20.565&0.030&&&mar 11&8.983&20.488&0.027 + jan 12&12.065&20.111&0.023&&&jan 13&11.121&20.063&0.012&&&mar 11&5.393&20.543&0.027&&&mar 11&9.031&20.455&0.028 + jan 12&12.215&20.077&0.016&&&jan 13&11.417&20.056&0.012&&&mar 11&5.441&20.526&0.028&&&mar 11&9.079&20.499&0.030 + jan 12&12.363&20.119&0.014&&&jan 13&11.490&20.038&0.012&&&mar 11&5.515&20.539&0.029&&&mar 11&9.129&20.447&0.032 + jan 12&12.688&20.045&0.016&&&jan 13&11.565&20.048&0.012&&&mar 11&5.564&20.526&0.028&&&mar 11&9.178&20.494&0.032 + jan 13&5.839&20.120&0.019&&&jan 13&11.640&20.059&0.012&&&mar 11&5.612&20.508&0.023&&&mar 11&9.226&20.458&0.029 + jan 13&5.987&20.140&0.015&&&jan 13&11.713&20.051&0.012&&&mar 11&5.660&20.527&0.025&&&mar 11&9.274&20.501&0.031 + jan 13&6.061&20.139&0.014&&&jan 13&11.787&20.043&0.012&&&mar 11&5.708&20.520&0.025&&&mar 11&9.323&20.462&0.032 + jan 13&6.136&20.150&0.014&&&jan 13&11.861&20.057&0.012&&&mar 11&5.758&20.496&0.022&&&mar 11&9.373&20.494&0.034 + jan 13&6.211&20.125&0.014&&&jan 13&11.935&20.043&0.012&&&mar 11&5.806&20.497&0.024&&&mar 11&9.421&20.504&0.037 + jan 13&6.285&20.119&0.014&&&jan 13&12.009&20.062&0.013&&&mar 11&5.951&20.476&0.024&&&mar 11&9.469&20.536&0.038 + jan 13&6.359&20.097&0.013&&&jan 13&12.083&20.063&0.013&&&mar 11&6.002&20.541&0.024&&&mar 11&9.518&20.499&0.036 + jan 13&6.433&20.105&0.013&&&jan 13&12.157&20.051&0.013&&&mar 11&6.050&20.542&0.024&&&mar 11&9.566&20.537&0.038 + jan 13&6.509&20.097&0.013&&&jan 13&12.231&20.060&0.013&&&mar 11&6.099&20.529&0.025&&&mar 11&9.616&20.492&0.037 + jan 13&6.583&20.100&0.013&&&jan 13&12.305&20.071&0.013&&&mar 11&6.147&20.515&0.025&&&mar 11&9.664&20.489&0.043 + jan 13&6.658&20.102&0.013&&&jan 13&12.379&20.097&0.014&&&mar 11&6.195&20.530&0.027&&&mar 11&9.712&20.532&0.040 + jan 13&6.731&20.112&0.013&&&jan 13&12.454&20.093&0.014&&&mar 11&6.243&20.499&0.024&&&mar 11&9.760&20.596&0.042 + jan 13&6.805&20.099&0.013&&&jan 13&12.528&20.099&0.014&&&mar 11&6.292&20.552&0.024&&&mar 11&9.809&20.567&0.041 + jan 13&6.879&20.089&0.013&&&jan 13&12.602&20.096&0.015&&&mar 11&6.340&20.545&0.025&&&mar 11&9.866&20.607&0.042 + jan 13&6.953&20.074&0.013&&&jan 13&12.676&20.080&0.015&&&mar 11&6.388&20.475&0.024&&&mar 11&9.915&20.579&0.039 + jan 13&7.027&20.099&0.013&&&jan 13&12.750&20.106&0.015&&&mar 11&6.437&20.503&0.023&&&mar 11&9.963&20.599&0.041 + jan 13&7.101&20.123&0.013&&&jan 13&12.899&20.111&0.016&&&mar 11&6.675&20.486&0.038&&&mar 11&10.011&20.533&0.040 + jan 13&7.176&20.072&0.013&&&jan 13&12.973&20.119&0.018&&&mar 11&6.722&20.506&0.025&&&mar 11&10.059&20.609&0.044 + jan 13&7.250&20.085&0.013&&&jan 13&13.047&20.111&0.016&&&mar 11&6.773&20.473&0.023&&&mar 11&10.109&20.537&0.046 + jan 13&7.323&20.067&0.012&&&jan 13&13.121&20.115&0.017&&&mar 11&6.821&20.462&0.023&&&mar 11&10.157&20.592&0.048 + jan 13&7.398&20.055&0.012&&&jan 13&13.195&20.130&0.020&&&mar 11&6.869&20.478&0.024&&&mar 11&10.206&20.582&0.049 + jan 13&7.472&20.071&0.013&&&jan 13&13.269&20.159&0.020&&&mar 11&6.917&20.479&0.022&&&mar 11&10.254&20.610&0.053 + jan 13&7.546&20.098&0.013&&&jan 13&13.344&20.142&0.020&&&mar 11&6.966&20.460&0.023&&&mar 11&10.302&20.561&0.051 + jan 13&7.620&20.102&0.012&&&mar 11&2.850&20.446&0.033&&&mar 11&7.016&20.475&0.024&&&mar 11&10.354&20.610&0.054 + apr 4&3.028&20.766&0.044&&&apr 4&4.973&20.710&0.025&&&apr 4&7.177&20.829&0.031&&&apr 6&4.040&20.890&0.028 + apr 4&3.071&20.767&0.027&&&apr 4&5.021&20.732&0.026&&&apr 4&7.225&20.853&0.031&&&apr 6&4.090&20.926&0.033 + apr 4&3.127&20.775&0.028&&&apr 4&5.069&20.691&0.026&&&apr 4&7.275&20.795&0.031&&&apr 6&4.138&20.889&0.027 + apr 4&3.176&20.773&0.028&&&apr 4&5.117&20.657&0.025&&&apr 4&7.323&20.842&0.032&&&apr 6&4.186&20.859&0.026 + apr 4&3.224&20.794&0.028&&&apr 4&5.166&20.711&0.029&&&apr 4&7.371&20.794&0.033&&&apr 6&4.235&20.910&0.026 + apr 4&3.275&20.723&0.028&&&apr 4&5.220&20.719&0.028&&&apr 4&7.419&20.844&0.033&&&apr 6&4.283&20.917&0.026 + apr 4&3.323&20.706&0.025&&&apr 4&5.268&20.727&0.027&&&apr 4&7.468&20.833&0.033&&&apr 6&4.333&20.893&0.025 + apr 4&3.371&20.738&0.027&&&apr 4&5.316&20.708&0.027&&&apr 4&7.522&20.802&0.034&&&apr 6&4.381&20.855&0.024 + apr 4&3.420&20.776&0.028&&&apr 4&5.364&20.713&0.027&&&apr 4&7.570&20.830&0.036&&&apr 6&4.429&20.858&0.024 + apr 4&3.468&20.758&0.029&&&apr 4&5.412&20.689&0.026&&&apr 4&7.618&20.820&0.037&&&apr 6&4.478&20.857&0.024 + apr 4&3.517&20.722&0.028&&&apr 4&5.468&20.735&0.027&&&apr 4&7.666&20.745&0.037&&&apr 6&4.526&20.862&0.025 + apr 4&3.566&20.772&0.027&&&apr 4&5.716&20.731&0.026&&&apr 4&7.715&20.824&0.042&&&apr 6&4.575&20.854&0.025 + apr 4&3.614&20.724&0.025&&&apr 4&5.781&20.739&0.026&&&apr 4&7.764&20.766&0.042&&&apr 6&4.624&20.826&0.025 + apr 4&3.662&20.764&0.027&&&apr 4&5.829&20.732&0.026&&&apr 4&7.812&20.797&0.037&&&apr 6&4.672&20.821&0.027 + apr 4&3.711&20.759&0.026&&&apr 4&5.878&20.760&0.026&&&apr 4&7.860&20.804&0.036&&&apr 6&4.855&20.829&0.033 + apr 4&3.760&20.769&0.026&&&apr 4&5.926&20.782&0.026&&&apr 4&7.908&20.835&0.038&&&apr 6&4.952&20.811&0.030 + apr 4&3.808&20.735&0.026&&&apr 4&5.974&20.769&0.026&&&apr 4&7.957&20.814&0.038&&&apr 6&5.243&20.747&0.028 + apr 4&3.857&20.750&0.025&&&apr 4&6.023&20.758&0.027&&&apr 4&8.006&20.855&0.037&&&apr 6&5.291&20.763&0.028 + apr 4&3.905&20.728&0.026&&&apr 4&6.072&20.784&0.026&&&apr 4&8.054&20.839&0.038&&&apr 6&5.341&20.740&0.027 + apr 4&3.953&20.784&0.027&&&apr 4&6.120&20.790&0.027&&&apr 4&8.102&20.773&0.034&&&apr 6&5.389&20.794&0.029 + apr 4&4.002&20.714&0.028&&&apr 4&6.168&20.764&0.026&&&apr 4&8.151&20.781&0.035&&&apr 6&5.437&20.745&0.036 + apr 4&4.051&20.756&0.028&&&apr 4&6.216&20.786&0.026&&&apr 4&8.199&20.797&0.037&&&apr 6&5.486&20.743&0.031 + apr 4&4.099&20.758&0.026&&&apr 4&6.305&20.790&0.026&&&apr 4&8.250&20.737&0.036&&&apr 6&5.534&20.743&0.025 + apr 4&4.147&20.710&0.025&&&apr 4&6.354&20.805&0.027&&&apr 4&8.298&20.763&0.038&&&apr 6&5.583&20.735&0.025 + apr 4&4.195&20.724&0.026&&&apr 4&6.402&20.795&0.028&&&apr 4&8.346&20.741&0.038&&&apr 6&5.631&20.696&0.024 + apr 4&4.244&20.744&0.028&&&apr 4&6.450&20.819&0.028&&&apr 4&8.394&20.750&0.053&&&apr 6&5.679&20.741&0.025 + apr 4&4.292&20.710&0.025&&&apr 4&6.498&20.817&0.028&&&apr 4&8.442&20.754&0.045&&&apr 6&5.728&20.731&0.025 + apr 4&4.341&20.719&0.024&&&apr 4&6.547&20.823&0.029&&&apr 4&8.498&20.798&0.051&&&apr 6&5.776&20.716&0.025 + apr 4&4.389&20.716&0.024&&&apr 4&6.595&20.850&0.030&&&apr 4&8.546&20.756&0.048&&&apr 6&5.825&20.707&0.025 + apr 4&4.438&20.684&0.026&&&apr 4&6.644&20.851&0.029&&&apr 4&8.594&20.751&0.049&&&apr 6&5.873&20.705&0.024 + apr 4&4.488&20.698&0.023&&&apr 4&6.692&20.835&0.029&&&apr 4&8.642&20.699&0.050&&&apr 6&5.921&20.719&0.025 + apr 4&4.536&20.687&0.023&&&apr 4&6.740&20.810&0.030&&&apr 4&8.740&20.729&0.055&&&apr 6&5.970&20.704&0.026 + apr 4&4.584&20.712&0.023&&&apr 4&6.790&20.855&0.031&&&apr 6&3.029&20.856&0.134&&&apr 6&6.018&20.728&0.027 + apr 4&4.632&20.724&0.023&&&apr 4&6.839&20.817&0.031&&&apr 6&3.646&20.858&0.027&&&apr 6&6.067&20.724&0.027 + apr 4&4.680&20.719&0.023&&&apr 4&6.887&20.850&0.032&&&apr 6&3.701&20.865&0.027&&&apr 6&6.115&20.727&0.027 + apr 4&4.730&20.727&0.024&&&apr 4&6.935&20.894&0.033&&&apr 6&3.749&20.920&0.027&&&apr 6&6.164&20.725&0.027 + apr 4&4.779&20.699&0.025&&&apr 4&6.983&20.877&0.031&&&apr 6&3.797&20.875&0.027&&&apr 6&6.212&20.715&0.029 + apr 4&4.827&20.672&0.026&&&apr 4&7.032&20.824&0.029&&&apr 6&3.847&20.888&0.025&&&apr 6&6.260&20.710&0.028 + apr 4&4.875&20.684&0.025&&&apr 4&7.080&20.867&0.030&&&apr 6&3.895&20.901&0.027 & & & + apr 4&4.923&20.711&0.025&&&apr 4&7.129&20.859&0.031&&&apr 6&3.992&20.901&0.028 & & & + [ table2 ] lcccccc 1987 apr 1&&&&0.3&@xmath1252&@xmath1246 + 1988 apr 28&8.932 @xmath0 0.001&8.931&8.935&0.5 0.8&@xmath122 @xmath1210&@xmath1218 @xmath1017 + 1994 nov 22&8.939 @xmath0 0.003&8.939&8.941&0.5&@xmath123 @xmath104&@xmath102 @xmath127 + 1999 may 20&8.941 @xmath0 0.002&8.939&8.945&0.5 0.7&@xmath109 @xmath1010&@xmath121 @xmath1021 + 2010 oct 25&8.950 @xmath0 0.002&8.948&8.955&0.5 0.8&@xmath1035 @xmath1043&@xmath1040 @xmath1013 + 2013 feb 24&8.948 @xmath0 0.001&8.948&8.950&0.2&@xmath1247&@xmath1248 + [ table3 ] [ lasttable ] ) for our data plotted as a function of ut on each night . the magnitudes have had absolute calibrations , extinction corrections , and comparison star corrections applied and are given in table [ table2 ] . the symbols are defined in the legend . note that while @xmath4 remained approximately constant during the observations , _ @xmath5 _ and @xmath6 increased , causing the brightness to steadily decrease from month to month.,width=332 ] and the magnitudes are normalized to ( 1,1,0 ) as described in the text . the 8.948 hr period is the synodic period expected in 2013 based on the 2010 prograde sidereal solution found in @xcite while the 8.953 hr period is the synodic period expected in 2013 based on the 2010 retrograde sidereal solution . the 8.953 hr period is clearly incompatible with the 2013 data , thus definitively ruling out retrograde rotation.,width=332 ]

ngc1851 is surrounded by a stellar component that extends more than ten times beyond the tidal radius . although the nature of this stellar structure is not known , it has been suggested to be a sparse halo of stars or associated with a stellar stream . we analyse the nature of this intriguing stellar component surrounding ngc1851 by investigating its radial velocities and chemical composition , in particular in comparison with those of the central cluster analysed in a homogeneous manner . in total we observed 23 stars in the halo with radial velocities consistent with ngc1851 , and for 15 of them we infer [ fe / h ] abundances . our results show that : _ ( i ) _ stars dynamically linked to ngc1851 are present at least up to.5 tidal radii , supporting the presence of a halo of stars surrounding the cluster ; _ ( ii ) _ apart from the ngc1851 radial velocity - like stars , our observed velocity distribution agrees with that expected from galactic models , suggesting that no other sub - structure ( such as a stream ) at different radial velocities is present in our field ; _ ( iii ) _ the chemical abundances for the-process elements sr and ba are consistent with the-normal stars observed in ngc1851 ; _ ( iv ) _ all halo stars have metallicities , and abundances for the other studied elements ca , mg and cr , consistent with those exhibited by the cluster . the complexity of the whole ngc1851 cluster+halo system may agree with the scenario of a tidally - disrupted dwarf galaxy in which ngc1851 was originally embedded . [ firstpage ] globular clusters : general individual : ngc1851 techniques : spectroscopy

1406.0944

recent discoveries on multiple stellar populations in globular clusters ( gcs ) have revealed that some of these old stellar systems show chemical inhomogeneities , not just in the light elements involved in the hot h - burning , but also in heavier elements and in the overall metallicity . to explain the large metallicity dispersion in @xmath2 centauri it has been suggested that this gc is the remnant of a dwarf galaxy disrupted through tidal interactions with the milky way , rather than a true gc ( e.g. norris et al.1996 ; bekki & freeman 2003 ; bekki & norris 2006 ) . in this scenario , @xmath2 centauri would be the dense nucleus of a dwarf galaxy , which was cannibalised by the milky way . the recent discoveries of more @xmath2 centauri - like gcs , with internal metallicity variations ( e.g. marino et al.2009 ; da costa et al.2009 ; 2014 ; yong et al.2014 ) , support the hypothesis that at least these `` anomalous '' gcs were as massive as small galaxies able of retaining fast supernovae ejecta . the relatively high fraction of metal - richer stars in these gcs also supports the idea that they were more massive . the possible gc - dwarf galaxy connection may have consequences for near - field cosmology and the hierarchical assembly of our galaxy . at least the gcs with internal variations in metallicity may contribute to the inventory of original satellites , along with existing dwarf spheroidals and ultra - faints , to alleviate the missing satellites problem of the @xmath3-cold dark matter scenario ( e.g. , kauffmann et al . 1993 , klypin et al . 1999 , moore et al . 1999 ) . the hypothesis that at least some gcs may constitute the surviving nuclei of tidally disrupted dwarf galaxies implies that the milky way has stripped the less bound external stars from these systems during successive passages through the galactic potential , leaving only their compact nuclei . a snapshot of this phenomenon may be m54 , as it shows an intrinsic fe dispersion ( bellazzini et al . 2008 , carretta et al . 2010 ) and lies at the centre of the sagittarius dwarf galaxy that is being tidally disrupted by the milky way ( ibata et al.1994 ) . while nearly all the galactic gcs have chemical variations in the light elements involved in the hot h - burning ( such as c , n , o , na , see e.g. , kraft 1994 ; gratton et al . 2004 ) , only a few of them are known to possess spreads in fe and @xmath1-element abundances ( @xmath2 cen , m22 , ngc 1851 , m2 , ngc5824 , ngc3201 , e.g. , smith et al . 2000 ; marino et al . 2009 , 2011a ; yong & grundahl 2008 ; lardo et al . 2013 ; yong et al . in prep . ; da costa et al . 2009 , 2013 ; simmerer et al . it is intriguing that the gcs with metallicity and @xmath1-element variations are generally the more massive ones . the complexity of the multiple stellar populations in these objects is puzzling and we do not have yet a coherent picture to explain the formation of their different generations of stars . in this context , ngc 1851 is one of the most intriguing targets . much effort has been dedicated to this gc after the discovery of a prominent bimodal sub - giant branch ( sgb , milone et al . the formation scenario for these two sgb components is still under debate . observational constraints for the sequence of events that led to the formation of these stellar groups can be inferred from their chemical compositions and their radial distributions in the cluster . the radial profile of the two sgb components has been found to not change significantly within 8 from the cluster center ( milone et al . in contrast , zoccali et al . ( 2009 ) did not observe the faint sgb out to @xmath02.4 in the southwest quadrant . by analysing high - resolution uves spectra for ngc 1851 red giants ( rgb ) , yong & grundahl ( 2008 ) discovered that this cluster hosts two groups of stars with different content of @xmath1-process elements ( see also villanova et al . 2010 ; carretta et al . these two stellar groups have been found to define two rgb sequences following on from the two different sgbs ( han et al . 2009 ; lee et al . 2009 ; lardo et al . the photometric split on the sgb observed by milone et al . ( 2008 ) has been theoretically interpreted as due to either a difference in age of @xmath01 gyr or to a possible dichotomy in the c+n+o ( cassisi et al . 2008 ; ventura et al . 2009 ; sbordone et al . the latter scenario has been supported by spectroscopic studies showing that @xmath1-enriched stars are also enhanced in their overall c+n+o content ( yong et al . 2009 ; and in prep . ) . we note , however , that villanova et al . ( 2010 ) did not find an abundance spread for c+n+o . to date , variations in the overall c+n+o have been found also in other two `` anomalous '' gcs , m22 ( marino et al . 2011b , 2012a , alves brito et al . 2012 ) and @xmath2 centauri ( marino et al . 2012b ) . interestingly , ngc1851 is surrounded by a diffuse stellar halo with a radius of more than 250 pc ( 67 from the cluster center ) and a mass of about 0.1% of the dynamical mass of ngc 1851 ( olszewski et al . the extension of this stellar structure is far beyond the tidal radius predicted by the king model ( king 1962 ) , that is the distance from the cluster center where cluster stars are expected to drastically disappear due to tidal interactions . the origin of this halo remains unknown , although various hypotheses exist . it could be the consequence of isolated cluster evaporation through tidal or disk shocking that may have originated a stellar tail . such processes are believed responsible for the streams observed in several gcs , such as pal5 ( odenkirchen et al . 2001 ; koch et al . 2004 ) . observations of ngc1851 are contradictory : while photometrically there is no evidence for tidal streams ( olszewski et al . 2009 ) , the presence of a possible tail of stars with radial velocity around @xmath0150 has been reported by sollima et al . ( 2012 ) . alternatively , the huge halo of ngc1851 could have formed from the destruction of a dwarf galaxy in which the cluster may have once been embedded . bekki & yong ( 2012 ) outlined a possible self consistent and dynamically plausible scenario for the formation of ngc1851 s multiple populations and its stellar halo . in their scenario , two gcs in a dwarf galaxy merge ( owing to the low velocity dispersion of the host dwarf ) and form a new nuclear star cluster surrounded by field stars of the host dwarf . the host dwarf galaxy is stripped through tidal interaction with the milky way leaving the stellar nucleus which is observed as ngc1851 . thus , the two stellar populations in ngc 1851 originate in the two gcs that merged to form the nucleus . bekki & yong ( 2012 ) predict that ngc 1851 s stellar halo contains three stellar populations : two from the original gcs that merged to form the nucleus and the remaining population is from field stars surrounding stellar nuclei . in the present study we investigate the nature of this intriguing stellar system , the gc ngc1851 plus its halo , by deriving radial velocities and , for the first time , chemical abundances for the halo stars . the chemical and dynamical properties of the halo stars will be compared with the ones observed within the tidal radius of the cluster . basic information for ngc1851 can be found in harris ( 1996 , 2010 edition , and references therein ) . its distance from the sun is @xmath012.1 kpc . the king tidal radius of ngc1851 was estimated to be 6.7 by mclaughlin & van der marel ( 2005 ) , which is lower than the value of 11.7 in the previous harris compilation derived by trager et al.(1993 ) . given the large uncertainties related to the determination of the tidal radius , we assume in this work the conservative value of 11.7 . assuming the @xmath4 from harris ( 2010 edition ) and a typical @xmath5 , we get a mass of @xmath6 , which places ngc1851 among the most massive gcs . we assumed for the center of the cluster the coordinates ( ra ; dec)@xmath7=(05:14:06.76 ; @xmath840:02:47.6 ) from goldsbury et al.(2011 ) . in this paper we used four distinct photometric data sets . first , we used stetson ( 2000 ) ground - based @xmath9 , @xmath10 , @xmath11 and @xmath12 photometry . this photometric catalog has been established from about 550 images taken at different telescopes , i.e. the max planck 2.2 m , the ctio 4 m , 1.5 m , and 0.9 m telescopes , and the dutch 0.9 m telescope in la silla . these data have been reduced by using procedures for the photometric and astrometric data reduction described by stetson(2005 ) , and have already been used in milone et al.(2009 ) . we refer the reader to the sect . 2.3 of milone et al.(2009 ) for further information on this dataset . in the present work we have complemented the stetson catalog with images collected with the wide field imager ( wfi ) of the max planck 2.2 m telescope at la silla ( [email protected] m ) through the @xmath13 filter under the sumo campaign . details on the wfi data in the @xmath13 band , analysed here for the first time , are provided in tab . [ tab : photometry ] . in summary @xmath14 photometry has been used for stars in a region between @xmath010 to the south and 13 to the north , and between @xmath015 to the west and 15 to the east , relatively to the center of ngc1851 . secondly , to study stars in the halo of ngc1851 , we collected @xmath15 images with [email protected] m of a field between @xmath010 and 35 to the south of the cluster center . photometry and astrometry for this dataset have been obtained by using the program img2xym_wfi and the procedure described by anderson et al.(2006 ) . details of the [email protected] m dataset which has been analysed for the first time in this paper , are listed in tab . [ tab : photometry ] . third and finally , to investigate the most crowded central regions we use _ hubble space telescope _ ( @xmath16 ) f606w and f814w photometry obtained with the wide field channel of the advanced camera for survey ( wfc / acs ) and f275w photometry collected with the ultraviolet and visual channel of the wide field camera 3 ( uvis / wfc3 ) . the wfc / acs photometry comes from go-10775 ( pi.a.sarajedini , see sarajedini et al.2007 and anderson et al.2008 ) and is presented in milone et al.(2008 ) . the uvis / wfc3 photometry comes from go12311 ( pi.g.piotto ) and is presented in piotto et al.(2012 , see also bellini et al.2010 for details on the data reduction ) . we adopt the following terminology for the various fields for which different photometries are available : ( 1 ) ` central field ' for the field of 3@xmath173 covered by the @xmath16 photometry ; ( 2 ) ` inner field ' is all the field inside the tidal radius , e.g. , @xmath16 photometry where available , stetson+sumo photometry otherwise ; ( 3 ) ` outer field ' is the field outside the tidal radius . left and right panels of fig . [ fig : cmdall ] show the inner and outer field @xmath10 versus @xmath18 cmds . on each cmd we superimpose the fiducial line of the cluster . the inner and outer field cmds have not been de - reddened . however , the reddening across the ngc1851 field of view is very low , @xmath19 , and the differential reddening should not be significant as it is much lower than internal uncertainties in the ground - based photometry . the reddening variations across a field of 2@xmath172 around the cluster center predicted from the schlegel et al . ( 1998 ) maps is also low with minimum and maximum values for @xmath20 0.0287 and 0.0351 , respectively . a maximum reddening variation of 0.0064 mag does not produce significant color changes to the inner and outer field cmds . the location on the sky of the ground - based photometry fields , as well as the central @xmath16 field , relative to the cluster center , are represented in the middle panel : the central @xmath16 field is delimited in red ; the region that we define inner field is located within the tidal radius , represented as a green circle ; while the region defined as outer field is comprised within the cyan square . ground - based photometric data have been used to estimate atmospheric parameters , as described in sect . [ sec : atm ] . hence , it is important to have an estimate of the internal photometric uncertainties . according to the photometric catalogs from stetson(2000 ) , the average @xmath21(mag ) for a star with @xmath2219.5 in the @xmath9 , @xmath10 , @xmath11 , @xmath12 bands is @xmath00.003 , @xmath00.002 , @xmath00.012 , @xmath00.003 mag , respectively . since each star has been typically observed in tens of images , the formal error should be significantly smaller . in the external field we obtained @xmath9 , @xmath10 , @xmath12 formal errors of @xmath00.005 , @xmath00.004 , @xmath00.006 mag , respectively . such small values are lower limits of the true photometric errors as demonstrated by the fact that the color spread of all the sequences of the cmds is significantly larger than few milli - magnitudes . photometry is indeed affected by a number of additional uncertainties . spatial variations of the photometric zero point along the field of view , introduced by small inaccuracy in the psf model , by the sky or bias determination , or by small reddening variations , are a very common property of any photometric catalog ( see anderson et al.2008 for a discussion on this issue ) . to estimate the error of target stars we started by measuring the color spread of ms stars as described in milone et al.(2009 , see their sect . briefly , we have verticalized the ms of ngc1851 by subtracting to the color of each star the color of the ms at the same @xmath10 magnitude , then we have determined the histogram distribution of the color difference ( @xmath23color ) , and , finally , we have fitted the histogram with a gaussian . we have repeated this procedure for @xmath18 , @xmath24 , @xmath25 , and @xmath24 colours . we assumed the @xmath21 of the least - squares best - fitting gaussian as our estimate of the color error . specifically , to estimate the errors associated to sgb stars in the central field , we used the color spread of ms stars with @xmath26 . we obtained @xmath27 , @xmath28 , @xmath29 , @xmath30 . we thus assumed 0.02 as typical magnitude uncertainty for stars in the internal field . in the case of stars in the external field , due to the small number of ms stars , we used a larger magnitude interval ( @xmath31 ) and accounted for field - star contamination by subtracting from the observed histograms of @xmath23color distribution , the corresponding histogram distributions for field stars . the later has been determined by using the besanon galactic model ( robin et al.2003 ) for stars within the same area as the external field . we found @xmath32 , @xmath33 , @xmath34 , thus assumed a typical magnitude uncertainty of 0.02 mag in each filter . these uncertainties will be considered when discussing the impact of photometric errors on the atmospheric parameters ( see sect . [ sec : atm ] ) . [ cols="<,^,^,^,^,^,^,^",options="header " , ] the mean chemical abundances for the ngc1851-rv like stars are listed in table [ tab : medie ] , along with those obtained for the central field . in the following , we first discuss results for the ngc1851 cluster stars in the internal field , so that the chemical composition of the ngc1851 halo stars can be compared with those of the central cluster . -@xmath35 ( from piotto et al . 2012 ) , @xmath36-@xmath37 ( from milone et al . 2008 ) , @xmath9-@xmath18 , and @xmath13-@xmath38 cmds ( milone et al . 2009 ) . stars assigned to the bsgb and fsgb have been represented in blue and red , respectively , stars with more uncertain location respect to the double sgb have been coloured in yellow.,width=317 ] abundances for the ngc1851 stars in all the stars in the inner field ( upper panel ) , the bsgb and fsgb ( middle panel ) , and for the ngc1851-rv like halo stars ( lower panel ) . , width=321 ] the chemical composition of ngc1851 is not _ standard _ for a gc , as it shows internal variations in @xmath1-process elements , that is a peculiarity of a few gcs such as those showing split sgbs ( e.g. , m22 and @xmath2 centauri ) . a proper comparison between the external and internal field of this gc requires accounting for these features . starting with the analysis of the cluster stars in the internal field , we have assigned each target to the faint sgb ( fsgb ) or the bright sgb ( bsgb ) , based on their position on the cmd . in fig . [ fig : sgb_phot ] we show the position of these central field targets on the cmd in various photometric bands . an inspection of this figure suggests that the separation between bsgb and fsgb is much more distinct by using @xmath39 and @xmath40 images from @xmath16 ( left panels ) , than by using standard broad - band photometry from ground ( right panels ) . as the photometry from space is more precise , we use the @xmath16 photometric information to assign each star to the bsgb or the fsgb , when available . in fig . [ fig : sgb_phot ] the bsgb and fsgb spectroscopic targets have been represented in blue and red , respectively . these color codes will be used consistently in the following discussion . in some cases , the association with one of the two sgb is not obvious for stars having only ground - based photometry , as some targets lie on the ms where the split is not visible . so , we define a third group of stars whose position on the cmd is ambiguous ( yellow dots in fig . [ fig : sgb_phot ] ) , and we do not consider them in the comparison with the halo stars chemical composition . in total we have identified 18 bsgb and 15 fsgb in the @xmath16 field , and 32 bsgb and 14 fsgb in the ground - based photometry , with 7 uncertain stars . to properly compare the abundances obtained in the halo with those of the stellar groups observed in the cluster we derive the average abundances of the bsgb and fsgb samples , as listed in table [ tab : medie ] . in the following discussion the comparison between the average abundances of the various analysed groups of stars ( bsgb , fsgb and halo ) has been performed by using as error the quadratic sum of the errors associated with the considered means . the [ fe / h ] histogram of the two sgb groups is shown in the upper panel of fig . [ fig : histocl ] . the total distribution of the entire sample , represented in green , is entirely within that expected from the error analysis of our relatively low resolution data . by dividing the sample into bsgb and fsgb , our mean [ fe / h ] values agree within observational errors , suggesting that the two populations have the same average metallicity within @xmath00.05 dex ( @xmath41 ) , confirming previous studies on the rgb based on higher quality data ( villanova et al . 2010 ; carretta et al . 2011 ) . a smaller difference between the iron content of the two sgbs can not be ruled out by our data . we recall here that the trends in the [ x / h ] abundances can not be interpreted in support of any overall metallicity variation as they can result from our observational errors . the strontium - to - iron ratio , [ sr / fe ] , exhibits a large star - to - star variation , exceeding the observational errors . by dividing stars in bsgb and fsgb the [ sr / fe ] mean abundance of the fsgb is larger than that inferred for the bsgb , with a mean difference @xmath23[sr / fe]@[email protected]@xmath430.04 ( a @xmath06@xmath21 significant difference ) . despite the large errors on [ ba / fe ] , the fsgb is also richer in barium by [email protected] dex ( with a significance of more than 6 @xmath21 ) . in summary , our results on the internal field confirm previous findings on the presence of two groups of stars in ngc1851 with different abundances in those elements mostly produced in the @xmath1-processes , like sr and ba ( e.g. yong & grundahl 2008 ; villanova et al.2010 ) , and that the fsgb is populated by stars that have undergone some kind of enrichment from these processes ( e.g. gratton et al . 2012 ) . in this respect , ngc1851 is very similar to m22 , which shows a bimodal distribution in @xmath1-elements ( marino et al.2009 ) and a fsgb populated by stars with higher @xmath1-element abundances ( marino et al.2012a ) . as a visual representation of the differences in sr and ba between the two sgbs , in fig . [ fig : synthesis_example ] we represent some portions of the analysed spectra around key spectral features for one bsgb star ( sgb - a - st.45 ) and one fsgb star ( sgb - b - st.16 ) . superimposed on the observed spectra are the best - fit synthesis ( solid red ) , two syntheses corresponding to sr , ba , and fe abundances varied by @xmath430.2 dex , and a synthesis computed with the average abundance of the fsgb and the bsgb ( dashed red ) for sgb - a - st.45 and sgb - b - st.16 , respectively . the high dispersion in [ c / fe ] , significantly exceeding observational errors , suggests that the abundance in this element shows significant internal variations . this is consistent with what was observed in other gcs , including the less complex ones ( e.g. , m 4 , marino et al . 2008 ) , and mostly interpreted as due to some kind of intra - cluster pollution from material that has undergone h - burning at high temperature . it is worth noting here that the high dispersion in [ c / fe ] observed in the complete sample does not diminish by dividing stars in bsgb and fsgb , being only marginally lower for the bsgb ( tab . [ tab : medie ] ) . the average [ c / fe ] is higher in the bsgb , with a mean difference of @xmath23([c / fe])@[email protected] , that is slightly higher than a 3 @xmath21 level . the chemical pattern of [ c / fe ] , its variation within different @xmath1-groups and the larger spread among the @xmath1-poor ( bsgb ) stars agrees with previous findings obtained first for m22 ( marino et al . 2011b ; 2012a ) and then for ngc1851 itself ( lardo et al . 2012 ; gratton et al . 2012 ) . we note that [ mg / fe ] appears to be slightly higher in fsgb stars ( at a level of @xmath03.5 @xmath21 ) . magnesium abundances for rgb stars in ngc1851 have been provided by yong et al . ( 2008 ) from uves spectra and carretta et al . ( 2010 ) from giraffe spectra . the average [ mg / fe ] abundances for their sample of rgb stars is @[email protected] ( @xmath21=0.07 ) and @[email protected] ( @xmath21=0.04 ) in yong et al . and carretta et al . , respectively . although a direct comparison with these two studies can not be done , because they use different spectral mg features and possible non - lte corrections may apply differently . we note that their values are consistent with the average abundance that we found for the total sample of sgb stars , [ mg / fe]@[email protected]@xmath430.02 ( @xmath21=0.16 ) , and for the bsgb stars , [ mg / fe]@[email protected]@xmath430.02 ( @xmath21=0.15 ) . on the other hand , the mg abundance of the fsgb appears to be higher , but a similar effect has not been found in previous studies on the rgb . given our large uncertainties associated with individual mg abundance measurements , the presence of this difference should be viewed with caution and needs to be investigated further . none of the other species appears to show differences among the two sgbs over a 3 @xmath21 level . lcccccc & @xmath23&@xmath23&@xmath23&@xmath23[a / h ] & @xmath48&@xmath49 + & @xmath43100 k & @xmath430.05 & @xmath430.20 & @xmath430.11 dex & & + + + + + @xmath50}$ ] & @xmath430.04 & @xmath430.01 & @xmath430.02 & @xmath510.02 & @xmath430.10/@xmath430.16 & 0.11/0.17 + @xmath52}$ ] & @xmath510.08 & @xmath510.01 & @xmath430.01 & @xmath430.01 & @xmath430.13/@xmath430.18 & 0.15/0.20 + @xmath53}$ ] & @xmath430.03 & @xmath430.00 & @xmath430.02 & @xmath510.02 & @xmath430.09/@xmath430.13 & 0.10/0.14 + @xmath54}$ ] & @xmath430.17 & @xmath510.01 & @xmath510.02 & @xmath510.02 & @xmath430.18/@xmath430.38 & 0.25/0.42 + @xmath55}$ ] & @xmath430.10 & @xmath510.02 & @xmath510.02 & @xmath510.10/@xmath510.07 & @xmath430.02/@xmath430.07 & 0.15/0.14 + @xmath56}$ ] & @xmath510.01 & @xmath430.02 & @xmath510.02 & @xmath430.02 & @xmath430.12/@xmath430.20 & 0.13/0.20 + @xmath57}$ ] & @xmath510.07/@xmath510.04 & @xmath510.11 & @xmath510.15 & @xmath430.02 & @xmath430.09/@xmath430.21 & 0.22/0.29 + + + + @xmath50}$ ] & @xmath430.04 & @xmath430.01 & @xmath430.03 & @xmath510.02 & @xmath430.06 & 0.08 + @xmath52}$ ] & @xmath510.08 & @xmath510.01 & @xmath430.00 & @xmath510.03 & @xmath430.13 & 0.16 + @xmath53}$ ] & @xmath430.04 & @xmath510.01 & @xmath430.02 & @xmath510.02 & @xmath430.08 & 0.09 + @xmath54}$ ] & @xmath430.06 & @xmath430.00 & @xmath510.01 & @xmath510.05 & @xmath430.19 & 0.21 + @xmath55}$ ] & @xmath430.10 & @xmath510.01 & @xmath510.03 & @xmath510.07 & @xmath430.03 & 0.13 + @xmath56}$ ] & @xmath510.04 & @xmath430.02 & @xmath430.01 & @xmath430.04 & @xmath430.07 & 0.09 + @xmath57}$ ] & @xmath510.03 & @xmath510.11 & @xmath510.17 & @xmath430.00 & @xmath430.11 & 0.23 + the chemical abundances and their averages inferred for the halo stars are listed in table [ tab : abundances ] and table [ tab : medie ] , respectively . although we could infer metallicities for a relatively large number of halo stars ( 15 ) , the abundances of the other elements were possible only in the stars with higher s / n spectra . the [ fe / h ] distribution of the ngc1851 halo stars is presented in the lower panel of fig . [ fig : histocl ] . an inspection of this figure immediately suggests that the distribution for the halo stars spans a range similar to the one observed in the internal field . the average [ fe / h ] for the halo stars is [ fe / h][email protected]@xmath430.02 dex ( rms=0.09 ) , and its difference with the mean values obtained for the internal field are : @xmath23[fe / h]@[email protected]@xmath430.03 dex ; @xmath23[fe / h]@[email protected]@xmath430.03 dex ; @xmath23[fe / h]@[email protected]@xmath430.03 dex . indeed , we conclude that , within observational errors , the [ fe / h ] in the stars analysed in the halo is consistent with the mean abundance obtained for the ngc1851 cluster stars . a comparison between all the abundances inferred for the halo with those obtained for the bsgb and fsgb in the inner field is shown in fig . [ fig : all ] . the first observation we make is that the [ sr / fe ] distribution for the halo is consistent with that shown by the bsgb , with just one star falling in the range spanned by the fsgb . although the internal error on [ ba / fe ] is much larger , the [ ba / fe ] range for the halo is also more similar to that spanned by bsgb stars than to that of the fsgb . note however that , in this case , some stars are also consistent with the fsgb range , but due to the large observational errors affecting barium abundances , we can not draw strong conclusions on the distribution of this element alone . to quantify the probabilities that the abundance distributions in sr and ba obtained for the fsgb , bsgb in the inner field and in those for the outer field derive from the same parent distribution , we performed some ks tests . the ks probabilities that the bsgb and the fsgb abundances derive from the same distribution are 0.0000 both for sr and ba . similarly low are the probabilities when we compare the fsgb and the halo , for which we obtain 0.0001 and 0.0047 for sr and ba , respectively . on the other hand , the probabilities that the bsgb and the halo sr and ba abundances derive from the same distribution are 0.8180 and 0.5622 , respectively . in fig . [ fig : synthesis_examplehalo ] we show the spectral features for sr and ba in three halo stars , with the best - fit synthesis ( blue ) and the synthesis corresponding to the mean chemical abundances for the fsgb observed in the inner field ( red ) . this figure illustrates well the limit of our observations : e.g. , while the chemical abundances of the two rgbs ( t207 and t198 ) is well distinct from the composition of the fsgb , the uncertainties associated with some ms stars ( e.g. , t138 ) are much larger . in the particular case of t138 , the star with more ambiguous position on the [ ba / fe]-[sr / fe ] plane ( see fig . [ fig : sproc ] ) , the abundances for barium and strontium are in fact only slightly lower than the fsgb average abundance . keeping in mind all the uncertainties we have , in particular for the ms stars , the most robust comparison between the halo and the cluster stars that we can make with our observations is by combining results for the two analysed @xmath1-process elements sr and ba . in fig . [ fig : sproc ] we show [ ba / fe ] as a function of [ sr / fe ] for the bsgb , the fsgb and the halo stars . stars with both sr and ba available clearly distribute in the same manner as the bsgb stars . the two stars not lying on the ngc1851 sequence ( see sect . [ subs : rvs10 ] ) have been indicated with black crosses . for completeness , in fig . [ fig : sproc ] we plot stars with only sr or ba available as cyan dots separated by the horizontal and vertical dashed lines , respectively . we comment briefly these four stars . as previously discussed , due to the large errors in [ ba / fe ] we can not draw conclusions on the three stars with only ba abundances available . indeed , the large uncertainties associated with the single ba measurements for halo ms stars , that is about 0.29 dex as listed in tab . [ tab : err ] , do not allow us to confidently assign the two out three stars with higher ba to neither the @xmath1-rich or the @xmath1-poor group . the location of the star with just sr is inconclusive . we can conclude that the analysis of our sample of halo stars thus does not provide strong evidence for the presence of @xmath1-rich stars corresponding to the fsgb population , insofar as all the stars for which ba and sr abundances are available have values compatible with those of the bsgb . on the probability of observing @xmath1-rich stars , considering that in the central field the fsgb contributes around 35% of the stars ( milone et al . 2009 ) , we would expect to observe 1.75/5 ( with sr and ba abundances available ) at the high @xmath1-process composition of the fsgb . we observe no such stars . as a further test , the binomial probability of observing none out of five stars in the @xmath1-poor group is 0.12 , and none out of seven stars ( if we include the two stars with anomalous position on the cmd ) is 0.05 . these probablilties are small , however we remark here that they are not zero , and future analysis of larger sample of stars are necessary to increase the sample of halo stars with available chemistry . as long as we refer to our analysed sample , the present results provide no strong evidence for the presence of @xmath1-rich stars , and indeed support a halo populated by just bsgb @xmath1-poor stars . to make this statement more conclusive we need to improve the statistics in the future and eventually reduce our 0.12 probability to @xmath00.00 . for these statistics we assume the bsgb and fsgb fractions based on @xmath16 and ground - based photometry from milone et al . ( 2009 ) that follow the sgb split from the cluster center out to 8 with no strong evidence for different radial distribution of the two branches . beyond this distance out to the tidal radius , there are too few sgb stars to identify the sequences and determine their contribution to the total cluster mass . on the other hand , zoccali et al . ( 2009 ) found that the fsgb dramatically drops at a distance of @xmath02.4 from the center . we note that since we have detected @xmath1-rich / fsgb stars up to @xmath05@xmath61 from the center , we confirm that there is no drop of fsgb stars at @xmath02.4@xmath61 . as abundances for sr and ba were possible for 7/23 stars , including two rgbs , we have 16 ms stars , including the four stars with only sr or ba abundance available , that do not have available abundances for both @xmath62-capture elements . to get chemical information for these 16 stars we combined their observed spectra by simply averaging them with the same weight , so that we obtained a spectrum with higher s / n . then , we constructed two moog synthetic spectra by averaging those corresponding to the atmospheric parameters of each star : one with the mean sr and ba abundances obtained for the bsgb ( mean bsgb synthetic spectrum ) , and the other with the mean sr and ba of the fsgb ( mean fsgb synthetic spectrum ) . the comparison between the mean observed halo spectrum and the mean bsgb and fsgb synthetic spectra is shown in fig . [ fig : synthesis_avehalo ] . the mean observed halo spectrum near the sr and ba spectral features is clearly best matched by the mean bsgb synthetic spectrum . the stars not analysed for individual abundances of sr and ba have predominantly the bsgb abundances . regarding the other elements , halo chemical composition agrees with that observed in the inner field of ngc1851 . as shown in fig . [ fig : all ] , the distributions of all the elements in the halo are consistent with those observed in both the bsgb and fsgb , not being any significant difference between the abundances obtained for the two sgbs . the halo stars are enhanced in @xmath63 elements ca and mg , and roughly solar - scaled [ cr / fe ] , although the rms in this latter species is high due to observational errors . carbon has been inferred for five stars , including t066 ( not on the cluster sequence ) , the two red giants ( t198 and t207 ) , and two ms stars ( t186 and t207 ) . the comparison with the c abundances in the inner field can be done for just the two ms stars , as rgb stars have undergone the first dredge - up bringing carbon into the internal layers , resulting in lower surface abundances for this element . the two ms stars t186 and t220 have [ c / fe][email protected] and @xmath80.15 , respectively , and the mean difference with the median values of the internal field are : @xmath23[c / fe]@[email protected]@xmath430.03 dex ; @xmath23[c / fe]@[email protected]@xmath430.03 dex ; @xmath23[c / fe]@[email protected]@xmath430.04 dex , that are consistent with both the bsgb and fsgb ( see tab . [ tab : medie ] ) . we conclude that , for all the analysed elements , the chemical composition of the halo is consistent with that observed in the cluster ( as shown in fig . [ fig : all ] ) . in particular the @xmath1-elements are consistent with the bsgb abundances . the similar abundance distributions for all the elements with available measurements , is an additional signature for cluster membership , and more in general , for the existence of a halo surrounding ngc1851 . having established the existence of a halo of stars beyond the tidal radius of ngc1851 , we can investigate the intrinsic rv dispersion ( @xmath64 ) of these outer stars in comparison with the inner field . to this aim we used the procedure described in mackey et al.(2014 ; see their sect . 4.2 ) that takes into account the contribution of observational errors to the rv dispersion . briefly , we used a maximum likelihood technique , assuming that the measured rvs for our stars are normally distributed around the average value according to their measurement uncertainties and the intrinsic cluster velocity dispersion . we can obtain numerical estimates for the intrinsic cluster velocity dispersion by maximising the logarithm of the joint probability function for the observed rvs . the procedure has been done at three different bins of radial distances from the cluster center : namely in the central field of 3@xmath173 covered by @xmath16 photometry , in the inner field covered by ground - based photometry at a distance from the center between @xmath04 and @xmath09 , and for the stars in the outer field that covers the halo . figure [ fig : rvdisp ] shows the obtained intrinsic velocity dispersions in our three radial bins . in the cluster field the velocity dispersion decreases with radius , from @xmath05 in @xmath16 field , to @xmath04 between 4 and 9 . there is no significant difference between the bsgb and fsgb . scarpa et al . ( 2011 ) determined the velocity dispersion radial profile from flames spectra using high - resolution setups for 184 stars along the upper sgb and the lower rgb in the inner field of ngc1851 . their results are plotted , along with ours , in fig . [ fig : rvdisp ] . our measurements agree with those reported by scarpa and collaborators within 1 @xmath21 . out of the tidal radius we find a dispersion comparable with that observed in the region between 4 and 9 . we note here that some physical and technical issues may affect our @xmath65 determination : * stellar masses : since most of the halo stars belong to the upper ms , while inner field stars are sgb , to properly compare the @xmath65 estimates at different radial distances we need to account for the different masses of the stars . lower mass stars should have a higher dispersion , typically , than higher mass stars . however , this effect is negligible in our sample , since the mass difference between upper ms and sgb is small . to quantify , under energy equipartition for two stars with masses @xmath66 and @xmath67 we have that @xmath68=@xmath69 , that for a sgb with mass @xmath700.8@xmath71 and a ms star with mass @xmath720.7@xmath71 would be @xmath73 larger by @xmath07% than @xmath74 ; * field contamination : although we exclude that this is an issue in the inner field where we have much higher statistics , a low degree of field contamination ( @xmath02 stars out of 23 ) may affect our sample of ngc1851 rv - like stars in the halo , and may affect the value of @xmath64 . to estimate this effect we determined the radial velocity dispersion ( @xmath75 ) by excluding each pair of stars ( _ i , j _ , with _ i_=1,23 and _ j_=1,23 ) from our ngc1851 halo sample . in the end , we obtained an average dispersion of 3.72 ( rms=0.22 ) with @xmath64 ranging from 2.57 to 3.96 ; * binaries : binary contamination can inflate the dispersion . as discussed in sect . [ sec : rv ] , we do not expect a large fraction of binaries in ngc1851 ( from milone et al2012 the fraction of binaries in the central field outside the half - mass radius is @xmath01.6% ) . of course , we can not exclude that the halo stars may have , for some reason , a larger binary fraction . for the ngc1851 halo sample we do not see any evidence for the presence of stars whose rms in rvs from different exposures significantly exceeds that introduced by the quality of the spectra ( see sect . [ sec : rv ] for more details ) ; * possible fiber - to - fiber systematic errors : while the plate - to - plate systematics are randomly distributed and are likely removed using our procedure , fiber - to - fiber effects may be there as the flames fibers configuration that we used is the same for every exposure ( that means each star is observed with the same fiber ) . such effects are small , and never exceed @xmath00.5 ( e.g. , sommariva et al . 2009 ) . with due consideration of all the issues affecting the @xmath64 estimates , we may speculate that the apparently continuous nature of the dispersion profile from the cluster to the halo provides yet more evidence that the halo is associated with the cluster . we can also exclude a truncation in the observed @xmath64 in the vicinity of the tidal radius , and that is contrary to the expectations from king - type theoretical models . the bsgb and fsgb observed in the inner field are chemically different in the @xmath1-process elements , but they look kinematically similar . ) as a function of the distance from the cluster center ( in logarithmic units ) for the stars in the @xmath16 central field ( within 3@xmath173 ) , the stars in the inner field covered by ground - based photometry ( between @xmath04 and 9 from the cluster center ) , and in the outer field ( from the tidal radius up to @xmath030 from the cluster center ) . the color codes are as follows : green is for the inner field stars of ngc1851 ; cyan for the halo of ngc1851 ; blue and red are the stars in the inner field divided in the bsgb and fsgb samples ( see sect . [ sec : abb_clu ] ) respectively . the black triangles are values derived in scarpa et al . the dashed line shows the location of the tidal radius . , width=309 ] we have provided a spectroscopic analysis of the halo that surrounds ngc1851 . in particular we have studied the nature of this intriguing stellar structure by analysing rvs and elemental abundances , providing for the first time a chemical inventory in a halo surrounding a gc . we have measured radial velocities and elemental abundances of fe , c , mg , ca , cr , sr and ba in a sample of both halo and cluster stars , from which we draw the following conclusions : * the rv distribution in the observed halo field shows a peak not expected from galactic models at the characteristic rv of ngc1851 , confirming the presence of a halo surrounding the cluster ; * fifteen stars in the halo field exhibit radial velocities and metallicities consistent with the cluster field ; * our observed rv distribution agrees ( apart from the rv range of ngc1851 ) with that expected from galactic models , suggesting that no other sub - structure , such as streams , is present in our field ; * the halo has the same metallicity distribution as ngc 1851 . none of the ngc 1851 rv - like stars found outside the tidal radius shows a [ fe / h ] content different from the range observed in the inner field ; * the halo stars for which we could estimate abundances for sr and ba show abundances consistent with those observed on the bsgb , i.e. , they have lower sr and ba compared with those of the fsgb ; * our sample does not exhibit any strong evidence for the presence of stars with @xmath1-element abundances compatible with the fsgb . within the multiple stellar population context , qualitatively , our observations are in agreement with a scenario in which : _ ( i ) _ the first stellar population of ngc1851 is made up of @xmath1-poor ( bsgb ) stars ; _ ( ii ) _ the second generation is expected to form in the central region of the cluster possibly enriched in @xmath1-elements ; _ ( iii ) _ while the less radially concentrated first generation is lost early in the cluster evolution due to the expansion and stripping of the cluster outer layers resulting from early mass loss associated with first generation supernova ejecta ( e.g. , dercole et al.2008 ) . if this scenario is correct , our observations constitute the first clear evidence that gcs lose primarily first - generation stars by evaporation mechanisms . the chemical composition of field stars in the halo , that is consistent with the first - generation stars in gcs , may mainly be due to the evaporation of first - generation gc stars into the field . although the central relaxation time of ngc1851 is small ( of the order of @xmath76 yr ) , in the halo any dynamical time will be very long due to the low density . so , while in the central part of the cluster the timescales are short enough that the population is mixed , in the external halo field , they are sufficiently long such that any initial gradient may have been preserved . it is worth noticing that in the cmds shown by milone et al . ( 2009 ) there is no obvious difference in the radial distributions of the bsgb and fsgb of ngc1851 up to @xmath08@xmath61 from the center . however , their observations are limited to a small region well within the tidal radius of the cluster , and we can not exclude different radial gradients at more external regions . if the fsgb is more centrally concentrated and the outer parts of the cluster are dominated by the bsgb , then since outer stars are more easily lost , that would be consistent with the bsgb dominance in the halo . we recall here that both the @xmath1-poor , and possibly also the @xmath1-rich groups in ngc1851 show internal variations in light elements ( e.g. , c , n , o , na ) . the presence of sub - populations within each main @xmath1-group makes the scenario more complex , challenging the identification of the sequence of the various stellar bursts that gave origin to the multiple stellar populations in this gc . an exhaustive study of the radial distributions of the ngc1851 stellar groups should account also for the presence of these sub - populations . the main question here is : why does ngc1851 possess a stellar halo with these kinematic and chemical properties ? in principle the presence of a halo may be simply consistent with a stellar system that is losing its external stars into the field . on the other hand , gcs are expected to follow a king profile , with the surface density drastically dropping at the tidal radius . it has been well established that while king models succeed in describing the internal stellar surface - brightness profiles , they often fail in the outer regions of clusters , including the milky way gcs ( e.g. , mclaughlin & van der marel 2005 ; carballo - bello et al . gcs surface - brightness profiles suggest that stellar clusters do not simply truncate on the king tidal radius , but have low density extended halos . all gcs should be stressed , to different extent , by interactions with the host galaxy . hence , deviations from the king profile could not be surprising . mclaughlin & van der marel ( 2005 ) have shown that the observed surface - brightness profiles of gcs is generally better reproduced by a modified isothermal sphere based on an _ ad hoc _ stellar distribution function developed for elliptical galaxies by wilson ( 1975 ) . the corresponding wilson tidal radius of mclaughlin & van der marel ( 2005 ) is larger than that derived assuming a king profile , being @xmath045 for ngc1851 . however , although we can formally fit the surface - brightness profiles of gcs in the regions beyond the king radius adopting _ ad hoc _ analytical templates , in many cases it remains unclear why we observe such halo envelopes . for ngc1851 the main question could hence not be why it possess a halo , but why it shows such a peculiarly uniform halo , without evident tidal features in the large photometric field analysed by olszweski et al.(2009 ) . the ngc1851 halo may be a more interesting case , as it extends for 67 from the cluster center , that is much farther than any estimate for its tidal radius present in the literature . its very low density , as determined in olswzeski et al . , would make it easily perturbed by passages through the disk . according to dinescu et al . ( 1999 ) , ngc1851 is at a distance from the galactic center of 16.0 kpc , its orbit has apogalacticon and perigalacticon distances of @xmath030 kpc and @xmath05 kpc , respectively , a high eccentricity of @xmath00.7 , and passes through the disk of the milky way five times per gyr . hence , it seems unlikely that an outer envelope of stars remains attached to the cluster for a hubble time or , alternatively , that the cluster could shed sufficient stars to make such a huge envelope in half an orbit . our spectroscopic sample of stars does not show any evidence of tidal streams . although our observations allow us to draw this conclusion only for a relatively small region of the halo , olswzeski et al . did not find any evidence for tidal streams in their large field of view . this apparent lack of streams in the olswzeski et al . analysis is puzzling . it is difficult to envisage a scenario that assumes that such an extended low - density halo is bound to ngc1851 , if the cluster has been in orbit around the milky way for a significant portion of a hubble time . however , even if it seems unlikely , we can not exclude that , if the cluster has been in orbit around the galaxy for a long time , we are seeing a particular phase of the ngc1851 evolution : e.g. , a very extended envelope of stars that is in the process of escaping from the cluster , but still bound to it . ngc1851 is currently at an intermediate distance from the galactic center between the apogalacton and perigalacticon , and it is at a distance of 7.1 kpc from the galactic plane ( djorgovski 1993 ) , that is around the maximum distance from the plane allowed by its orbit ( dinescu et al.1999 ) . the halo may have formed after the last passage through the disk , and will then be largely swept away next time the cluster crosses the disk and , presumably , recreated as the cluster moves through the halo in the half - orbit time before the next disk crossing . future dynamical simulations may be enlightening in this regard , and should prove if such a scenario is plausible for the formation of the ngc1851 halo . alternatively , it is tempting to speculate that we really are seeing a system that has been captured relatively recently and the envelope does represent the former dwarf galaxy . in this case , there just have not been enough orbits yet for the envelope to be completely stripped off by passages through the disk . hence it is possible that the halo population could look like the first generation in the cluster in the same way that the first generation in most globular clusters looks like the field halo population at the same [ fe / h ] . if this hypothesis is correct , then we would expect that the halo population does not host stars enriched in na because dwarf galaxies do not show na - o anticorrelations . the first cluster proposed to be the nucleus of a dwarf tidally disrupted through milky way interactions was @xmath2 centauri ( norris et al . assuming a similar scenario , it would be less problematic to understand the large chemical variations , also in heavy elements , displayed by this cluster ( e.g. , norris & da costa 1995 ; johnson & pilachowski 2010 ; marino et al . although less pronounced , ngc1851 shows chemical and photometric peculiarities similar to those of @xmath2 centauri ( e.g. , milone et al . 2008 ; yong & grundahl 2008 ; villanova et al . 2010 ; carretta et al . in contrast to ngc1851 , for the case of @xmath2 centauri there is no evidence for stellar halo envelopes , probably because its location relatively close the galactic centre and its tightly bound orbit . these ensure that galactic tidal forces at its location are relatively strong so that any loosely bound population is quickly removed preventing the survival of any such structure to the present - day ( see da costa & coleman 2008 ) . if the accretion of @xmath2 centauri has not occurred recently there has been time to remove the loosely bound outer envelope that could have originally surrounded the cluster . on the other hand , m54 , that is the gc associated with the nucleus of the sagittarius dwarf galaxy , preserves an external envelope , that is the field population of the sagittarius . collectively , these facts may reinforce the conjecture that _ anomalous _ gcs could represent the surviving nuclei of dwarf galaxies disrupted by tidal interactions with the milky way , and we can speculate that the halo surrounding ngc1851 is the remnant of the parent dwarf galaxy . to test the feasibility of this hypothesis , dynamical simulations taking into account the tidal interactions between the milky way and dwarfs are fundamental . recent simulations presented in bekki & yong ( 2012 ) show that the halo in ngc1851 can be reproduced if this object formed in a dwarf galaxy environment , through the merger of two clusters ( corresponding to the @xmath1-rich and @xmath1-poor groups ) that sink into the center of the host galaxy . from their simulations , they expect to have three stellar populations in the halo : the @xmath1-poor and the @xmath1-rich stars , plus a third population representing the field of the host dwarf . our results seem to not strongly favour this idea . assuming that each of the three populations should be equally represented in the halo , we have not found any clear evidence for the presence of either the @xmath1-rich stars or a third population differing from the metal distribution observed in ngc1851 to be associated with a field population from the host galaxy . if , following the bekki scenario , we assume that the three populations are equally represented ( e.g. , fsgb=33% ; bsgb=33% , and field=33% ) , the probability to not observe any star with the ba and sr abundances of the fsgb population of ngc1851 , even if not null , is low ( see sect . this suggests that , even with our modest statistics , it is unlikely that there are three populations equally represented in the field , as in the bekki & yong scenario . furthermore , even if an internal metallicity variation of the order of 0.10 dex in the halo ( that we can not see because of observational errors ) can not be ruled out , the metal distribution of the external field ( 15 stars ) is very similar to that observed within the tidal radius . this favours the idea that the halo does not host stellar populations with metallicities too distinctive with respect to ngc1851 , but we can not completely exclude that the lack of such stars is instead introduced by photometric selection effects that can affect our sample . we note that olswzeski et al . found evidence for the presence of at least the two sgb populations ( bsgb and fsgb ) in the cmd of the ngc1851 halo . however , it is possible that their fsgb may not correspond to the one observed in the cluster ; given they found a significantly larger separation between the two sgbs than that found out to 8 from the cluster center ( milone et al . 2009 ) . as previously discussed , the halo population we are actually observing could be either the @xmath1-poor population observed in the cluster or the field population of the host galaxy , that may show the same metallicity and chemical properties of the cluster bsgb . in the latter case , we do nt have to see bekki & yong s postulated 3rd population , as it may be the same as the cluster first generation and we may have to look at a lot of halo stars to see a different [ fe / h ] distribution from that of the cluster first generation . it is also worth noting that the dynamical simulations by bekki & yong assume ngc1851 is the result of a merger between two clusters . it might be interesting to see if similar simulations assuming the ngc1851 various groups formed in a self - pollution scenario are able to better reproduce the observations . the lack of a drop in the velocity dispersion profile is another piece of evidence that makes ngc1851 hardly compatible with the king model . a similar lack is intriguingly seen in m54 , the gc lying at the center of sagittarius . this gc shows a velocity dispersion profile that falls off with the distance from the cluster center , and increases outside , due to either contamination from sagittarius in the external part of the cluster or to tidal harassment from the host galaxy ( bellazzini et al.2008 ) . in the hypothesis that the ngc1851 halo represents the parent galaxy population , we would qualitatively expect a relatively high velocity dispersion for the halo . on the other hand , if the halo is simply formed by stars evaporating from ngc1851 , it may be that the stars in the region outside the tidal radius are undergoing some kind of heating , due to either tides and/or unseen ( dark ) matter or modified newtonian dynamics ? we are not able to answer this question here . we only note that , according to dynamical simulations presented in kpper et al . ( 2010 ) , the velocity dispersion of unbound escaping stars outside gc tidal radii should significantly increase and flattening due to tidal interactions with the milky way . the orbit of ngc1851 suggests that it is not at its perigalacticon and it is at its largest distance from the galactic plane , but according to kpper et al . this effect , even if magnified at the perigalacticon , seems to be present at any location on the orbit . we note however that our sample of 23 stars outside the tidal radius may still be bound to the cluster . future investigations on the velocity dispersions in the halo of ngc1851 may give information on the nature of the ngc1851+halo system , assuming that measurements will be available at different radial locations in the halo and with better statistics . we conclude that the presence of such an extended halo in ngc1851 may suggest that the complex chemical enrichment in this gc has taken place in a dwarf whose nucleus is ngc1851 . interestingly , the location of ngc1851 coincides with the disk of satellites , a relatively thin , highly inclined plane defined by the distribution of luminous milky way satellite galaxies ( kroupa et al.2005 ; metz et al.2007 ; pawlowski et al.2012 ; pawlowski & kroupa2013 ) . it would be interesting in the future to see whether ngc1851 is co - orbiting with this structure or not . it is conceivable that ngc1851 s host galaxy was tidally disrupted by the collision and its stars were dispersed into the milky way halo in a similar way as what we observe with m54 and the sagittarius dwarf today . the ultra low density halo of ngc1851 could therefore be the last trace of its ancient host . our spectroscopic analysis provides , for the first time , chemical abundances from low resolution data and the identification of two low rgb halo stars lying on the fiducial of ngc1851 , that may be useful to follow - up at higher resolution . the present analysis is confined to a relatively small sample , that does not show strong evidence for @xmath1-enrichment . we stress however that in the future , we need to extend this analysis to larger fields to increase the statistics and possibly confirm the lack / deficiency of fsgb stars in the halo that result from the present study . we warmly thank the referee christian moni bidin for his helpful comments that improved the quality of the paper . afm , ad and ma have been supported by grants fl110100012 and dp120100991 . apm , hj , gdc and jen acknowledge the financial support from the australian research council through discovery project grant dp120100475 . anderson , j. , sarajedini , a. , bedin , l. r. , et al . 2008 , aj , 135 , 2055 arlandini , c. , kppeler , f. , wisshak , k. , et al . 1999 , apj , 525 , 886 bekki , k. , & freeman , k. c. 2003 , mnras , 346 , l11 bekki , k. , & norris , j. e. 2006 , apj , 637 , l109 bekki , k. , & yong , d. 2012 , mnras , 419 , 2063 bellazzini , m. , ibata , r. a. , chapman , s. c. , et al . 2008 , aj , 136 , 1147 bellini , a. , bedin , l. r. , piotto , g. , et al . 2010 , aj , 140 , 631 bessell , m. , & murphy , s. 2012 , pasp , 124 , 140 carballo - bello , j. a. , gieles , m. , sollima , a. , et al . 2012 , mnras , 419 , 14 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , apj , 345 , 245 carretta , e. , bragaglia , a. , gratton , r. g. , et al . 2010 , a&a , 520 , a95 cassisi , s. , salaris , m. , pietrinferni , a. , et al . 2008 , apj , 672 , l115 castelli , f. , & kurucz , r. l. 2004 , arxiv : astro - ph/0405087

assume is a random sample of uniform , independent points from a triangle . the longest convex chain , , of is defined naturally ( see the next paragraph ) . the length of is a random variable , denoted by . in this paper we determine the order of magnitude of the expectation of . we show further that is highly concentrated around its mean , and that the longest convex chains have a limit shape .

0906.5452

let @xmath6 be a triangle with vertices @xmath7 and let @xmath8 be a finite point set . a subset @xmath9 is a _ convex chain _ in @xmath2 ( from @xmath10 to @xmath11 ) if the convex hull of @xmath12 is a convex polygon with exactly @xmath13 vertices . a convex chain @xmath3 gives rise to the polygonal path @xmath14 which is the boundary of this convex polygon minus the edge between @xmath10 and @xmath11 . the _ length _ of the convex chain @xmath3 is just @xmath4 . for most part of this paper we assume that @xmath15 is a random sample of @xmath1 random , uniform , independent points from @xmath2 . let @xmath5 be the length of a longest convex chain in @xmath0 . the random variable @xmath5 is a distant relative of the `` longest increasing subsequence '' problem , cf . @xcite . in this paper we establish several properties of @xmath5 . the first concerns its expectation , @xmath16 . [ limit ] there exists a positive constant @xmath17 for which @xmath18{n}}={\alpha}\,.\ ] ] theorem [ limit ] , together with some geometric arguments based on theorem [ mobius ] below , implies that the longest convex chains have a limit shape @xmath19 in the following sense . let @xmath20 be the collection of all longest convex chains from @xmath0 . for every @xmath21 @xmath22 where @xmath23 stands for the hausdorff distance . in fact , the statement of theorem [ limsh ] is much stronger , because there @xmath24 also converges to @xmath25 . the limit shape turns out to be the unique parabola arc @xmath26 that is tangent to the sides @xmath27 at @xmath10 and @xmath28 at @xmath11 , see figure [ parab ] a ) . the parabola arc @xmath19 will be called _ the special parabola _ in @xmath2 . the proof of the limit shape result is based on the following theorem , saying that @xmath5 is highly concentrated around its expectation . [ conc1 ] for every @xmath29 there exists a constant @xmath30 , such that for every @xmath31 @xmath32 for the quantitative version of the limit shape theorem we fix our triangle @xmath2 as @xmath33 . [ limsh ] let @xmath34 and define @xmath35 . then there exists @xmath36 , depending on @xmath37 , such that for every @xmath31 , @xmath38 when choosing one random point in triangle @xmath2 , the underlying probability measure is the normalized lebesgue measure on @xmath2 . most of the random variables treated in this paper ( e.g. @xmath5 ) are defined on the @xmath1th power of this probability space , to be denoted by @xmath39 . in this case @xmath40 denotes the @xmath1th power of the normalized lebesgue measure on @xmath2 . throughout the paper , @xmath41 stands for the ( lebesgue ) area measure on the plane . so when choosing @xmath1 independent random points in @xmath2 , the number of points in any domain @xmath42 is a binomial random variable of distribution @xmath43 . hence the expected number of points in @xmath44 is @xmath45 . for binomial random variables we have the following useful deviation estimates , which are relatives of chernoff s inequality , see @xcite , theorems a.1.12 and a.1.13 , pp 267 - 268 . if @xmath46 has binomial distribution with mean value @xmath47 and @xmath48 , then @xmath49 on the other hand , for @xmath50 , @xmath51 we will use ( [ hoeffding ] ) often , mainly with @xmath52 . = the special parabola arc @xmath19 in @xmath2 is characterized by the fact that it has the largest affine length among all convex curves connecting @xmath10 and @xmath11 within @xmath2 . ( for the definition and properties of affine arc length see @xcite or @xcite . ) this is a consequence of the following theorem from @xcite . assume that a line @xmath53 intersects the sides @xmath54 $ ] resp . @xmath55 $ ] at points @xmath56 and @xmath57 . let @xmath58 be a point on the segment @xmath59 $ ] and write @xmath60 resp . @xmath61 for the triangle with vertices @xmath62 resp . @xmath63 , see figure [ mobiusfig ] . [ mobius]@xcite under the above assumptions @xmath64{{\mathrm{a}}(t_1)}+\sqrt[3]{{\mathrm{a}}(t_2 ) } \leq \sqrt[3]{{\mathrm{a}}(t)}.\ ] ] equality holds here if and only if @xmath65 and @xmath53 is tangent to @xmath19 at @xmath58 . the equality part of the theorem implies the following fact . assume that @xmath66 are points , in this order , on @xmath19 . let @xmath67 be the triangle delimited by the tangents to @xmath19 at @xmath68 and @xmath69 , and by the segment @xmath70 $ ] , @xmath71 ; see figure [ parab ] b ) . [ areasum ] under the previous assumptions @xmath72{{\mathrm{a}}(t_i)}= \sqrt[3]{{\mathrm{a}}(t)}$ ] . in particular , when @xmath73 for each @xmath74 and @xmath75 , then @xmath76{{\mathrm{a}}(t)/t}<k$ ] . we will need a strengthening of theorem [ mobius ] . assume @xmath56 resp . @xmath57 divides the segment @xmath54 $ ] resp . $ ] in ratio @xmath77 and @xmath78 , see figure [ mobiusfig ] . = 0.55 [ mobi+ ] with the above notation @xmath64{{\mathrm{a}}(t_1)}+\sqrt[3]{{\mathrm{a}}(t_2 ) } \leq \sqrt[3]{{\mathrm{a}}(t)}- \sqrt[3]{{\mathrm{a}}(t)}\frac 13 ( a - b)^2.\ ] ] let @xmath79 be a number between @xmath25 and @xmath80 so that @xmath58 divides the segment @xmath59 $ ] in ratio @xmath81 . then , writing @xmath82 for the area of the triangle with vertices @xmath83 , @xmath84 showing @xmath85 . similarly , @xmath86 hence we have to prove the following fact : @xmath87 implies @xmath88{abc } - \sqrt[3]{(1-a)(1-b)(1-c ) } \geq \frac{1}{3 } ( a - b)^2.\ ] ] denote @xmath89 the left hand side of ( [ abc ] ) . by computing the derivative of @xmath89 with respect to @xmath79 yields that for fixed @xmath90 and @xmath91 , @xmath89 is minimal when @xmath92 it is easy to see that with this @xmath79 @xmath64{abc } + \sqrt[3]{(1-a)(1-b)(1-c ) } = \left(\sqrt{ab}+\sqrt{(1-a)(1-b)}\right)^{2/3}.\ ] ] now , denote @xmath93 by @xmath94 , so @xmath95 we claim that @xmath96 : this is the same as @xmath97 which is just the inequality between the arithmetic and geometric means for the numbers @xmath98 . therefore , using @xmath99 , @xmath100 theorems [ mobius ] and [ mobi+ ] imply the following [ erinto ] if @xmath101 and @xmath53 is tangent to @xmath19 at @xmath58 , then with the above notations , @xmath102 . it is clear that the underlying triangle @xmath2 can be chosen arbitrarily , since an affine transformation does not influence the value of @xmath5 . our standard model for @xmath2 is the one with @xmath103 , @xmath104 , @xmath105 as the vertices of @xmath2 . in this case the special parabola @xmath19 has equation @xmath106 . there are several choices for the underlying finite set @xmath107 . for instance , consider the lattice @xmath108 where @xmath109 is the usual lattice in @xmath110 and @xmath111 is large , and set @xmath112 . clearly , @xmath113 as @xmath114 . write @xmath115 for a longest convex chain in @xmath2 . it is shown in @xcite that , as @xmath114 ( or @xmath116 ) , @xmath117{t^2 { \mathrm{a}}(t)}(1+o(1))=\frac 6{(2\pi)^{2/3}}n^{1/3}(1+o(1)).\ ] ] this result is analogous to theorem [ limit ] , except that in the lattice case the value of the constant is known to be @xmath118 , while in the present paper only the existence of the limit @xmath17 is shown , together with @xmath119 , see section [ expect ] . this is similar to the longest increasing subsequence problem , @xcite , where it is easy to see that the expectation is of order @xmath120 , but proving the precise asymptotic formula @xmath121 turned out to be difficult , cf . @xcite and @xcite . in our case , numerical experiments suggest that @xmath122 and we venture to conjecture that this is the actual value of @xmath17 . more generally , let @xmath123 be a convex compact set with nonempty interior , and set @xmath124 . a set @xmath125 is said to be _ in convex position _ if no point of @xmath3 lies in the convex hull of the others . in other words , the convex polygon @xmath126 has exactly @xmath4 vertices . let @xmath127 be a maximum size subset of @xmath128 which is in convex position and set @xmath129 . it is shown in @xcite that @xmath130 where @xmath131 denotes the supremum ( actually , maximum ) of the affine perimeter that a convex subset of @xmath46 can have . the main difficulty lies in the case of triangles , that is , proving ( [ tria ] ) . these results can be extended , quite easily , to the present case when @xmath0 is a random sample of @xmath1 uniform independent points from @xmath46 . for instance , writing @xmath132 for the maximum size subset of @xmath0 in convex position , one can show the following . [ positio]under the above conditions @xmath133 here @xmath17 is the constant from theorem [ limit ] . one can also prove that @xmath134 has a limit shape , namely , the unique convex subset of @xmath46 whose affine perimeter is equal to @xmath131 . the proofs are almost identical to those used in @xcite , so we do not repeat them here , instead we rather explain what is different and more interesting . another random model is when @xmath107 comes from a homogeneous planar poisson process @xmath135 of intensity @xmath136 . given a domain @xmath44 in the plane , @xmath137 , the number of points in @xmath44 , has poisson distribution with parameter @xmath138 , i.e. @xmath139 we can also think of the poisson model as follows : for a domain @xmath44 , we first pick a random number @xmath140 according to the corresponding poisson distribution , and then choose @xmath140 random , independent , uniform points in @xmath44 . the advantage of the poisson model is that the number of points of @xmath135 in disjoint domains are independent random variables , unlike in the uniform model . as is well known , the uniform model @xmath0 and the poisson model @xmath135 behave very similarly . in particular , theorems [ limit ] , [ conc1 ] , and [ limsh ] remain valid for the poisson model as well , with essentially the same quantitative estimates . the proofs are quite standard , and we do not go into the details . actually , the proof of theorem [ limsh ] is simpler in the poisson model since there the subtriangles behave the same way as any other triangle . the longest increasing subsequence problem has been almost completely solved by now , see @xcite . in this respect , our results only constitute the first , and perhaps the simplest , steps in understanding the random variable @xmath5 . the main target of this section is to prove of theorem [ limit ] . we also establish upper and lower bounds for the constant involved . we start with an upper bound on @xmath16 : @xmath141{n } } \leq \sqrt[3]{2}e = 3.4248\dots.\ ] ] it is shown in @xcite , equation ( 5.3 ) ( cf . @xcite as well ) that the probability of @xmath142 uniform independent random points in @xmath2 forming a convex chain is @xmath143 therefore the probability that a convex chain of length @xmath142 exists is at most @xmath144 . in other words @xmath145 we use this estimate and stirling s formula to bound @xmath16 . assume @xmath146{2}e$ ] . then @xmath147{n } + \sum_{k > \gamma \sqrt[3]{n } } { \mathbb{p}}(l_n\ge k ) \\ & \leq \gamma \sqrt[3]{n } + \sum_{k > \gamma \sqrt[3]{n } } { n \choose k } \frac{2^k}{k ! \,(k+1)!}\\ & \leq \gamma \sqrt[3]{n } + \sum_{k > \gamma \sqrt[3]{n } } \frac{(2n)^k}{(k!)^3 } \\ & \leq \gamma \sqrt[3]{n } + \sum_{k > \gamma \sqrt[3]{n } } \frac{1}{\sqrt{(2\pi \gamma)^3 n } } \left(\frac{2\,e^3}{\gamma^3}\right)^k\\ & \leq \gamma \sqrt[3]{n } + n^{-1/2 } c,\end{aligned}\ ] ] where @xmath148 is a positive constant . since this holds for arbitrary @xmath146{2}\,e$ ] , ( [ felso ] ) is proved . next we establish a lower bound for @xmath16 . we use the second half of corollary [ areasum ] with @xmath149 . so we have triangles @xmath67 of area @xmath150 for @xmath151 , and the last [ intro]triangle @xmath152 of area less than @xmath150 . by ( [ areasum ] ) @xmath153{n/2}$ ] . let @xmath0 be the uniform independent sample from @xmath2 . let @xmath154 be a point of @xmath155 , provided that @xmath156 . the collection of such @xmath154 s forms a convex chain . hence the expected length of the longest convex chain is at least the expected number of non - empty triangles @xmath67 , so @xmath157{\frac{n}{2}}-1\right ) \ , \left(1-e^{-2}\right ) \approx 0.6862 \ , n^{1/3}.\end{aligned}\ ] ] what we have proved so far is that @xmath158 we show next that the limit exists . suppose on the contrary that @xmath159 . the idea of the proof is to use the second half of corollary [ areasum ] again , with the longest convex chain in the small triangles having length close to the limsup , while in the large triangle , @xmath16 is close to the liminf . for convenience , we suppose that @xmath160 choose a large @xmath1 with @xmath161{n}$ ] , and an @xmath30 much larger than @xmath1 with @xmath162{n}$ ] . here @xmath24 is a suitably small positive number . define @xmath163 so that the equation @xmath164 holds . choose @xmath30 uniform , independent random points from triangle @xmath2 . define @xmath165 . hence the expected number of points in a triangle ( contained in @xmath2 ) of area @xmath150 is @xmath163 . apply the second half of corollary [ areasum ] with this @xmath150 . then the number of triangles , @xmath142 , satisfies @xmath166{n/ n_1}$ ] . denote by @xmath167 the number of points in @xmath67 , and by @xmath168 the expectation of the length of the longest convex chain in @xmath67 . clearly @xmath167 has binomial distribution with mean @xmath163 , except for the last triangle where the mean is less than @xmath163 . since the union of convex chains in the triangles @xmath67 is a convex chain in @xmath2 between @xmath169 and @xmath170 , by estimate ( [ hoeffding ] ) we have @xmath171{n}\\ & \geq \left(\sqrt[3]{n/ n_1}-1\right)\left(1- n_1^{-1/2}\right ) ( 1-{\varepsilon } ) \,\overline{{\alpha } } \sqrt[3]{n}\\ & = \overline{{\alpha}}\ , \sqrt[3]{n}(1-{\varepsilon})\left(1-n_1^{-1/2}\right)\left(\sqrt[3]{n / n_1}-\sqrt[3]{n / n}\right)\\ & \ge \overline{{\alpha } } \,\sqrt[3]{n}(1 - 2{\varepsilon}),\end{aligned}\ ] ] where the last inequality holds if @xmath1 is chosen large enough and @xmath30 is chosen even larger with @xmath172 very small . thus @xmath173 which , for small enough @xmath24 , contradicts our assumption @xmath174 . * remark . * the lower bound @xmath175 is probably the easiest to prove . a better estimate , also mentioned by enriquez @xcite , can be established as follows . assume @xmath2 is the standard triangle and let @xmath44 denote the domain of @xmath2 lying above @xmath176 . then @xmath177 , so the expected number of points in @xmath44 is @xmath178 , and the number of points is concentrated around this expectation . the affine perimeter of @xmath44 is @xmath179{1/2}$ ] ( see @xcite ) , and a classical result of rnyi and sulanke @xcite yields that expected number of vertices of @xmath180 is about @xmath181{\frac{2}{3 } } \left(\frac{1}{3 } \right)^{-1/3 } 2\sqrt[3]{1/2 } \ , \sqrt[3]{2n/3 } \approx 1.5772 \,\sqrt[3]{n}\ ] ] since most vertices are located next to the parabola , the majority of them form a convex chain , and so @xmath182{n } } \geq 1.5772\dots.\ ] ] this sketch can be completed with standard tools . from now on , we will use this estimate . also , @xmath183 will always refer to the limit constant of theorem [ limit ] . the concentration results proved here are consequences of talagrand s inequality from @xcite which says the following . suppose @xmath3 is a real - valued random variable on a product probability space @xmath184 , and that @xmath3 is 1-lipschitz with respect to the hamming distance , meaning that @xmath185 whenever @xmath186 and @xmath187 differ in one coordinates . moreover assume that @xmath3 is _ @xmath188-certifiable_. this means that there exists a function @xmath189 with the following property : for every @xmath186 and @xmath91 with @xmath190 there exists an index set @xmath191 of at most @xmath192 elements , such that @xmath193 holds for every @xmath187 agreeing with @xmath186 on @xmath191 . let @xmath140 denote the median of @xmath3 . then for every @xmath194 we have + @xmath195 and @xmath196 when applied to @xmath5 , these inequalities prove concentration about the median , to be denoted by @xmath197 . theorem [ conc1 ] concerns the mean of @xmath5 . however , concentration ensures that the mean and the median are not far apart , in fact , @xmath198 . first we need a lower bound on @xmath197 . [ median ] suppose that @xmath199 . then @xmath200{3n/\log n}.\ ] ] since this is a special case of lemma [ mediansub ] from the next section , the proof will be given there . the statement cries out for the application of talagrand s inequality . the random variable @xmath5 satisfies the conditions with @xmath201 , since fixing the coordinates of a maximal chain guarantees that the length will not decrease , and changing one coordinate changes the length of the maximal chain by at most one . write @xmath202 for the median in the present proof . setting @xmath203 where @xmath204 is an arbitrary positive constant , we have @xmath205 define now @xmath206 with a constant @xmath48 , which will be fixed at the end of the proof in order to give the correct estimate . if @xmath207 , then @xmath208 , and the denominator in the exponent is at most @xmath209 . thus @xmath210 on the other hand , for @xmath211 we have @xmath212 next , we compare the median and the expectation of @xmath5 . @xmath213 the range of @xmath5 is @xmath214 $ ] , so the integrand is @xmath25 if @xmath215 . substitute @xmath216 , and divide the integral into two parts at @xmath217 : @xmath218 where @xmath219 and @xmath220 by lemma [ median ] , @xmath221 , so @xmath222 . since @xmath197 goes to infinity as @xmath1 increases ( again by lemma [ median ] ) , the bound on @xmath223 is eventually much smaller than the one on @xmath224 : @xmath225 for all large enough @xmath1 . hence @xmath16 is of the same order of magnitude as @xmath197 , and we obtain @xmath226 for fixed @xmath37 and for large enough @xmath1 , ( [ exp - med ] ) implies @xmath227 using @xmath228 from ( [ felso ] ) and ( [ medianlim ] ) , it is easy to see that @xmath229 since for large enough @xmath1 , @xmath230 , ( [ betasmall ] ) finally implies @xmath231 with ( [ medianlim ] ) and the choice of @xmath232 . * remark . * the constant in the exponent is far from being best possible . we have made no attempt to find its optimal value . in general , talagrand s inequality is too general to give the precise concentration , see talagrand s comments on this in @xcite . for the proof of theorem [ limsh ] we need to consider subtriangles @xmath233 of @xmath2 , that is , triangles of the form @xmath234 with @xmath235 , while @xmath0 is still a random sample from @xmath2 . we will need to estimate the concentration of the longest convex chain from @xmath0 in @xmath233 . since this random variable depends only on the relative area of @xmath233 , we may and do assume that @xmath2 is the standard triangle and @xmath236 . thus @xmath237 . write @xmath238 for the length of the longest convex chain in @xmath233 from @xmath239 to @xmath240 , and @xmath241 for its median . in the following statements , we consider the situation when @xmath242 , the expected number of points from @xmath0 in @xmath233 , tends to infinity . as in the proof of theorem [ conc1 ] , we need two estimates : a lower bound for the median guarantees that the mean and the median are close to each other , while an upper bound for the expectation ( or for the median ) is needed for deriving the inequality in terms of @xmath1 . here comes the lower bound ; the case @xmath243 is lemma [ median ] . [ mediansub ] suppose that @xmath244 . then @xmath245{3ns/\log ( ns)}.\ ] ] set @xmath246 , and apply the second half of corollary [ areasum ] to the triangle @xmath233 . the number of triangles is @xmath142 with @xmath64{3ns/\log ( ns)}<k \leq \sqrt[3]{3ns/\log ( ns ) } + 1.\ ] ] for any @xmath247 } , the probability that @xmath67 contains no point of @xmath0 is @xmath248 hence the union bound yields @xmath249{3ns/\log ( ns)}\,\big ) & \ge 1-{\mathbb{p}}(t_i\cap x_n = \emptyset \mbox { for some } i \le k ) \\ & \ge 1-k\,(ns)^{-1/3}\\ & \ge 1- \big ( \sqrt[3]{3/\log ( ns ) } + ( ns)^{-1/3 } \big),\end{aligned}\ ] ] which is greater than @xmath250 by the assumption . obtaining an upper bound for the mean is slightly more delicate ; note that in the lemma below @xmath251 need not be fixed . [ meansub ] assume @xmath252 . then @xmath253 where @xmath183 is the same constant as in theorem [ limit ] . take any @xmath21 and choose @xmath254 ( depending on @xmath24 ) so large that for every @xmath255 , @xmath256 . the random variable @xmath257 has binomial distribution with mean @xmath258 . when @xmath258 is large enough , @xmath259 , and we use ( [ hoeffding ] ) for a lower estimate : @xmath260 for the upper bound , jensen s inequality applied to @xmath261{x}$ ] comes in handy : @xmath262{k } \\ & \leq n_0 \ , + { \mathbb{p}}(k \geq n_0 ) \ , ( 1+{\varepsilon } ) \ , \alpha \,\left ( \sum_{k = n_0}^{n } \frac{{\mathbb{p}}(k = k)}{{\mathbb{p}}(k \geq n_0 ) } \ k \right)^{1/3}\\ & \leq n_0 \ , + { \mathbb{p}}(k \geq n_0)^{2/3 } \ , ( 1+{\varepsilon } ) \ , \alpha \ , ( { \mathbb{e}}\ , k)^{1/3}\\ & \leq n_0 + ( 1+{\varepsilon } ) \ , \alpha \ , ( ns)^{1/3}\leq ( 1 + 2 { \varepsilon } ) \ , \alpha \ , ( ns)^{1/3}. \qedhere\end{aligned}\ ] ] next , we derive the strong concentration property of @xmath238 , the analogue of theorem [ conc1 ] . [ concsub ] suppose @xmath263 is a constant with @xmath264 . then for every @xmath29 there exists a constant @xmath30 , such that for every @xmath31 and every @xmath265 , @xmath266 this proof is almost identical with that of theorem [ conc1 ] . since @xmath238 is a random variable on @xmath267 , we can apply talagrand s inequality with the certificate function @xmath201 in the same way as in the proof of theorem [ conc1 ] . write again @xmath140 for @xmath241 , the median of @xmath238 . define @xmath268 with @xmath232 , then the estimates ( [ betasmall ] ) and ( [ betabig ] ) remain valid with @xmath238 in place of @xmath5 . just as before , @xmath269 where @xmath224 and @xmath223 are defined the same way as in ( [ i1 ] ) and ( [ i2 ] ) . moreover , @xmath224 satisfies the inequality ( [ i1 ] ) . with @xmath223 we have to be a bit more careful . note that @xmath270 with @xmath271 guarantees that lemma [ mediansub ] is applicable for @xmath272 . as @xmath273 is monotone increasing for @xmath274 , @xmath275{\frac{3ns}{\log ( ns ) } } \ge \sqrt[3]{\frac{3n^{1-\tau}}{(1-\tau ) \log n } } > \sqrt[3]{\frac{n^{1-\tau } } { n^{(1-\tau)/2 } } } = n^{(1-\tau)/6}\ ] ] for large enough @xmath1 , and therefore by ( [ i2 ] ) @xmath276 where of course @xmath277 . lemma [ mediansub ] implies that @xmath278 , thus the bound on @xmath223 is much smaller than the one on @xmath224 for large enough @xmath1 . therefore , just as in ( [ exp - med ] ) , @xmath279 hence @xmath280 is of the same order of magnitude as @xmath281 . since @xmath282 , we can use lemma [ meansub ] , obtaining that for large enough @xmath1 , @xmath283{ns}.\ ] ] again for fixed @xmath37 and for large enough @xmath1 , @xmath284 and by ( [ submedfelso ] ) , @xmath285 since for large enough @xmath1 , @xmath286 , ( [ betasmall ] ) applied to @xmath238 and ( [ submedfelso ] ) finally implies @xmath287 * remark . * the proof also yields that for any @xmath288 , there exists @xmath30 ( depending on @xmath289 and @xmath290 only ) , such that the inequality of theorem [ concsub ] holds for any @xmath291 $ ] and for every @xmath31 . for the proof of theorem [ limsh ] we need further preparations . we start by assuming that @xmath46 is a convex compact set in the plane and @xmath292 , and @xmath293 is a random sample of @xmath1 uniform and independent points from @xmath46 . we need to estimate the probability that @xmath293 is in convex position , that is , no point of @xmath293 is contained in the convex hull of the others . we denote this probability by @xmath294 . [ l : convpos]if @xmath46 is as above , @xmath295 let @xmath296 be the smallest area parallelogram containing @xmath46 . as is well known , @xmath297 . let @xmath298 be a random sample of @xmath1 uniform and independent points from @xmath296 . in this case a ( surprisingly exact ) result of valtr @xcite says that @xmath299 now we have @xmath300 where the last step is a straightforward estimate . > from now on we work exclusively with the standard triangle @xmath2 . assume next that @xmath46 is a convex subset of the triangle @xmath2 , and let @xmath0 be random sample of @xmath1 uniform and independent points from @xmath2 . we define @xmath301 as the random variable @xmath302 > from theorem [ positio ] it is not hard to determine what the asymptotic expectation of @xmath301 is . but what we need is that @xmath301 is large with small probability . this is the content of the next lemma . [ l : m > m ] let @xmath46 be a convex subset of @xmath2 . then for any positive integers @xmath1 and @xmath303 satisfying @xmath304 , @xmath305 if @xmath306 , then @xmath307 contains a subset of size @xmath303 which is in convex position . lemma [ l : convpos ] and the union bound imply that @xmath308 the random variable @xmath309 has binomial distribution . thus we have @xmath310+\sum_{k = k_0}^{n}[ .. ].\end{aligned}\ ] ] here we choose @xmath311 to be equal to @xmath312 . then @xmath313\le \sum_{k < k_0}\left(\frac { 240\,e\,k_0}{\mu ^3}\right)^{\mu } < k_0 2^{-\mu } < \mu ^3 2^{-\mu}.\ ] ] since @xmath314 is decreasing for @xmath315 , and the condition on @xmath303 guarantees that @xmath316 , @xmath317 & \le & n{n \choose k_0}(2a(k))^{k_0}(1 - 2a(k))^{n - k_0}\\ & \le & n\left(\frac{ne}{k_0}\right)^{k_0}(2a(k))^{k_0 } = n \left(\frac{2ea(k)n}{k_0}\right)^{k_0}\\ & < & n2^{-k_0}=n2^{-\mu ^3/(480e)}.\end{aligned}\ ] ] for the proof of theorem [ limsh ] we will consider other parabolas that are similar to @xmath19 . let @xmath318 be the parabola defined by the equation @xmath319 where the parameter @xmath320 . the graph of @xmath318 is the homothetic copy of @xmath19 with ratio of homothety @xmath321 , and center of homothety at the origin , see figure [ parab2 ] a ) . assume the point @xmath322 is on @xmath19 . then the point @xmath323 is on @xmath318 , and the tangent line to this point on @xmath318 is given by the equation @xmath324 it follows that the distance between parallel tangent lines to @xmath19 and @xmath318 is @xmath325 define now @xmath326 here @xmath24 comes from theorem [ limsh ] . this definition immediately implies the following fact . [ pr : dist]if a convex chain @xmath14 lies between @xmath327 and @xmath328 , then @xmath329 . we need one more piece of preparation . assume @xmath53 is a tangent to @xmath318 , at the point @xmath330 . with the notations of section [ prelim ] , let @xmath60 and @xmath61 denote the two triangles determined by @xmath53 and @xmath330 , see figure [ parab2 ] a ) . let @xmath0 be a random sample of @xmath1 points from @xmath2 and let @xmath331 denote the length of the longest convex chain in @xmath67 , @xmath332 . = [ l:2chains ] for sufficiently large @xmath1 , if @xmath333 , then @xmath334{n}.\ ] ] let @xmath335 for @xmath332 . we want to apply theorem [ mobi+ ] . it is not hard to see ( using corollary [ erinto ] for instance ) that what is denoted by @xmath336 there , is equal to @xmath337 here . consequently @xmath338{t_1 /2 } + \sqrt[3]{t_2 /2 } \leq \sqrt[3]{1/2 } -\sqrt[3]{1/2 } \ ; \frac 1 3 \ , r^2.\ ] ] write @xmath331 for the longest convex chain in the triangle @xmath67 . by affine invariance @xmath331 has the same distribution as @xmath339 ( from section [ subtri ] ) for @xmath332 . we need to estimate @xmath340 from below . for four points @xmath341 , @xmath58 , @xmath57 and @xmath342 in this order on @xmath19 , denote by @xmath343 the triangle delimited by the tangents to @xmath19 at @xmath344 , and by the segment @xmath70 $ ] , @xmath345 ; see figure [ parab2 ] b ) . choose @xmath58 and @xmath57 so that @xmath346 and @xmath347 . then corollary [ areasum ] and ( [ terhiany ] ) imply that @xmath64 { { \mathrm{a}}(s_3 ) } \geq \sqrt[3]{1/2}\ ; \frac 1 3 \ , r^2.\ ] ] let now @xmath348 denote the length of a longest chain in @xmath343 for @xmath345 . for @xmath349 and @xmath350 , @xmath348 has the same distribution as @xmath339 ( and as @xmath331 ) . therefore @xmath351 for @xmath332 . further , @xmath352 follows from concatenating the longest convex chains in the triangles @xmath343 . thus we have @xmath353 the random variable @xmath354 has binomial distribution with mean @xmath355 which is at least @xmath356 . set @xmath357 . thus we obtain that for all large enough @xmath1 , @xmath358 and @xmath30 tends to infinity with @xmath1 . using the estimates ( [ hoeffding ] ) and ( [ also ] ) , again for large @xmath1 we have @xmath359n.\end{aligned}\ ] ] hence , by ( [ osszegkicsi ] ) @xmath360n.\ ] ] after the preparations in the previous sections we finally prove theorem [ limsh ] , that is , all chains in @xmath361 lie in a small neighbourhood of @xmath19 with high probability . note that similar limit shape results have been proved for convex chains @xcite ; however , they are of different character than the present case . we fix the constant @xmath362 . every result in this chapter holds for large enough @xmath1 , depending only on @xmath363 . we will not always mention this . for this proof we set @xmath364 . the strong concentration result of theorem [ conc1 ] directly shows that @xmath365 we call a convex chain @xmath366 _ long _ if its length is at least @xmath367 . we will show that all long convex chains lie between the parabolas @xmath368 and @xmath327 with high probability , where high means @xmath369 . in view of proposition [ pr : dist ] this suffices for the proof . let @xmath233 be the triangle with vertices @xmath370 , and define @xmath371 to be the event that there is a long convex chain @xmath372 having a point in @xmath233 . we prove first the following simple fact . [ l : h ] for @xmath1 large enough , @xmath373 let @xmath3 be a long convex chain with a point in @xmath233 , and let @xmath187 be a point of @xmath3 where the tangent to @xmath14 has slope @xmath80 . clearly @xmath374 . let @xmath375 be the part of @xmath3 between @xmath376 and @xmath187 , and @xmath377 be the part between @xmath187 and @xmath378 . then @xmath375 resp . @xmath377 are convex chains in the triangle @xmath379 and @xmath380 . as @xmath3 is a long convex chain , @xmath381 where @xmath331 denotes the length of the maximal chain in @xmath343 ( @xmath332 ) , @xmath382 . as @xmath383 , the limit of @xmath384 resp . @xmath385 is @xmath17 and @xmath386{0.1}$ ] . this follows from theorem [ limit ] and lemma [ meansub ] . so @xmath387{0.1})>1/10 $ ] , implying that for large enough @xmath1 @xmath388n > 3b= 3{\gamma}n^{1/6}\sqrt{\log n}.\ ] ] so we have @xmath389 the triangle @xmath343 is of area @xmath390 so theorem [ concsub ] shows that @xmath391 after this first step , we estimate the probability of the existence of a long convex chain not lying between @xmath327 and @xmath368 . first , we deal with the case when the chain goes below this region . we define a set of parabolas . let @xmath392 , @xmath393 , and @xmath394 note that @xmath395 . here we define @xmath396 by the conditions @xmath397 but @xmath398 is not contained in @xmath233 . thus the case when a long chain goes below @xmath399 is covered by lemma [ l : h ] . clearly @xmath396 is limited by @xmath400 . thus @xmath401 , say . the convex polygonal chains @xmath14 can be considered as functions defined on @xmath402 $ ] . we extend the definition of @xmath318 as 0 on the interval @xmath403 $ ] if @xmath404 , and consider this new `` parabola '' @xmath318 as a function defined on @xmath402 $ ] . a parabola is said to be _ below _ , resp . _ above _ @xmath14 if the corresponding function is smaller ( larger ) than the one corresponding to @xmath14 . the following lemma is important . [ l : finite ] there are points @xmath405 with @xmath406 , such that the upper envelope of the tangent lines @xmath407 of @xmath398 at @xmath408 is a broken polygonal path lying above @xmath409 . the line @xmath410 , which is tangent to @xmath398 at @xmath411 , intersects the graph of @xmath409 in two points . let @xmath412 denote the segment connecting these two points . it is not hard to check that the length of the segment , @xmath413 , decreases as @xmath330 moves away from the center point of @xmath398 . a simple computation reveals that @xmath414 where @xmath330 only moves up to the point when both endpoints of @xmath415 lie in @xmath409 . now choose @xmath416 on @xmath398 so that the lower endpoint of @xmath417 is the intersection of @xmath409 with the @xmath186-axis . once @xmath408 has been defined , we let @xmath418 be the point in @xmath398 for which the lower endpoint of @xmath419 coincides with the upper endpoint of @xmath420 , see figure [ parab3 ] a ) . the length of @xmath421 is smaller than @xmath422 . so the process of choosing the @xmath408 stops after @xmath423 steps . this finishes the construction of the points @xmath408 . the upper envelope of the tangent lines @xmath407 is a convex polygonal path that lies between @xmath409 and @xmath398 with edges @xmath420 . = now we define @xmath424 to be the event that there is a long convex chain @xmath366 with @xmath425 below @xmath14 but @xmath409 not below @xmath14 , @xmath426 . we split these events further . let @xmath427 be the event that there is a long convex chain @xmath3 with the parabola @xmath425 below @xmath14 but the line @xmath407 not below @xmath14 ; here @xmath428 comes from lemma [ l : finite ] . this implies that @xmath429 . [ l : gij]for every @xmath430 and every @xmath431 , @xmath432 . before the proof we state ( and prove ) the following corollary . [ cor1 ] the probability that there is a long convex chain @xmath366 such that @xmath14 is not above @xmath327 is at most @xmath433 . this is quite easy : if there is such a chain , then either @xmath371 , or some @xmath424 ( @xmath426 ) occur . since @xmath434 , @xmath435 and @xmath436 , the corollary follows from lemmas [ l : gij ] and [ l : h ] . let @xmath437 be the two triangles determined by @xmath408 and @xmath407 as usual , and let @xmath438 be the convex set between @xmath420 and @xmath425 , see figure [ parab3 ] b ) . we estimate @xmath439 as follows . a simple calculation as in ( [ eq : laq ] ) yields that the diameter of @xmath46 is at most @xmath440 , and @xmath46 is between the line @xmath407 and the parallel line tangent to @xmath441 . the distance of these lines is at most @xmath442 as one can easily check using ( [ eq : dist ] ) . then @xmath443 . a long convex chain @xmath444 which is above @xmath425 but not above @xmath407 splits into 3 parts : @xmath445 , @xmath446 , and @xmath447 . here @xmath448 are convex chains in @xmath60 ( from @xmath376 to @xmath408 ) and in @xmath61 ( from @xmath408 to @xmath378 ) , and @xmath449 is in convex position in @xmath46 . so with the notations of the previous section we have @xmath450 since @xmath3 is a long convex chain , @xmath451 . this implies that @xmath452 . we are going to show that this event has small probability . we apply lemma [ l : m > m ] with @xmath453 . for large enough @xmath1 it implies that @xmath454 since the condition @xmath455 is satisfied as @xmath456 and @xmath457 . next , @xmath458 now lemma [ l:2chains ] implies that @xmath459n$ ] , and hence @xmath460n -1.2\,b)\\ & \le \sum_{i=1,2}{\mathbb{p}}(l^i\ge { \mathbb{e}}l^i+0.26 r_{i-1}^2\sqrt[3]n -0.6\ , b)\nonumber \\ & \le \sum_{i=1,2}{\mathbb{p}}(l^i\ge { \mathbb{e}}l^i+4b).\nonumber\end{aligned}\ ] ] here the last step is justified by observing that @xmath461 and so for large enough @xmath1 @xmath462n & \geq 0.26\ , n^{1/3}\big(3 \sqrt{2 } { \gamma}^{1/2}n^{-1/12}(\log n)^{1/4 } - n^{-1/3}\sqrt{\log n } \big)^2 \nonumber \\ & > 4.6 { \gamma}\ , n^{1/6}\sqrt{\log n}=4.6 \ , b.\end{aligned}\ ] ] next , we estimate @xmath463 . when @xmath464 , we use theorem [ concsub ] with @xmath465 : @xmath466 the last inequality holds because of the remark following theorem [ concsub ] , since @xmath467 finally , when @xmath468 , the expected number of points in @xmath67 is @xmath469 . so for the random variable @xmath470 inequality ( [ hoeffdingfelso ] ) implies that @xmath471 for large enough @xmath1 , and hence @xmath472 thus @xmath473 for @xmath332 in all cases . now we handle the case of parabolas going above @xmath328 . set @xmath474 where @xmath475 . we define another series of parabolas : @xmath476 where @xmath188 is limited by @xmath477 . thus @xmath478 , say . the following geometric lemma is similar to lemma [ l : finite ] . [ l : finite+ ] there are points @xmath479 with @xmath480 such that the following holds . for each convex chain @xmath444 with @xmath481 above @xmath14 but @xmath482 not above @xmath14 , there is a @xmath483 such that the line @xmath484 is below @xmath14 . for such a long chain @xmath3 there is a smallest @xmath485 with @xmath486 above @xmath14 . then @xmath14 and @xmath486 have a common point and a common tangent @xmath53 at that point ( because both @xmath14 and @xmath486 are convex ) . let @xmath487 be the point on @xmath482 such that the line @xmath488 , tangent at @xmath487 to @xmath482 , is parallel with @xmath53 . it is evident that @xmath14 is above @xmath488 . let @xmath489 denote the set of lines that are tangent to @xmath482 and that have both @xmath169 and @xmath170 above it . we will construct a set of points @xmath490 such that each line in @xmath489 is above the segment @xmath491 for some @xmath492 . this construction then guarantees what the lemma requires . we need one more piece of notation . given @xmath483 let @xmath493 $ ] be the segment @xmath494 , with @xmath495 on the @xmath186-axis and @xmath496 on the @xmath187-axis . we shall construct the sequence of the @xmath495 s and @xmath496 s . the construction starts with @xmath497 at the midpoint of @xmath498 and we define first the other @xmath483 with @xmath499 closer to the origin than @xmath495 . see figure [ parab4 ] a ) . assume @xmath483 has been found . there is a unique tangent , @xmath53 , to @xmath482 passing through @xmath496 . let @xmath500 be the intersection point of @xmath53 with the @xmath186-axis , and @xmath501 the common point of @xmath498 with the tangent to @xmath498 through @xmath500 . the construction is finished when we reach @xmath502 , here @xmath503 denotes the @xmath186-coordinate of @xmath495 . corollary [ erinto ] implies that @xmath504 since @xmath505 , we reach @xmath502 after at most @xmath506 steps . the construction satisfies what we need : if a tangent to @xmath482 intersects the triangle in the segment @xmath507 $ ] with @xmath289 on the @xmath186 axis and @xmath508 $ ] , then @xmath289 is between @xmath500 and @xmath495 for some @xmath509 , and the segment @xmath507 $ ] is above the segment @xmath510 . the construction is extended to the other half of @xmath498 symmetrically , and @xmath511 follows . = next we define @xmath512 ( @xmath513 ) to be the event that there is a long convex chain @xmath444 such that @xmath481 is above @xmath14 but @xmath482 is not above @xmath14 , @xmath513 . further , let @xmath514 be the event there is a long convex chain @xmath366 with @xmath14 below @xmath481 but not below @xmath484 ( remember that @xmath515 ) . here @xmath516 and @xmath517 . we have now the following result , similar to lemma [ l : gij ] . [ l : fij]for every @xmath516 and every @xmath518 , @xmath519 . this lemma immediately implies the following corollary . [ cor2 ] the probability that there is a long convex chain @xmath366 such that @xmath14 is not below @xmath368 is at most @xmath520 . the proof follows from the facts that @xmath521 , @xmath478 , @xmath522 , and @xmath436 . now we give the proof of lemma [ l : gij ] which is analogous to that of lemma [ l : gij ] . let @xmath488 be the unique tangent to @xmath425 which is parallel with @xmath484 , and @xmath487 be the common point of @xmath488 and @xmath441 , see figure [ parab4 ] b ) . let @xmath437 be the two triangles determined by @xmath487 and @xmath488 , and let @xmath438 be the part of @xmath2 that lies between @xmath484 and @xmath488 . since the distance of these two lines is at most @xmath523 , @xmath524 . a long convex chain @xmath444 which is below @xmath481 but not below @xmath484 splits into 3 parts : @xmath445 , @xmath446 , and @xmath447 . here @xmath448 are convex chains in @xmath60 ( from @xmath376 to @xmath487 ) and in @xmath61 ( from @xmath487 to @xmath378 ) , and @xmath449 is in convex position in @xmath46 . so @xmath450 since @xmath3 is a long convex chain , @xmath525 , and so @xmath452 . we are going to show that this event has small probability . we apply lemma [ l : m > m ] again with @xmath453 . for sufficiently large @xmath1 the condition @xmath455 is satisfied , since @xmath526 and @xmath527 . so we have , just as in ( [ eq : last ] ) , @xmath528 therefore the estimate ( [ eq : start ] ) applies without change : @xmath529 now lemma [ l:2chains ] implies that @xmath530n$ ] , and just as in ( [ eq : cont ] ) , @xmath531n -0.6 \ , b)\\ & \leq \sum_{i=1,2}{\mathbb{p}}(l^i\ge { \mathbb{e}}l^i+4b).\end{aligned}\ ] ] here the last step is justified just as in ( [ eq : folytat ] ) except that this time @xmath532 . finally , we bound @xmath463 the same way as in the proof of lemma [ l : gij ] to obtain @xmath533 considering proposition [ pr : dist ] , we have to estimate the probability that there is a _ longest _ convex chain not lying between @xmath327 and @xmath368 . this event splits into two parts : either the longest convex chain is not long , or there is a long convex chain not between @xmath327 and @xmath368 . the probability of the first event is estimated by theorem [ conc1 ] , while the second part is handled via corollaries [ cor1 ] and [ cor2 ] . therefore the probability in question is at most @xmath534 * remarks . * in this proof one can avoid using the estimate on @xmath535 . in fact , choosing @xmath536 and @xmath537 small enough , the set @xmath46 contains more than @xmath538 points of @xmath0 with very small probability . so , with high probability , it can not add much to the size of a long convex chain . there are more events @xmath424 and @xmath427 , which has a minor effect on the final result . also , the triangle @xmath233 in lemma [ l : h ] is to be chosen much smaller . an important step in our proof is lemma [ l:2chains ] , essentially implying that if the distance between @xmath19 and the farthest point of a convex chain from @xmath19 is `` large '' , then the chain can not be too long . conditioning on the location of this farthest point would allow an elegant conditional expectation argument . however , fixing the farthest point modifies the underlying probability space and therefore the estimate coming from lemma [ l:2chains ] is no longer valid . to eliminate this difficulty , we chose to define finitely many subcases and estimate them separately , which can also be considered as a finite approximation of the continuous conditional expectation . in the final section we summarize the observations obtained by computer simulations . the search for the longest convex chains can be accomplished by an algorithm which has running time @xmath539 . this algorithm works as follows . we order the points by increasing @xmath186 coordinate , and then recursively create a list at each point . the @xmath142th element on the list at point @xmath487 contains the minimal slope of the last segment of chains starting at @xmath10 and ending at @xmath487 whose length is exactly @xmath142 , and a pointer to the other endpoint of this last segment . for creating the list at the next point @xmath487 , we have to search the points before @xmath487 , and see if @xmath487 can be added to the chains while preserving convexity . this algorithm can be speeded up with some ( not fully justified but useful ) tricks . first of all , theorem [ limsh ] guarantees that we have to search only among the points close to @xmath19 . the simulations show that most longest convex chains are located in a small neighbourhood of @xmath19 , whose radius is in fact of order approximately @xmath540 , much smaller than the width of order @xmath541 given by theorem [ limsh ] . therefore the search can be restricted to a subset of the points with cardinality of order @xmath542 . second , when looking for the longest chain , we have to search only points relatively close to @xmath487 , and chains which are already relatively long , thus reducing memory demands . [ tabla ] c| c | c | c | c @xmath1 & @xmath543 ' '' '' & @xmath544 & distance@xmath545 & deviation + 1000 ' '' '' & 2.532 & 4 & 0.270 & 1.254 + 10000 & 2.768 & 5 & 0.200 & 1.383 + 15625 & 2.813 & 5 & 0.150 & 1.293 + 50000 & 2.885 & 5 & 0.100 & 1.411 + 75000 & 2.906 & 5 & 0.070 & 1.580 + 100000 & 2.917 & 5 & 0.060 & 1.431 + 125000 & 2.926 & 5 & 0.050 & 1.637 + 421875 & 2.959 & 5 & 0.012 & 1.732 + 1000000 ' '' '' & 2.976 & 6 & 0.012 & 2.023 + + with the above method , the search can be executed for up to @xmath546 active points , in which case examining one sample takes about 2 minutes . as the experiments show , this provides a good approximation for @xmath1 s up to order @xmath547 . in each experiment , we increased the width of the searched neighbourhood until the increment did not generate a significant change in the average length of the longest convex chain . the results obtained by this method , although giving only a lower bound for @xmath16 , are heuristically close to it . @xmath548 . = 0.4 our largest search has been done for @xmath549 . the number of samples was @xmath550 except for the cases @xmath551 and @xmath552 , where we used @xmath553 samples in order to model the distribution of @xmath5 ( see figure [ eloszl ] ) . the results below well illustrate what the proof of theorem [ limit ] suggests , namely , that @xmath554 is increasing with @xmath1 . also , the data seem to confirm that @xmath548 . on table 1 we list the results obtained by the program . the first column is the number of points chosen in @xmath2 , the second is the average of @xmath555 . the third column contains the half - length of the interval of the values of @xmath5 , that is , latexmath:[$d_n= \lfloor \max @xmath552 . in the fourth column we list @xmath557 times the radius of the neighbourhood of parabola we used for the search ( the term @xmath558 comes from a transformation of coordinates ) . the last data are the standard deviation of the set of values of @xmath5 , ie . the square - root of its variance . figure [ varh ] illustrates the linear interpolation of @xmath559 as a function of @xmath560 . it is based on the data shown on table 1 . as we know from theorem [ conc1 ] , @xmath5 is highly concentrated near its expectation . this phenomenon is well recognizable on figure [ eloszl ] , where we plot the distribution in the cases @xmath561 and @xmath552 with @xmath553 samples . we express our special thanks to gbor tusndy for his constant attention and interest in this piece of work , for valuable ideas concerning computer simulations , and in particular for pointing out an error in the earlier version of this paper . we also thank zoltn kovcs for his suggestions regarding the implementation of the program . the second author was supported by hungarian national foundation grants t 60427 and t 62321 . finally , we dedicate this piece of work to the memory of the late professor sndor csrg , whose zest for life and enthusiasm for mathematics will always be a constant inspiration to us . gergely ambrus + department of mathematics + university college london + gower street , london wc1e 6bt + england , u.k . + and + bolyai institute + university of szeged + aradi vrt . tere 1 , 6720 szeged + hungary + e - mail : [email protected] imre brny + rnyi institute of mathematics + hungarian academy of sciences + po box 127 , 1364 budapest + hungary + and + department of mathematics + university college london + gower street , london wc1e 6bt + england , u.k . + e - mail : [email protected]

* abstract * we consider the compatibility of dama / libra , cogent , xenon10 and xenon100 results for spin - independent ( si ) dark matter weakly interacting massive particles ( wimps ) , particularly at low masses ( gev ) , assuming a standard dark matter halo . the xenon bounds depend on the scintillation efficiency factor for which there is considerable uncertainty . thus we consider various extrapolations for at low energy . with the measurements we consider , xenon100 results are found to be insensitive to the low energy extrapolation . we find the strongest bounds are from xenon10 , rather than xenon100 , due to the lower energy threshold . for reasonable choices of and for the case of si elastic scattering , xenon10 is incompatible with the dama / libra 3 region and severely constrains the 7 - 12 gev wimp mass region of interest published by the cogent collaboration .

1006.0972

the nature of the dark matter that comprises a quarter of the universe is one of the big unanswered questions in astrophysics and particle physics . perhaps the best motivated candidates are weakly interacting massive particles ( wimps ) which have weakly interacting cross sections and masses in the gev10 tev range . in recent months , there have been new data releases from many experiments that have engendered a great deal of excitement . of particular interest is a low mass region @xmath3 gev which at first sight seems to be compatible with a number of different experiments . the goal of this paper is to examine some of the issues regarding the question of such compatibility . in this paper , we will restrict ourselves here to considering wimps with spin - independent ( si ) interactions . spin - dependent and mixed couplings will be examined in a future work . it was initially the dama / nai experiment @xcite , looking for annual modulation @xcite of a wimp signal , that found such a possible low mass region . wimps with si interactions in the mass range 59 gev were found to be compatible with the dama / nai results and all negative results from other searches that existed at the time @xcite . the situation changed after the publication of the first dama / libra results @xcite ( see _ e.g. _ ref . @xcite and reference therein ) . for si interactions , ref . @xcite found that the best fit dama regions were ruled out to the 3@xmath2 c.l . but ref . @xcite also found that for wimp masses of @xmath08 gev , some parameters outside these regions still yielded a moderately reasonable fit to the dama data and were compatible with all 90% c.l . upper limits from negative searches , when ion channeling in the dama experiment as understood at the time was included ( see section [ sec : channeling ] below ) . the strongest bounds at the time came from cdms @xcite and xenon10 @xcite . since then many new data sets have been released and a reexamination of the light wimp region is now necessary . we will focus on the following three factors : a possible dark matter signal for 7 - 12 gev wimp s found by the cogent collaboration @xcite ; the existence of new better upper limits , in particular the bounds set by the xenon10 @xcite and xenon100 @xcite collaborations ; and the recognition that the effect of channeling in nai(tl ) crystal is less important than previously assumed @xcite ( see section [ sec : channeling ] ) . cdms @xcite has released its full data set , with tighter bounds and two unexplained events at low energy that may be compatible with background . cdms constraints will be included but are not a focus of this paper . our focus is on elastic scattering of spin - independent wimps from a standard maxwellian halo ; recent examinations comparing experimental studies in this case include @xcite . just prior to our paper being submitted , a revised version of ref . @xcite appeared that added an examination of xenon100 and @xmath1 in the context of scalar wimps . the halo model parameters and @xmath1 models used in that paper differ somewhat from ours , but the results are qualitatively similar . the interpretation of the xenon10 and xenon100 results requires the ability to reliably reconstruct the nuclear recoil energy from the observed signal . this depends on the scintillation efficiency factor @xmath1 for which there is considerable uncertainty at low energies ( see section [ sec : leff ] ) . a large part of this paper is devoted to examining the @xmath1 dependence of the two xenon constraints . in this paper we focus on comparing the following experimental results : the combined modulation signal @xcite as well as the total rate @xcite from dama / nai and dama / libra ; the combined cdms 5-tower results @xcite ; the recent first results from xenon100 @xcite ; and the older but lower threshold xenon10 reanalysis results @xcite . constraints for cdms , xenon10 , and xenon100 are determined using the maximum gap method @xcite , while the parameters compatible with dama are determined via the goodness - of - fit of their observed modulation signal with the theoretically expected signal . details of these statistical analyses may be found in ref . @xcite . for the two xenon experiments , we assume the energy resolution is primarily limited by a poisson distribution in the small number of photoelectrons ( pe ) expected at low recoil energies . interactions in the liquid xenon comprising the xenon10 and xenon100 detectors give rise to a prompt scintillation signal ( @xmath4 ) followed by a delayed secondary scintillation signal ( @xmath5 ) . the quantities @xmath4 and @xmath5 are discussed in the following section and the various thresholds and data cuts are described in _ e.g. _ ref . the efficiencies for xenon10 and xenon100 are taken from refs . @xcite and @xcite , respectively . however , for xenon10 , the @xmath4 peak finding efficiency factor @xmath6 , which was not included in ref . @xcite ( where it was not particularly relevant ) , must also be taken into account . for this @xmath6 factor , we take the more conservative of the two cases found in ref . @xcite . a second issue must also be taken into account . as explained in detail below , as the recoil energy decreases , so do the average @xmath4 and @xmath5 signals . at low enough energies , a sizable fraction of the events may fail to produce enough @xmath5 signal and/or to fall in the proper @xmath7 range ( the nuclear recoil band cut ) , even if a fluctuation gives a high enough @xmath4 . at higher recoil energies , the relative size of the fluctuations get smaller and the average @xmath5 is too high for any significant fraction of the @xmath5 fluctuations to fall below the @xmath5 threshold . however , somewhere below an @xmath4 average of 1 photoelectron ( pe ) in xenon10 , an event would need not only an upward fluctuation in @xmath4 , but an upward fluctuation in @xmath5 to pass both thresholds and it would require particular ranges of fluctuations to fall within the required range for @xmath7 . thus , to avoid issues with the @xmath5 threshold and nuclear recoil band cuts , we ignore recoil energies that give on average less than 1 pe in the @xmath4 signal . we find that , with poisson fluctuations included , low energy recoils in xenon10 tend to pass the various thresholds and cuts at a higher rate than predicted by the efficiencies we use ; these efficiencies can thus be considered conservative over the range at which they are applied . additional events passing these various cuts can arise from low energy recoils that give an average @xmath4 signal @xmath8 that falls below our imposed cutoff of @xmath9 pe . accounting for these events will strengthen the xenon10 constraints at low wimp masses . our @xmath8 cutoff is simply due to the efficiency being unknown for these low energy recoils at this time . we are examining these low energy recoil efficiencies for a future work , though a comprehensive treatment of these efficiencies in xenon10 and xenon100 has appeared in the meantime @xcite . in addition to constraints for the above experiments , we also show the 7 - 12 gev wimp mass region suggested by cogent as an explanation for excess events seen at low energies in their detector @xcite . we perform our own statistical analyses of all data sets except cogent , for which we simply use their published region . to allow for direct comparison , all the other experimental constraints are determined using the same 600 km / s galactic halo escape velocity as used by cogent in their analysis @xcite . this escape velocity falls within the 90% confidence interval of 498 km / s to 608 km / s that was determined by a recent analysis of high velocity stars @xcite , but is somewhat above the median likelihood of 544 km / s that was found in that analysis . similarly , we use throughout the value 220 km / sec ( used by cogent ) for the rotation velocity of the galactic disk in the vicinity of the sun . we note that new measurements suggest that the rotation velocity might be higher , @xmath10 km / s @xcite , the effect of which is to shift the best fit in all experiments to slightly lower wimp masses @xcite . * comparison with version one of this paper : * we have made two substantial changes , as discussed in the previous paragraphs . first , we have included the effects of @xmath6 , the @xmath4 peak finding efficiency factor . with this factor included , the xenon10 constraints shift upwards in scattering cross - section by at most @xmath0 @xmath111.7 at any given wimp mass , not by several orders of magnitude ( as was speculated in ref . second , in the current version we do not include recoil energies which yield on average less than 1.0 pe in the @xmath4 signal . it is this second issue that is by far the bigger effect . a more complete discussion can be found in the appendix . in addition , in the appendix we respond to various critiques of our work . in this section we discuss two important experimental issues . first , the scintillation factor in xenon @xmath1 is extremely important in interpreting results and yet is not well known . second , the channeling effect in dama / libra , again not well known , may change the location of the regions in wimp parameter space that are compatible with the data . the interpretation of the xenon10 and xenon100 results requires the ability to reliably reconstruct the nuclear recoil energy from the observed signal . this reconstruction depends on the scintillation efficiency factor @xmath1 for which there is considerable uncertainty at low energies . here we discuss this factor and present three models for @xmath1 at low energy . interactions in the liquid xenon comprising the xenon10 and xenon100 detectors give rise to a prompt scintillation signal , @xmath4 , followed by a delayed secondary scintillation signal , @xmath5 . the @xmath4 signal arises from a rapid relaxation of excited xenon states produced as a result of the interaction . the @xmath5 signal arises from ionized electrons also produced in the interaction ; these drift through the liquid xenon under an applied electric field , but once they reach the liquid surface they are extracted into a xenon gas phase where they emit proportional scintillation light . the drift time of the electrons causes this secondary scintillation ( the @xmath5 signal ) to be observed later than the @xmath4 signal , allowing both scintillation signals to be measured separately . the @xmath4 signal can be used to determine the energy of the interaction , while the combination of both signals allows discrimination between nuclear recoil events ( possibly wimp interactions ) and electron recoil events ( necessarily background interactions ) . the ratio of @xmath5 to @xmath4 is much higher in the case of electron recoils than in the case of nuclear recoils . interpretation of the xenon results requires the ability to reliably reconstruct the nuclear recoil energy @xmath12 from the observed @xmath4 signal . calibration of the nuclear recoil energy dependence of @xmath4 often involves gauging the detector s response to electron recoils at higher energies ; parts of the detector s response ( _ e.g. _ the fraction of scintillation photons that yield photoelectrons ( pe ) in the photodetectors ) are more easily determined in this case than with nuclear recoils at lower energies . taking @xmath4 to be normalized to the number of pe , @xmath4 and @xmath12 are related by an equation involving the higher energy electron recoil calibrations : @xmath13 here , @xmath14 is the light yield in pe / kevee for 122 kevee @xmath15-rays . @xmath1(@xmath12 ) is the scintillation efficiency of nuclear recoils relative to 122 kevee @xmath15-rays in zero electric field ; this factor is a function of the nuclear recoil energy . since there is an applied electric field in the experiment , which reduces the scintillation yield by quickly removing charged particles from the original interaction region , two additional factors must be taken into account : @xmath16 and @xmath17 are the suppression in the scintillation yield for electronic and nuclear recoils , respectively , due to the presence of the electric field in the detector volume . the quantities @xmath16 , @xmath17 , and @xmath14 are detector dependent ; @xmath1 is not . recent comments have drawn attention to the role that @xmath1determinations play in setting experimental constraints for xenon - based detectors @xcite . a variety of @xmath1 measurements have been made over the years @xcite , but limited statistics and systematics issues have so far prevented a clear picture from emerging as to the behavior of @xmath1 at low recoil energies . there are two primary issues in debate : ( 1 ) which of the @xmath1 measurements should be used as a basis for analyzing direct detection results ? and ( 2 ) measurements of @xmath1 have only been made at energies above some minimum ; what is the behavior of @xmath1 at low energies , where no measurements have as yet been made ? for the first issue , the xenon100 collaboration has chosen to use a global fit to multiple @xmath1 measurements in their analysis , whereas ref . @xcite suggests that the recent measurements by manzur _ et al . _ @xcite should be used ; in both cases , @xmath1 measurements are based upon fixed energy neutron scatters . we do not contribute to the debate as to which @xmath1 data sets are most appropriate ; however , in the interest of examining the most conservative xenon constraints , we use the manzur _ et al . _ data alone in our analyses . the choice of @xmath1 measurements to use in the xenon analyses has a significant impact on the resulting constraints for low wimp masses . the manzur _ et al . _ data yield the lowest values for @xmath1 among the fixed - energy neutron scatter measurements , implying the highest recoil energy thresholds and therefore the lowest sensitivity for the xenon detectors to low mass wimps ( which generate only low energy recoils ) . the manzur _ et al . _ data is shown in figure [ fig : leff ] . a comment is in order about the lower zeplin - iii @xmath1 measurement @xcite represented as a band in fig . 1 of @xcite . zeplin fits a nonlinear @xmath1 model to their broad spectrum nuclear recoil calibration data to obtain @xmath1 curves that were used in their analysis . these fits suggest a constant @xmath1 at recoil energies above @xmath030 kevnr , with @xmath1 sharply falling at energies below @xmath020 kevnr and approaching zero at @xmath07 - 8 kevnr ; see figure 15 of ref . @xcite and the accompanying text . thus the suggestion has been made by nonmembers of the zeplin team @xcite that @xmath1 should be taken to be zero below @xmath08 kevnr as a conservative model of @xmath1 . this @xmath1 model would yield significantly weaker xenon constraints relative to what we have referred to as conservative models based on the manzur _ _ measurements of @xmath1 @xcite . however , the members of the zeplin experiment themselves do not advocate their fits as being an indicator of @xmath1 behavior at recoil energies below @xmath0 8 kevnr @xcite . in addition , the dependence of these curves on statistical and systematic uncertainties has not been fully determined , where these uncertainties can significantly impact the lowest recoil energy portion of their fits . the zeplin - iii dark matter analysis is , in fact , mainly insensitive to the low recoil energy portion of their @xmath1 curves . further discussion of the zeplin data can be found in the appendix . moreover , as explained in detail in the appendix , below the recoil energies of 7 kevnr , the manzur _ et al . _ measurements are incompatible with @xmath18 at far more than the 3@xmath2 level . we consider the manzur _ _ measurements more reliable than the zeplin - iii estimate of @xmath1 at low energies . the second issue in debate is how @xmath1 behaves at energies below where measurements have been made . most @xmath1 measurements are at recoil energies above 5 kevnr ; manzur _ et al . _ have a measurement at @xmath19 kevnr . the @xmath1 behavior below these energies is unclear from an experimental and theoretical standpoint , at least at the precision necessary for use in a wimp constraint analysis . the xenon collaboration has suggested that @xmath1 measurements are consistent with @xmath1 being effectively constant at low recoil energies , at least at energies where recoils may contribute to their signal @xcite . various @xmath1 measurements are also consistent with an @xmath1 that decreases as one goes to lower recoil energies ; see _ e.g. _ sect . v of ref . @xcite which provides a theoretically motivated empirical model of such a decreasing @xmath1 . furthermore , ref . @xcite states that `` the mechanisms behind the generation of _ any _ significant amount of scintillation are still unknown and may simply be absent at the few kevnr level . '' given this uncertainty we use three different extrapolations of @xmath1 at low energies : constant , decreasing as one goes to lower recoil energies , or just zero . we choose as our fiducial @xmath1 model a piecewise linear interpolation between the central manzur _ _ values at their measured energies , shown in figure [ fig : leff ] . in addition , we will also examine similarly constructed @xmath1 models using the 1@xmath2 uncertainties in the manzur _ _ measurements are added in quadrature with the upper and lower uncertainties averaged ; uncertainties in the corresponding recoil energies for those measurements have been neglected . ] . the choice of linear interpolation vs. a quadratic interpolation or spline fit to the @xmath1 points has a negligle impact on the generated constraints compared to that from the 1@xmath2 variations in the @xmath1 points themselves . below recoil energies of 3.9 kevnr , the lowest manzur _ et al . _ measurement at 3.9 kevnr , where the two errors in the first case are the statistical and systematic uncertainties , respectively , and the second case is the combined uncertainty as described in the previous footnote . ] , we examine three behaviors for @xmath1 , also shown in figure [ fig : leff ] : ( 1 ) a constant @xmath1 , ( 2 ) an @xmath1 that goes linearly to zero at zero recoil energy , and ( 3 ) an @xmath1 that is strictly zero . even if the scintillation goes to zero at some low but finite recoil energy , there is no reason to expect this to occur above @xmath02 - 3 kevnr ; the measurements of @xmath1 provide no indication of an abrupt ( rather than gradual ) falling of @xmath1 at energies just below where the measurements exist . as such , the third case is perhaps unrealistically conservative , but never - the - less provides the most conservative case . in addition , the use of this case will allow us to examine the contribution of low energy recoils in generating constraints . the average @xmath4 signals as a function of the nuclear recoil energy @xmath12 that correspond to these @xmath1 models are shown in figure [ fig : s1enr ] for xenon10 and xenon100 . in the interest of examining the most conservative xenon constraints , we base all three cases on the data from manzur _ et al . _ @xcite . of the existing data sets , the manzur _ et al . _ data yield the lowest values for @xmath1 , implying higher recoil energy thresholds for the xenon experiments , and thereby reducing the sensitivity of xenon to low mass wimps . the channeling effect is of crucial importance when considering the compatibility of dama with other experimental results as this effect has the potential to significantly alter the wimp masses and cross - sections which are compatible with the dama modulation signal . channeling and blocking effects in crystals refer to the orientation dependence of charged ion penetration in crystals . in the `` channeling effect , '' ions incident upon a crystal along symmetry axes and planes suffer a series of small - angle scattering that maintain them in the open``channels '' in between the rows or planes of lattice atoms and thus penetrate much further into the crystal than in other directions . channeled incident ions do not get close to lattice sites , where they would be deflected at large angles , and they lose energy almost exclusively into electrons . the `` blocking effect '' consists in a reduction of the flux of ions originating in lattice sites along symmetry axes and planes , creating what is called a `` blocking dip '' in the flux of ions exiting from a thin enough crystal as a function of the exit angle with respect to a particular symmetry axis or plane . the potential importance of the channeling effect for direct dark matter detection was first pointed out by h. sekiya _ et al . _ @xcite and subsequently for nai(tl ) by drobyshevski @xcite and by the dama collaboration @xcite . when na or i ions recoiling after a collision with a dark matter wimp are channeled , their quenching factor is the ratio of ionization or scintillation produced by a nuclear recoil event in a crystal relative to that produced in an electron recoil event of the same energy . this is analogous to the @xmath1 factor in liquid xenon and is likewise used to reconstruct the nuclear recoil energy from the observed ionization / scintillation of an event . ] is approximately @xmath20 instead of @xmath21 and @xmath22 , since they give their energy to electrons . the dama collaboration @xcite estimated the fraction of channeled recoils and found it to be large for low recoiling energies in the kev range . using this evaluation of the channeling fraction , the regions in cross - section versus mass of acceptable wimp models in agreement with the dama data were found to be considerably shifted towards lower wimp masses and cross - sections . however , the dama calculation of the channeling fraction did not take into account that the recoiling lattice ions start initially from lattice sites ( or very close to them ) and , therefore , blocking effects are important . in fact , as argued originally by lindhard @xcite , in a perfect lattice and in the absence of energy - loss processes , the probability of a particle starting from a lattice site to be channeled would be zero . the argument uses statistical mechanics in which the probability of particle paths related by time - reversal is the same . in a perfectly rigid lattice , the fraction of channeled recoils would , in fact , be zero . however , the atoms in a crystal are actually vibrating about their equilibrium positions in the lattice . it is this displacement from equilibrium that allows for a non - zero channeling probability of recoiling ions . the vibration amplitude increases with the temperature , thus the effect is temperature dependent : in general the channeling fraction increases with temperature . upper bounds to the recoiling channeling fractions in nai(tl ) crystals at 20@xmath23c were obtained in ref . @xcite , using analytic models of channeling developed since the 1960 s , when channeling was discovered ( see for example refs . @xcite and references therein ) . these upper bounds on the channeling fractions were obtained with temperature effects taken into account not only through the vibrations of the colliding nucleus but also in the lattice . the latter depend on the parameter @xmath24 ( see ref . @xcite for details ) which in the relevant literature is found to be a number between 1 and 2 , with 1 giving the largest channeling fractions ( see figure [ fig : frac - roomt ] , reproduced from ref . @xcite ) . the fractions shown in figure [ fig : frac - roomt ] are also an upper bound in that no dechanneling mechanism has been taken into account to compute them . the collisions with tl impurities would take channeled ions out of their channel , and this process is not included ( see ref . @xcite for further explanations ) . in figure [ fig : dama ] , we show the wimp masses and si cross - sections compatible with the dama modulation signal both with and without channeling included ; contours are shown for regions compatible at the 7@xmath2 , 5@xmath2 , 3@xmath2 , and 90% level ( in order from larger to smaller regions ) . for the channeling , we use the largest channeling fractions shown in figure [ fig : frac - roomt ] as they provide the largest potential effect on the dama constraints . figure [ fig : dama ] shows that even in this case there is negligible difference between the channeling and non - channeling scenarios except for regions incompatible with dama at greater than the 5@xmath2 level . even in these cases , the difference lies only at wimp masses below 4 gev and at relatively high si cross - sections . as channeling is a negligible effect , we do not further include it . compared to our previous analysis in @xcite , the current study takes advantage of additional recently released dama data . the effect of the additional data has been to sharpen the regions in parameter space that match the data . for example , at @xmath25 , there are now two completely separate regions ( peaked at different wimp masses ) that were previously joined . in our current work we also display a @xmath26 contour in which the two regions are again connected . we remind the reader that we are using the goodness - of - fit statistic described in detail in ref . @xcite . our main results are shown in figs . ( [ fig : constantleff])-([fig : zeroleff ] ) , corresponding to the three cases for the behavior of @xmath1 at low recoil energies . the solid gray contours indicate the wimp parameters compatible with the dama modulation within the 5@xmath2 , 3@xmath2 , and 90% level ; the 5@xmath2 dama region is also shaded light gray . the ( filled ) pink contour corresponds to the 7 - 12 gev wimp mass region suggested by cogent ( we reiterate that we have not reanalyzed their data and simply display their published region here ) . cdms , dama ( total events ) , xenon10 , and xenon100 curves indicate regions for which the wimp parameters are excluded at the 90% level ( the parameters above these curves are excluded ) . the solid green region for xenon10 and solid purple region for xenon100 do _ not _ indicate regions compatible within a given level ( as opposed to the dama and cogent regions ) ; they instead indicate how the 90% exclusion constraints vary with the 1@xmath2 level uncertainties in the @xmath1 measurements . overlapping xenon10 and xenon100 1@xmath2 regions are shown in blue . for the fiducial ( central value ) @xmath1 model in the case where it is constant below 3.9 kevnr , shown in figure [ fig : constantleff ] , the xenon100 constraint excludes all of the dama 3@xmath2 region , but only the portion of the cogent region with wimp masses above 9 gev . we note that , because we use only the manzur _ et al . _ @xmath1 data @xcite , this constraint is weaker than that presented by xenon100 @xcite . if the 1@xmath2 uncertainties in @xmath1 are included , xenon100 could exclude nearly all of the dama 5@xmath2 region and the entire cogent region , if the largest value of @xmath1 in the 1@xmath2 region is taken . on the other hand , it might exclude only the cogent region above 11 gev and not even all of the dama 90% region , if the lowest value of @xmath1 in the 1@xmath2 region is used . however , the cdms constraint , unaffected by the issues with @xmath1 , constrains the same cogent region as the fiducial xenon100 case here , with a slighter weaker constraint on the dama region ( incompatible with the dama 2@xmath2 region , not shown ) . for the fiducial ( central value ) @xmath1 model in the case where it falls linearly to zero at zero recoil energy , shown in figure [ fig : fallingleff ] , the xenon100 constraint again excludes nearly all of the dama 3@xmath2 region and the portion of the cogent region with wimp masses above 9 gev . the 1@xmath2 variations in the @xmath1 measurements also yield a similar variation in the xenon100 constraint as they did in the constant @xmath1 case . the most extreme case , taking @xmath1 to be zero below 3.9 kevnr , yields similar xenon100 constraints as the other two cases , as seen in figure [ fig : zeroleff ] , although the constraint using the 1@xmath2 upper values of @xmath1 does not quite exclude the full cogent region , leaving a narrow window at wimp masses of 7 - 8 gev . it should be emphasized , however , that the linearly falling @xmath1 case is already conservative and taking @xmath1 to be zero below 3.9 kevnr is perhaps unrealistically conservative . the xenon100 constraints are nearly identical in the dama and cogent regions for all three cases of low energy @xmath1 behavior . in fact , the constraints based on the central and 1@xmath2 lower values of @xmath1 _ are _ identical ; only when using the upper 1@xmath2 @xmath1 values do the constraints differ . there are two main reasons for the similarity among the constraints : ( 1 ) the imposed @xmath9 pe cutoff and ( 2 ) the small potential contribution from recoil events with energies below 3.9 kevnr where the @xmath1 models differ . as can be seen in figure [ fig : s1enr ] , a recoil energy of 3.9 kevnr yields an average @xmath4 signal of 1.0 pe in xenon100 when using any of the three fiducial @xmath1 models . with the @xmath8 cutoff , there is no contribution from recoils at energies below 3.9 kevnr where the fiducial @xmath1 models differ ; thus , these constraints are identical . when using the 1@xmath2 lower values of @xmath1 , no recoils below 5.9 kevnr are included , so the lesser constraining portion of the 1@xmath2 xenon100 constraint bands shown in the figures are likewise identical . on the other hand , when using the 1@xmath2 upper values of @xmath1 , the @xmath9 pe cutoff corresponds to recoil energies of 2.5 , 3.1 , and 3.9 kevnr for the constant , linearly falling , and zero low energy @xmath1 models , respectively . in this case , low energy recoils contribute to the constraints . however , these low energy recoils can make only a small contribution to the observed signal , as will be discussed below . as the potential effect of the low energy @xmath1 behavior on the xenon100 constraints is masked by the @xmath9 pe cutoff , we show in figure [ fig : extended ] the xenon100 constraints for the three fiducial @xmath1 models when this cutoff is relaxed . in this figure , we have arbitrarily assumed the nuclear recoil band cut efficiency is constant at low recoil energies . in reality , this efficiency should fall at very low recoil energies and this approximation becomes inappropriate at recoil energies that yield @xmath8 somewhere below 1 pe . for this reason , these constraints should not be taken to be valid constraints ; we show them only to illustrate the potential effect of low energy recoils and the low energy @xmath1 behavior . with all cut efficiencies properly taken into account , the true constraints would lie somewhere between the constraints shown in this figure and those shown in the previous figures . with the relaxing of the @xmath8 cutoff , the xenon100 constraints are still nearly identical in the dama and cogent regions for all three cases of low energy @xmath1 behavior . these constraints are very similar to the ones found in the previous figures , when a @xmath8 cutoff was included , and are actually identical for the zero @xmath1 model as this model has no contributions from recoils with @xmath27 pe anyways . the three cases only begin to differ significantly in the low mass , high cross - section parameter space located around and above the dama regions in this figure . this can be explained by the xenon100 s1 analysis threshold of 4 pe s ( the full analysis range is 4 - 20 pe s ) . in the absence of a finite energy resolution , this corresponds to a nuclear recoil energy of 9.5 kevnr in our fiducial @xmath1 models , well into the energy range where the @xmath1 behavior is known . with a poisson fluctuation in the number of observed pe s , recoils at lower energies have a finite chance of producing 4 or more pe s and falling into the analysis range , even if the average number of pe s for events at those energies is below 4 . however , at 3.9 kevnr , the average expected number of pe s is 1.0 ; only 1.9% of such events yield 4 or more pe s . recoils of 3 kevnr yield an average number of expected pe s of 0.79 and 0.61 for the constant and falling @xmath1 cases , respectively , with corresponding probabilities of being observed ( 4 + pe s ) of 0.87% and 0.36% ( the third case , zero @xmath1 , produces no pe s at these energies ) . the small fraction of recoil events with energies below 3.9 kevnr that will be observed in the analysis range means that their contribution is only significant when there are essentially no events at higher energies ( due to low wimp masses and a finite escape velocity in the halo ) , but to produce a sufficient number of events to fall into the analysis range requires a very large number of wimp scatters in the @xmath01 - 4 kevnr range , which requires a high wimp cross - section . thus , even when using an overly optimistic nuclear recoil band efficiency , the three @xmath1 cases can only result in different constraints in the low mass , high cross - section region . this is not necessarily the case for @xmath1 curves based on measurements that yield values higher than manzur _ et al . _ , as this would push the analysis range corresponding to 4 - 20 pe s to lower recoil energies ; such cases , however , inevitably move the xenon100 constraints to the left . in any case , the most significant issue in the xenon100 analysis is the choice of @xmath1 measurements used to determine the @xmath1 dependence , not the @xmath1 behavior at low energies . we now turn to the xenon10 bounds . we have reanalyzed the xenon10 results in terms of the same @xmath1 models as used for xenon100 and discussed in the previous section ; our results differ from those shown by the xenon collaboration due to the difference in @xmath1 used in their analyses and ours . the xenon10 results are important because of the lower @xmath4 threshold of about 2 pe s used in that analysis , which corresponds to 4.6 kevnr nuclear recoil energies in our fiducial @xmath1 models ( neglecting poisson fluctuations ) , much lower than the 9.5 kevnr of the xenon100 4 pe threshold . because of the lower threshold , the behavior of @xmath1 at low recoil energies is relevant in producing the xenon10 constraints as poisson fluctuations allow for a non - trivial probability of seeing 2 + pe s for recoil energies below 3.9 kevnr . for the constant @xmath1 case shown in figure [ fig : constantleff ] , the lower threshold allows for a stronger sensitivity to lower wimp masses for xenon10 relative to xenon100 . the fiducial case excludes at the 90% c.l . all of the cogent region and dama to the 5@xmath2 contour . when the 1@xmath2 uncertainties in the @xmath1 measurements are taken into account , the constraints relax : dama is excluded to only the 3@xmath2 contour and the some of cogent region at wimp masses below 9 gev survive . the xenon10 constraints mildly weaken if @xmath1 is taken to fall linearly to zero below 3.9 kevnr , as seen in figure [ fig : fallingleff ] . the fiducial case still excludes all of the cogent region and dama to about the 4@xmath2 contour ( not shown ) . the @xmath1 1@xmath2 band here allows the same dama and cogent regions to survive as with the constant @xmath1 case . for the case where @xmath1 is zero below 3.9 kevnr , shown in figure [ fig : zeroleff ] , the xenon10 constraints further weaken and approach the xenon100 constraints as the low energy events are essentially turned off and the lower xenon10 threshold becomes less relevant . again , we note that this last case ( zero @xmath1 at low recoil energies ) is an extremely conservative case . as with xenon100 , the potential effect of the low energy @xmath1 behavior on the xenon10 constraints is limited by the imposed @xmath9 pe cutoff . we also show in figure [ fig : extended ] the xenon10 constraints for the three fiducial @xmath1 models when this cutoff is relaxed . the same caveats apply : the nuclear recoil band cut efficiency that is used is not appropriate for the full recoil energy range that it is applied over , so these do not represent valid constraints . again , these constraints are only used to illustrate the potential impact of the low energy @xmath1 behavior on xenon10 constraints . the actual constraints when all efficiencies are accounted for properly would fall somewhere between the constraints shown in figure [ fig : extended ] and those shown in figs . ( [ fig : constantleff])-([fig : zeroleff ] ) . with the @xmath8 cutoff relaxed , figure [ fig : extended ] shows how the low threshold allows for a strong xenon10 sensitivity to lower wimp masses . this is particularly evident with the constant @xmath1 case where , as can be seen in figure [ fig : s1enr ] , recoils of energy 1 kevnr yield an average @xmath4 signal of 0.4 pe ; @xmath06% of such recoils will produce the necessary 2 + pe . for the falling @xmath1 case , that same average @xmath4 signal of 0.4 pe occurs at a higher recoil energy of 2 kevnr , but this energy is still sufficiently low to provide sensitivity to low mass wimps . the presence of these non - trivial poisson fluctuations at low recoil energies leads to a very strong dependence of the xenon10 constraints on the low energy @xmath1 behavior . this should remain the case even when the various efficiencies are handled properly , though not quite to the degree shown in figure [ fig : extended ] . in particular , when the proper efficiencies are included , the xenon10 constraints in the constant and falling @xmath1 cases should gain an upward curve at low wimp masses , as seen with the other constraints , rather than the current linear appearance . these linear portions of the constraints at low wimp masses ( as they appear with the logarithmic scaling of the figure ) continue to arbitrarily low wimp masses ; however , they arise from the poisson tails of increasingly smaller energy events that would be suppressed when using the proper efficiencies . though we have not included it in this work , cdms silicon data may provide further constraints on the dama and cogent regions and should be considered in a full discussion of compatibility between the various experimental results . the cdms silicon results will be considered in future work . in summary , we have examined a number of subtleties relevant to direct detection studies of low mass wimps . in the interest of examining the most conservative xenon constraints , we have used the manzur _ et al . _ @xcite data alone in our analyses of the existing data sets , the manzur _ et al . _ data yield the lowest values for @xmath1 , implying higher recoil energy thresholds for the xenon experiments , and thereby reducing the sensitivity of xenon to low mass wimps . we find that , when basing the @xmath1 curves on these manzur _ et al . _ measurements , the behavior of @xmath1 at low energies ( less than 3.9 kevnr ) has negligible effect on the xenon100 constraints in the regions of interest for dama and/or cogent . for xenon100 , the choice of data sets upon which the @xmath1 dependence is based is more important than the extrapolated behavior of @xmath1 at low recoil energies . the strongest bounds are from xenon10 , rather than xenon100 , due to the lower energy threshold . for reasonable choices of @xmath1 and for the case of spin independent elastic scattering , we find that xenon10 is incompatible with the dama / libra 3@xmath2 region and severely constrains the cogent 7 - 12 gev wimp mass region . is grateful for financial support from the swedish research council ( vr ) through the oskar klein centre . g.g . was supported in part by the us department of energy grant de - fg03 - 91er40662 , task c. p.g . was supported in part by the nfs grant phy-0456825 at the university of utah . k.f . acknowledges the support of the doe and the michigan center for theoretical physics via the university of michgian . , p.g . , and c.s . thank the galileo galilei institute for theoretical physics for the hospitality and the infn for partial support during the completion of this work . we thank also e. aprile , k. arisaka , d. hooper , and k. zurek for helpful conversations ; p. sorensen for bringing to our attention the xenon10 @xmath4 peak finding efficiency factor that was missing in the first version of this paper ; l. baudis , a. manalaysay , g. plante , and p. sorensen for discussions regarding the xenon detectors and analysis ; and b. edwards and t. sumner for discussions regarding the zeplin - iii @xmath1 estimates . shortly after the first version of our paper was released , ref . @xcite appeared commenting on it . we include here a detailed explanation of the xenon10 efficiencies and cuts we are using , as well of our use of the manzur _ _ measurements as a conservative choice for @xmath1 ( instead of the zeplin - iii measurements ) which we believe are relevant in view of the comments expressed in ref . @xcite . the @xmath4 peak finding efficiency factor @xmath6 had not been included in the xenon10 analysis in the first version of our paper ( as correctly pointed out in ref . @xcite ) and has been accounted for in this revised version . however , this leads to only a moderate weakening of the xenon constraints by shifting the bound on the cross section upward by less than a factor of two . it is important to point out that including this effect does not weaken the constraints upwards in cross - section by 2 - 3 orders of magnitude , contrary to the claim in the critique given in ref . one can understand the small magnitude of the effect with the following reasoning . a valid signal event in the xenon detectors is required to produce coincident scintillation in at least two pmts . the @xmath6 factor accounts for experimental limitations in identifying and reconstructing at least two pmt contributions to the overall @xmath4 signal of a recoil event . the @xmath4 signal is determined from the area under the peaks produced in the electronic readout of the pmts . due to digitization of the signal and intrinsic pmt performance , the size and shape of the peaks will vary signal from a single pe as determined from the area of these peaks is @xmath28 pe . thus , @xmath4 can take on non - integer values even though it is given in terms of number of pe . ] ; see fig . 14 of ref . @xcite for an example . small or poorly shaped peaks may fail to be properly tagged as a pe peak in a pmt . using the more conservative estimate found in ref . @xcite , only @xmath29 of 2 pe events will have both pe peaks properly tagged and identified as coincident . as the number of pe s increases , the probability of passing the two - fold pmt requirement rapidly rises to 100% . the significance of the @xmath6 factor on xenon10 constraints can be easily estimated . if this factor is conservatively assumed to be 60% over the entire @xmath4 analysis range ( 2 + pe ) instead of just at low @xmath4 , the expected number of events passing the xenon10 cuts at all energies should fall to 60% of the number of events expected without @xmath6 applied . as the recoil rate at a particular wimp mass is proportional to the scattering cross - section , the original recoil spectrum can be exactly reproduced by shifting the cross - section upwards by a factor of @xmath30 . thus , if a wimp mass and cross - section is excluded by xenon10 at some given cl without the @xmath6 factor applied , a wimp with the same mass and @xmath111.7 higher cross - section would yield the same excluded spectrum with the @xmath6 factor included . in this case , the xenon10 exclusion curves shift upwards in cross - section by @xmath111.7 . as @xmath6 is higher than 60% for @xmath4 larger than 2 pe , the shift in cross - section will be even smaller at heavier wimps where a significant number of high @xmath4 events are expected . we demonstrate in figure [ fig : etas1 ] the weakening of the xenon10 constraints when @xmath6 is included . we show only the three fiducial ( central ) @xmath1 models described in our paper ; constraints without and with @xmath6 included are given by solid and dashed curves , respectively . for a given wimp mass , the constraint increases by @xmath0 @xmath111.7 in the scattering cross - section at low wimp masses and by less at higher wimp masses . this is a relatively modest weakening of the constraint given the logarithmic scaling of the exclusion curve figures . please recall that we have used the more conservative estimate of @xmath6 found in ref . @xcite . had we used the @xmath6 factor given in ref . @xcite , the change would have been even milder . we will clarify an issue that is perhaps the source of what appears to be an erroneous application of the @xmath6 factor in ref . @xcite that yielded a greater weakening in the xenon10 results than expected . given a recoil at some energy @xmath12 , eqn . ( [ eqn : s1 ] ) gives the _ average _ expected @xmath4 scintillation signal @xmath8 . given the measured @xmath4 in an event , eqn . ( [ eqn : s1 ] ) can be inverted to obtain a reconstructed recoil energy @xmath31 . due to the discreteness of the produced pe s and the variation in the @xmath4 peaks , the observed prompt scintillation signal @xmath4 is a random value whose expectation value is @xmath8 . thus , the reconstructed recoil energy @xmath31 is also a random value and provides only an estimate of the ( unknown ) true recoil energy , @xmath12 . the distinction between @xmath31 and @xmath12 is important . with the pe fluctuations and @xmath4 peak variations , the true recoil energy can not be precisely determined for any given recoil event . @xmath31 is simply an estimate of the likely recoil energy that produced an event and is understood to have some intrinsic uncertainty attached to it . at high recoil energies , @xmath32 and @xmath31 can be taken as a good approximation of the actual recoil energy . at low recoil energies , @xmath31 may differ significantly from @xmath12 on an event by event basis . the @xmath6 factor is sometimes given as a function of recoil energy . however , as @xmath6 is really a function of the @xmath4 signal , this recoil energy refers to the reconstructed energy @xmath31 of an event ( recall @xmath31 and @xmath4 have a 1:1 mapping through eqn . ( [ eqn : s1 ] ) ) and does _ not _ refer to the actual recoil energy @xmath12 . neglecting the @xmath4 peak variations and efficiencies other than @xmath6 , the fraction of events @xmath33 that exceed the 2 pe threshold from scatters at a recoil energy that yields an average @xmath4 signal @xmath8 is given by : @xmath34 where @xmath6 is taken as a function of @xmath4 and @xmath35 is the poisson probability of seeing @xmath36 pe with an average of @xmath37 . if the @xmath6 factor were assumed to be a function of the actual recoil energy @xmath12 , then @xmath38\ ] ] would erroneously be taken as the fraction of events exceeding the @xmath4 threshold . use of eqn . ( [ eqn : fbad ] ) in place of eqn . ( [ eqn : f ] ) will improperly yield greatly weakened xenon10 constraints at low wimp masses . a second change has been made in the present version of this paper . in the first version , we assumed the same nuclear recoil band efficiency given for @xmath39 in ref . @xcite applied to all poisson fluctuated events that appeared in that range . this was an optimistic assumption that may overestimate the xenon10 sensitivity at low wimp masses . there are two data cuts relevant for low energy recoils that yield @xmath39 : a recoil event must have @xmath40 pe ( @xmath5 threshold ) and @xmath41 ( nuclear recoil band ) to be accepted as a valid event . the latter requirement is a cut designed to exclude electron recoil background events which tend to produce higher values of @xmath42 than nuclear recoils ; the range of values here accept the lower @xmath045 - 50% of the distribution of @xmath42 expected for nuclear recoils , as determined from calibration data . at very low recoil energies , an event that produces an upward fluctuated @xmath43 pe and would otherwise be considered a valid event might fail to produce enough @xmath5 signal to pass the @xmath5 threshold . the efficiency can thus be expected to fall at low recoil energies . to avoid problems with these two data cuts , we have added a cutoff to our analysis and ignore recoil energies for which @xmath27 pe . we also take the same 47% nuclear recoil band efficiency in the 2 - 5 pe @xmath4 bin for all events in this @xmath4 range stemming from low energy recoils . the addition of this cutoff significantly weakens the xenon10 constraints for wimp masses below @xmath06 gev for the constant and falling @xmath1 models discussed in the paper , as seen by the dotted curves in figure [ fig : etas1 ] . this cutoff has a far greater impact on the constraints than the inclusion of the @xmath6 factor and is the dominant source of the change in xenon10 constraints from the first version of this paper . the third @xmath1 model , zero below recoil energies of 3.9 kevnr , is unaffected by this cutoff as it does not yield any observable events at low energy anyways . the @xmath9 pe cutoff and 47% assumed efficiency we have adopted here are conservative for two reasons : ( 1 ) the @xmath4 and @xmath5 fluctuations are independent . this means that the poisson fluctuated events that yield higher @xmath4 than the average will have lower @xmath42 ratios on average and are more likely to pass the nuclear recoil band cut . the upward poisson fluctuated events for @xmath44 can survive this cut as much as 70 - 80% of the time , higher than the assumed 47% . ( 2 ) our choice of a cutoff at @xmath45 pe is due to our limited ability to examine the efficiencies at lower recoil energies , not due to an expected lack of events at these low energies . we note that there will also be fluctuations in the number of ionization electrons that lead to the @xmath5 signal threshold of @xmath0300 pe corresponds to about 12 ionization electrons , each of which produce @xmath025 pe . ] ; even at very low recoil energies , there may be a non - zero probability of both @xmath4 and @xmath5 exceeding their respective thresholds and producing events in the analysis region that pass all cuts . for the constant @xmath1 case at a wimp mass of 5 gev and scattering cross - section of 10@xmath46 pb , a point not excluded by the conservative bounds in figure [ fig : etas1 ] , about 20,000 recoils would be expected for recoil energies corresponding to @xmath47 . even with a very small efficiency for such recoils , some number of these events would be expected to have fluctuations that put them in the xenon10 analysis region . if the nuclear recoil band efficiency were only @xmath01% over this range , 10 + events would be expected to pass the various cuts ; the lack of such events in the data would rule out these wimp parameters . @xcite and @xcite suggest using the the zeplin - iii @xmath1 measurement @xcite , represented as a band in fig . 1 of ref . @xcite , as a conservative choice of @xmath1 . thus , we discuss here this possibility . zeplin fits a nonlinear @xmath1 model to their broad spectrum nuclear recoil calibration data to obtain @xmath1 curves that were used in their analysis . these fits suggest a constant @xmath1 at recoil energies above @xmath030 kevnr , with @xmath1 sharply falling at energies below @xmath020 kevnr and approaching zero at @xmath07 - 8 kevnr ; see figure 15 of ref . @xcite and the accompanying text . the suggestion has been made @xcite that @xmath1 should be taken to be zero below @xmath08 kevnr as a conservative model of @xmath1 . this @xmath1 model would yield significantly weaker xenon constraints relative to what we have referred to as conservative models based on the manzur _ _ measurements of @xmath1 @xcite . we have several reservations about using this zeplin inspired @xmath1 model . first , zeplin does not provide estimates of @xmath1 below @xmath07 kevnr and , in fact , states they are limited in constraining @xmath1 at these low energies ( in the caption of their fig 15 , it is said that `` the constraints become very weak outside the energy ranges shown . '' ) . second , neither the technical details of zeplin s curve - fitting nor estimates of the statistical and systematic uncertainties are provided in their paper . without these , it is unclear with which degree of certainty the low energy ( @xmath07 - 10 kevnr ) end of the zeplin - iii curves should be treated . several sources of error may contribute to the level of uncertainty in these @xmath1 estimates . one is the errors associated with the monte carlo used to compare with data , which are difficult to quantify since they can arise from inaccuracies in the inelastic neutron scattering data used in the gant4 code and obtained from an international library of such data . another issue are systematic errors associated with uncertainties in the position of the neutron source near the detector during calibration runs . if the uncertainties are large , then zeplin does not have the statistical power to determine the @xmath1 behavior at low recoil energies and one should base the @xmath1 curves on measurements that have a better statistical power to analyze this @xmath1 behavior , such as the manzur _ et al . if the uncertainties are small so that the zeplin curve is expected to be an accurate representation of the @xmath1 behavior at low energies , then there is a very serious discrepancy between the zeplin and manzur _ _ results ( see below ) . the zeplin collaboration claims no leverage on @xmath1 below about 8 kevnr and do not suggest that it goes to zero in that energy range or 90% c.l . ) in the uncertainties . ] the recoil energy interval in the zeplin analysis of 10.730.2 kevnr does not include the low energy , low @xmath1 end of their @xmath1 curves . in this analysis region , the zeplin @xmath1 curves are compatible with the manzur _ et al . _ measurements within the 1 - 2@xmath2 level . furthermore , the zeplin inspired model proposed by ref . @xcite , taking @xmath18 below recoil energies of @xmath08 kevnr , is strongly incompatible with the manzur _ _ measurements as well as other fixed energy neutron scattering measurements such as those of aprile _ _ @xcite . we reiterate that our choice to use the manzur _ et al . _ data in our analysis was due to the lower @xmath1 values that give more conservative xenon constraints and should not be construed as an indication that we regard this data as the most accurate available . we make no contribution to the discussion regarding the accuracy of the various low energy recoil measurements from neutron beam scattering . see ref . @xcite for a discussion of the various potential issues which may affect the @xmath1 determinations in these experiments . in any case , much of the discussion below regarding the incompatibility of the manzur _ et al . _ data with @xmath18 applies also to the aprile _ et al . _ measurements . given the manzur _ et al . _ measurement of _ @xmath48 ( statistical and systematic errors , respectively ) at a recoil energy of @xmath19 kevnr , one might naively conclude this measurement is consistent with @xmath18 at the @xmath02 - 3@xmath2 level . however , the @xmath1 value and errors are determined from a @xmath49 fit to the data as a function of @xmath1 . the @xmath49 versus @xmath1 curves , such as shown in fig . 11(c ) of ref . @xcite for a recoil energy of 6 kevnr , are not symmetric about the minima ( which provide the central values ) , but rise very rapidly as @xmath1 becomes small , effectively diverging at @xmath18 . the result is that the manzur _ _ measurements ( four of which are below recoil energies of 7 kevnr ) are incompatible with @xmath1 @xmath50 recoil energies of @xmath08 kevnr , as this would yield @xmath49 values that are imcompatible at far higher than the 3@xmath2 level . we consider the manzur _ _ measurements more reliable than the zeplin - iii estimate of @xmath1 at low energies . to illustrate the incompatibility of the manzur _ et al . _ data with @xmath18 , one can understand the @xmath49 behavior as follows . et al . _ measured @xmath1 at several recoil energies by observing the scintillation response in a xenon detector for neutrons from a beam of fixed energy ( 2.8 mev ) that scatter in a fixed direction ; the recoil energy is fixed by the angle of scatter measurement arise mainly from this range . ] . if @xmath1 is non - zero at the recoil energy corresponding to a particular scattering angle , a histogram of the @xmath4 from observed events will generate a peak due to scintillation from the single scatter events in the detector . the primary background is double scattering neutrons which can reach the neutron scintillator located at the fixed angle from the neutron beam , but scatter with a different energy than expected for single scatter neutrons that reach the neutron scintillator ; one of the two scatters must occur outside of the active volume for the event to be mistaken as a single scatter . these double scattering neutron events are expected to occur at an almost negligible rate compared to the single scatter signal events . in addition , these double scatter events produce a fairly flat @xmath4 distribution , not a peak ( see _ e.g. _ figure 9 of ref . figures 11(a ) and 11(c ) of ref . @xcite show an @xmath4 histogram and corresponding @xmath49 versus @xmath1 curve , respectively , for recoils centered at 6 kevnr . the histogram shows a peak of @xmath0700 events ( consistent with scintillation from single scatter signal events ) with a negligible flat contribution ( expected from any significant double scatter background ) . the @xmath49 in this case is minimized at @xmath51 . at lower @xmath1 , the @xmath49 grows rapidly for this reason : it is difficult to account for this @xmath0700 scintillation event peak if the nuclear recoils produce little to no scintillation . note the systematic issues discussed in manzur _ _ generally affect the interpretation of which @xmath1 would produce such a peak , but do not provide significant alternate mechanisms for producing such a peak aside from the scintillation from low energy recoils ( scintillation that can only be produced if @xmath52 ) . thus , while determining the value of @xmath1 based on the size and shape of such @xmath4 peaks may be limited by statistics or biased by systematic effects , the simple presence of such a peak is strong evidence that @xmath1 is non - zero . given that manzur _ _ have observed scintillation peaks at multiple recoil energies over @xmath0 4 - 8 kevnr , we consider the `` conservative '' @xmath1 model suggested in refs . @xcite and @xcite , with @xmath18 below @xmath08 kevnr , to be grossly incompatible with existing data and therefore unrealistically conservative . g. gelmini and p. gondolo , arxiv : hep - ph/0405278 . p. gondolo and g. gelmini , phys . d * 71 * , 123520 ( 2005 ) [ arxiv : hep - ph/0504010 ] . r. bernabei _ et al . _ [ dama collaboration ] , eur . j. c * 56 * , 333 ( 2008 ) [ arxiv:0804.2741 [ astro - ph ] ] . c. savage , g. gelmini , p. gondolo and k. freese , jcap * 0904 * , 010 ( 2009 ) [ arxiv:0808.3607 [ astro - ph ] ] . z. ahmed _ et al . _ [ cdms collaboration ] , phys . rev . lett . * 102 * , 011301 ( 2009 ) [ arxiv:0802.3530 [ astro - ph ] ] . j. angle _ et al . _ [ xenon collaboration ] , phys . lett . * 100 * , 021303 ( 2008 ) [ arxiv:0706.0039 [ astro - ph ] ] . c. e. aalseth _ et al . _ [ cogent collaboration ] , arxiv:1002.4703 [ astro-ph.co ] . j. angle _ et al . _ [ xenon10 collaboration ] , phys . d * 80 * , 115005 ( 2009 ) [ arxiv:0910.3698 [ astro-ph.co ] ] . e. aprile _ et al . _ [ xenon100 collaboration ] , arxiv:1005.0380 [ astro-ph.co ] . n. bozorgnia , g. b. gelmini and p. gondolo , arxiv:1006.3110 [ astro-ph.co ] . z. ahmed _ et al . _ [ the cdms - ii collaboration ] , arxiv:0912.3592 [ astro-ph.co ] ; z. ahmed _ et al . _ [ the cdms - ii collaboration ] , science * 327 * , 1619 ( 2010 ) . a. bottino , f. donato , n. fornengo and s. scopel , phys . d * 81 * , 107302 ( 2010 ) [ arxiv:0912.4025 [ hep - ph ] ] . j. kopp , t. schwetz and j. zupan , jcap * 1002 * , 014 ( 2010 ) [ arxiv:0912.4264 [ hep - ph ] ] . a. l. fitzpatrick , d. hooper and k. m. zurek , arxiv:1003.0014 [ hep - ph ] . s. chang , j. liu , a. pierce , n. weiner and i. yavin , arxiv:1004.0697 [ hep - ph ] . f. petriello and k. m. zurek , jhep * 0809 * , 047 ( 2008 ) [ arxiv:0806.3989 [ hep - ph ] ] . a. bottino , f. donato , n. fornengo and s. scopel , phys . d * 78 * , 083520 ( 2008 ) [ arxiv:0806.4099 [ hep - ph ] ] . s. chang , a. pierce and n. weiner , phys . d * 79 * , 115011 ( 2009 ) [ arxiv:0808.0196 [ hep - ph ] ] . d. hooper , f. petriello , k. m. zurek and m. kamionkowski , phys . d * 79 * , 015010 ( 2009 ) [ arxiv:0808.2464 [ hep - ph ] ] . s. yellin , phys . d * 66 * , 032005 ( 2002 ) [ arxiv : physics/0203002 ] . p. sorensen , presented at the 2010 light dark matter workshop , uc davis , http://particle.physics.ucdavis.edu/seminars/data/media/2010/apr/sorensen.pdf ; see the blue curve on page 6 . m. c. smith _ et al . _ , mon . not . * 379 * , 755 ( 2007 ) [ arxiv : astro - ph/0611671 ] . m. j. reid _ et al . _ , astrophys . j. * 700 * , 137 ( 2009 ) [ arxiv:0902.3913 [ astro-ph.ga ] ] . c. savage , k. freese , p. gondolo and d. spolyar , jcap * 0909 * , 036 ( 2009 ) [ arxiv:0901.2713 [ astro - ph ] ] . f. arneodo _ et al . _ , nucl . instrum . a * 449 * , 147 ( 2000 ) . r. bernabei _ et al . _ , eur . j. direct c * 3 * , 11 ( 2001 ) . d. akimov _ et al . _ , phys . b * 524 * , 245 ( 2002 ) [ arxiv : hep - ex/0106042 ] . e. aprile _ et al . _ , phys . d * 72 * , 072006 ( 2005 ) [ arxiv : astro - ph/0503621 ] . v. chepel _ et al . _ , astropart . * 26 * , 58 ( 2006 ) . p. sorensen _ et al . _ , nucl . instrum . a * 601 * , 339 ( 2009 ) [ arxiv:0807.0459 [ astro - ph ] ] . e. aprile _ et al . _ , phys . c * 79 * , 045807 ( 2009 ) [ arxiv:0810.0274 [ astro - ph ] ] . a. manzur , a. curioni , l. kastens , d. n. mckinsey , k. ni and t. wongjirad , phys . c * 81 * , 025808 ( 2010 ) [ arxiv:0909.1063 [ physics.ins-det ] ] . v. n. lebedenko _ et al . _ , phys . d * 80 * , 052010 ( 2009 ) [ arxiv:0812.1150 [ astro - ph ] ] . t. sumner for the zeplin collaboration , private correspondence . h. sekiya , m. minowa , y. shimizu , y. inoue and w. suganuma , phys . b * 571 * , 132 ( 2003 ) [ arxiv : astro - ph/0307384 ] ; h. sekiya , m. minowa , y. shimizu , w. suganuma and y. inoue , arxiv : astro - ph/0405598 ; h. sekiya , m. mionwa , y. shimizu , w. suganuma and y. inoue , arxiv : astro - ph/0411215 . e. m. drobyshevski , mod . a * 23 * ( 2008 ) 3077 [ arxiv:0706.3095 [ physics.ins-det ] ] .

we report the results of a systematic search for signatures of metal lines in quasar spectra of the sloan digital sky survey ( sdss ) data release 3(dr3 ) , focusing on finding intervening absorbers via detection of their doublet . here we present the search algorithm , and criteria for distinguishing candidates from spurious lyman forest lines . in addition , we compare our findings with simulations of the lyman forest in order to estimate the detectability of doublets over various redshift intervals . we have obtained a sample of 1756 doublet candidates with rest - frame equivalent width 0.05 in 855 agn spectra ( out of 3702 objects with redshifts in the accessible range for detection ) . this sample is further subdivided into 3 groups according to the likelihood of being real and the potential for follow - up observation of the candidate . the group with the cleanest and most secure candidates is comprised of 145 candidates . 69 of these reside at a velocity separation 5000 km / s from the qso , and can therefore be classified tentatively as intervening absorbers . most of these absorbers have not been picked up by earlier , automated qso absorption line detection algorithms . this sample increases the number of known absorbers at redshifts beyond z 2.7 substantially .

0707.1700

our understanding of the nature of the intergalactic medium ( igm ) and its evolution as traced by the absorption features observed in the sightlines towards luminous objects like quasars , has benefitted tremendously within the past few years both from observational and theoretical advances . high - resolution studies of the lyman @xmath0 forest have been extended to very low column densities due to high - resolution echelle spectrographs on powerful 8 m class telescopes like hires ( keck ) and uves ( vlt ) @xcite . at the same time , theoretical models incorporating gas dynamics , radiative cooling , and photoionisation , have been developed that can reproduce the majority of the observed properties of the quasar absorption spectra @xcite . in this picture , baryonic gas can be encountered in a wide variety of physical conditions . within the lyman @xmath0 forest , at least at high redshift where the forest is the main repository for baryons , most of the gas resides in relatively cool ( t@xmath3 k ) low - to - medium overdensity structures that are not in dynamical or thermal equilibrium and are mostly governed by photoionisation . + the igm is expected to be highly ionised with a neutral fraction of hydrogen f ( ) @xmath4 to @xmath5 . in such a medium , metals are also highly ionised . thus , oxygen , as the most abundant intergalactic metal , may exist in the state . since the 1032/1038 doublet is observable from ground - based telescopes for redshifts beyond z@xmath6 2.0 , these transitions constitute a primary tool for studying the characteristics of the metals in the igm at high redshift . the density of the igm is sufficiently low as to allow the production of by photoionisation from the intergalactic uv and soft x - ray radiation field @xcite . indeed , absorbers seen towards higher redshifts can usually be modelled by structures that are photoionised @xcite . + due to the different wavebands required for detections at different redshifts and the strong evolution of the lyman forest , a variety of different techniques and observational strategies have to be applied for finding and identifying absorbers . detection of the 1032/1038 doublet from ground based telescopes is only possible beyond redshifts of @xmath72.0 due to the 3000 atmospheric cutoff , and thus studies of the absorber statistics below this threshold need to rely on space - based instruments . @xcite conducted the first systematic survey of absorbers at z @xmath8 1.0 , employing fos on hst to study a sample of 11 qsos , and found 12 suitable candidates of which 9 were expected to be real , thus establishing that the number density per redshift interval of these absorbers is similar to , if not greater than , that for and absorbers at the same redshift . + towards higher redshifts , despite the advantange of shifting the lines into the range for ground - based instruments , the ever increasing density of the lyman forest renders unambigious identifications of absorbers in its midst more difficult than detections made longward of the lyman @xmath0 forest . thus , most studies have relied upon high - resolution and high - signal - to - noise ratio spectra . around redshifts of z @xmath8 2.0 - 3.0 , a variety of surveys find evidence for a population of absorbers residing in low to medium overdense regions , mostly photoionised by the ambient extragalactic radiation field @xcite . the metal enrichment of the igm inferred by theses surveys ranges from @xmath9 to @xmath10 of the solar value for structures around the mean density of the universe at @xmath11 2.0 . using uves spectra for 2 lensed quasar systems , @xcite find size constraints for such intervening absorbers with lower boundaries on the kpc scale , where no or only very little variation in different lines of sights are seen this is broadly consistent with other studies of binary qso sightline inferring correlation lengths of the absorbers up to several tens of kpc @xcite . + an interesting method to statistically infer the abundance from measuring the mean optical depth associated with all pixels of the lyman @xmath0 forest with similar optical depths , was pioneered by @xcite . the authors find a general correlation of @xmath12 with @xmath13 when @xmath14 1.0 . this method was applied to by @xcite , @xcite and recently @xcite who assert the existence of in gas where the optical depth is as low as 0.1 . there is , however , some ambiguity in the interpretation of the results of such statistical pixel - to - pixel correlations . for an extensive discussion of the method see @xcite , and @xcite for a critical analysis of the results . + the mechanism for enriching the igm with metals to the level inferred by these surveys remains somewhat unclear , yet there are at least good candidates : early and wide - spread dissemination of metals by population iii star formation on pre - galactic structures at very high redshifts @xcite or winds and superwinds from starbursts within galaxies at later stages @xcite . for the latter mechanism , driving these winds out of the dense environments into the low density igm remains a crucial point of contention , and a variety of methods have been suggested ranging from supernovae @xcite , to ejection by mergers @xcite , or photoevaporation @xcite . + table 1 gives an overview of different surveys for direct detection in a variety of environments and over a wide redshift range . @xcite find 6 different systems and a further 8 potential , but blended , candidates in a single object ( hs0747 + 4259 ) at 1.46 @xmath15 1.81 , deriving a redshift path density of @xmath16 13 . @xcite analyse 230 lyman @xmath0 forest lines with a hydrogen column density n@xmath17 @xmath18 , for 7 qsos with @xmath19 2.6 , and retrieve a total of about 50 accompanying systems with secure identifications . using the capabilities of vlt and uves within the large programme : the cosmic evolution of the igm , @xcite survey 10 bright qso at 2.1 @xmath20 2.8 , and find 136 candidates with 12.7 @xmath21 14.6 in 51 systems . @xcite focus on two z@xmath22 = 2 qsos and hydrogen absorbers with log n@xmath23 14.0 . they identify 7 individual lines in 2 intervening systems in one case , and find 13 individual lines in 10 intervening systems in the other . @xcite study the frequency of absorbers associated with 35 damped and sub - damped lyman @xmath0 systems , and report 12 detections of which 9 are intervening . @xcite present 10 intervening absorbers seen towards two lensed qso pairs at a median redshift of z@xmath24 = 2.3 . our sample , residing at z@xmath25 , therefore greatly increases the redshift range of known absorbers at high redshifts . it will allow us to test expectation based upon photoionisation models like the ones presented in @xcite , thereby possibly constraining the physical conditions of the igm and the metagalactic uv / x - ray background at high - redshifts . + in this paper , we have decided to apply a direct pixel - by - pixel search for signatures of strong 1032/1038 features seen in the spectra of sdss qsos . this is , to our knowledge , the first systematic survey at high redshifts ( 2.7 @xmath205.0 ) at low resolution ( r@xmath81800 ) and low signal - to - noise ratio . we demonstrate that there is a redshift window of opportunity where the density of the lyman forest is not yet too high to obliterate the hope for finding and identifying relatively narrow metal absorbers , and that the pathlength covered by the combined sdss sample is high enough to expect finding a reasonable number of such absorbers given conservative assumptions about the metallicity of the igm , the abundance of due to photoionisation and the average signal - to - noise ratio of the qso data sample . + what type of absorbers do we expect to detect via such a direct search without a priori selecting on other transitions ? if a sightline passes through a galaxy , and if there are analogues to local examples surveyed early by @xcite and @xcite for highly ionised metals in absorption , we anticipate to detect in such cases mixed with in the gas phase inside that galaxy , like in the disk of our galaxy . furthermore , a recent , thorough examination of fuse spectra @xcite revealed the occurence of local absorbers in almost all of the more than 100 sightlines probed by extragalactic background objects , indicating a large covering fraction when a sightline passes the galaxy . while the majority of these absorbers are probably located within the galaxy @xcite , especially the high - velocity ( relative to the local standard of rest ) traces a variety of different environments and phenomena , most likely including warm / hot gas interaction in an extended galactic corona and even truly intergalactic gas within the local group @xcite . as an aside , the gas residing in the background agn host galaxy is subject to a strong uv radiation field by the central engine itself , and thus might contain more than normal galaxies . and while we are per se not interested in such absorbers for the study of intergalactic gas , we have included intrinsic absorbers , as they are being found with our algorithm as well . + from hydrodynamic simulations @xcite assert , however , that the majority of detectable lines with rest frame equivalent widths ew@xmath26 do _ not _ reside in virialised regions like galaxies , groups and clusters , but trace intergalactic gas at only moderate overdensities ( @xmath810 - 40 ) . although their estimated redshift path density of about 5 such absorbers per redshift of unity drops rapidly by a factor of at least ten when increasing the equivalent width limit to 0.35 , even that lower rate of incidence allows us to detect an appreciable number of such strong intergalactic absorbers , owing to the large redshift path length probed by the collective sdss agn sample . + to summarise , we expect to detect the signals of absorbers in a variety of environments . while this first step of finding rests upon a blind doublet search plus an alignment with lyman @xmath0 and @xmath27 lines , in order to ensure high purity of the candidate sample , we may then , in a future effort , use independent line detections of different ions to better characterise the physical stae and potentially the nature of the absorbers found here . because there are a variety of possible origins of the ovi gas , including outflows , intervening galaxies , and the contribution of the agn host galaxy , this next step is needed to arrive at a sample of igm absorbers . there is evidence for a population of absorbers that contain little , if any detectable hydrogen , as pointed out by @xcite , and it is clear that our search algorithm presented here can not detect those . however , as we anticipate the need for high - resolution follow - up of our candidates in order to confirm their nature , we might then be able to learn about this oxygen - rich subset ( type 0 of @xcite ) of absorbers as well . + the paper is organised as follows : section [ prelim ] details our preliminary analyses of whether the signal - to - noise ratio of the sdss spectra and the density of the lyman forest enables our search program . in section [ agn_sample ] we present our sdss qso sample selection , and section [ search_strategy ] gives an overview of the search algorithms we applied . before we summarise and conclude in section [ summary ] , we present the results of our search in section [ candidate_list ] . the search method developed here can be easly tailored to a variety of different ionic species . we stress that follow - up observations of the candidates we retrieve with high s / n and high resolution is neccessary to determine the reliability of our search algorithm by determining the nature of the features we detect . + throughout this study we use a cosmology with h@xmath28=71 km s@xmath29mpc@xmath29 , @xmath30 = 0.27 and @xmath31=0.73 . abundances are given by number relative to hydrogen , and vacuum wavelength are being used , if not otherwise noted . with rest - frame wavelengths below the lyman @xmath0 wavelength of 1215.67 , the 1032/1038 doublet absorption lines inevitably fall into the lyman alpha forest of systems located along the same sightline with slightly lower redshifts . thus , two effects may prohibit our attempts to find such absorbers : first , the ubiquity of the ly@xmath0 forest lines , especially at higher redshifts , can lead to blending of the oxygen lines with lines , thus rendering a correct identification of the doublet impossible . second , two different ly @xmath0 lines at different redshifts and the accidental occurence of optical depth ratios expected for can mimic an doublet , leading to a false identification . in order to estimate the severity of the latter effect , we have created mock catalogues of lyman forests lines , using the line density estimate of @xcite and the column density distribution function of @xcite : @xmath32 @xmath33 where @xmath34 denotes the number of lyman @xmath0 lines per redshift interval dz , and f(n@xmath35 ) is the probability distribution of obtaining a line with column density of neutral hydrogen n@xmath35 within the interval n@xmath35 , n@xmath35+dn@xmath35 . we then analysed these catalogs , and retrieved all line pairs that exhibit the correct wavelength ratio ( within a given tolerance ) as well as the correct ratio of optical depths at the line center : @xmath36 @xmath37 @xmath38 is independent of redshift since both lines get shifted by the same factor ( 1+z ) , and the result for @xmath39 follows from atomic physics : @xmath40 where n is the column density , @xmath41 the wavelength of the transition , c the speed of light , a@xmath42 the einstein - coefficient for the transition from the upper ( k ) to the lower ( j ) level , b@xmath43 the line broadening factor . the statistical weights @xmath44 of the two transitions is equal to two . + with the given resolution of sdss spectra ( r@xmath81800 ) and the conservative assumption that line centroids can be determined to about 1/3 of the resolution element , we expect a tolerance level for the wavelength ratio of @xmath45 = 3.0@xmath4610@xmath47 . furthermore , we can rule out all pairs that do not exhibit the range of flux decrements , expected from the ratios of the two line strengths convolved with the instrument capabilities . the efficiency of this procedure depends crucially on the ability to precisely measure the equivalent width ( ew ) of lines , but even a conservative estimate of a 20% accuracy on a single pixel flux measurement , corresponding to a low signal - to - noise ratio of about 5 , leads to a reduction of the number of false detections of roughly a factor of eight at all redshifts . + an analysis of the mock data sets spanning the complete redshift range available to us ( 2.7 @xmath48(sdss ) ) and varying the thresholds on the desired precision for @xmath38 and @xmath39 yields for the number of expected interlopers : @xmath49 where @xmath50 is the absorber redshift , and @xmath51 is the number of expected absorbers per spectrum . n@xmath52 is a strongly increasing function of the redshift : @xmath53 note that the coefficients and exponents in equations [ interlopers ] and [ g_z ] are derived from analysing the mock data set , without taking additional absorption from metal lines into account . nor did we analyse the contribution of random noise features mimicking doublets . the latter problem , however , is obviated by the fact that we are going to focus on lines that are clearly above the noise limit , as we will explain in section 3 . + we focus on absorbers accompanied by strong lyman absorption [ o / h ] @xmath54 0 ) . such absorbers , classified by bergeron & herbert - fort as type 0 , could potentially slip through our search criterion as they need not be accompanied by strong absorption . a priori we can not determine which fraction of these absorbers we might lose , some of them exhibit strong enough absorption to pass our criteria ( right panel fig . 2 of bergeron & herbert - fort ( 2005 ) ) . ] . therefore , we rule out all lines that do not show accompanying transitions . given the line density estimate of @xcite , the lower limit for the average velocity separation of two lyman @xmath0 absorbers is @xmath55 where we used the relation @xmath56 to obtain the lower limit in the last step , and equated @xmath57 . + the velocity separation of one pixel in sdss spectra , however , amounts to @xmath58 thus , the probability to have at least one line fall right onto the pixel corresponding to the redshift of an `` pixel '' is @xmath59 note that the @xcite line density includes lyman @xmath0 absorbers with column densities as low as log n@xmath35 = 13.64 . restricting ourselves to absorbers that show at least an optical depth at line centre for the lyman @xmath27 line of @xmath60(lyman @xmath27 ) = 1.0 , leads to another factor of f@xmath61 0.4 in the exclusion of spurious interlopers therefore , we can further reduce the number of spurious interlopers by @xmath62 , at least , by requiring the candidate feature to be accompanied by strong lyman @xmath0 and @xmath27 absorption . since higher order series lines become rather weak , adding additional components probably does not follow the same simple multiplicative propagation of probabilities , and so we have considered in this feasibility study only the first two terms of the lyman series . + of the absorber . the rapid increase for the lines up to log ( 1+z@xmath63 ( = z@xmath6 3.4 ) can mainly be attributed to the nearly linearly growing redshift pathlength , whereas the rapid increase of interlopers is due to the redshift evolution of the line density . we assume an [ o / h ] = -2.0 to calculate the number of absorbers by integrating over the column density distribution by hu et al . ( 1995 ) , after assessing a correction factor for the fraction of lines retrievable ( for details see section 2.2 ) . the red , dashed curve represents the expected number of interlopers that exhibit less than a fiducial value for the deviations in both line position and line strength ( for details see text ) , whereas the dotted line takes the additional requirement into account that the interlopers have to show associated , strong absorption . it is obvious that our search efficiency peaks around a redshift of z@xmath6 3.2 , and rapidly deteriorates for redshifts z@xmath64 . thus , we expect our window of opportunity to be between 2.8 @xmath65 4.0 , where the ratio of detections of real systems to spurious interlopers is greater than unity.,title="fig:"][expected_lines ] figure 1 shows in red the number of expected interlopers without ( dashed ) or with ( dotted ) the additional requirement of accompanying lyman @xmath0 and @xmath27 absorption . for these calculations we have used the fiducial values of @xmath66 and @xmath67 . introducing the additional requirement significantly reduces the number of false positives , as figure 1 demonstrates . + finally , we can also disqualify a potential candidate as belonging to an doublet if it is clear that it is part of a lyman series itself , i.e. we can check whether there are lower ( or also higher ) lyman transitions at the same redshift with similar velocity structures . this cut becomes possible for all lines above 4500 when the lyman @xmath27 line redshifts into the lower wavelength regime of sdss . thus , we expect to even further reduce the number of falsely classified absorbers beyond a redshift of z@xmath68 3.37 by a large fraction . another criterion that can be used to correctly identify a metal line doublet is the shape of the lines : both are expected to have the same velocity structure and should be rather narrow , in stark contrast to the lyman lines that tend to be broader than 20 km / s . due to the low resolution of sdss spectra , however , this criterion can not be implemented here unless the line is so strong that its width exceeds one resolution element . such strongly saturated or even damped lines are rare , and would have been picked up by other automatic search algorithms already , as we describe later . + we will discuss the implementation and success of these and various other criteria to distinguish real lines from features mimicking metal lines in section 3 . in order to estimate whether blending and/or noise renders the identification of existing doublets in the lyman forest impossible , we have created 100 mock spectra of a typical lyman forest , following the number and column density distributions from above , and adjusting the spectral resolution and pixel size to the sdss fiducial value . for the doppler broadening parameter , we have assumed a gaussian distribution of width 8 km / s around a mean of @xmath69 , the exact values for the distribution are , however , not important at this stage of modelling , as the instrumental resolution broadens the line profiles over more than the average velocity width . in this forest we placed doublets at a redshift of about 3.5 , added noise according to the appropriate ( worst - case ) sdss scenario ( i.e. we have chosen a value of s / n = 3 here , which is below the nominal s / n @xmath70 4 sdss limit , see section 3.3 for details of the sdss sample s s / n ) , and checked carefully how often we could reliably retrieve the doublet , when varying the strength of the absorption feature ( until it completely vanished in the noise ) . figure 2 shows two examples : in the first , both components are clearly identified , whereas in the second , both are severely blended . of course , stronger lines can be identified more easily . from this analyis , we conclude that at least 15% of systems with an ew = log ( ew@xmath71 can be unambigiously identified at a redshift of z @xmath8 3.5 , and hence higher fractions at lower z or greater strength . because the modelling at this stage is only suppposed to render an initial estimate of the line detectability , we have not included second - order effects like line clustering , or a detailed description of the effects of continuum placement errors . + forest features , generated according to the column density distribution and line densities discussed in the text , an doublet with ew@xmath72 = 0.15 was placed at redshift z@xmath6 3.5 . the positions of the two lines are indicated by the dotted line . then the worst case level of noise for the sdss spectra was added . * upper panel : * in this example both absorption lines can be easily identified and they maintain the expected optical depth ratio.*lower panel : * in this example , both components are blended with lyman forest lines , changing the optical depth ratio drastically . note also how noise affected the 1038 component and shifted its flux upwards . clearly , such a scenario renders the identification of the doublet impossible.,title="fig:"][mock_spectrum1 ] how many such absorbers do we expect per sightline ? the above limit for the ew lies only slightly higher than the line observability index ( lox for an absorption line of a given species by modelling gas irradiated with a given x - ray / uv background with cloudy @xcite . for our purposes , lox @xmath8 log @xmath73 , cf . @xcite ) derived by @xcite for a metallicity of [ o / h]@xmath74 = log [ n(o)/n(h ) ] - log [ n(o)/n(h)]@xmath75 . integrating the @xcite distribution over the interval @xmath76 log n@xmath77 , where the lox is of order 1.5 and above ( cf . fig.2 of @xcite ) , and along the sightline to a qso to a redshift z@xmath24 of the absorber , we obtain the number n(z@xmath24 ) of such lines associated with detectable absorption : @xmath78 where @xmath79 is the redshift path length , and r(lox , z@xmath22 ) is the estimate for the fraction of lines correctly identifiable at the emission redshift of the qso . here we have assumed for simplicity a constant r(lox ) = 0.15 , a value based upon our simulation at z@xmath80 3.5 . the rapid rise of the number of expected lines in figure 1 is a result of the almost linearly growing redshift pathlength @xmath81dz . it is clear from figure 1 that beyond redshifts of z@xmath6 4.2 , the sheer number of interlopers will make unambigious detections of features very difficult , and the search efficiency will peak below that redshift . from the simulations and estimates above , we conclude that we will be able to retrieve a large enough sample of absorbers in the spectra of sdss quasars to obtain meaningful results for a statistical analysis . furthermore , we will be able to exclude false identifications via a variety of methods , so that we can hope to clean , to a certain degree , our sample of system candidates up to redshifts @xmath19 4.0 . the sloan digital sky survey ( sdss ) is an imaging and spectroscopic survey of the sky @xcite using a dedicated wide - field 2.5 m telescope @xcite at apache point observatory , new mexico . imaging is carried out in a drift - scan mode using five broad @xmath82 bands spanning the range from 3000 to 10,000 @xcite . the 95% completeness limit in the @xmath83 band is 22.2 mag ( on the ab system ) . for more specific information on the imaging and quality control thereof see @xcite and @xcite . details of the astrometry can be found in @xcite . a variety of algorithms selects objects from the imaging data for spectroscopy . these targets are arranged on series of plug plates , called tiles , with radius 1.49@xmath84 @xcite , each containing provision for a total of 640 targets and calibration stars . optical fibers at the focal plane feed the light from holes in aluminium plates to a pair of double spectrographs . the resulting spectra range from 3800 to 9200 with a resolution of r = @xmath85 . the s / n of the sdss spectra entering the public data base are required to be above 4.0 at a fibre magnitude of 19.9 in the red sdss spectrum and a g magnitude of 20.2 in the blue @xcite . specific information on sdss data products can be found in @xcite and @xcite as well as @xcite . quasars within the sdss survey are mainly selected by colour criteria from the imaging campaign , and subsequently verified by spectroscopic follow - up studies @xcite . quasar spectra with absorption lines have been published by a number of groups : @xcite , mainly focusing on broad absorption features . several teams have developed pipelines isolating and identifying absorption lines and systems in qso spectra @xcite . following the format of the catalogue by @xcite a comparison of these pipeline - created catalogues with lists made by visual inspection reveal a completeness of above 95% for such an automation . the incompleteness results mainly from blending of the line , key to the detection of qso absorption systems , most importantly with night sky emission . + the main catalogue of @xcite presents lists of significant absorption lines , equivalent widths , various line parameters and possible redshifts . discrete absorption line systems are obtained by matching lines of different transitions of redshifts within a certain range , depending on the s / n of the spectra . different criteria such as number and quality of components allow for a grading of the systems - all class a systems e.g. need to contain at least four , unblended lines above a 4@xmath86 detection limit matching in redshift , free of artefacts and blends , occuring redward of the lyman @xmath0 forest @xcite . the qsoals catalogue used here is based on the sdss data release 3(dr3 ) @xcite . at rest - frame wavelengths of 1031.912 and 1037.613 , the doublet starts to become visible in the spectra of sdss quasars only at redshifts of z@xmath6 2.7 . we have thus retrieved all normalised qso spectra in the qsoals database beyond this lower redshift . our sample consists of 3702 quasars with redshifts ranging from 2.70 to 5.413 @xcite . figure 3 shows the redshift distribution of the sample . more than 60% of the sources are below a redshift of z@xmath87 3.5 , where we expect to maximise the search efficiency for doublet signatures and the cleanliness of the candidate sample . as figure 3 demonstrates , the majority of the sources have @xmath88 band magnitudes between 19.5 and 20.5 . as the reliability of an absorption line detection is a function of the local signal - to - noise ratio , we have determined the average signal - to - noise ratio of all spectra within the region of interest for us . this average value is usually below the overall signal - to - noise ratio of the spectrum since the flux levels within the lyman @xmath0 forests are much lower than beyond it . figure 3 shows the distribution of these average signal - to - noise ratio values with the emission redshifts of the sample sources . especially towards higher redshifts , it is apparent that the increasing density of the forest and the general faintness of the sources lead to a severe decrease in the signal - to - noise ratio that will hinder unambigious identifications of absorption lines . however , below z@xmath80 3.5 there are enough quasars with signal - to - noise ratio above the conservative limit we used in section 2.1 for the feasibility study : 359 spectra are of objects with z@xmath89 and have a signal - to - noise ratio within the forest of s / [email protected] . distribution of our sdss qsoals sample of 3702 quasars . note that more than 60% of the sources are below a redshift of z@xmath87 3.5 , where we expect to maximise the search efficiency and cleanliness of the candidate sample . 359 of these have a s / n @xmath705.0 . * middle panel :* the @xmath88 mag distribution ( taken from @xcite ) of our sdss qsoals sample of 3702 quasars . the majority of the sources exhibit fluxes between 19.5 @xmath90 20.5 . * bottom panel :* the average signal - to - noise ratio per pixel in the lyman @xmath0 forest for each source versus the emission redshift z@xmath22 . due to the decreased flux in the forest , this value for the signal - to - noise ratio often falls substantially below the nominal lower limit of signal - to - noise ratio=4.0 ( indicated by the dashed line ) for the complete spectrum to enter the sdss data set , especially towards higher redshifts for which the sources become fainter , in general , and also the density of the forest increases substantially.,title="fig:"][z_distribution ] the absorption systems and line lists of the sdss qsoals catalogues primarily contain securely identified features , resting primarily on key absorption lines like the or doublets . in many cases , identification of such systems is made possible by the fact that the absorption occurs redward of the lyman @xmath0 forest and thus in relatively clean parts of the spectra . for our search of signatures of the doublet , however , it is clear that all of the candidates must be found in the forest . furthermore , these systems need not be accompanied by other metal absorption systems like civ or mgii that would allow unambigious identification outside the lyman forest ( cf . thus , we can not expect the automated pipelines used to create the sdss qsoals catalogues to have picked up on the relatively weak features within the lyman forest in which we are interested . therefore , we had to develop our own algorithms to construct a complete list of absorbers at intervening redshifts . after completing our search with this method , we performed a cross - check with the qsoals database to see if the putative candidates also show transitions of other ions at the same redshift . the results of these checks are summarised in section 5.1 . for a median doppler broadening parameter @xmath91 = 16 km / s , obtained by simcoe et al . ( 2002 ) , it is obvious that nearly all of the lines that we expect to be present in sdss spectra will not be resolved , as 1 ( i.e. the rough pixel size ) represents about 70 km / s in velocity space at 4000 , where we expect to have the best search efficiency . nonetheless , we implemented a simple pixel by pixel comparison for all spectra , characterising each single pixel within the wavelength range from the blue sdss limit of 3800 up to the maximum wavelength allowed for the 1032 component @xmath92 by its normalised flux and thus optical depth . for each pixel , we derive its hypothetical absorption redshift z@xmath24 as if it were the 1032 component . then we identify the corresponding pixels in the spectra that would belong to the second 1038 component as well as the lyman @xmath0,@xmath27 and , if possible , @xmath93 lines for that same redshift . naturally , the overlap between the redshifted wavelength range of the 1032 component pixel and the actual wavelength range of the pixel containing the corresponding 1038 component is not perfect . and thus , the absorption structure of the 1038 transition corresponding to its 1032 counterpart can span in principle two pixels , and hence the _ ratio _ of the absorption measured in such pairs of 1032/1038 pixels may not be a reliable quantity for identifying anymore . the estimates of the line widths above , however , demonstrate that the velocity width of a line is most likely going to be only a small fraction of the width of a single pixel - and so this scenario will occur only in a small number of cases . furthermore , by coincidence , the wavelength separation of the pixels in sdss spectra and wavelength ratio specific to the doublet create an overlap at the level of 0.94 for all spectra and all pixels . we did , additionally , check for potential candidates in the pixel pairs that do overlap by only 6% , but the number of candidates retrieved in this way agreed well with just the expectations from random interlopers . we therefore estimate that the loss of valid detections due to this complication is minimal . in order to assess whether an absorption feature exhibits the correct ratio of optical depths while not being able to fully resolve the structure of the absorption lines , we need to derive a quantitative measure of that ratio from the observed normalised fluxes per pixel . assume that a ) the complete profile of an transition fits into one resolution element , and b ) that this line is the only absorption line within that element ( i.e. no blending with lyman lines ) . then , it is possible to extract the equivalent width of the in that pixel from the measured normalised flux as @xmath94 where @xmath95 is the equivalent width of the line in that pixel in the observer s frame , @xmath96 is the pixel size in wavelength units , and @xmath97 is the normalised flux of the pixel assumed to contain the 1032 transition . from this estimate of the equivalent width @xmath95 , we can in turn derive the expected flux @xmath98 at the corresponding pixel : @xmath99 where @xmath100 is the ratio of equivalent widths of the two transitions . this ratio depends on the strength and velocity width of the absorber . for unsaturated lines it remains fairly close to 2 , and drops to unity for saturated lines . we have modelled the 1032 and 1038 transitions with voigt profiles of various doppler parameters @xmath101 , and are thus able to compute : @xmath102 as a function of the estimated equivalent width ew@xmath103 . this introduces a model dependence into our derivation as we need to assume a value for the doppler parameter @xmath101 . we find , however , that even large variations of the doppler parameter , @xmath104 , only affect @xmath105 on the 30% level . here we take @xmath106 as fiducial value for the doppler parameter . + the following criteria were used to classify pixels as candidates belonging to an doublet . 1 . both pixels of the doublet candidate need to exhibit normalised fluxes lower than 1.0 . this simply ensures that absorption features rather than pixels whose noise pushed them over the flux level of unity are selected . 2 . the expected flux @xmath98 is allowed to deviate from the actually measured flux @xmath107 less than a certain threshold @xmath108 where @xmath86 is the standard deviation of this flux difference , estimated from the signal - to - noise ratio of each pixel and the uncertainty on the line width mentioned above . the effects of varying @xmath109 will be described in the next section . the inferred equivalent width of the stronger 1032 component has to be above a minimum limit @xmath110 this enables us to clear our candidate lists from very weak features , i.e. spurious lyman @xmath0 forest lines with low column densities which we expect to be the main contaminant at high redshifts . note that this criterion is usually automatically fulfilled for candidates in spectra with low signal - to - noise ratio passing criterion 4 mentioned below , and thus is mostly useful for cases where the signal - to - noise ratio is significantly above the 4.0 threshold for entering the sample . 4 . the significance of absorption in the normalised units . the ratio of this quantity to the noise in the same units , i.e. @xmath111 , is our metric for the significance . ] in each pixel should be above a low threshold @xmath112 while this criterion is already implicitly in place for the expected and measured flux difference and the lower limit to the equivalent width , this formulation puts a stronger limit on the number of candidates . requiring a high significance for each pixel allows us to focus on absorption lines with a high likelihood of being real , rather than noise artifacts . in contrast to criterion 3 , this cut proves most useful in spectra with low signal - to - noise ratios . note that this criterion introduces a lower threshold for the equivalent width that is signal - to - noise ratio - dependent : at a fiducial average signal - to - noise ratio of 5.0 ( 8.0 ) and a significance level of 4.0 @xmath86 ( 2.0 @xmath86 ) , it corresponds to a rest - frame equivalent width limit for the sdss set - up of ew@xmath113 = 0.195 ( 0.06 ) . since we expect to be accompanied by lyman features , we require @xmath114 where the x represents either lyman @xmath115 or - if applicable - @xmath93 . due to a lack of information regarding the metallicity and the physical state of the absorbing gas , there is , a priori , no straight forward way of calculating the absorption expected for a given value of @xmath97 . therefore , we introduce these hard cuts on the flux for the associated lyman series transitions , and resort to this simple , yet efficient method of checking for accompanying absorption . we reject all candidates that show signs of being part of lyman series absorption , i.e. we check ( by eye ) whether the feature in question could be a lyman line , accompanied by higher ( or lower ) series members elsewhere in the spectrum with a similar velocity structure . this turns out not to be a severe cut : only 4 of altogether 1760 candidates passing all other tests had to be eliminated by this procedure . thus , we find it unnecessary to apply a more rigorous test than this check by eye . it is clear that the efficiency of detecting candidates and filtering out spurious lyman series absorbers is a complicated function of redshift and signal - to - noise ratio of each spectrum . thus , a more flexible approach , tailored to each object , could result in a higher overall yield . we have , nevertheless , applied the aforementioned rigorous , yet simple strategy to the complete sample regardless of redshift and signal - to - noise ratio of each source . this allows for a more straightforward comparison of the results with expectations based upon our initial feasibility study , and certainly facilitates quantitative analyses like completeness estimates . and it is certainly the conservative approach . + we note that possible blends of the two 1032/1038 transitions with other lines at very similar rest - wavelengths constitute an additional complication . specifically , the presence of strong lines of might have the 1036 line of overlapping with 1038 , and potentialy shift this blend to a lower wavelength . also , any system that contains appreciable amounts of molecular hydrogen , h2 , will also produce a blend with the 1038 transition , but leave the 1032 component untainted . at the redshift range of interest for us , however , the separation between the and 1038 line centres amounts to at least 6 (observed frame ) , too much to seriously affect our search criteria . hence , we have not implemented any correction factors due to the presence of these features into our search algorithms . + interestingly enough , none of our good candidates appear in systems that are bona - fide damped - lyman @xmath0 absorbers ( dlas ) . while fox et al . ( 2007 ) argue that potentially all dlas contain appreciable amounts of absorption , we caution that in their sample of 35 high - resolution spectra of dlas and sub - dlas covering also the ovi doublet , about 60% of the sightlines show blended features due to the high density of forest lines . obviously , the ability to retain most real absorbers in the candidate list , while filtering out the contaminating spurious features mimicking metal lines , depends crucially on the choice of the parameters k@xmath116 and @xmath117(limit ) from our search criteria . we have extensively tested the effects of different choices for these parameters on the numbers and the quality of the candidates . . for all panels , the data points in black result from the fully implemented search algorithm for candidates , whereas data points in red are obtained by searching for candidates with random redshift separations instead of the fixed one based upon the wavelength ratio , while still maintaining all the other criteria . the dotted , vertical lines indicate the choices for the cut criteria that minimise the spurious interloper contribution , while simultaneously maximising the number of real candidates kept in the sample . * criterion 4 * ( upper left panel ) the ratio of the numbers of real to spurious candidates increases up to a significance level of @xmath118 . * criterion 2 * ( upper right panel ) the average number of candidates retrieved is a strongly increasing function of the allowed flux deviation of the 1038 pixel from the expected value . the uncertainty in the doppler width of the absorbers contributes significantly to the error allowance in the estimate for the expected flux . * criterion 3 * ( lower left panel ) the lower limit on the rest - frame equivalent width of the 1032 component appears well before the turn - over to a much lower retrieval rate of real candidates ( @xmath8 0.3 for 2.9 @xmath119 3.0 ) . criterion 4 is a stronger selector for the low signal - to - noise ratio spectra , thus we have chosen this weaker selector mainly to avoid very weak interlopers for candidates in high signal - to - noise ratio spectra . * criterion 5 * ( lower right panel ) requiring a low flux value for the accompanying lyman transitions , removes effectively a large fraction of the random lines . note that applying even stricter limits will throw out good candidates in low signal - to - noise ratio spectra when noise can push a lyman pixel over the threshold limit.,title="fig:"][cut_effects ] figure 4 demonstrates how changing one of the cut criteria while keeping all others constant affects the average number of candidates retrieved from the 380 spectra within the redshift range @xmath120 . these spectra have on average about 300 usable pixels . the data points in black indicate the average number of candidates per spectrum when applying the full search strategy , while the points in red are derived by introducing a random shift of @xmath121(random ) rather than the true @xmath122 for the doublet separation , but keeping the other selection criteria ( accompanying lyman transitions , line significance and equivalent widths ) . this procedure effectively measures the frequency of interloper incidences . note that in all cases , we retrieve more candidates for the real search than for the randomised case . we have chosen the following parameters for filtering our dataset 1 . criterion 2 @xmath123 + while at first hand , it seems unnecessary to apply such a strict criterion , an inspection of the candidate lists produced with softer limits revealed that too many spurious candidates could pass . this is mainly due to the uncertainty in the estimated , expected ew of the 1038 feature when not knowing the doppler width of the absorbers . while obviously potentially eliminating good candidates in slightly more noisy spectra , the gain in clearing the sample from absorbers mimicking clearly overweighs . criterion 3 @xmath124 + the lower limit on the equivalent width of a line introduced by this cut occurs well before the turn - over to a much lower retrieval rate ( at ew @xmath8 0.3 for the case of 3.9 @xmath54 z@xmath125 4.0 , see figure 4 ) , but still eradicates effectively a higher number of spurious very low column - density absorbers that tend to become ubiquitous at higher redshifts . we have chosen this low limit mainly to keep the ( few ) cases of spectra with very high signal - to - noise ( for sdss standards , i.e. signal - to - noise ratio@xmath126 10.0 ) . note that the lower ew limit introduced by this cut depends on the wavelength coverage of a pixel , and is thus slightly redshift dependent . 3 . criterion 4 @xmath127 s / n(pixel)@xmath128 . + the number of candidates is a strongly declining function of the detection significance , as demonstrated by figure 4 . however , above a significance level of @xmath129 the curve shows signs of flattening , before , finally , at even higher values one simply runs out of pixels above the required signal - to - noise ratio . thus we have decided to apply the above limit , which turns out to be the strongest selector in cutting down the numbers of candidates . criterion 5 : flux limit for lyman lines * @xmath130 * @xmath131 ( if applicable ) * @xmath132 ( if applicable ) + limiting ourselves only to candidates that are accompanied by potential lyman features of such depth maximises the ratio between real absorbers and random interlopers , as can be seen from figure 4 . requiring even lower flux values throws out mainly potential candidates in low signal - to - noise ratio spectra , when noise pushes a lyman pixel over the threshold , while softening the flux criterion leaves too many spurious interlopers in the sample . 5 . in a final step , all of the candidates having passed the above criteria , were checked for possible higher ( and also lower ) associated lyman series lines . only 4 ( out of 1756 ) candidates had to be eliminated this way , which we take as a sign of having implemented already rigorous cuts to the sample with the criteria discussed above . figure [ results ] shows the results of applying the search algorithm with the cut criteria mentioned in section 4 to the full dataset of 3702 spectra . for each redshift bin , the black data points in the lower panel represent the average number of candidates retrieved per spectrum . this number includes both real candidates and spurious interlopers . the red dotted line is a linear fit to the number of such random interlopers derived by averaging over 15 runs of the search algorithm when applying a random redshift offset to the location of the doublet line ( properly at 1038 ) . the redshift evolution of these ( n@xmath133 ) agrees very well with the expectations from the initial feasibility study ( cf . equation 7 and fig . 1 ) . note that the number of real candidates plus spurious interlopers for the full search is always higher than in the random search process , lending further credibility to the robustness of our search strategy . the average approximate ratio of real to random candidates per redshift interval is presented in the upper panel of fig . [ results ] . the number of real candidates is assumed to be the difference between the black data points and the estimate for the interloper frequency ( red dotted line in lower panel of that figure ) . as expected , the ability to produce a clean sample diminishes rapidly beyond redshifts of @xmath134 4.0 . + forest density and the metallicity of the igm are presented . * lower panel : * for each redshift bin , the black data points represent the average number of candidates retrieved per spectrum . this number includes both real candidates and spurious interlopers . the red dotted line is a linear fit to the number of such random interlopers derived by averaging over 15 runs of the search algorithm when applying a random redshift offset to the location of the doublet line ( properly at 1038 ) . the resulting redshift evolution of these interlopers agrees very well with the estimates from the feasibility study in paragraph 1 . for details * upper panel :* the average approximate ratio of real to random candidates per redshift interval . the number of real candidates is assumed to be the difference between the black data points and the estimate for the interloper frequency ( red dotted line in lower panel ) . note the decreasing efficiency to produce a clean sample with increasing redshift . as expected from figure 1 , the `` window of opportunity '' lies roughly between 0.55 @xmath1350.7 (= 2.7 @xmath136 4.0 ) , when the ratio real / random rises above unity.,title="fig:"][results ] [ ew ] altogether , we retrieve 1756 candidates in 855 different agn spectra ( out of a total of 3702 ) passing all of the above criteria . these candidates still exhibit a wide range of properties . we examined each case manually , and grouped them into three categories depending on the quality and likely interpretation . * * high quality sample :* + the highest quality , most unambigious candidates for absorbers . all lines are strong and the ovi doublet plus ly @xmath0 and ly @xmath27 transitions are not obviously blended . the putative absorption is both stronger than the feature and exhibits the expected line ratios for lyman @xmath0 to lyman @xmath27 . line to be blended or even non - existent due to its inaccessibility in the spectra for absorbers of the lowest redshift . maintaining a strict non - blending criterion for ly @xmath93 , which is much weaker than ly @xmath0 and @xmath27 , would have left us with too few good candidates . ] 145 candidates could be placed in this group . in 76 cases , the absorption redshift is so close to the qso s emission redshift that these absorbers are tentatively classifiable as intrinsic ( @xmath137 km / s ) . the remaining group of 69 good candidates will be the prime sample for possible follow - up studies . * * medium quality sample :* + these candidates generally exhibit a secure identification of many of the features in question , but also at least one characteristic that leaves some doubt , e.g. a blending of one line , weak features in comparison to the or excessively broad lines . 586 ( 30% ) candidates fall into this category . * * low quality sample :* + in this class of the lowest quality candidates , we encounter mostly examples with very low overall signal - to - noise ratio within the forest or features that are probably strong lyman @xmath0 absorbers at redshifts low enough such that we can not see lyman @xmath27 to apply our exclusion criterion . 1029 of the 1756 candidates ( @xmath138 ) fall into this category , which is considerably higher than the fraction we expected to be spurious interlopers from our initial feasibility studies , and thus leads us to the conclusion that even in this group there are real absorbers . figures 7 , 8 and 9 highlight examples of each of the categories . feature , which we require to be present , but allow for it to be blended ) . the putative absorption is both stronger than the feature and exhibits the expected line ratios for the lyman series ( @xmath0 to @xmath27 ) . the position of the doublet and the associated lyman @xmath139 lines are indicated by the dotted lines . the plate , fiber and julian date identification for the sdss spectrum , as well as the emission and absorption redshifts together with the i magnitude and the average signal - to - noise ratio within the lyman forest are given at the top of the graph . note that the local s / n , used to compute the significance of a feature , may be different from the average value given here , and is always taken from the sdss pipeline error array estimate.,title="fig:"][figure1 ] lines are indicated by the dotted lines . the plate , fiber and julian date identification as well as the emission and absorption redshifts together with the i magnitude and the general signal - to - noise ratio within the lyman forest are given at the top of the graph . these candidates generally exhibit a secure identification of many of the features in question , but also at least one characteristic that leaves some doubt , e.g. a blending of one line ( as in the case here for the 1032 component ) , weak features in comparison to the or excessively broad lines . note that the example here also highlights a case where the absorber redshift is too low to have the lyman @xmath93 transition in the spectrum.,title="fig:"][figure2 ] lines are indicated by the dotted lines . the plate , fiber and julian date identification as well as the emission and absorption redshifts together with the i magnitude and the general signal - to - noise ratio within the lyman forest are given at the top of the graph . these candidates generally allow only for very insecure identifications of many of the features in question , mostly because of low overall signal - to - noise ratio in the forest ( as in the case presented here ) , but also blending and even a few cases where a strong lyman @xmath0 system could not be identified as such by our method due to the lack of accompanying higher order lyman series transitions because of its low redshift . about 60% of the candidates fall into this category.,title="fig:"][figure3 ] having constructed the candidate list of absorbers with the search algorithm described above , we performed a cross - check with the qsoals database in order to determine whether there are additional ions seen at the same redshift . here , we have focused first on the 69 candidates for intervening absorbers of the highest quality , but will extend the cross - check fully in a forthcoming paper . we have checked for other metal lines accompanying the candidates within a range of @xmath140 km / s . for 10 of these candidates the qsoals database lists secure identifications of absorber systems at redshifts very close to value determined from the doublet . in all of these cases , the velocity difference between the absorber determined by our search algorithm and the listed qsoals redshift for the absorber system is less than 300 km / s . these absorbers show transitions in a variety of other ions , most prominently 1548/1550 , 1335 , 1394/1402 , 1671 , and , in 2 cases of low enough redshift not to push the lines beyond the sdss spectral window , even 1855/1863 . not only does the existence of the other transitions in these absorbers greatly increase the likelihood for the absorbers to be real , but it will also enable us to study the physical state of the absorbing medium via the full , established arsenal of astrophysical diagnostics . the in - depth analysis of these absorbers will be the central theme of forthcoming papers . + while the remaining 59 candidates have no absorption systems listed in the qsoals database , there are cases where the database indicates the detection of an absorption feature at the location expected from our redshift , but does not assign an identification , probably due to a failure to securely identify the transition . we have checked for the presence of the following ions : 1548/1550 , 1335 , 1394/1402 , 1671 , and 1855/1863 . these are the primary transitions that are expected to be strong enough and , due to their rest wavelengths , can occur beyond the lyman forest in the part of the spectrum where the detection is thus much easier . for 36 out of the 59 candidates , we do not find any securely detected absorption feature in the qsoals catalogues . in 11 cases , there is one of the transitions listed , and in 8 cases 2 absorption features are present . for another two absorbers , we found 3 features each , and in the two best cases , even 4 of the 8 transitions in question could be retrieved . given the low likelihood that these features are noise artifacts , especially for the cases where they are redward of the lyman forest , the additional presence of up to 4 absorption lines lends further credibility to the reality of the candidate . + we note that the non - detection of a transition in the qsoals catalogue can not be equated with the non - existence of the feature in the spectrum . in order to enter the qsoals catalogue , a feature needs to pass certain threshold criteria , set by the automatic program detailed in @xcite . specifically , the requirement of a four - sigma detection of a feature results in an equivalent width sensitivity which depends on the signal - to - noise ratio , but is at least of order 50 - 200 m in the absorber restframe @xcite . we are going to manually recheck each spectrum separately at the wavelengths where we expect potential absorption features , and derive upper limits for lines not detected . these can also help with the interpretation of the physical state of the absorbing gas . + a more difficult task is the assessment of the completeness level of our sample . we have noticed that , in some cases , the qsoals database indicates the presence of absorbing systems with which is not being picked up by our search algorithm . we have examined closely the results for both search algorithms for the 134 qso sightlines with 3.2 @xmath20 3.23 . we find that we retrieve 12 out of 37 grade a , b and c absorption systems ( containing 4 , 3 or 2 secure line identifications at the same redshift , respectively ) with doublet identification in the qsoals , whereas none of the 5 grade d systems with is detected by our algorithm . on the other hand , we obtain 24 candidates in that subsample that do not appear in the qsoals database , or only show one of the two doublet components . + what is the reason for the apparently low ( @xmath8 30 % ) success rate for our algorithm compared to the qsoals search program ? in most of the cases where we do not detect a system , one or even both of the doublet transitions are blended , and in some cases there was a lack of accompanying lyman @xmath27 absorption . furthermore , in a significant fraction of the qsoals candidates , there is a velocity difference between various metal components greater than the sdss pixel separation , as pointed out by york et al . ( 2006 ) , which leads to an exclusion of the candidate in our approach . thus , it is clear that the rigidity of our search criteria does lead to a certain fraction of lost real candidates . in order to meaningfully constrain the properties of the igm at high redshift , the estimate of this lost fraction is absolutely crucial . we postpone a full monte - carlo analysis of the search efficiency to the next paper in the series that will deal with this issue in depth for the absorbers with the highest likelihood of being real . + the results of this crosscheck with the qsoals database are summarised in table 3 . [ summary_of_surveys ] .overview of searches for intervening absorbers at high redshift . [ cols="^,^,^,^ " , ] we have performed the search for ovi absorbers based upon the information in the single pixels of the original sdss spectra . the sdss detector is built in such a way that spectra are slightly oversampled compared to the resolution element achievable with the nominal r=1800 resolution . hence , the question arises : could we have done better by smoothing the spectra according to the oversampling before searching , hereby gaining potentially significantly in s / n ? the answer to this question depends to a degree on the goals that are desired for the sample of candidates . in this appendix , we compare the pros and cons of the two possible approaches , and explain why we have chosen to use the single pixel search method . + let us begin by clearly stating our criterion for a successful search algorithm . we want to obtain a sample of good quality ovi absorber candidates . with such a sample we want to be able to perform two different actions : a. the selection of follow - up candidates with higher resolution and s / n . note that we have set out to construct a sample at higher redshift than has ever been tried before , and b. derive basic statistical quantities such as dn / dz and density / metallicity estimates for this sample . hence we need to try both to maximise the number of absorbers found and the cleanliness of the sample . it is immediately obvious that increasing the s / n of the basic search bin ( pixel vs. resolution element ) could indeed be very helpful to increase the number of absorbers we can hope to find , given the often very low quality of the spectra with s / n per pixel of the order of a few . on the other hand , smoothing over a larger bin size increases the probability of hi interlopers falling within the bin distorting the flux measurement in a good candidate , or creating new bad candidates by mimicking an ovi doublet . therefore , it is a priori not clear which method is favourable . + in order to decide upon a quantitative argument , we have indeed smoothed all 3702 spectra to the nominal resolution of the sdss spectrograph by running a gaussian filter with the appropriate fwhm over each spectrum . the oversampling factor is about 2.5 , and hence we gain roughly a factor of @xmath141 in s / n for each pixel . then we re - ran just the same search algorithm over these new spectra . the results compared to the original search can be summarised as follows : * in the single pixel search , we retrieve 1756 unique candidates for absorbers , i.e. correcting for finding double or multiple candidates within one or two pixels of each other in the same spectrum . the search in the smoothed spectra yields 5018 such candidates , when applying the same search criteria . * there is some overlap : 543 candidates appear in both lists , i.e. a fraction of 31% of the single - pixel candidates are found again in the search after smoothing . * the ratio of interlopers to all candidates ( real and interlopers ) , as estimated by the random search , rises to a much higher value for the search after smoothing than for the single - pixel search : while we estimate about 780 of the 1756 systems found via the latter method to be random interlopers ( 44% ) , the same criteria for the results obtained with the former method yield a fraction of almost 85%(4230 out of 5018 ) , as detailed below . it is this latter point that clearly favours the implementation of the one - pixel search strategy , given the goal of a candidate list that is as clean as possible under the given noise and resolution constraints of sdss . note that the number of detected candidates and random interloper estimates depends strongly on the cut criteria introduced in section 4.1.1 . the specific values we have adopted were chosen to achieve an optimal result for the single - pixel method . is it possible to change those parameters for the new search in order to achieve a lower fraction of interlopers ? we have tested this by varying the cut values over a wide range of the parameter space , just like we did for the original one - pixel search as detailed in section 4.1.2 . , but could not find any combination of them that had the interloper fraction drop below 75% . this is highlighted in figure 10 , that can directly be compared to figure 4 : it is evident that the number of detections rises by a factor of about 5 - 8 for all cut criteria , but the fraction of random interlopers increases even more dramatically all across the board ( red lines in the figure ) . this result is somewhat surprising given that the initial feasibility analysis showed that by increasing the s / n the ratio of real - to - random candidates should also increase . a potential complication may be the clustering of ( stronger ) lyman forest features , that could increase the number of interlopers . the overlap fraction between candidates found in both the single - pixel approach and the search after smoothing may appear on first glance low . specifically , of the 1756 candidate systems from the former method only 543 remain after smoothing ( 31 % ) , while our estimate of the fraction of real absorbers vs. spurious interlopers and noise artifacts indicates that 69 percent of the the candidates should show a true ovi profile and hence extend over more than one pixel , i.e. are suitable for a successful resolution detection . in the following , we investigate how these two seemingly contradicting numbers can be reconciled . while binning the spectra to the instrumental resolution of sdss , i.e. combining the flux from 2 - 3 adjacent pixels , certainly reduces the noise , but since the absorption in neighbouring pixels is typically weaker one may actually lower the signal - to - noise ratio of the candidate . a priori , it is difficult to assess how the detectability of lines is affected by this modification of the search routine . hence , we have resorted to a monte - carlo analysis , creating new data sets of mock absorbers in resolution space based upon the candidates specifications obtained in the single pixel search . these new sets of candidates were scrutinised by the same algorithm we used to search for absorbers in the real , smoothed spectra . the recipe for creating one such new data set is as follows : we begin by extracting for each candidate the relevant flux and noise value at each pixel in question for the search criteria ( i.e. ovi 1032 & 1038 , ly @xmath115 , and @xmath93 ( when possible ) ) . in order to estimate the expected flux values in adjacent pixels , we simply create a voigt profile which needs to reproduce the flux in the central pixel and has an appropriate doppler parameter ( fixed to b @xmath8 14 km / s for the ovi , and b @xmath8 35 km / s for hi ) . to these expected fluxes in the 2 adjacent pixels , we add noise at the level found for that central pixel . then we combine the information of the new set of pixels according to the method used for the smoothing of the data , i.e. adding and averaging the fluxes while adding the noise in quadrature only . hence , we derive for each candidate of the one pixel search a new set of flux and noise values in the relevant transitions . in this way , we construct a new sample of artificially created potential resolution candidates , based upon the assumption that the one - pixel flux values found in our search represent the depression of flux in the centre of the line . the new sample is then subjected to the same search algorithm we used for the real rebinned spectra , and the fraction of candidates passing the search criteria is noted . we repeat this procedure for 10,000 versions of such mock lists , in order to estimate the expectation value and its scatter for the number of candidates remaining . from this simple procedure , we derive an average `` retention rate '' of 38 @xmath142 3 percent , only slightly higher than the value for the overlap percentage in the real data sets . it seems plausible that allowing for more complicated scenarios ( line centres not exactly at central pixel , correlated noise , range of doppler parameters for underlying profiles , etc ... ) the fraction could be pushed even lower , and hence we conclude that it is not a sign of our search algorithms being unreliable when finding a low overlap percentage .

we discuss the spectral asymptotics of some open subsets of the real line with random fractal boundary and of a random fractal , the continuum random tree . in the case of open subsets with random fractal boundary we establish the existence of the second order term in the asymptotics almost surely and then determine when there will be a central limit theorem which captures the fluctuations around this limit . we will show examples from a class of random fractals generated from dirichlet distributions as this is a relatively simple setting in which there are sets where there will and will not be a central limit theorem . the brownian continuum random tree can also be viewed as a random fractal generated by a dirichlet distribution . the first order term in the spectral asymptotics is known almost surely and here we show that there is a central limit theorem describing the fluctuations about this , though the positivity of the variance arising in the central limit theorem is left open . in both cases these fractals can be described through a general crump - mode - jagers branching process and we exploit this connection to establish our central limit theorems for the higher order terms in the spectral asymptotics . our main tool is a central limit theorem for such general branching processes which we prove under conditions which are weaker than those previously known . + * msc : * 28a80 ( primary ) , 60j80 , 35p20 ( secondary ) .

1503.06028

let @xmath0 be a non - empty bounded open subset of @xmath1 for @xmath2 and let @xmath3 be the dirichlet laplacian on @xmath0 . then the spectrum @xmath4 of @xmath5 is discrete and forms a positive increasing sequence @xmath6 where the eigenvalues are repeated according to their multiplicity . interest in the geometric information about @xmath0 encoded by @xmath4 started a little over 100 years ago and was crystallised by kac in his paper @xcite entitled ` can one hear the shape of a drum ? ' or more precisely , does @xmath4 determine @xmath0 up to isometry ? the answer to that question is no in general , as shown in @xcite ; see also @xcite for a concise presentation of a family of counterexamples . however some geometric information about @xmath0 can be recovered . weyl s theorem shows that the eigenvalue counting function @xmath7 defined by @xmath8 has asymptotic expansion @xmath9 as @xmath10 , for some constant @xmath11 depending only on @xmath12 , where @xmath13 denotes the @xmath12-dimensional lebesgue measure . aside from prompting kac s question this result has led to a large body of work on the behaviour of the eigenvalue counting function and we now give a very brief description of the results that have motivated the work we will present here . as a first extension it is natural to ask about the second order term in this expansion . if @xmath14 is smooth , then under some assumptions , that there are not too many periodic geodesics , the expansion has a second order term @xmath15 as @xmath10 , for some other constant @xmath16 depending only on @xmath12 . the reader is referred to @xcite and references therein for more information . this means that , under some regularity conditions , we can recover the size of the domain and that of the boundary from the spectral asymptotics ; in particular , using the isoperimetric inequality , we can determine whether or not @xmath0 is an open ball . interest in the second term of the expansion of @xmath7 grew further when berry studied the spectral asymptotics of domains with a fractal boundary in @xcite . he conjectured that the hausdorff dimension of @xmath14 should drive the second order term . brossard and carmona in @xcite studied the associated partition function , a smoothed version of the eigenvalue counting function , and showed that the minkowski dimension , @xmath17 , was the relevant notion of dimension for the second order term in the short time expansion of this function . for the counting function itself a general result of lapidus @xcite shows that , if @xmath18 , the second order term is of order @xmath19 provided the minkowski content of the boundary is finite . in general it is difficult to determine the precise order of growth for the second order term for arbitrary boundaries , however for one - dimensional domains @xcite it was shown that the minkowski dimension captures the order of growth of the second term in the asymptotics and the minkowski content , the constant , when they exist . the problem of determining the spectral asymptotics has also been considered for sets which are themselves fractal . for some classes of fractal , such as the sierpinski gasket , or more generally p.c.f . self - similar sets @xcite or generalised sierpinski carpets @xcite , a laplacian can be defined and shown to have a discrete spectrum . the exponent for the leading order growth rate in the eigenvalue counting function is called the spectral dimension and differs from the hausdorff or minkowski dimension of the set . if the fractal has enough symmetry , such as for instance the sierpinski gasket , then a weyl type theorem is no longer true @xcite , @xcite in that the rescaled limit of the eigenvalue counting function does not converge . however the weyl limit does exist for ` generic ' deterministic p.c.f . self - similar sets @xcite and also for random sierpinski gaskets @xcite and it is natural to ask about the growth of the second order term in these settings . our aim is to consider some random fractals where we anticipate more generic behaviour of the counting function . we will consider both domains with fractal boundaries and fractal sets here . firstly we will consider the case of open subsets with fractal boundaries in the one - dimensional case of a so called fractal string . our second case will be an example where the set itself is a fractal , the continuum random tree . in both cases the first order terms in the spectral asymptotics due to the fractal structure are understood and we will focus on the behaviour of the second order terms . a fractal string is a set obtained as the complement of a cantor set in the unit interval , so can be thought of as a sequence of intervals of decreasing length @xcite . the dirichlet laplacian is then the union of the dirichlet laplacians on each interval . some discussion of the spectral asymptotics of random fractal strings can be found in @xcite where it is shown that for cantor sets constructed via random iterated function systems , the second order term due to the boundary exists almost surely . we will consider a suitable subset of these random fractal strings and determine when the order of the fluctuations about the boundary term is given by a central limit theorem ( clt ) . this turns out to be a subtle question and the existence of a clt is determined by the rate of convergence in an associated renewal theorem . we will give a precise statement after introducing all the terminology in theorem [ thm::spectralasympstring ] . we will then show that when the fractal is generated using a dirichlet distribution , the existence of a central limit theorem depends on the particular dirichlet distribution considered . an example of what we are able to show is the following . let @xmath20 , for @xmath21 , be the random fractal string obtained as the complement of the random cantor set generated by subdividing any interval of length @xmath22 into three , retaining two intervals of size @xmath23 , and removing one of length @xmath24 , where the pair @xmath25 is independent for each interval and distributed as dirichlet(@xmath26 ) ( that is a beta(@xmath26 ) distribution in this simple case ) and @xmath27 . we write @xmath28 for the probability law for the random fractal string and @xmath29 for expectation with respect to @xmath28 . we note that @xmath30 will be the minkowski dimension of the random cantor set @xmath28-almost surely , that is the dimension of the boundary of the string . we write @xmath31 for the associated eigenvalue counting function . [ thm : ex1 ] ( i ) for all @xmath32 and @xmath33 there is a strictly positive deterministic constant @xmath34 such that as @xmath35 @xmath36 ( ii ) if @xmath37 , then there exists a strictly positive deterministic constant @xmath38 such that as @xmath35 @xmath39 where @xmath40 is normally distributed with mean 0 and variance @xmath41 . + ( iii ) there exists an @xmath42 and a @xmath33 such that : if @xmath43 , then there exists a not - identically - zero periodic function @xmath44 such that @xmath45 where @xmath46 , @xmath47 and , for this range of @xmath48 we have @xmath49 . in particular @xmath50 does not converge in distribution as @xmath35 . \(1 ) the first result gives the almost sure behaviour of the second term in the counting function asymptotics and is true for random fractal strings constructed using a wide class of distributions on the simplex . + ( 2 ) in part ( iii ) we conjecture that it is possible to take @xmath51 and any @xmath52 . indeed , towards proving the above result , we first provide conditions under which a clt holds ( see theorem [ thm::spectralasympstring ] and section [ subsec::numericalexample ] ) , and explain when one will not ( see remark [ rmk::sharpspeedconvergence ] ) . this distinction is determined by the rate of convergence in a related renewal theorem and depends on the values of the roots of @xmath53 , which we solve numerically ( we can also solve this equation analytically for small values of @xmath48 ) . these computations demonstrate that we can take @xmath54 to be at least @xmath55 . furthermore , although we are not able to prove it rigorously , the monotonicity of the results suggests that @xmath54 can be taken arbitrarily large . + ( 3 ) we also conjecture that , in the case where there is no clt , i.e. @xmath56 , the size of the second order term is determined by @xmath57 , in that , @xmath28 almost surely for @xmath58 , @xmath59 where @xmath60 and @xmath61 as @xmath62 . + ( 4 ) the proof of the above result shows that the period of @xmath63 is given by @xmath64 , where @xmath65 is one of the complex conjugate pair of roots whose real part gives @xmath57 . observe that , as @xmath48 increases , the beta@xmath66 distribution becomes closer to the distribution given by a delta measure at the point ( 1/2,1/2 ) . if we take @xmath67 , then we anticipate that our random fractal string should converge ( in a suitable sense ) to the cantor string ( the string formed as the complement of the classical ternary cantor set ) as @xmath48 goes to infinity . it is known that for the cantor string there is a non - constant periodic function that appears in the second order term in the counting function asymptotics @xcite . thus our result suggests that there is a non - trivial transition in the parameter space from the case where there is ` enough randomness ' for a clt about the second order term , to the case where there is not , through to the limit , where there is not even a strong law of large numbers for this term . we will also consider the case of the brownian continuum random tree , a random self - similar fractal . it was shown in @xcite that there was a weyl limit for the counting function in this case . it was also shown that the second order term for this fractal set was of order 1 in mean which would be anticipated as the boundary of the tree is just two points , a 0-dimensional set . in this paper we show that there is a clt about the almost sure asymptotics . however at this point we have not shown strict positivity of the variance due to the complexity of the correlation structure in the variance of the limit of the rescaled counting function . we conjecture that there will be a non - trivial clt for this counting function . this will show that the randomness in the construction means the second order term in the spectral asymptotics is determined by the fluctuations about the leading order term , as these are much greater than the effects due to the boundary of the set . the main technical tool we develop is a central limit theorem for the general crump - mode - jagers branching process . in our setting the random fractal sets , the random cantor set boundary of the string , or the continuum random tree , can be encoded as general branching processes . we are able to use a characteristic associated with these processes to determine the behaviour of the counting function . in this case there may be dependence on the offspring of an individual and we obtain a clt in this more general setting , extending the work of @xcite . we also remark that the techniques used here can easily be applied to geometric counting functions or other functions associated with heat flow , such as the partition function or heat content of the set . we anticipate similar behaviour in the fluctuations of these quantities about their almost sure limits . the paper is organised as follows . in section [ bpcltsec ] , we recall the definition of the general branching process and some laws of large numbers for such processes . we then prove our central limit theorem for the general branching process using a taylor expansion proof . in section [ sec::dirichletweights ] , we restrict ourselves to general branching processes where a suitable function of the birth times is chosen to lie on an @xmath68-dimensional simplex , which will ensure that the limit of the usual branching process martingale is a constant . we will call such processes @xmath69-gbps and discuss extensively how to establish the conditions required for the central limit theorem in this setting as this will allow us to illustrate when we do and do not have a central limit theorem for the associated general branching process . in section [ sec::cantorstrings ] , we define a family of open subsets @xmath70 of @xmath71 $ ] whose random boundary is a statistically self - similar cantor set built using scale factors on the simplex . we are then able to show our main result which gives conditions for the existence of a central limit theorem . in section 5 we consider some examples where the law of the @xmath69-gbp is given by a dirichlet distribution . we show that , for some dirichlet weights , the eigenvalue counting function of the set @xmath70 satisfies a central limit theorem . as a consequence we will be able to establish theorem [ thm : ex1 ] . in section 6 we turn to the continuum random tree . we recall that this tree can be viewed as a random self - similar set and how to construct a laplace operator on it . we then show that the conditions for the general branching process central limit theorem hold and hence there is a clt in the spectral asymptotics . for convenience , we will use the shorthand notation @xmath72 with @xmath73 to mean some positive constant whose value is fixed for the length of a proof or a subsection . in this subsection , we introduce the general or c - m - j branching process . the presentation is inspired by @xcite , to which the reader is referred for further information . in the general branching process , the typical individual @xmath74 is born at time @xmath75 , has offspring whose _ birth times _ are determined by a point process @xmath76 on @xmath77 , a _ lifetime _ modelled as a non - negative random variable @xmath78 , and a ( possibly random ) cdlg function @xmath79 on @xmath80 called a _ characteristic_. we index the individuals of the general branching process using the address space @xmath81 the ancestor @xmath82 is born at time @xmath83 , and individual @xmath74 has @xmath84 offspring whose birth times @xmath85 satisfy @xmath86 where @xmath87 is the dirac measure and @xmath88 is the concatenation of @xmath74 and @xmath89 . the trace of the underlying galton - watson process is a random subtree of @xmath90 which we denote by @xmath91 . we write @xmath92 for the set of infinite lines of descent in the process . for @xmath93 we also use the notation @xmath94 if there exists a sequence @xmath95 with @xmath96 with @xmath97 such that @xmath98 . similarly , for @xmath99 , @xmath100 we write @xmath94 if there exists a sequence @xmath101 with @xmath102 such that @xmath103 . a cut - set @xmath104 of @xmath91 is a collection of @xmath105 such that @xmath106 and @xmath107 for all @xmath108 and @xmath109 there is an @xmath110 such that @xmath94 . it is customary to assume that the triples @xmath111 are i.i.d . but we allow @xmath79 to depend on the progeny of @xmath74 ; we also do _ not _ make any assumptions on the joint distribution of @xmath112 . when discussing a generic individual , it is convenient to drop the dependence on @xmath74 and write @xmath113 . we will write @xmath28 for the associated probability law and @xmath29 for its expectation . we define @xmath114 ) , \quad \nu(dt ) = \be \xi(dt ) , \quad \xi_\gamma(dt ) = e^{-\gamma t } \xi(dt ) , \quad \text{and } \quad \nu_\gamma(dt ) = \be \xi_\gamma(dt),\ ] ] for @xmath115 . furthermore , we will always assume that the general branching process has _ malthusian growth _ , i.e. that there exists a _ malthusian parameter _ @xmath116 for which @xmath117 . this implies , in particular , that the general branching process is super - critical . we denote the moments of the probability measure @xmath118 by @xmath119 in all cases of interest to us , @xmath120 will be finite . note , however , that some convergence results still hold when that is not the case , as explained in @xcite . the presence of the characteristic @xmath121 in the population is captured using the _ characteristic counting process _ @xmath122 defined as @xmath123 where the @xmath124 are i.i.d . copies of @xmath122 . an important example in the study of random fractals is the characteristic @xmath125 , whose corresponding counting process @xmath122 has the property that @xmath126 is the number of offspring born after time @xmath127 to parents born up to time @xmath127 . later , we will define characteristics that count eigenvalues of the dirichlet laplacian . there are two central elements in the study of the asymptotics of the counting process . the first is that the functions @xmath128 satisfy the well - studied renewal equation @xmath129 see @xcite for a classic exposition and @xcite for alternative results . the second is the process defined by @xmath130 where @xmath131 is the set of individuals born after time @xmath127 to parents born up to time @xmath127 . the process @xmath132 is a non - negative cdlg @xmath133-martingale with unit expectation , where @xmath134 we call it the _ fundamental martingale _ of the general branching process . the martingale convergence theorem shows that @xmath135 as @xmath136 , almost - surely , for some random variable @xmath137 . furthermore , under an @xmath138 condition standard in the theory of branching processes , @xmath132 is uniformly integrable . more precisely , in @xcite , doney proved the following result . the following are equivalent : 1 . @xmath139 < \infty$ ] ; 2 . @xmath140 ; 3 . @xmath141 ; 4 . @xmath142 almost surely on the set where there is no extinction ; 5 . @xmath132 is uniformly integrable . otherwise , @xmath143 almost surely . for technical reasons , it is often easier to apply renewal theory under the assumption that @xmath121 vanishes for negative times . when that is not the case , we can set @xmath144 so that @xmath145 this means that @xmath146 , the counting process of the characteristic @xmath147 , and we can then work with @xmath148 instead of @xmath122 because @xmath147 vanishes for negative times and @xmath148 and @xmath122 obviously have the same asymptotics as @xmath136 . as discussed in @xcite , the general branching process provides a natural way to encode statistically self - similar sets . we outline this connection now . to build a statistically self - similar set @xmath149 , we start with the address space @xmath90 defined in and a non - empty compact set @xmath150 . to each @xmath151 , we associate a random collection @xmath152 , where @xmath153 is a natural number and @xmath154 are contracting similitudes whose ratios we write @xmath155 . we assume that the collection is i.i.d . in @xmath74 . the random numbers @xmath156 generate a random subtree @xmath91 of @xmath90 defined by @xmath157 and @xmath158 for @xmath159 , define @xmath160 where @xmath161 is the length of the word @xmath74 . the set @xmath149 has the intuitive property that it can be written as scaled i.i.d . copies of itself , namely , @xmath162 where @xmath163 are i.i.d . copies of @xmath149 . let us emphasise that the choice of @xmath150 is not unique in general . however , for technical reasons discussed in @xcite , we make the following two assumptions . first , we assume that the sets @xmath164 _ form a net _ , i.e.@xmath165 and also @xmath166 the analogue of the _ open set condition _ for self - similar sets . second , we assume that the construction of @xmath149 is proper . i.e. that every cut - set @xmath104 of @xmath91 satisfies the condition : for every @xmath167 , there exists a point in @xmath168 that does not lie in any other @xmath169 with @xmath170 . the hausdorff dimension of statistically self - similar sets is almost surely constant on the event that it is not empty and was calculated in @xcite . it is given in the following result by a formula , the random analogue of that due to moran @xcite and hutchinson @xcite familiar from the deterministic setup . let @xmath149 be a statistically self - similar set . write @xmath171 for the number of similitudes and their ratios . then , on the event that the set @xmath149 is not empty , @xmath172 to specify a general branching process corresponding to the random set @xmath149 , we set @xmath173 and @xmath174 . with this parametrisation , the set @xmath168 in the construction of @xmath149 corresponds to an individual born at time @xmath75 and has size @xmath175 . furthermore , since @xmath176 the malthusian parameter @xmath30 is equal to the hausdorff dimension of @xmath149 by definition . before we can prove our central limit theorem for the general branching process , we state nerman s laws of large numbers , proved in @xcite . they are proved for non - negative characteristics with progeny dependence . in applications , if this is not the case , it suffices to write the characteristic as the difference of its positive and negative parts . we start with the weak law of large numbers . recall that a measure is said to be lattice if its support is contained in a discrete subgroup of @xmath80 and non - lattice otherwise . [ thm::nermanwlln ] let @xmath111 be a general branching process with malthusian parameter @xmath30 , where @xmath177 and @xmath178 for @xmath179 . assume that @xmath180 is directly riemann integrable and that @xmath118 is non - lattice . assume further that , for every @xmath127 , @xmath181 < \infty.\ ] ] then , @xmath182 where @xmath120 is defined in , and @xmath183 as @xmath136 , where @xmath137 is the almost sure limit of the fundamental martingale of the general branching process . furthermore , if @xmath132 is uniformly integrable , then the convergence also takes place in @xmath184 . the strong law of large numbers requires the following additional regularity condition . [ cond::ghfuncondition ] there exist non - increasing bounded positive integrable cdlg functions @xmath185 and @xmath186 on @xmath187 such that @xmath188 < \infty \quad \text{and } \quad \be \left [ \sup_{t \geq 0 } \frac{e^{-\gamma t } \phi(t)}{h(t ) } \right ] < \infty.\ ] ] this first part of the condition is satisfied if there exists a non - increasing bounded positive function @xmath185 such that @xmath189 is finite , because then @xmath190 which has finite expectation . as such , this can be thought of as a moment condition that is weaker than imposing that @xmath118 have a finite second moment ; take @xmath191 . in particular , if the expected number of offspring is finite , this part of the condition is satisfied since , with the latter choice of @xmath185 , @xmath192 we can now state the strong law of large numbers . [ thm::nermanslln ] let @xmath111 be a general branching process with malthusian parameter @xmath30 , where @xmath177 and @xmath178 for @xmath179 . assume that @xmath118 is non - lattice . assume further that condition [ cond::ghfuncondition ] is satisfied . then , @xmath182 where @xmath120 is defined at , and @xmath193 as @xmath136 , where @xmath137 is the almost sure limit of the fundamental martingale of the general branching process . furthermore , if @xmath132 is uniformly integrable , then the convergence also takes place in @xmath184 . similar results have been proved by gatzouras in the lattice case . we will not use them here and refer the reader to @xcite . in @xcite , jagers and nerman proved a central limit theorem for the general branching process under the assumptions that the characteristics are i.i.d . we now give a taylor expansion proof of a similar result , but continue to allow @xmath79 to depend on the progeny of @xmath74 . we start by introducing some additional notation . consider the general branching process @xmath194 with malthusian parameter @xmath30 . we assume that @xmath195 is such that @xmath196 has zero expectation . in applications , @xmath195 is typically a suitably centred version of some characteristic @xmath121 ; we will discuss examples in section [ sec::dirichletweights ] . we will use the rescaled version @xmath197 of @xmath198 defined by @xmath199 where @xmath200 . finally , to have a proxy for the variance , we define @xmath201 where @xmath202 as @xmath203 satisfies an equation of the form which leads to the renewal equation , the functions @xmath204 and @xmath205 defined by @xmath206 satisfy the renewal equation @xmath207 our central limit theorem requires two technical conditions which we discuss now . [ cond::integrabilitycondclt ] there exists @xmath208 such that @xmath209 as @xmath136 . this is a regularity condition on @xmath195 . in applications , we typically expect @xmath195 to satisfy a weak law of large numbers . therefore , the sum should grow like @xmath210 and we can expect that the condition is satisfied . [ cond::momentcondclt ] there exists @xmath211 such that @xmath212 this is a moment condition . in applications , it is convenient to check it for the third moment , i.e. when @xmath213 , because that can be done using renewal arguments . [ thm::clt ] let @xmath194 be a general branching process with malthusian parameter @xmath30 , where @xmath195 is such that @xmath214 for every @xmath127 . assume that @xmath204 is bounded and that @xmath215 some finite constant , as @xmath136 . assume further that conditions [ cond::integrabilitycondclt ] and [ cond::momentcondclt ] hold . then , @xmath216 as @xmath136 , where the distribution of @xmath217 is characterised by @xmath218 = \be\left[e^{- \frac 12 \theta^2 v(\infty ) m_\infty}\right].\ ] ] in the proof , we will use that if @xmath219 and @xmath220 are complex numbers whose modulus is bounded by @xmath221 , then @xmath222 a proof of this may be found in ( * ? ? ? * lemma 3.4.3 ) . for @xmath223 , iterating from , the definition of @xmath197 , we get @xmath224 the first sum appearing in can be rewritten as @xmath225 which , by condition [ cond::integrabilitycondclt ] , converges to 0 in probability as @xmath136 if we choose @xmath226 appropriately small . for the rest of the proof , we fix such a choice of @xmath226 we now consider the other sum appearing in , and show that it converges in distribution to @xmath217 as @xmath136 . the result will then follow from slutsky s lemma . in other words , for @xmath227 , we want to show that @xmath228 \to 0,\ ] ] as @xmath136 . to do this , write , for an @xmath229 that will be chosen below , @xmath230 and split as @xmath231\\ \label{eq::splitsecondbit } & + \be \left [ e^{i \theta \sum_{x \in \lambda_{\epsilon t } } e^{-\gamma \sigma_x/2 } \tilde z_x(t- \sigma_x ) } - e^{- \frac 12 \theta^2 v(\infty ) \sum_{x \in \lambda_{\epsilon t } } e^{-\gamma \sigma_x } } ; a_{\epsilon , t}\right]\\ \label{eq::splitthirdbit } & + \be \left[e^{- \frac 12 \theta^2 v(\infty ) \sum_{x \in \lambda_{\epsilon t } } e^{-\gamma \sigma_x } } - e^{- \frac 12 \theta^2 v(\infty ) m_\infty } ; a_{\epsilon , t}\right ] . \end{aligned}\ ] ] we will show that each of these terms converge to @xmath232 as @xmath136 . fix @xmath227 and @xmath233 . let @xmath234 be such that @xmath235 whenever @xmath236 satisfies @xmath237 . and let @xmath238 be such that for @xmath239 , @xmath240 and @xmath241 where @xmath242 is given by condition [ cond::momentcondclt ] . let us start by dealing with . for @xmath239 , @xmath243 which , by markov s inequality , is bounded by @xmath244\\ & \leq x_0^{-(2 + \kappa ) } |\theta|^{2 + \kappa}e^{-\gamma \epsilon \kappa t/2 } \be \left [ \sum_{x \in \lambda_{\epsilon t}}e^{-\gamma \sigma_x } \be\{| \tilde z_x(t- \sigma_x)|^{2+\kappa}| \cf_{\epsilon t}\ } \right]\\ & \leq x_0^{-(2 + \kappa ) } |\theta|^{2 + \kappa } \sup_{u \in \r } \be \{|\tilde z(u)|^{2 + \kappa}\ } e^{-\gamma \epsilon \kappa t/2}\\ & \leq \delta , \end{aligned}\ ] ] where we have used that @xmath245 , that @xmath246 is independent of @xmath247 and that @xmath248 . ( this is where we need to control the moment of order @xmath249 for some @xmath250 . ) therefore , is dominated by @xmath251 since @xmath87 is arbitrary , it follows that @xmath252 \to 0,\ ] ] as @xmath136 , as required . to deal with , recall that @xmath253 as @xmath136 . therefore , dominated convergence implies that @xmath254 \to 0,\ ] ] as @xmath136 , as required . we will spend the rest of the proof dealing with , which we rewrite as @xmath255\ ] ] using the conditional independence built into the branching structure , where @xmath256 by , the term inside the expectation satisfies @xmath257 a taylor expansion and the second order exponential estimate yield that @xmath258 furthermore , @xmath259 now , since @xmath260 = 0,\ ] ] reasoning as in shows that @xmath261 for @xmath239 . and , since for @xmath262 we have @xmath263 a taylor expansion and the first moment exponential estimate yield that @xmath264 notice that , for @xmath265 , @xmath266 together , to show that is dominated by @xmath267 m_{\epsilon t}\right ] + \be\left[\theta^2 ( 2 + \delta ) \|v\|_\infty \sum_{x \in \lambda_{\epsilon t } \cap \lambda_{t/2 } } e^{-\gamma \sigma_x}\right].\ ] ] fix @xmath268 large . for @xmath127 large enough , by lemma 3.5 of @xcite , @xmath269 \leq \be\left [ \sum_{x \in \lambda_{\epsilon t } \cap \lambda_{\epsilon t + c } } e^{-\gamma \sigma_x } \right ] \to \mu^{-1}{\int_c^\infty(1- \nu_\gamma(s))ds},\ ] ] as @xmath136 . the limiting expression can be made arbitrarily small by choosing @xmath270 large enough . therefore , the second term in converges to 0 . since @xmath248 , the first term can also be made arbitrarily small by adjusting @xmath87 . the result follows . the statement of theorem [ thm::clt ] requires us to show that @xmath271 . this is typically done using renewal theory . recall that if @xmath118 is lattice , supported on @xmath272 , say , then this convergence does not occur . instead we typically get that , for every @xmath273 , @xmath274 as @xmath275 , where @xmath276 is a @xmath277-periodic function . in this case , the proof presented above shows that , for every @xmath273 , @xmath278 as @xmath275 , where the distribution of @xmath279 is characterised by @xmath280 = \be \left [ e^{-\frac 12 \theta^2 g(y ) m_\infty}\right].\ ] ] in this section , we discuss applications of the central limit theorem to the general branching process under the assumption that the number of offspring @xmath281 is constant and equal to @xmath68 , say , and that the birth times @xmath282 are distributed such that , for some fixed constant @xmath283 , we have @xmath284 is a distribution on the simplex @xmath69 in @xmath285 , i.e. @xmath286 the latter assumption ensures that the fundamental martingale @xmath287 . we call a branching process with offspring distribution defined on the simplex in this way a @xmath69-general branching process . for simplicity , we also suppose throughout this section that @xmath118 is non - lattice , though as discussed in section [ lattice ] , we expect subsequential versions of the results below to hold if this is not the case . a basic example that we will return to is the case where the distribution of @xmath284 is dirichlet@xmath288 . in this case our general branching process @xmath111 , with malthusian parameter @xmath30 , will satisfy a weak law of large numbers , i.e.@xmath289 as @xmath136 . we wish to describe the random fluctuations around the limit . to do this , we study the expression @xmath290 = e^{-\gamma t/2}\left[z^\phi(t ) - e^{\gamma t } z^\phi(t)\right ] + e^{\gamma t/2 } [ z^\phi(t ) - z^\phi(\infty)].\ ] ] our aim is to apply theorem [ thm::clt ] to the first part of this expression . we will see that conditions making this possible ensure that the second term converges to 0 . this will then produce a result on the fluctuations of @xmath126 thanks to slutsky s lemma . an outline of this section is that we begin by centring our characteristic in order to apply the central limit theorem of section 2 . we then show that in order to control the centred characteristic we need to control the rate of convergence in the associated renewal theorem . to start , notice that , in the notation of section [ bpcltsec ] , @xmath291 is a centred characteristic counting process , if we define @xmath195 by @xmath292 , \end{aligned}\ ] ] where we have used that @xmath293 . this representation can be used to get the following control on @xmath195 under an assumption on @xmath294 . [ lem::convergencerate ] let @xmath111 be a @xmath69-general branching process . assume that @xmath295 for some positive constants @xmath296 and @xmath297 $ ] . then , @xmath298 notice that @xmath299 the result follows upon noticing that @xmath300 and proceeding similarly for the other sum . this lemma and the decomposition in show that understanding the rate of convergence of @xmath301 to its limit in the renewal theorem is helpful for estimating the fluctuations of @xmath122 . consider the renewal equation @xmath302 where @xmath303 is a non - lattice distribution function on @xmath187 . the key to solving the renewal equation is the renewal measure @xmath304 where @xmath305 denotes the @xmath68-fold convolution of @xmath303 with itself , because @xmath306 is typically given by @xmath307 see @xcite among others . the renewal theorem of @xcite states that if @xmath303 has a finite mean @xmath120 , then @xmath308 where @xmath309 means that @xmath310 as @xmath311 . under some regularity conditions on @xmath312 , e.g. if @xmath312 is directly riemann integrable , this can be used to show that @xmath313 as @xmath136 ; see @xcite . the error in the linear approximation of the renewal function @xmath314 has been studied extensively . when @xmath303 has a finite second moment @xmath315 , @xmath316 as @xmath136 . the rate of convergence of @xmath317 to its limit in this case can be studied using the fourier transform @xmath318 of @xmath303 defined by @xmath319 we refer the reader to @xcite and references therein for proofs ; see also appendix b of @xcite for a discussion of the lattice case . in particular , stone proved the following theorem in @xcite . [ thm::stoneexpconvergencerate ] suppose that there exists @xmath320 such that @xmath321 is analytic and @xmath322 when @xmath323 . then , for every @xmath324 , @xmath325 as @xmath136 . since we aim to apply lemma [ lem::convergencerate ] , we will be particularly interested in exponential rates of convergence of @xmath326 to its limit . this rate of convergence is connected to that in theorem [ thm::stoneexpconvergencerate ] by the following lemma adapted from @xcite which we include for convenience . [ lem::convergencegtoz ] let @xmath306 , @xmath312 and @xmath303 satisfy the renewal equation . suppose that @xmath327 as @xmath136 . then , @xmath328 it suffices to notice that @xmath329 and use the definition of @xmath330 . discussing the conditions of the central limit theorem involves finding growth estimates for @xmath195 which can be verified using lemma [ lem::convergencerate ] . to be compatible with the framework of nerman s law of large numbers , we will focus on the situation where the characteristic vanishes for negative times ; extensions will be discussed where needed . we start by looking at condition [ cond::integrabilitycondclt ] . [ lem::checkintegrabilitycondition ] let @xmath111 be a @xmath69-general branching process and let @xmath195 be defined as in . assume that @xmath178 for @xmath179 and that there exists a constant @xmath331 such that @xmath332 for some @xmath333 . then , condition [ cond::integrabilitycondclt ] is satisfied . let @xmath334 . then , for @xmath335 , @xmath336 by theorem [ thm::nermanwlln ] , @xmath337 as @xmath338 . so the result follows since @xmath339 . coupled with lemma [ lem::convergencerate ] , this result hints that we should not expect to have a central limit theorem when @xmath340 does not decay at least as fast as @xmath341 , the threshold for the second term in the right - hand side of to converge to 0 . we will explain more precisely why this is sharp in remark [ rmk::sharpspeedconvergence ] . let us now discuss the moment condition for the central limit theorem . the next lemmas discuss a set of sufficient assumptions for condition [ cond::momentcondclt ] to be satisfied for the @xmath69-gbp . we introduce the function @xmath342 and note that @xmath343 and that @xmath344 is strictly decreasing in @xmath345 . [ lem::birthtimessumcontrol ] let @xmath111 be a @xmath69-general branching process . then , we have @xmath346 = \sum_{k = 0}^\infty \psi(y)^k,\ ] ] and the sum is finite if @xmath347 . by monotone convergence , @xmath346 = \sum_{k = 0}^\infty \be \left [ \sum_{|x| = k } e^{- y \gamma \sigma_x } \right].\ ] ] further , conditioning on the birth times , we get that @xmath348 = \be \left [ \sum_{|x| = k-1 } \sum_{i = 1}^n e^{- y \gamma \sigma_x } e^{- y \gamma ( \sigma_{x , i } -\sigma_x ) } \right]= \psi(y ) \be \left [ \sum_{|x| = k-1}e^{- y \gamma \sigma_x } \right].\ ] ] iterating this and summing over @xmath349 proves the equality with the infinite sum . noting that @xmath350 for @xmath347 completes the proof . relying on this , we can produce the following estimate on the third moment of the scaled process @xmath197 . [ lem::thirdmomentpositivetimes ] let @xmath111 be a @xmath69-general branching process . assume that @xmath178 for @xmath179 , that @xmath351 and that @xmath204 is bounded . then , @xmath352 notice that @xmath353 where @xmath354 therefore , it is clear that @xmath355 is bounded by @xmath356 to prove the lemma , it is sufficient to check that @xmath357 times the expectation of each of these terms is bounded , which we do now . to deal with , note that @xmath358 thanks to lemma [ lem::birthtimessumcontrol ] . notice that our assumption on the boundedness of @xmath204 implies that @xmath359 therefore , we can control the term corresponding to using that @xmath360^{1/2 } \right ] \\ & \leq c_1 ^ 2 c_2 \be \left [ \sum _ { x \in \sigma } e^{- \gamma \sigma_i } \sum_{i = 1}^n e^{-\gamma \sigma_{x , i } /2 } \right]\\ & \leq n c_1 ^ 2 c_2 \sum_{k = 1}^\infty \psi(3/2)^k\\ & < \infty , \end{aligned}\ ] ] using that @xmath361 and lemma [ lem::birthtimessumcontrol ] . to deal with , we proceed similarly to get that @xmath362 the similar argument needed to bound the term corresponding to relies on the observation that , if @xmath363 and @xmath349 are not all equal , then , assuming without loss of generality that @xmath89 is different , we have @xmath364)^{1/2 } ( \be [ \bar z_j(t- \sigma_j)^2| \cf _ \emptyset])^{1/2 } ( \be [ \bar z_k(t- \sigma_k)^2|\cf _ \emptyset])^{1/2}. \end{aligned}\ ] ] using this and reasoning as above then shows that @xmath365 the proof is complete . these results are easily combined to produce a version of the clt for the case with weights on the simplex . let @xmath111 be a @xmath69-general branching process such that @xmath178 for @xmath366 . assume that @xmath367 for some @xmath368 , that @xmath369 for some @xmath370 , and that @xmath204 is bounded with @xmath371 as @xmath372 . then the clt holds in that @xmath373 as @xmath136 , where the distribution of @xmath217 is normal with mean 0 and variance @xmath374 . it follows from lemmas [ lem::convergencerate ] , [ lem::checkintegrabilitycondition ] and [ lem::thirdmomentpositivetimes ] that conditions [ cond::integrabilitycondclt ] and [ cond::momentcondclt ] are satisfied . this , with our other assumptions , gives the required conditions for theorem [ thm::clt ] . in this section , we discuss the spectral asymptotics of a family of open subsets of @xmath71 $ ] whose boundary is a random self - similar cantor set generated using a distribution on the simplex , called @xmath69-random cantor strings . spectral asymptotics for a variety of cantor strings have been studied extensively in @xcite and references therein . we specialise the discussion to @xmath69-random cantor strings here so that we can study the fluctuations of the spectrum using the results of the previous section . choose @xmath52 , @xmath375 and consider the random vector @xmath376 with a probability law on the simplex . start with the unit interval @xmath377 $ ] and replace it by @xmath68 equally spaced intervals of length @xmath378 . replace each of these intervals by @xmath68 intervals created independently with the same procedure . iterating indefinitely , we obtain a decreasing sequence of compact sets @xmath379 , and @xmath380 is a statistically self - similar cantor set . indeed , in the notation of subsection [ subsec::appgbptofractals ] , it suffices to set @xmath381 the maps @xmath382 can easily be deduced from this . the corresponding general branching process @xmath383 ( no characteristic just yet ) is obtained as described in subsection [ subsec::appgbptofractals ] . by construction this is a @xmath69-general branching process and its malthusian parameter is @xmath30 . figure [ fig::cantorstring ] depicts the first 4 iterations in the construction of the set @xmath149 with the distribution @xmath384 . the set in which we are interested in this chapter is @xmath385\setminus k$ ] , whose boundary is @xmath149 by construction . thanks to the lindelf property , @xmath70 is a countable union of intervals . as such , @xmath70 is a _ random string _ in the sense of @xcite and references therein . consider the set @xmath70 defined above and recall that the hausdorff dimension of @xmath386 is @xmath30 . using theorem [ thm::nermanslln ] , we now show that the minkowski dimension of @xmath387 exists and is also equal to @xmath30 almost surely . the minkowski dimension of the @xmath69-random cantor string @xmath149 is almost surely equal to @xmath30 . consider the characteristic function defined by @xmath388 the corresponding counting process @xmath122 counts the number of offspring born after time @xmath127 to parents born up to time @xmath127 . as such , @xmath126 is an upper bound for the covering number @xmath389 of @xmath149 with balls of radius @xmath390 . by the strong law of large numbers , we have @xmath391 as @xmath136 . this is easily used to check that @xmath392 almost surely . the result follows since @xmath393 almost surely . recall that the eigenvalue counting function for a domain @xmath0 ( or a countable union of domains ) of @xmath80 is defined by @xmath394 following @xcite , we define @xmath395 the function @xmath396 has the property that if @xmath397 and @xmath398 are disjoint , then @xmath399 furthermore , for @xmath400 , a change of variables shows that @xmath401 in our applications of the central limit theorem , we will rely on the following assumption , which was discussed in the previous section . [ ass::convergencerate ] the rate of convergence in the renewal theorem satisfies @xmath402 for large @xmath127 , for some @xmath368 we may now use the general branching process to study @xmath403 . [ thm::spectralasympstring ] let @xmath149 be a @xmath69-random cantor string with dimension @xmath30 and consider the string @xmath385 \setminus k$ ] . then , @xmath404 as @xmath10 , for some strictly positive constant @xmath405 . furthermore , if assumption [ ass::convergencerate ] holds , then @xmath406 as @xmath10 , where @xmath40 has a normal distributions with mean 0 and variance @xmath407 for some strictly positive constant @xmath408 . define the random variable @xmath409 by @xmath410 by construction and the properties in and , we have @xmath411}(\lambda ) + \sum_{i = 1}^n \bar n_{r_i u_i}(\lambda ) = ( n-1 ) \bar n_{[0,1]}(s^2 \lambda ) + \sum_{i=1}^n \bar n_{u_i } ( r_i^2 \lambda),\ ] ] where @xmath412 are i.i.d . copies of @xmath70 . now recall that the eigenvalues of @xmath413 for the unit interval @xmath71 $ ] are @xmath414 . therefore , @xmath415}(\lambda ) = \pi^{-1 } \lambda^{1/2 } - \lfloor \pi^{-1 } \lambda^{1/2 } \rfloor,\ ] ] which is bounded by @xmath416 . to use the general branching process , set @xmath417}(s^{2 } e^{2 t } ) \quad \text{and } \quad z^\phi(t ) = \bar n_u(e^{2t})\ ] ] so that @xmath418 where @xmath419 are i.i.d . copies of @xmath122 and @xmath122 is the counting process of the characteristic @xmath121 . furthermore , notice that @xmath420 for some positive constants @xmath421 and @xmath422 . to establish the first statement of the theorem , we use and set @xmath423 which is bounded thanks to . thanks to theorem [ thm::nermanslln ] , this implies that @xmath424 as @xmath136 . by definition , this means that @xmath425 as @xmath10 , as required . let us now prove the second part of the theorem under assumption [ ass::convergencerate ] . consider @xmath426 and @xmath427 defined as in using @xmath121 and @xmath147 . notice that @xmath428 are equal for @xmath335 . in particular , the corresponding variance functions @xmath429 and @xmath430 are equal for @xmath335 . furthermore , applying the fact that @xmath431 is i.i.d . together with the bounds from lemma [ lem::convergencerate ] , it follows that there are positive constants @xmath432 such that @xmath433 these observations and the renewal theorem of @xcite imply that @xmath434 say . since @xmath147 is bounded and @xmath435 by assumption [ ass::convergencerate ] and lemma [ lem::convergencegtoz ] , the conditions of theorem [ thm::clt ] are satisfied . this implies that @xmath436 as @xmath136 . using the definition of @xmath148 , the decomposition in , ( [ zchi ] ) and slutsky s lemma completes the proof . [ rmk::sharpspeedconvergence ] the arguments in the proof can be used to show that there can not be a central limit theorem when the rate of convergence in the renewal theorem is not fast enough . suppose that @xmath437 as @xmath136 , for some real constant @xmath421 . then notice that the centring @xmath426 introduced in and used in the proof of theorem [ thm::spectralasympstring ] satisfies @xmath438\\ & = \phi_\emptyset(t ) + c_1 e^{(\gamma- \beta_1 ) t } \left ( \sum_{i = 1}^n e^{-(\gamma - \beta_1 ) \sigma_i } - e^{-\gamma\sigma_i } \right ) + o(e^{(\gamma- \beta_1 ) t})\\ & = \phi_\emptyset(t ) + c_1 e^{(\gamma- \beta_1 ) t } r + o(e^{(\gamma- \beta_1 ) t } ) , \end{aligned}\ ] ] say , as @xmath136 ( where the remainder is deterministic ) . notice that @xmath439 is a strictly positive random variable . in particular , for some @xmath440 , we have @xmath441 . therefore , there exists @xmath442 such that , for @xmath443 , @xmath444 this implies that , for @xmath445 , @xmath446 if @xmath447 $ ] , then , by renewal theory , we can not have a finite limiting variance in and in particular , no central limit theorem . we now consider the case where the distribution on the simplex is the dirichlet distribution and perform some explicit calculations to show the range of behaviour that is possible in our set up . in this set up the birth times are distributed as @xmath448 where @xmath449 denotes the dirichlet distribution . we will describe this situation by saying that the @xmath69-general branching process has dirichlet weights @xmath450 and write @xmath451 as usual . in general , it is difficult to determine the solutions to @xmath452 needed to use theorem [ thm::stoneexpconvergencerate ] . in some cases with dirichlet weights which we discuss now , however , we can use the properties of the gamma function to study @xmath318 and deduce convergence rates for the renewal theorem . let @xmath453 be a random vector in @xmath454 . it is well - known that , for every @xmath89 , @xmath455 where @xmath456 denotes the beta distribution . recall that if @xmath457 then @xmath458 where @xmath459 is the beta function , i.e.@xmath460 therefore , we get that @xmath461 = \frac{\gamma(\alpha_0)}{\gamma(\alpha_0 + \theta ) } \sum_{i=1}^n \frac{\gamma(\alpha_i + \theta)}{\gamma(\alpha_i)},\ ] ] where the equation defines @xmath462 . for the general branching process with dirichlet weights defined above , it follows that @xmath463 = \psi(1- w/\gamma).\ ] ] if @xmath464 for every @xmath89 , we can use that @xmath465 to reduce the function @xmath462 to a rational function which may be simpler to analyse . notice that this assumption implies in particular that there exist some @xmath466 and @xmath467 such that , for every @xmath89 , we have @xmath468 . therefore , @xmath469 from which it follows that @xmath470 . by definition of @xmath330 in and writing @xmath471 , we have @xmath472 denoting by @xmath185 the fourier transform of @xmath330 and @xmath318 that of @xmath471 , we thus get that @xmath473 applying and , for the general branching process with dirichlet weights @xmath474 satisfying @xmath464 for every @xmath89 , the function @xmath318 can be written @xmath475 where @xmath476 is a polynomial of degree @xmath477 . it follows that @xmath478 say . it is easy to see that @xmath479 now , decompose @xmath185 into partial fractions and write @xmath480 where @xmath481 are the roots of @xmath482 with corresponding multiplicities @xmath483 and @xmath484 are polynomials with @xmath485 for every @xmath89 ; in particular , @xmath486 recall that , for @xmath487 and @xmath488 , @xmath489 and therefore that @xmath490 using this , it is easy to check that @xmath491 where the @xmath492 are polynomials determined using and satisfying @xmath493 . of course , since @xmath303 is supported on @xmath187 , so is @xmath494 and therefore , by definition of @xmath330 , we have @xmath495 putting this together shows that @xmath496 which we can integrate to study the asymptotics of @xmath330 . a particular example which will guide us below is given in the following lemma . [ lem : simpleroots ] assume that all the roots of @xmath482 are simple . then , @xmath497 where @xmath498 , the residue of @xmath185 at @xmath499 . since in this case none of the singularities can be removed and all have order one , all the @xmath72 are non zero . furthermore , since if @xmath499 is a root with residue @xmath72 then @xmath500 is a root with residue @xmath501 as @xmath502 , roots with the same real part can not cancel out . in particular , in this case , the result of theorem [ thm::stoneexpconvergencerate ] is sharp . since all the roots of @xmath482 are simple , we have @xmath503 where we must have @xmath498 . integrating then shows that @xmath504 since @xmath303 has a second moment , the result now follows from after letting @xmath136 . let us first discuss how the observations above enable us to establish the desired rate of convergence for some simple cases of dirichlet weights . + _ example 1 . _ [ lem::caseswhereworks ] consider the general branching processes with dirichlet weights @xmath474 described above . assume that @xmath505 then the fourier transform @xmath321 of @xmath118 is analytic and @xmath322 when @xmath506 $ ] . in particular , @xmath507 as @xmath136 . letting @xmath508 a direct calculation gives @xmath509 there is always a solution to @xmath510 at @xmath511 and all we require is that the other solutions are less than 0 to establish , via , that @xmath321 is analytic and @xmath322 on @xmath512 . for @xmath513 , the only solution to @xmath514 is @xmath515 . for @xmath516 , the other solution to @xmath514 is given by @xmath517 . for @xmath518 , the other solutions to @xmath514 are @xmath519 and @xmath520 has solutions to @xmath514 at @xmath521 thus the real parts of all the solutions are less than zero and we have the required analyticity . the rest of the statement follows from theorem [ thm::stoneexpconvergencerate ] . analytic solutions for the solutions to the equation @xmath510 do not appear to be available for larger values of @xmath349 . + _ example 2 . _ + here we discuss the general branching process derived from the class of examples mentioned in the introduction in which the dirichlet weights are of the form @xmath522 with @xmath523 . we will establish the theorem from the introduction . thanks to lemma [ lem::caseswhereworks ] , we know that if @xmath524 , as @xmath525 , then the fourier transform @xmath318 of @xmath526 defined in can be used to show that the rate of convergence in the renewal theorem is sufficiently fast for the requirements of theorem [ thm::clt ] . in other words , the applicability of theorem [ thm::clt ] depends on the regularity of the characteristic @xmath121 . more generally we need to solve the equation @xmath527 as @xmath528 this is a polynomial equation and hence we seek roots of @xmath529 by letting @xmath530 the rate of convergence in the renewal theorem is given by the root of @xmath531 with smallest strictly positive real part . we have computed these values numerically . the numerical evidence shows that when @xmath48 increases , some roots of @xmath532 get close to the imaginary axis . this phenomenon is illustrated in figure [ fig::somephaseplots1 ] which contains phase plots of @xmath531 for different values of @xmath48 ; we rescaled for convenience . to highlight this more clearly , figure [ fig::somephaseplots2 ] contains some close - ups of phase plots showing the absence or presence of such roots of @xmath532 for different values of @xmath48 . in particular , when @xmath533 , the two non - zero roots of @xmath531 closest to the imaginary axis are @xmath534 when @xmath535 , they are @xmath536 and when @xmath537 , they are @xmath538 we have in fact computed the real part of the relevant root for all values of @xmath48 from 1 to 80 these are plotted in figure [ fig::somephaseplots3 ] . numerically , this establishes that @xmath539 is the smallest integer value for which @xmath531 has roots with real part @xmath540 . our computations also show that for @xmath541 the roots of @xmath531 are all simple and occur as complex conjugate pairs except for the root at 0 . to summarise , this numerical evidence shows that the general branching process with dirichlet weights @xmath542 admits a central limit theorem of the type described when @xmath37 , but not when @xmath543 . moreover , the monotonicity of the plot in figure [ fig::somephaseplots3 ] suggests that the range for which there is not a central limit theorem extends to all @xmath544 . we note that we see similar results in the asymmetric case with dirichlet weights @xmath545 , @xmath546 with @xmath547 . in this case the polynomial equation becomes @xmath548 here is a table showing for a given @xmath549 the values of @xmath550 below which we are in the central limit theorem regime . [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] we would like to thank mohsin javed for help with the numerics and charles stone for providing useful references . the first author was supported by a berrow foundation scholarship and the swiss national science foundation . d. aldous . the continuum random tree . ii . an overview . in _ stochastic analysis ( durham , 1990 ) _ , volume 167 of _ london math . lecture note ser . _ , pages 2370 . cambridge univ . press , cambridge , 1991 . distribution of modes in fractal resonators . in _ structural stability in physics ( proc . symposia appl . catastrophe theory and topological concepts in phys . tbingen , tbingen , 1978 ) _ , volume 4 of _ springer ser . synergetics _ , pages 5153 . springer , berlin , 1979 . berry . some geometric aspects of wave motion : wavefront dislocations , diffraction catastrophes , diffractals . in _ geometry of the laplace operator ( proc . sympos . pure math . hawaii , honolulu , hawaii , 1979 ) _ , proc . pure math . , xxxvi , pages 1328 . soc . , providence , r.i . , 1980 . m.l . fractal drum , inverse spectral problems for elliptic operators and a partial resolution of the weyl - berry conjecture . 325:465529 , 1991 . lapidus and m. van frankenhuysen . . birkhuser boston inc . , boston , ma , 2000 . complex dimensions of fractal strings and zeros of zeta functions .

known bright s stars , recognized as such by their enhanced s - process abundances and c / o ratio , are typically members of the asymptotic giant branch ( agb ) or the red giant branch ( rgb ) . few modern digital spectra for these objects have been published , from which intermediate resolution spectral indices and classifications could be derived . for published s stars we find accurate positions using the two - micron all sky survey ( 2mass ) , and use the fast spectrograph of the tillinghast reflector on mt . hopkins to obtain the spectra of 57 objects . we make available a digital s star spectral atlas consisting of 14 spectra of s stars with diverse spectral features . we define and derive basic spectral indices that can help distinguish s stars from late - type ( m ) giants and carbon stars . we convolve all our spectra with the sloan digital sky survey ( sdss ) bandpasses , and employ the resulting magnitudes together with 2mass mags to investigate s star colors . these objects have colors similar to carbon and m stars , and are therefore difficult to distinguish by color alone . using near and mid - infrared colors from iras and akari , we identify some of the stars as intrinsic ( agb ) or extrinsic ( with abundances enhanced by past mass - transfer ) . we also use band and 2mass magnitudes to calculate a temperature index for stars in the sample . we analyze the proper motions and parallaxes of our sample stars to determine upper and lower limit absolute magnitudes and distances , and confirm that most are probably giants .

1107.4029

when a low to intermediate mass star exhausts its core supply of hydrogen , contraction and heating of the core causes the outer layers of the star to expand and cool , whereby the star becomes a red giant ( rg ) with an average effective temperature of 5000k or lower . once the core temperature rises to about 3@xmath3k , helium burning begins , and the stellar surface contracts and heats creating the horizontal branch ( for population ii ) or red clump ( for population i ) stars on the color - magnitude diagram . completion of core helium burning and the start of helium shell burning returns the star to a luminous , cooler phase as it evolves onto the asymptotic giant branch ( early- or e - agb ) where the star greatly expands and the hydrogen shell is ( almost ) extinguished . later , as the he - burning shell approaches the h - he discontinuity , a phase of double shell burning begins . recurrent thermal instabilities in this thermally - pulsing asymptotic giant branch ( tp - agb ) phase are driven by helium shell flashes , periodic thermonuclear runaway events in the he - shell the energy released by these pulses expand the star , whereby the hydrogen shell is again basically extinguished during some time . a third dredge up might then occur , and afterward the hydrogen shell will re - ignite ( and a new thermal pulse will occur ) . during these short - lived pulses , nucleosynthesis products from combined h - shell and he - shell burning are dredged up to the outer layers of the star . the complex details of the dredge - up and its outcome are strongly mass - dependent ( and references therein ) , but generally speaking , surface abundance enhancements result in c , he , and the @xmath4-process elements , including ba , la , zr , and y. s and carbon ( c ) stars are traditionally thought to be members of the tp - agb . in addition to possibly elevated c / o ratios , these stars exhibit other unique spectral characteristics as a result of the dredge - up . while m giants often exhibit titanium oxide ( tio ) absorption bands , both carbon and zirconium have a higher affinity to free oxygen . therefore , a higher c / o ratio also results in the disappearance of tio bands and the appearance of zirconium oxide ( zro ) bands . s stars are therefore distinguished primarily by their zro bands , while c stars show strong c@xmath5 and cn bands . the s star spectral class is divided into three subtypes ( in order of increasing carbon abundance ) : ms , s , and sc stars . classification into one particular class is difficult , but is generally based on the c / o ratio and the relative strengths of tio , zro , and cn molecular absorption bands . overall , a star in the s star class is generally defined as having a c / o ratio ( r ) of @xmath6 and strong zro absorption features . pure s stars are those that show only zro bands , and no tio bands @xcite . classification is complicated by the fact that increased intensity of the zro bands may be due to an excess of zr , rather than being exclusively tied to the c / o ratio @xcite . furthermore , many agb s and c giants are highly variable , and since the strength of molecular bands depends on the effective temperature and surface gravity of the star , spectral type can vary significantly with time . s stars whose spectra show lines of the short - lived element @xmath4-process technetium ( tc ) are known as `` intrinsic '' s stars , and are most likely be members of the tp - agb . @xmath7tc is only produced by the @xmath4-process , while the star is on the agb . since the half - life of @xmath7tc is only 2.13 x 10@xmath8 years , and the average duration of the agb phase of stellar evolution is roughly 1 myr , stars exhibiting significant abundances of technetium are almost certainly members of the agb . by contrast , technetium - poor s stars are known as `` extrinsic '' s stars . these stars are likely to be part of binary systems and their unusual chemical abundances probably originated in a past mass - transfer episode . the present extrinsic s star likely once accreted @xmath4-process rich material from its companion , a tp - agb star at the time that has since evolved into a white dwarf . extrinsic s stars therefore display enriched @xmath4-process elements , despite having never produced these elements themselves . most known extrinsic s stars are assumed to be members of the red giant branch ( rgb ) . they can sometimes be distinguished from intrinsic s stars on the basis of color , because they do not necessarily show the red excess characteristic of stars on the agb . while all currently known extrinsic s stars are thought to be giants , these definitions open up the possibility of extrinsic s stars on the main sequence ( hereafter referred to as dwarf s stars ) that have previously accreted @xmath4-process material and carbon from a companion but whose tc has since decayed . analogous to the dwarf c stars @xcite , a faint s star could be shown to be a dwarf if it has a sufficiently large parallax or proper motion ; a large proper motion predicts a tangential velocity greater than the galactic escape speed unless the object is faint and nearby . s stars are very similar to c stars , with the major difference being a slightly lower c / o ratio . faint high galactic latitude carbon ( fhlc ) stars were once assumed be giant stars at large distances . in 1977 , however , g77 - 61 was discovered to have high proper motion and an upper limit absolute magnitude of + 9.6 , therefore making it the first known carbon dwarf star @xcite . in this case , the term dwarf refers to a star on the main sequence . @xcite discovered four other dwarf carbon ( dc ) stars using techniques that selected known carbon stars for high proper motion . now , well over 100 dwarf carbon stars are known @xcite . the local space density of dcs is far higher than all types of c giants combined @xcite . therefore , contrary to previous assumptions , dwarf c stars are the numerically dominant type of carbon star in the galaxy . since dc stars are now known to be so common , we have begun a search for dwarf s stars . there is little reason to believe that dwarf s stars should not exist . in fact , they may be more abundant than dc stars , since less carbon enhancement is required to produce an s dwarf as compared to a c dwarf . on the other hand , if the range of abundance ratios that produces the characteristic features of s stars is narrow , they may be quite rare . furthermore , the range of c / o ratios that produces s star spectral features may be different for giants and dwarfs , given the higher gravities and larger temperature ranges in dwarfs . we use medium - resolution digital spectra of known s stars ( which are probably giants ) to generate corresponding sloan digital sky survey ( sdss ) colors , to see whether their colors might be sufficiently distinctive to search for additional s stars in the sdss with reasonable efficiency . we also use the fast spectra to investigate molecular band spectral indices for classification . our paper is organized as follows : in [ sec : obs ] we briefly review the observations and our data reduction and processing procedures . in section [ sec : spatlas ] we describe the production of a spectral atlas of s stars , and derive spectral indices from the molecular band strengths in [ sec : specind ] . in [ sec : colors ] we describe the generation of sdss colors and analyze the overall colors of the fast sample in order to determine whether an efficient color selection for s stars exist . in [ sec : intext ] we discuss techniques for distinguishing intrinsic and extrinsic s stars and apply them to the fast sample . in [ sec : motion ] we analyze the parallaxes and proper motions of s stars in the sample to characterize upper and lower limit magnitudes and distances . we use colors to generate approximate temperature indices for s stars in [ sec : temp ] . in [ sec : sum ] we summarize our results and present our conclusions . the largest spectral catalog of s stars is stephenson s second edition ( 1984 ) , which includes 1347 objects with poorly known positions . the stars in this catalog were originally identified from blue , red , and infrared objective prism plates . the majority of the s stars were discovered on the basis of the red system of zro absorption bands , with a bandhead around 6474 . since the stephenson catalog has a limiting v magnitude of approximately 11.5 , most of the objects are highly saturated in the sdss and reliable u , g , r , i , and z magnitudes have not been found for any known s stars . we assembled a sample of known s giants in order to generate likely g and r colors for s giants and dwarfs . these constraints on colors of s giants and dwarfs could allow us to search the sdss and 2mass catalogs for potential dwarf s star candidates . to choose the stars making up the sample , we selected high galactic latitude ( @xmath9 deg ) , northern ( @xmath10 deg ) s stars from stephenson s catalog . this ensures that our sample is accessible from mt . hopkins , and also decreases problems of object confusion or reddening . we then correlated the sample from stephenson s catalog with 2mass to find improved positions for most objects ; accuracy of the original positions is typically @xmath11 , but commonly as poor as 20 . the final sample list , which includes 57 objects , can be found in table[tab : stars ] . each object is accompanied by its identification in both the stephenson and 2mass catalogs , as well as its position . since many of these stars are known to be variable , we include the date of observation . we include other common identifiers for each target and a published spectral type , if available . we include our calculated temperature index for the star , which is discussed further in [ sec : temp ] . spectral types and temperature indices of the s stars in the sample may vary as a function of time , because the stars themselves do . when spectral types for different epochs were available , we match the @xmath2 magnitudes of the epochs to those listed in table[tab : mags ] to find the spectral type . finally , we include identifications of some stars as intrinsic or extrinsic ( denoted respectively by i and e ) from @xcite or from our own calculations based on akari magnitudes . for more information regarding the intrinsic / extrinsic identifications , see [ sec : intext ] . spectra were obtained using the fast instrument on the 1.5 m tillinghast reflector on mt . hopkins . fast produces medium - resolution optical wavelength spectra , spanning a wavelength range from 3474 to 7418 . all data were taken using the 3 " slit and a 300linesmm@xmath12 grating , producing a resolution of @xmath131,000 at 6,000 . exposure times varied widely based on the magnitude of the star , but ranged from @xmath143 to 300 s. the two - dimesional spectra were bias - subtracted and flat - fielded by the observatory standard data reduction procedure . after extraction , and the one - dimensional spectra were wavelength calibrated . we used standard star spectra taken each night of the observations to produce sensitivity curves and flux calibrated spectra . an extinction correction file from kitt peak national observatory was also applied to correct for major atmospheric effects . the standard calibration stars used to create the sensitivity curve are bd+26 2606 , bd+33 2642 , bd+17 4708 , feige 34 , and hd 19445 . flux calibration and extinction corrections were performed using the _ onedspec _ package and the _ calibrate _ task in iraf . we used a combination of observatory logs and night - to - night comparison of the sensitivity function to determine whether the calibrated target spectra were reliable enough to use in color analysis . we rejected some of the initial data sample based on concerns about non - photometric conditions and atmospheric distortion . other nights are included in the final data for analysis because while the data may not be entirely photometric , it remains reliable enough for color analysis because of the non - wavelength dependent nature of most remaining distortions . we discuss the errors in our derived sdss colors further in [ sec : colors1 ] . many of the targets are luminous enough in red wavelengths that longer exposure times cause parts of the spectrum to saturate . the wavelengths from 6000 onward are of particular concern . however , shorter exposure times limited the amount of information we could derive from wavelengths below 4500 because of low signal - to - noise ratios . to create an s star spectral atlas , when possible we interpolated between two different exposure times : a shorter exposure time that produced accurate , unsaturated spectra at the longer wavelengths , and a longer exposure that resulted in higher signal - to - noise at shorter wavelengths . generally , we determined by hand which spectra were reliable in a particular wavelength region by comparing multiple exposure times and examining long exposures for nonlinear behavior in areas from 6000 onward . for stars with only one available exposure time , we rejected a star as saturated if the counts exceeded 60,000 . after interpolating between spectra to eliminate saturated regions and rejecting some data , we were left with 46 s star spectra . we use s star line identifications @xcite to identify major spectral features and select objects with unique characteristics to include in the final spectral atlas , which includes 14 objects marked in table[tab : stars ] . the objects were chosen to be a representative sample of the larger fast sample , which includes many fundamentally similar objects . two sample spectra with associated molecular line identifications are plotted in figure[fig : speclines ] . while the spectra display the same molecular bands , most notably those associated with zro , they have different overall colors , which may be due to differences in abundances and/or effective temperatures . we also note that these spectra show some tio bands , indicating that they are not pure s stars , but instead probably belong to the ms classification . the appearance of tio bands indicates the possibility of a c / o ratio slightly less than 0.95 . the creation of a spectral atlas allows for greater investigation into the similarities and differences within the s star class . for instance , some members of the spectral atlas exhibit , in addition to prominent zro features , h@xmath15 , h@xmath16 , and h@xmath17 emission lines . the presence of these emission lines indicates that these objects are almost certainly agb stars , e.g. , miras . the spectral atlas is made available digitally as 14 individual fits files via this journal . we can immediately use the first digital spectral atlas of s stars to derive some medium - resolution spectral indices , potentially useful to classify stars as m , s , or c even where they may have very similar broadband colors . apart from our own fast spectra of 48 s stars , we use m giants from the indo - us spectral atlas @xcite . we use eight carbon star spectra also obtained with the fast for a separate project ( green et al . 2011 , in preparation ) . these latter spectra span a wide range of spectral band strengths , but are selected from the sdss color wedge defined by @xcite . we defined three indices , one each for tio , zro , and c@xmath5 . we follow the basic premise of using ratios of mean flux per ngstrm across key molecular bands , so that spectra of differing resolution should yield similar results . we use a neighboring comparison region for each band , so that sensitivity to broadband flux calibration is minimized . for the zro index , we divide the mean flux across 6400 6460 ( zro off - band ) by that between 6475 6535 ( zro band ) . for the tio index , we use mean @xmath12 flux from 7065 7175 ( tio band ) divided by 6965 7028 ( tio off - band ) . the c index is derived from 5025 5175 ( c@xmath5 band ) and 5238 5390 ( off - band ) . thus , lower flux in the denominator caused by strong molecular bands yields a larger spectral index value . we note that the c index near 5100 can also be substantially contaminated by tio , so we label it `` ct51 '' hereafter . at least within this representative sample of stars , a combination of these indices yields a preliminary s star classification , as shown in figure [ fig : specind ] . the objects most reliably classified as s stars would have zro index @xmath18 and c@xmath5 index @xmath19 . to use the sdss database to find likely candidates for both s giants and dwarfs , we need reliable colors for our objects in as many sdss bandpasses as possible . the spectra from fast cover the bandpasses for the @xmath20 and @xmath21 filters completely , and we are thus able to convolve our spectra with these bandpasses and generate sdss colors . each of the fast spectra contains 2681 data points , with approximate wavelength separation of 1.47 . to convolve the spectra with the filter transmission curves , we use linear interpolation between transmission values to match a transmission constant to each data point in the spectrum once we find a transmission constant for each wavelength , we normalize the transmission curve and then convolve the spectrum with the transmission curve . we use the iraf task _ calcphot _ in the _ synphot _ package of _ stsdas _ to convolve the spectra with the bandpasses and obtain sdss colors from our spectra . the sdss bandpass curves are taken from information made available by the sdss consortium @xcite . as with all sdss magnitudes , the resultant magnitudes are in the ab magnitude system . by including both sdss and 2mass bandpasses , we are able to calculate 10 unique colors . we also conduct error analysis on the derived sdss colors using the standard star spectra and the known g - r colors and r magnitudes for bd+26 2606 , bd+33 2642 , bd+28 4211 , and bd+174708 . these colors are published by the sdss consortium , and allow us to constrain our errors in both the magnitudes and colors @xcite . we calculate the difference between measured and standard @xmath20 and @xmath21 magnitudes , and find a standard deviation of @xmath22 mag for the @xmath20 band and @xmath23 mag for the @xmath21 band . while these seem high and would suggest , via propagation of errors , a correspondingly high error in our colors ( on the order of 0.4 mag ) , we find a standard deviation of only @xmath24 mag for the color . the value was calculated by comparing the derived @xmath25 colors for the standard stars to the known @xmath25 colors from the sdss data release 5 standard star network @xcite . the low value in the uncertainty of our colors is probably due to the fact that while our spectra may not be absolutely photometrically calibrated , many of the remaining distortions are not wavelength dependent and therefore cancel out during color calculation . examples of remaining distortions include slit losses , an incomplete cancellation of the telluric spectrum or a thin cloud layer that dims the overall flux at all wavelengths . so while individual magnitudes remain unreliable at best , the errors in the derived color are comparable to color errors in 2mass and are thus reliable enough for our analysis . the 2mass magnitude errors for our sample have a wide range dependent upon the reliability of the photometry for a particular night , but generally range from 0.02 magnitudes up to about 0.2 magnitudes . colors calculated from 2mass magnitudes thus have errors comparable to the derived sdss colors . the initial aim of the project was to select for s stars based on their sdss and 2mass colors . to achieve this goal , we derive @xmath20 and @xmath21 magnitudes for the stars making up our sample and combine with the known 2mass magnitudes to construct a total of 10 unique colors . the @xmath20 , @xmath21 , and 2mass magnitudes , along with @xmath2 band , iras , and akari photometry where available , are presented in table[tab : mags ] . many of these colors , however , are ultimately unhelpful for analysis purposes because the colors of known s stars cover several magnitudes . such a wide spread in magnitudes does not allow for efficient discovery of the s stars based on color . in other color combinations , the s stars are not well distinguished from the stellar locus , so selection based on color is contaminated by large numbers of stars from the main sequence . we find that the most useful color combinations are @xmath25 , @xmath26 , and @xmath27 because in these colors , the fast s stars sample is well distinguished from the stellar locus . we compare these colors with those found for general sdss stars , by overplotting some 300,000 sdss/2mass matches of @xcite , which they used to parametrize the stellar locus in terms of color . in figure[fig:2mass_sdsscolors ] , we present two color - color diagrams in which the s giants from the fast sample are best separated from the stellar locus . the first shows the comparison of @xmath25 color and @xmath26 color , while the second shows the comparison @xmath28 and @xmath29 color . however , the s giants that comprise the fast sample , by virtue of their location on the tp - agb or rgb , have similar colors and effective temperatures to m giants . in figure[fig:2mass_sdsscolors ] , we also examine colors for m giants and dwarfs matched in the sdss and 2mass surveys . while the m giants are very few in number ( due to the depth of both of these surveys , many nearby m giant targets are saturated ) , we can immediately see that s stars and m stars lie in essentially the same area of color space . we also include c stars on the color - color diagrams , since they follow evolutionary paths similar to s stars and also have enhanced abundances . the s stars of the fast sample , while well - distinguished from the overall stellar locus on both color - color diagrams , are particularly difficult to distinguish from c stars on the @xmath25 versus @xmath26 diagram . on the plot comparing @xmath28 to @xmath29 , s stars can be distinguished reasonably well from the bulk of the m star population using @xmath30 and @xmath31 . given the biases in the current sample of s stars , it is difficult to determine the efficiency of this selection , but this region on the color - color plot appears to be the least contaminated by m and c stars . spectroscopy for a sample of perhaps 100 stars in this color region could directly test the efficiency of this color selection . distinguishing intrinsic and extrinsic s stars is difficult without high - resolution spectroscopy , where direct detection of tc is possible . however , as presented in van eck and jorrissen ( 2000 ) , there are other measurements which serve to segregate the two types with reasonable statistical accuracy . since extrinsic s stars are generally on the rgb ( or early agb ) , they cluster closer to the main sequence than intrinsic s stars in terms of color . stars on the agb generally show a red excess which is absent on the rgb . however , van eck and jorissen ( 2000 ) find that blue excursions of mira variables , which make up a significant proportion of intrinsic s stars , make the two classes difficult to distinguish based on distance from the main sequence alone . additionally , intrinsic and extrinsic s stars have been shown to segregate well in a ( @xmath32 $ ] , @xmath33 $ ] ) diagram . these colors are based on iras photometry , which is available for many of the stars in the fast sample . iras magnitudes in these bands are presented in table[tab : mags ] for 37 stars ( with two bands available for 29 ) . identifications of many of the s stars in the fast sample as intrinsic or extrinsic are included in table[tab : stars ] . all such identifications not marked with an asterisk are from @xcite . classification of most of the s stars in our sample is inconclusive because of their intermediate positions in the color - color diagrams . however , this ir technique remains the most efficient photometric classification method known to determine whether s stars are intrinsic or extrinsic . we also use results from vanture et al . ( private communication ) , which indicate that s stars presenting li in their spectra are intrinsic s stars . lithium in stars is destroyed at fairly low temperatures @xcite but can be produced by the cameron - fowler mechanism , in high luminosity ( @xmath34 ) tp - agb stars via hot bottom burning . since , as mentioned above , extrinsic s stars are often less evolved along the giant branch than intrinsic s stars , they generally lack any evidence for the resonance line li 6707 ( vanture et al . 2011 , in preparation ) . we compare the list of s stars with and without li detections to our fast sample and analyze them for the li 6707 feature . unfortunately , we find that this line can not be reliably detected with spectra of our resolution and s / n . the akari satellite surveyed much of the sky in both near and far infrared bands . we use flux data collected by akari in the s9w and l18w bandpasses to find infrared magnitudes for six objects in the fast sample without available iras colors . the s9w and l18w bandpasses are qualitatively similar to the iras 12 and 25 micron bandpasses . all available akari magnitudes are presented in table[tab : mags ] , for 52 stars ( with two bands available for 35 ) . since the zero magnitude fluxes for these bandpasses are not well established , all magnitudes are presented on the ab system . in figure[fig : nearmidircolors ] , we show two color - color diagrams comparing intrinsic and extrinsic stars using @xmath35 $ ] , @xmath36 $ ] , and @xmath37-[18]$ ] colors . s stars previously classified as intrinsic vs. extrinsic are shown in black , with others plotted in red . from the akari magnitudes and plots , we are able to identify an additional two stars in the sample as intrinsic or extrinsic . we identify 22315839 + 0201206 as an extrinsic s star . this star has a @xmath36 $ ] value of @xmath38 mag and a @xmath39-[9]$ ] value of @xmath40 mag . this places it in the extreme lower left corner of the color - color diagram comparing @xmath35 $ ] color to @xmath36 $ ] color , far away from any known intrinsic stars but close to many known extrinsic stars . similarly , we are able to identify 19364937 + 5011597 as an intrinsic s star on the basis of color . this star has a @xmath35 $ ] value of only @xmath41 mag , and a @xmath36 $ ] value of @xmath42 mag . this places it in the upper right - hand corner of the plot comparing @xmath35 $ ] and and @xmath36 $ ] color , far away from the cluster of extrinsic s stars . the four other stars for which we have no classification ( 03505704 + 0654325 , 10505517 + 0429583 , 13211873 + 4359136 , and 16370314 + 0722207 ) cluster close to the transition between intrinsic and extrinsic s stars in all three color comparisons , meaning that we can not confidently identify these with either class of stars . we include our additional identifications in table[tab : stars ] , marked with an asterisk to distinguish them from those from @xcite . as mentioned before , s stars cover a wide range of spectral types and effective temperatures . however , they generally have effective temperatures similar to those of m giants . the starting point for classifying the spectral type and temperatures of our s stars , therefore , is to calculate a temperature index using m - giant criteria @xcite . @xcite use a grid of stellar models to calibrate a relation between color and temperature index for m stars . their analysis uses cit / ctio colors @xmath43 and @xmath44 . we find @xmath2 magnitudes for some of the fast sample using simbad . we also convert the 2mass k@xmath45 magnitudes and @xmath29 colors using the following relations from @xcite : @xmath46 we then use the temperature index / color relation to assign a preliminary temperature index to the stars in the fast sample . where possible , we use the @xmath47 relation , since this is specified by the authors as the most temperature sensitive . when there is no @xmath2 magnitude available , we use the @xmath48 relation @xcite . results of this analysis are presented in table[tab : stars ] . these temperature indices should be treated with caution , since many of the stars in the fast sample are highly variable . furthermore , differences in c / o ratios , may lead to temperature errors up to 400k @xcite . however , the preliminary classification gives us a rough idea of the relative effective temperatures of the s stars . we find that the average temperature index of the stars in the sample is 5 , which in m giants corresponds to an effective temperature of roughly 3500 k @xcite . in table[tab : stars ] , we also present known spectral types from @xcite , which include temperature indices for the s stars . we find reasonably good agreement between our calculated temperature indices and those presented as part of the spectral type . differences between our temperature index and those included as part of the published spectral types are within the range of variability of the star . we also note that since we do not have a reliable color / temperature relation for stars with a temperature index greater than 7 , we present all of these classes as 7 + . we also find reasonably good agreement ( within one temperature index ) between the values using @xmath47 and those derived using @xmath48 . we also conduct a preliminary analysis on possible correlations between the intrinsic / extrinsic distinction and the temperature indices . we find that the average temperature index of the intrinsic stars is @xmath49 , while the average temperature index of extrinsic stars is @xmath50 . therefore , while the mean temperature index shows some differences , this measurement alone is not statistically significant . a larger sample would be needed to determine the validity of differences between average effective temperature between intrinsic and extrinsic s stars . the temperature indices of s stars in general are insufficient to classify them as intrinsic or extrinsic , except for the reddest objects . four of the stars in the original fast sample have reliable parallax measurements ( defined as a parallax detection at the @xmath51 level ) available from the hipparcos survey , so calculation of the absolute magnitude of these stars in the @xmath20 and @xmath21 bands is possible . we calculate the error in our absolute magnitudes using the presented error in the parallax measurements and assuming apparent magnitude error values of @xmath22 and @xmath23 . results of this analysis are given in table[tab : plx ] . we note the calculated absolute @xmath20 and @xmath21 magnitudes , as well as the associated errors . objects are identified using the 2mass identifier , and the parallax and associated error as given by the hipparcos catalog are also included . we can immediately identify each of these four objects as asympotic or red giant branch members : if they were dwarf stars of the same spectral type , we would expect magnitudes in the @xmath20 and @xmath21 of + 5 or higher , as opposed to the derived values , which center around 0 . we also find that some stars in the original fast sample display significant ( again , @xmath51 ) proper motions . the proper motions are available in the tycho-2 catalogue . of these , we have reliable @xmath20 and @xmath21 photometry for 16 objects . given the proper motion and assuming that all objects are moving at total space velocities less than the galactic escape velocity , we can derive an upper limit on the distance , and thus also constrain the absolute magnitudes of the objects . to derive an approximate distance ( @xmath52 ) , we use the equation : @xmath53 where @xmath54 is the proper motion of the object measured in radians per year , @xmath55 is the transverse velocity of the object , and @xmath56 is the galactic escape velocity . we take the galactic escape velocity to be 545 km / s @xcite , which is equivalent to 0.0056 parsecs per year . the distance constraint is only an upper limit because we neglect radial velocities . however , we note that our rvsao measurements generally yield small radial velocities ( @xmath5750km / s ) for the s star sample . we then use the same formulas presented above to derive @xmath20 and @xmath21 absolute magnitudes and associated errors . these @xmath20 and @xmath21 absolute magnitudes essentially represent the brightest the object concerned could be , considering its proper motion and assuming that all observed objects are gravitationally bound members of the milky way . results of the analysis are presented in table[tab : pm ] . we can immediately see from these lower bound magnitudes that these objects could be rgb or agb stars - even considering that the actual magnitudes are probably significantly fainter . for the sample of 16 stars , we find that the average lower limit on the absolute @xmath20 magnitude @xmath58 is @xmath59 mag , correspondingly @xmath60 mag . cool dwarf stars generally have @xmath20 and @xmath21 absolute magnitudes greater than + 8 mag we conclude that all 16 objects displaying significant proper motions could be giants . to further bear out this analysis , we also determine a lower limit on the distance for these objects by assuming ( conversely to above ) that they _ do _ have a dwarf magnitude . we take the approximate absolute @xmath20-band magnitude for an m dwarf star to be @xmath58 = 10 mag and the absolute @xmath21-band magnitude to be @xmath61 = 9 mag . this method can be applied to the entire sample , because it does not rely on proper motion or parallax measurements . we find that most of these objects , were they to have typical dwarf magnitudes , would be within 25 pc of us . this distance would almost guarantee that these objects would have detectable parallax and proper motion . for instance , assuming extrinsic s stars are a mostly spheroid population ( like dc or ch stars : see @xcite and @xcite ) with velocity of @xmath62km / s relative to the sun @xcite , the typical proper motion for a star 25 pc away would be on the order of 2.25 mas / yr . additionally , such stars would have a parallax on the order of 40 mas . since such a parallax would be easily detectable by hipparcos , these lower limit distances therefore also suggest that the stars of the fast sample are likely to be giants or agb stars . using a sample of 57 medium - resolution s star spectra taken with the fast spectrograph , we create a spectral atlas of s stars comprised of 14 objects that span a range of spectral types within the ms , s , and sc classes . the atlas is published as a collection of 1-d fits files via this journal . after generating @xmath20 and @xmath21 sdss magnitudes from the spectra , we find that the sdss magnitudes in the @xmath20 and @xmath21 bands , combined with @xmath63 , @xmath64 , and @xmath65 magnitudes from the 2mass catalog for s stars may with further confirmation allow for reasonably efficient color selection of these objects from the sdss and 2mass catalogs . we use previously published data to identify some of the stars in the sample as intrinsic or extrinsic stars , and find that the fraction of extrinsic s stars in the sample is approximately @xmath66 . we also assign temperature indices to the stars in the sample based on the m star scale of temperature indices . we find that much of the s star sample falls at temperature index 4 or above , meaning that the effective temperatures of most of these stars are well below 5000 k. we also analyze objects in the fast sample with detectable parallaxes and proper motions to generate absolute magnitude limits , as well as lower limit distances based on assumed dwarf magnitudes . this analysis bears out our initial assumption that the fast sample is primarily composed of giant stars , either on the agb or rgb . many thanks to the referee for a thorough reading . we are grateful to kevin covey for the use of the high - quality sample of sdss/2mass matches . also , thanks to warren brown and mukremin kilic , who provided many helpful discussions and hints on color selection and the sdss photometry . many thanks to doug mink for all his invaluable help with _ rvsao _ and sdss correlations , and to bill wyatt for his help with accessing sdss dr-7 spectra , all of which we hope to use in upcoming publications . we gratefully acknowledge bob kurucz , andrew vanture and george wallerstein for illuminating discussions . thanks also to verne smith and gw for compiling an initial list of li 6707detections in s stars . i am grateful to the sao reu program organizers - marie machacek , christine jones , and jonathan mcdowell - for all their support . finally , this work is supported in part by the national science foundation research experiences for undergraduates ( reu ) and department of defense awards to stimulate and support undergraduate research experiences ( assure ) programs under grant no . 0754568 and by the smithsonian institution . ake , t. b.1979 , , 234 , 538 bergeat , j. , knapik , a. , & rutily , b.2002 , , 385 , 94 bodenheimer , p.1965 , , 142 , 451 carpenter , j. m.2001 , , 121 , 2851 covey , k. r. , et al.2007 , , 134 , 2398 dahn , c. c. , liebert , j. , kron , r. g. , spinrad , h. , & hintzen , p. m.1977 , , 216 , 757 downes , r. a. , et al.2004 , , 127 , 2838 fukugita , m. , ichikawa , t. , gunn , j. e. , doi , m. , shimasaku , k. , & schneider , d. p.1996 , , 111 , 1748 gray , r. o. , corbally , c. j.stellar spectral classification , princeton university press , 2009 green , p. j. , margon , b. , & macconnell , d. j.1991 , , 380 , l31 green , p. j. , margon , b. , anderson , s. f. , & cook , k. h.1994 , , 434 , 319 green , p. j. , margon , b. , anderson , s. f. , & macconnell , d. j.1992 , , 400 , 659 groenewegen , m. a. t.1993 , , 271 , 180 herwig , f. 2005 , , 43 , 435 hg , e. , et al.1997 , , 323 , l57 houdashelt , m. l. , bell , r. a. , sweigart , a. v. , & wing , r. f.2000 , , 119 , 1424 iben , i. , jr . , & renzini , a. 1983 , , 21 , 271 ishihara , d. , et al.2010 , , 514 , a1 jorissen , a. , frayer , d. t. , johnson , h. r. , mayor , m. , & smith , v. v.1993 , , 271 , 463 keenan , p. c. , & boeshaar , p. c.1980 , , 43 , 379 keenan , p. c.1954 , , 120 , 484 margon , b. et al . , 2002 aj , 124 , 1651 murakami , h. , et al.2007 , , 59 , 369 perryman , m. a. c. , et al.1997 , , 323 , l49 plez , b. , van eck , s. , jorissen , a. , edvardsson , b. , eriksson , k. , & gustafsson , b.2003 , modelling of stellar atmospheres , 210 , 2p sirko , e. , et al.2004 , , 127 , 914 skrutskie , m. f. , et al.2006 , , 131 , 1163 smith , j. a. , et al.2002 , , 123 , 2121 smith , m. c. , et al.2007 , , 379 , 755 stephenson , c. b.1984 , publications of the warner & swasey observatory , 3 , 1 valdes f. , gupta r. , rose j.a . , singh h.p . , bell d.j . 2004 , astrophys . . ser . , 152 , 251 - 259 van eck , s. , & jorissen , a.2000 , , 360 , 196 van eck , s. , et al . 2010 , arxiv:1011.2092 vyssotsky , a. n.1947 , publications of the leander mccormick observatory , 9 , 197 wyckoff , s. , & clegg , r. e. s.1978 , , 184 , 127 yang , x. , chen , p. , wang , j. , & he , j.2006 , , 132 , 1468 cccclclcc 9 & 006.008240 & + 38.577049 & 00240197 + 3834373@xmath67 & hd 1967/r and&09/15/09 & s5e zr5 ti2 & 7 & i + 22 & 016.300978 & + 19.197842 & 01051223 + 1911522 & hd 6409/cr psc & 09/15/09 & & 4/5 & e + 32 & 021.413226 & + 21.396074 & 01253917 + 2123458 & rx psc & 09/15/09 & & 0/1 & i + 45 & 028.582236 & + 21.889090 & 01541973 + 2153207 & bd+21 255 & 09/15/09&s3 zr1 ti3 & 3/4 & e + 57 & 036.476487 & + 38.122776 & 02255435 + 3807219 & bi and & 09/15/09 & s8 zr7 ti4 & 7 & i + 73 & 052.040388 & + 17.679628 & 03280969 + 1740466 & & 09/20/09 & & 4/5 & + 74 & 052.936921 & + 04.695488 & 03314486 + 0441437@xmath67 & & 09/20/09 & & 6 & + 80 & 055.889249 & + 22.437023 & 03433341 + 2226132 & bd+21 509 & 09/16/09 & & 5/6 & e + 83 & 057.737708 & + 06.909048 & 03505704 + 0654325 & & 09/16/09 & & 6 & + 94 & 066.090802 & -02.532759 & 04242179 - 0231579 & bd-02 891 & 09/20/09&s2 zr2- ti2 & 4 & e + 106 & 073.937675 & + 79.999931 & 04554504 + 7959597 & bd+79 156 & 09/20/09&s4 zr2- ti3- & 4/5 & e + 134 & 080.836187 & -04.570626 & 05232068 - 0434142 & hd 35273/v535 ori&09/20/09 & & 7 + & i + 312 & 108.717017 & + 68.804321 & 07145208 + 6848155@xmath67 & hd 54587/aa cam&11/10/09&m5s@xmath68 & 4/5 & i + 339 & 111.489470 & + 62.591026 & 07255747 + 6235276@xmath67 & & 11/10/09 & & 4 & + 347 & 112.048398 & + 45.990597 & 07281161 + 4559261 & hd 58521/y lyn & 11/10/09&m6s@xmath68 & 7 + & i + 403 & 117.325749 & + 23.734451 & 07491817 + 2344040 & hd 63334/t gem & 11/10/09&s3e zr2.5 ti2 & 4/5 & i + 405 & 117.681704 & + 47.003967 & 07504360 + 4700142 & & 11/10/09 & & 4/5 & + 413 & 118.221777 & + 34.614128 & 07525322 + 3436508 & bd+34 1698 & 11/10/09 & & 6 & i + 418 & 118.369741 & + 17.780506 & 07532873 + 1746498 & hd 64209 & 11/10/09 & & 4/5 & e + 431 & 119.233374 & + 31.167286 & 07565600 + 3110022@xmath67 & ao gem & 11/10/09 & & 7 & i + 460 & 121.836937 & + 11.321875 & 08072086 + 1119187 & & 03/10/10 & & 4/5 & + 471 & 122.761565 & + 08.139297 & 08110277 + 0808214 & & 03/10/10 & & 5/6 & e + 494 & 125.428453 & + 17.285120 & 08214282 + 1717064 & hd 70276/v cnc & 03/10/10&s0.5e zr0 & 1 & i + 589 & 137.661696 & + 30.963114 & 09103880 + 3057472 & hd 78712/rs cnc&03/10/10&m6s@xmath69 & 7 + & i + 612 & 143.901826 & + 69.157043 & 09353643 + 6909253 & bd+69 524 & 03/11/10 & & 3/4 & e + 707 & 162.729912 & + 04.499532 & 10505517 + 0429583 & & 03/11/10 & & 6 & + 722 & 166.970132 & + 68.366440 & 11075283 + 6821591 & hd 96360/hl uma&03/11/10 & & 4/5 & e + 803 & 190.986134 & + 61.093254 & 12435667 + 6105357 & hd 110813/s uma&03/11/10&s6e zr6 & 5/6 & i + 819 & 200.328046 & + 43.987118 & 13211873 + 4359136 & av cvn & 03/11/10 & s3 zr2 ti1 & 3/4 & + 833 & 207.141756 & + 31.999107 & 13483402 + 3159567 & & 03/11/10 & & 0 & + 836 & 208.005660 & -03.480082 & 13520135 - 0328482 & hd 120832 & 03/11/10 & & 4/5 & e + 856 & 214.462975 & + 83.831696 & 14175111 + 8349541@xmath67 & hd 1272266/r cam&06/24/09&s2e zr2 & 3 & e + 855 & 216.950026 & -03.079711 & 14274800 - 0304469 & bd-02 3848 & 05/03/09 & & 4 & e + 892 & 232.384097 & + 00.189076 & 15293218 + 0011206 & & 05/03/09 & & 2/3 & + 902 & 238.100360 & -23.352097 & 15522408 - 2321075 & & 05/03/09 & & 6 & e + 903 & 237.694254 & + 48.483078 & 15504662 + 4828590@xmath67 & hd 142143/st her&05/03/09&m6.5s zr1 ti6 + & 5/6 & i + 926 & 245.401029 & + 56.877048 & 16213624 + 5652373@xmath67 & hd 147923 & 05/03/09 & & 2 & e + 932 & 249.263106 & + 07.372439 & 16370314 + 0722207@xmath67 & bd+07 3210 & 05/03/09 & & 4/5 & + 958 & 256.694247 & -02.017769 & 17064661 - 0201039 & & 05/03/09 & & 6 & + 981 & 260.529640 & + 23.817570 & 17220711 + 2349032 & bd+23 3093 & 05/03/09&s3.5 zr3 + ti4 & 4/5 & e + 986 & 261.200788 & + 33.918953 & 17244818 + 3355082 & & 05/03/09 & & 3 & + 1002&267.252799 & + 21.710087 & 17490067 + 2142363@xmath67 & & 05/03/09 & & 4 & e + 1065&279.398924 & + 43.256588 & 18373574 + 4315237 & & 05/03/09 & & 3 & e + 1087&282.875526 & + 48.911934 & 18513012 + 4854429@xmath67 & tu dra & 05/03/09 & & 3/4 & i + 1150&294.205739 & + 50.199917 & 19364937 + 5011597 & hd 185456/r cyg&05/03/09&s5e zr5 ti0 & 7 & i * + 1152&294.345155 & + 67.188873 & 19372283 + 6711199 & & 05/03/09&s5- zr4.5 & 5/6 & i + 1165&297.641369 & + 32.914139 & 19503392 + 3254509@xmath67 & hd 187796/chi cyg&05/03/09&s6+e zr2 ti6.5 & 7 + & i + 1222&307.913302 & + 04.061802 & 20313919 + 0403424@xmath67 & & 05/23/09 & & 5 & + 1247&313.355676 & + 06.539369 & 20532536 + 0632217 & & 05/23/09 & & 2/3 & + 1262&317.601674 & + 01.606119 & 21102440 + 0136220 & & 05/27/09 & & 4/5 & + 1264&318.232437 & + 09.548880 & 21125578 + 0932559 & & 07/17/09 & & 3/4 & + 1290&331.995242 & + 29.665623 & 22075885 + 2939562 & & 07/17/09 & & 4/5 & e + 1299&337.993328 & + 02.022399 & 22315839 + 0201206 & & 07/17/09 & & 5 & e * + 1304&340.954936 & + 33.926231 & 22434918 + 3355344 & hd 215336 & 07/17/09 & & 1/2 & e + 1315&343.648463 & + 16.941910 & 22543563 + 1656308 & hr peg & 07/19/09 & s4 + zr1.5 ti4 & 4/5 & i + 1328&349.035110 & + 28.863674 & 23160842 + 2851492@xmath67 & & 07/19/09 & & 3 & e + 1340&356.540618 & + 34.783325 & 23460974 + 3446599 & & 07/19/09 & & 3 & + crrrrrrrrrr 00240197 + 3834373&8.40&6.37&7.39&2.02&0.77&0.12&&@xmath70&3.17&@xmath71 + 01051223 + 1911522&9.38&7.79&&3.67&2.76&2.48&6.68&1.75&8.28&1.63 + 01253917 + 2123458&16.53&14.40&8.80&6.27&5.39&4.98&8.90&3.91&10.16 & + 01541973 + 2153207&10.18&8.60&9.00&5.29&4.48&4.29&8.57&3.59&10.26&3.61 + 02255435 + 3807219&11.90&9.63&&3.50&2.43&1.92&6.14&1.10&7.54&0.95 + 03280969 + 1740466&12.81&11.14&&7.91&7.02&6.72&11.09 & & & + 03314486 + 0441437&12.68&11.05&&6.93&6.02&5.65&9.92 & & & + 03433341 + 2226132&10.74&9.05&9.69&5.03&4.03&3.80&8.14&3.12&9.72&3.23 + 03505704 + 0654325&12.40&10.54&&6.42&5.47&5.12&9.32&4.32&10.52 & + 04242179 - 0231579&10.42&8.78&9.33&5.62&4.78&4.44&8.80&3.96&10.24&3.85 + 04554504 + 7959597&10.68&9.04&9.66&5.67&4.87&4.55&8.97&4.04&10.18&3.99 + 05232068 - 0434142&11.23&9.04&10.00&3.90&2.80&2.36&&0.38&6.47&@xmath72 + 07145208 + 6848155&8.85&7.33&&2.60&1.67&1.39&5.74&0.70&6.92&0.08 + 07255747 + 6235276&12.49&10.92&&7.67&6.83&6.53&10.79 & & & + 07281161 + 4559261&&&7.37&0.65&@xmath73&@xmath74&3.43&@xmath75&4.15&@xmath76 + 07491817 + 2344040&13.37&11.10&8.00&4.12&3.22&2.71&7.16&2.20&8.60&2.05 + 07504360 + 4700142&11.83&10.26&&7.32&6.43&6.17&10.43 & & & + 07525322 + 3436508&11.34&9.79&&4.99&3.99&3.71&7.99&2.95&9.38&2.81 + 07532873 + 1746498&&&8.56&4.82&3.84&3.47&7.76&2.79&9.23&2.83 + 07565600 + 3110022&16.96&14.34&&6.60&5.66&5.13&8.81&3.73&10.11&2.97 + 08072086 + 1119187&10.93&9.43&&7.37&6.52&6.22&10.43 & & & + 08110277 + 0808214&10.19&8.30&&5.45&4.58&4.17&8.42&3.52&9.84&3.35 + 08214282 + 1717064&&&7.50&5.08&4.00&3.59&6.67&1.46&8.07&1.46 + 09103880 + 3057472&6.88&4.94&6.08&@xmath77&@xmath78&@xmath79&&@xmath80&2.72&@xmath79 + 09353643 + 6909253&10.24&8.78&9.30&5.55&4.69&4.42&8.84&3.84&10.08&3.89 + 10505517 + 0429583&12.65&10.90&&5.36&4.54&4.05&8.28&&9.46 & + 11075283 + 6821591&9.13&7.73&8.10&4.07&3.17&2.77&7.36&2.37&8.81&2.26 + 12435667 + 6105357&9.69&7.69&8.87&4.46&3.43&3.02&7.21&2.06&8.39&1.71 + 13211873 + 4359136&11.00&9.37&9.78&6.28&5.43&5.16&9.53&&10.79 & + 13483402 + 3159567&13.30&11.80&&9.03&8.43&8.22 & & & & + 13520135 - 0328482&&&9.59&5.50&4.76&4.35&8.75&3.78&10.35 & + 14274800 - 0304469&10.26&8.77&9.47&5.74&4.93&4.59&9.01&4.05&10.66 & + 14175111 + 8349541&10.58&8.81&6.97&3.82&2.90&2.46&6.90&1.83&8.38&1.66 + 15293218 + 0011206&&&&7.47&6.70&6.40&10.77 & & & + 15522408 - 2321075&&&&6.67&5.76&5.38&9.68&4.79 & & + 15504662 + 4828590&8.41&6.50&&0.74&@xmath81&@xmath82&3.09&@xmath83&3.83&@xmath84 + 16213624 + 5652373&8.50&7.05&7.59&4.71&3.65&3.46&7.66&2.82&9.18&2.64 + 16370314 + 0722207&10.50&8.95&9.53&5.50&4.71&4.38&8.64&&9.98 & + 17064661 - 0201039&12.73&10.94&&7.35&6.36&6.07 & & & & + 17220711 + 2349032&10.91&9.37&10.10&5.73&4.98&4.59&8.98&4.09&10.40&3.99 + 17244818 + 3355082&12.28&10.73&&7.73&6.92&6.63&10.98 & & & + 17490067 + 2142363&12.71&11.22&&7.78&6.94&6.64&10.95&5.98 & & + 18373574 + 4315237&12.42&10.93&&7.46&6.61&6.36&10.61&5.86 & & + 18513012 + 4854429&17.02&14.59&10.00&5.82&5.14&4.74&7.99&2.52&8.78&2.01 + 19364937 + 5011597&12.21&9.58&8.15&2.25&1.38&0.86&3.94&&4.87 & + 19372283 + 6711199&11.18&9.23&9.70&5.07&4.19&3.76&7.88&2.79&9.30&2.64 + 19503392 + 3254509&12.55&9.79&6.80&0.17&@xmath85&@xmath86&&@xmath87&&@xmath88 + 20313919 + 0403424&12.43&10.90&&7.48&6.61&6.25&10.64 & & & + 20532536 + 0632217&12.78&11.29&&7.99&7.17&6.90&11.14 & & & + 21102440 + 0136220&&&&7.60&6.79&6.44&10.71 & & & + 21125578 + 0932559&&&&7.94&7.09&6.81&11.15 & & & + 22075885 + 2939562&&&&6.56&5.71&5.39&9.64&4.90&10.87 & + 22315839 + 0201206&&&&6.46&5.56&5.21&9.44&&11.13 & + 22434918 + 3355344&&&7.82&4.98&4.07&3.79&8.10&3.18&9.60&3.02 + 22543563 + 1656308&7.02&5.43&6.47&2.31&1.24&1.04&5.09&@xmath89&6.63&@xmath90 + 23160842 + 2851492&13.62&12.10&&9.11&8.32&8.03 & & & & + 23460974 + 3446599&12.02&10.53&&7.49&6.63&6.40&10.70 & & & + ccccccccc 09103880 + 3057472&8.06&0.98&125&15&1.41&0.33&@xmath91&0.33 + 15504662 + 4828590&3.07&0.75&326&80&0.85&0.56&@xmath92&0.56 + 16213624 + 5652373&2.31&0.62&433&117&0.32&0.61&@xmath93&0.61 + 22543563 + 1656308&3.31&0.93&302&85&0.02&0.64&@xmath94&0.64 + clcccccc 00240197 + 3834373&35.89&2.46&3.22&8.4&6.37&@xmath954.14&@xmath956.17 + 01051223 + 1911522&12.65&0.95&9.13&9.38&7.79&@xmath955.42&@xmath957.01 + 03505704 + 0654325&10.66&2.50&10.8&12.4&10.55&@xmath952.77&@xmath954.62 + 04242179 - 0231579&7.50&1.27&15.4&10.41&8.78&@xmath955.53&@xmath957.16 + 07145208 + 6848155&18.44&1.07&6.26&8.85&7.33&@xmath955.13&@xmath956.65 + 09103880 + 3057472&34.22&0.85&3.38&6.8&4.94&@xmath955.84&@xmath957.70 + 09353643 + 6909253&8.54&1.67&13.5&10.24&8.78&@xmath955.42&@xmath956.88 + 11075283 + 6821591&22.65&1.15&5.10&9.13&7.73&@xmath954.41&@xmath955.81 + 12435667 + 6105357&13.62&1.31&8.48&9.69&7.69&@xmath954.95&@xmath956.95 + 13211873 + 4359136&11.31&2.18&10.2&10.99&9.37&@xmath954.06&@xmath955.68 + 14175111 + 8349541&7.26&1.62&15.9&10.58&8.82&@xmath955.43&@xmath957.19 + 14274800 - 0304469&13.18&1.89&8.77&10.26&8.77&@xmath954.45&@xmath955.94 + 15504662 + 4828590&22.38&1.11&5.16&8.41&6.5&@xmath955.15&@xmath957.06 + 16213624 + 5652373&12.32&1.14&9.37&8.5&7.05&@xmath956.36&@xmath957.81 + 19372283 + 6711199&9.75&2.35&11.8&11.18&9.24&@xmath954.19&@xmath956.13 + 22543563 + 1656308&15.75&0.99&7.33&7.02&5.43&@xmath957.31&@xmath958.90 +

we present _ nustar _ high energy x - ray observations of the pulsar wind nebula ( pwn)/supernova remnant 21 . we detect integrated emission from the nebula up to kev , and resolve individual spatial features over a broad x - ray band for the first time . the morphology seen by _ nustar _ agrees well with that seen by _ xmm - newton _ and _ chandra _ below 10 kev . at high energies _ nustar _ clearly detects non - thermal emission up to kev that extends along the eastern and northern rim of the supernova shell . the broadband images clearly demonstrate that x - ray emission from the north spur and eastern limb results predominantly from non - thermal processes . we detect a break in the spatially integrated x - ray spectrum at kev that can not be reproduced by current sed models , implying either a more complex electron injection spectrum or an additional process such as diffusion compared to what has been considered in previous work . we use spatially resolved maps to derive an energy - dependent cooling length scale , with . we find this to be inconsistent with the model for the morphological evolution with energy described by . this value , along with the observed steepening in power - law index between radio and x - ray , can be quantitatively explained as an energy - loss spectral break in the simple scaling model of , assuming particle advection dominates over diffusion . this interpretation requires a substantial departure from spherical magnetohydrodynamic ( mhd ) , magnetic - flux - conserving outflow , most plausibly in the form of turbulent magnetic - field amplification .

1405.3239

a pulsar - wind nebula ( pwn ) is a bubble of relativistic particles and magnetic field inflated by a rotation - powered pulsar , emitting centrally peaked synchrotron radiation . young pwne are frequently found inside shell supernova remnants ( snrs ) , where the relativistic wind of electron - positron pairs ( and perhaps ions ) experiences a wind termination shock close to the pulsar due to the pressure of the snr interior . beyond this wind shock , the relativistic fluid can radiate synchrotron emission from the radio to the x - ray band , and inverse - compton ( ic ) emission at higher energies . a pwn / snr combination is often called a ` composite ' snr . pwne can outlive their snr and interact directly with the interstellar medium . see @xcite for a review . the detailed characterization of pwn spectra gives information on the particle energy distribution produced by the pulsar , and also on the nature of acceleration in relativistic shocks . however , spectral structure present immediately downstream of the wind shock can be altered by propagation effects including diffusive transport and radiative losses , which may depend on the evolution of the entire pwn . the ability to use pwne as laboratories in which to study the behavior of relativistic pair plasmas depends largely on the extent to which these various effects can be disentangled . models studying pwne with time - dependent , one - zone , homogeneous approximations , provide insights into the evolution of the spectrum across all energy bands of the nebula as a whole . broadband spectral - energy distributions ( seds ) , coupled with spatially resolved spectroscopy , are required for this purpose . radially - dependant models also provide valuable insights by addressing detailed spatial and spectral structures of pwne . the classic work of @xcite , which invokes a particular hydrodynamic model , predicts the spectral break between the optical and x - ray energy bands , as well as the behavior of size with photon energy for the crab nebula . implicit in that calculation is a prediction for the radial dependence of the spectrum . @xcite assume particle transport by pure advection in a spherical geometry ; this situation predicts a roughly uniform nebular spectrum with radius until the ( energy - dependent ) nebular edge , where radiative energy losses sharply steepen the spectrum . this behavior is generally inconsistent with observations @xcite , in particular in several objects a fairly uniform spectral steepening with radius is observed ( see * ? ? ? the cause of spectral structure is of interest beyond the specific pwn context ; if its origin is in the physics of particle acceleration at relativistic shocks , the results have implications for other objects which share the same shock mechanisms . however , the structure may be due instead to transport and evolutionary effects in the pwn post - shock flow . at the same time , spatially integrated spectra of pwne are not well understood , especially in their most salient feature a large steepening of the spectrum between the flat radio emission and the considerably steeper spectra observed from ir through x - rays . @xcite documents differences between radio and x - ray spectral indices of @xmath5 ( @xmath6 ) for seven out of the eight pwne modeled , compared to radiative energy losses in homogeneous steady sources which can produce , at most , @xmath7 . if this break is not due solely to evolutionary effects and instead is due to at least , in part , to intrinsic spectral structure in the particle distribution injected at the wind shock , it is implying something important and interesting about particle acceleration in pulsar winds . 21 was discovered in 1970 @xcite in the radio band , and first observed in the x - ray band in 1981 @xcite . it is a classic example of a crab - like pwn : it has a filled , mostly symmetric spherical morphology centered on a pulsar . measurements spanning 44 years show a well - characterized flat spectrum in the radio regime ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , with recent observations reporting a spectral index of @xmath8 @xcite . the flux density at 1ghz is @xmath9 jy @xcite , and @xmath10co and h analyses determined the distance to the nebula to be @xmath11 kpc @xcite . in this paper we adopt a distance of @xmath12 kpc . observations with _ chandra _ and _ xmm - newton _ find that the x - ray emission is dominated by a centrally peaked core that contains @xmath13 of the 2 8 kev flux . the spectrum is described by a non - thermal power law with no evidence of line emission . the total unabsorbed nebular flux is @xmath14 kev@xmath15 erg @xmath16 s@xmath17 @xcite . safi - harb and coworkers detected spectral steepening in the nebula , indicative of synchrotron cooling . the innermost @xmath18 radius region has a power law photon index of @xmath19 ( @xmath20 ) which softens to @xmath21 at a radius of @xmath22 consistent with the edge of the nebula . this spectral softening is also visible in hardness ratio images @xcite . the photon index of @xmath23 yields @xmath24 . the associated pulsar , psr j1833 - 1034 , was discovered in 2005 , 35 years after the initial detection of the pwn . psr j1833 - 1034 has a @xmath25 ms period , @xmath26 , @xmath27 kyr , and @xmath28ergs@xmath17 @xcite . despite the pulsar s high spin - down luminosity , pulsations have not been detected in the x - ray band . _ chandra _ and _ xmm - newton _ observations detected a region of diffuse , uniform emission extending from the edge of the pwn at @xmath22 out to @xmath29 ( e.g. * ? ? ? * ) , with a non - thermal power law spectrum of @xmath30 . the flux is substantially dimmer than the pwn . this symmetric emission was proposed to be an extension of the pwn itself @xcite , but the absence of coincident radio emission , and the recognition of the importance of dust scattering for the large column density toward g21.5 - 0.9 , led @xcite and @xcite to model the smoothly distributed halo emission as due to dust scattering . additionally , two regions of brightened emission were discovered near the outer edges of the halo , referred to as the north spur and eastern limb @xcite that could not be explained by dust scatter . these regions , located approximately @xmath31 and @xmath32 , respectively , from the center of the nebula , have week but detectable x - ray emission @xcite . deep radio observations have not detected any emission from either the halo or the eastern limb , while the north spur can be clearly seen in a 1.4 ghz image @xcite . @xcite , and later @xcite , found the spectrum of the north spur comprises of a weak , low - temperature thermal component and a non - thermal continuum . this knot of emission has been interpreted as the result of ejecta interacting with the h - envelope of the sn . the eastern limb , in contrast , was found to have only a non - thermal spectrum , and has spectral and morphological features that imply it is the limb - brightened region of the sn shell @xcite . however , the low surface brightness of both the eastern limb and north spur prevent their @xmath33 kev continuua from being characterized more specifically . in this paper we present the first sub - arcminute x - ray images above 10 kev and corresponding x - ray spectroscopic studies of 21 . @xmath34 discusses the _ nustar _ observations . @xmath35 presents our spectral analysis , while @xmath36 presents our image analysis and @xmath37 details our search for the pulsar in the high - energy x - ray band . lastly , in @xmath38 we discuss the spectral and spatial studies and how _ nustar _ can shed light on the physics in a pwn and the natures of the north spur and eastern limb . _ nustar _ observed 21 on four separate occasions for a total of 281 ks : 2012 july 29 ( obsid 10002014003 ) , 2012 july 30 ( obsid 10002014004 ) , 2013 february 26 ( obsid 40001016002 ) , and 2013 february 27 ( obsid 40001016003 ) . the center of the remnant was located approximately @xmath39 from the on - axis position for all observations so that the majority of the pwn was located on one of the four detector chips . _ nustar _ , which contains two co - aligned optic / detector focal plane modules ( fpma and fpmb ) , has a half - power diameter ( hpd ) of @xmath40 , an angular resolution of @xmath41 ( fwhm ) over its @xmath42 kev x - ray energy range , and a characteristic fwhm energy resolution of 400 ev at 10 kev . the field of view is @xmath43 at @xmath44 kev as defined by the full width at half intensity . the _ nustar _ nominal reconstructed coordinates are accurate to @xmath45 ( @xmath46 confidence level ) @xcite . prior to our imaging and spectral analysis , we registered individual observations to j2000 coordinates using the central peak position of the pwn measured by _ chandra _ : ra(j2000)@xmath47 , dec(j2000)@xmath48 @xcite . note that the _ chandra _ centroid position is @xmath49 offset from the radio pulsar position @xcite . the _ nustar _ field of view for 21 is devoid of any visible point sources including the foreground star ss 397 , which is located @xmath50 sw of the center of the pwn . we determined the centroid position of the inner @xmath51 radius region of the pwn in the @xmath52 kev band using the idl routine * gcntrd * so that the _ nustar _ image is not contaminated by substructures such as the eastern limb and north spur . the centroiding errors ( 90% confidence level ) are @xmath53 in both ra and dec . we subsequently confirmed by fitting a circular 2-d gaussian convolved with the _ nustar _ psf ( section 4.1 ) that the centroid position of the pwn overlaps with the radio pulsar position within our uncertainty . lccccccc[ht ! ] @xmath54 & @xmath55 & @xmath56 & @xmath57 & @xmath58 & @xmath59 & @xmath60 & @xmath61 + @xmath62 & @xmath63 & @xmath64 & @xmath63 & @xmath65 & @xmath63 & @xmath66 & @xmath63 + @xmath67 & @xmath63 & @xmath68 & @xmath63 & @xmath69 & @xmath63 & @xmath70 & @xmath63 + @xmath71 & @xmath72 & @xmath73 & @xmath63 & @xmath63 & @xmath63 & @xmath63 & @xmath63 + @xmath74 & @xmath75 & @xmath76 & @xmath63 & @xmath63 & @xmath63 & @xmath63 & @xmath63 + @xmath77 & @xmath78 & @xmath79 & @xmath80 & 0.99 ( 3557 ) & @xmath81 & @xmath82 & @xmath83 + [ tab : specfit ] kev from the entire pwn ( extraction radius=@xmath84 ) . eight _ nustar _ spectra and their residuals are shown . the middle panel depicts the residuals fo for the absorbed power - law fit , while the lower panel depicts the absorbed broken power - law fit . , scaledwidth=45.0% ] we performed _ nustar _ spectral analysis integrated over the pwn region as well as spatially resolved spectroscopy which we present in subsequent sections . we applied the same analysis procedures , described below , for all _ nustar _ spectra . prior to spectral fitting we generated _ nustar _ response matrix ( rmf ) and effective area ( arf ) files for an extended source using _ nuproducts _ ( nustardas v1.1.1 ) , then grouped the spectra to @xmath85 counts per bin using ftools * grppha * ( heasoft 6.13 ) . we fit the _ nustar _ spectra using * xspec * version 12.8.0 @xcite , with the atomic cross sections set to those from @xcite and the abundances to @xcite . we fit the power - law model * pegpwrlw * and the broken power - law model * bknpower*. we multiplied the continuum models by the interstellar absorption model , * tbabs * , with @xmath86 frozen to @xmath87@xmath16 from @xcite . we fit the spectra from each observation and each module jointly by linking all parameters except the continuum normalization , to take into account small calibration uncertainties between the two modules flux normalization . we generated background spectra using _ nulyses _ , _ nustar_-specific software that accounts for detector background and cosmic x - ray background ( cxb ) for a given extraction region of the source spectrum . _ nulyses _ creates source - free background maps from the _ nustar _ blank - sky survey data taken less than a month from the 21 observations . this method was applied to all subsequent spectral analyses . the conventional way of extracting background spectra from a nearby region is not applicable to our analysis since the pwn , broadened by the _ nustar _ psf , covers most of the detector chip and the detector background varies among the different chips . we fit the _ nustar _ spectra in the @xmath88 kev band , above which the detector background dominates . we extracted _ nustar _ data using a @xmath84 radius circular region centered at the pulsar psr j1833@xmath891034 position . this region represents the largest circle within the same detector chip ( to ensure that _ nustar _ detector background is the same and predictable ) , and it encloses over 93% of the pwn photons . we first fit the _ nustar _ spectra with a single absorbed power - law model in the @xmath88 kev energy band . we obtained a best - fit power - law index of @xmath90 and an x - ray flux of @xmath91 erg s@xmath17 @xmath16 . these results are consistent within their uncertainties to the parameters reported by @xcite ( @xmath92 , @xmath93 erg s@xmath17 @xmath16 ) , who analyzed 21 for the purposes of cross - calibrating x - ray instruments operational at that time . all uncertainties ( 90% confidence level ) quoted in the text and tables include both systematic and statistical errors . a more detailed comparison of 21 spectral fitting between _ nustar _ and other x - ray telescopes as well as systematic error estimates will be addressed in a separate calibration paper . a single power - law fit is not satisfactory , with reduced @xmath94 , with clear residuals evident between 5 and 10 kev ( see the left panel of figure [ fig : specfit ] ) . we then fit a broken power - law model ( * bknpower * ) to the _ nustar _ spectra , yielding reduced @xmath95 . statistical tests using the f - distribution ( * ftest * in xspec ) show the broken power - law model is statistically favored over the single power - law model with high significance , so that a spectral break is unambiguously required . the best - fit power law indices were @xmath96 and @xmath97 , with the best - fit break energy of @xmath98 kev . in order to investigate any instrumental effects that mimic a break around 9 kev , we analyzed _ nustar _ observations of the known power - law source 3c273 . we followed the same procedures used for 21 , and found no systematic residuals around 9 kev in the _ nustar _ data . a spectral break in another pwn , the crab nebula , has recently been detected by _ nustar _ at similar energies of @xmath99 kev @xcite . dust scattering can also mimic the low - energy spectral behavior observed in the _ nustar _ recent analysis of two low - mass x - ray binaries with similar absorption columns of @xmath100 @xmath16 , gx5 - 1 and gx13 + 1 , present radial profiles of models of the dust scattering halos . these profiles indicate that @xmath101 of the source photons at @xmath102 kev are enclosed between @xmath103 @xcite . including the energy - dependant halo intensity profile relationship @xmath104 @xcite , a radius of @xmath105 encloses over @xmath106 of the source photons in both @xmath107 kev and @xmath108 kev . similar analysis reveals that the @xmath109 extraction region , presented in section 3.2 , has a scattering loss of @xmath110 and @xmath111 of source photons in @xmath107 kev and @xmath108 kev , respectively . loss of photons by dust scattering therefore has a negligible effect , and can not explain the spectral break in the _ nustar _ spectrum . additionally , because our scattering estimates are based on point sources , we can not conclusively rule out the possibility of a very small effect of dust scattering on the radial behavior of the low - energy spectrum presented in section 3.2 . an x - ray spectral break at @xmath112 kev was first suggested by @xcite . they extrapolated the spectral fits from the hard and soft x - ray bands reported by @xcite , and noted that a break at @xmath112 kev was likely . _ nustar _ not only confirms the break , but provides the first measurement of the @xmath2 kev break in broad - band 3 45 kev spectra of 21 in a single x - ray telescope . obtaining a well - calibrated x - ray spectrum from 21 is particularly valuable when creating pwn theoretical models . specifically , @xcite presented synchrotron emission from radio to gev energies by modeling the electron spectrum emanating from the pulsar of 21 . they incorporated adiabatic and energy losses in a one - zone model . figure [ fig : sed ] shows the _ nustar _ best - fit broken power - law data overplotted on figure 3 from @xcite . more details are presented in section 6.1 . a number of young pwne , including 21 , exhibit spectral softening from the center of the pwn outward . this is the classic signature of synchrotron burn - off . @xcite and later @xcite extracted _ chandra _ spectra at various annuli of 21 and showed spectral softening from @xmath19 in the central @xmath113 radius circle to @xmath21 in the outer annulus at a radius of @xmath114 . _ nustar _ has the ability to probe the synchrotron burn - off effects in the hard x - ray band above 10 kev . we extracted _ nustar _ spectra from the central @xmath51 region as well as from four nested annuli , each @xmath51 in width . after experimenting with various annulus widths , we found that the annulus width of @xmath51 is the best compromise between the _ nustar _ angular resolution and the expected spatial variation from the _ chandra _ results . the broad - band @xmath88 kev fit results from both the absorbed * powerlaw * and * bknpower * models are shown in table [ tab : specfit ] . the power - law index fit with a single power - law model increases from @xmath115 in the @xmath51 inner region to @xmath116 in the outer @xmath117 radius annulus , confirming the spectral softening discovered by _ chandra_. however , the residuals for the inner @xmath51 and @xmath118 regions also show a spectral break around 9 kev . we fit a broken power - law model , and the results are shown in table [ tab : specfit ] and plotted in figure [ fig : ggraph ] . we conclude that the spectral break at 9 kev is detected with high significance in the inner @xmath51 radius circular region and the @xmath119 annulus . interestingly , as is shown in figure [ fig : ggraph ] , the second ( high energy ) power - law component shows no spatial variation . the spectral softening is observed only below 9 kev . , dec(j2000)@xmath48 . the blue points illustrate the photon index @xmath54 below the spectral break of @xmath2 kev . the red points illustrate the photon index @xmath62 above the break.,scaledwidth=45.0% ] in the imaging analysis , we applied common procedures for image preparation . specifically , we generated the following ; 1 ) mosaic images from the four observations , 2 ) exposure maps with vignetting effects included and 3 ) an effective psf that takes into account the various off - axis positions , orientation angles , and exposure times of the observations . after proper source registration as described in section 2 , we selected photon events in different energy bands using * dmcopy * ( ciao v.4.4 ) . we chose energy bands to ensure sufficient photon statistics in each image , as well as to minimize the effects of averaging the energy - dependent vignetting function over the energy range . we then summed all four _ nustar _ raw images in each energy band to create mosaic images using * ximage * ( heasoft 6.13 ) . for each observation , we generated an exposure map with the optics vignetting effects using _ nuexpomap _ ( nustardas v 1.1.1 ) and summed those as well . we then applied exposure - corrections to the raw mosaic images in * ximage * and generated energy - dependent flux images . figure [ fig : rawprofiles ] shows the @xmath120 kev flux image on the left , clearly showing the centrally peaked pwn , broadened by the wings of the _ nustar _ psf . in order to investigate the radial profile and detect subtle features possibly buried in the raw images , we generated an effective psf used for image deconvolution and 2-d forward image fitting , as described in the subsequent sections . we estimated the background level using the _ nulysis _ software described in section 3 . first , we reproduced the background levels in the three detector chips where the contribution of 21 is negligible . we found that the background for the 21 observation is largely due to the stray - light cxb component below 20 kev and the internal detector background above 20 kev . we were able to reliably produce the background count rates on the detector chip containing the 21 image , and found that the source emission is dominant up to 30 kev . hereafter we present imaging analysis below 30 kev . a measurement of the pwn radius in different energy bands is complementary to the radially dependent spectral analysis . the angular resolution of _ nustar _ is comparable to the pwn size ( @xmath121 ) . therefore it is essential to take into account the blurring effects by the psf convolution . we adopted a forward - folding method to measure the size of the pwn using * sherpa*,ciao s modelling and fitting package @xcite . the use of * sherpa * allows us to use the goodness - of - fit test ( c - statistics ) to find the best - fit parameters . we convolved an assumed source model , a circular 2-d gaussian profile , with the effective psf described in section 3 , and fit the _ nustar _ flux images . we also ran the * conf * command in * sherpa * to determine 90@xmath122 confidence intervals for all fit parameters . we confined our fitting range to within a radius of @xmath123 . in general , a circular 2-d gaussian provides a good fit to the entire pwn , leaving only small residuals . the best - fit gaussian fwhms in five energy bands are plotted in figure [ fig : rve ] . the trend of decreasing pwn size with energy is evident , confirming the synchrotron burn - off effect at energies above @xmath44 kev . in 6.2 we employ the fwhm radius to measure the synchrotron cooling length . we fit the data with a power - law @xmath124 . the best fit of @xmath125 is shown in figure [ fig : rve ] and yields an index of @xmath126 . these results are used in 6.2 to constrain physical conditions in the nebula . , with @xmath126 . , scaledwidth=45.0% ] the eastern limb and north spur are only partially resolved in the raw _ nustar _ mosaics due to the small size of 21 compared to the _ nustar _ psf , combined with the low surface brightness of these features . to detect these features with high confidence , we applied two different methods . first , we confirmed the detection of the eastern limb and northern spur in the flux images in each energy band ( section 4.2.1 ) . second , we sharpened the _ flux images using the lucy - richardson deconvolution algorithm @xcite . in order to detect the eastern limb and north spur in the raw _ nustar _ image , we analyzed line intensity profiles of 6 image pixels ( a total of @xmath127 ) in width . we chose projection axis lines along different orientation angles around the center of the pwn . the left image in figure [ fig : rawprofiles ] contains two lines , colored red and blue , that indicate the positions along which the flux profiles were taken . the red line was chosen to fall along the regions of brightest eastern limb emission , while the blue line overlays a region of 21 that does not contain any emission from either the eastern limb or the north spur . the intensity profiles on the right of figure [ fig : rawprofiles ] correspond to the lines drawn in the left of figure [ fig : rawprofiles ] . the red profile shows clear and significant excess emission when compared to the blue profile , confirming that the eastern limb is detected in the @xmath120 kev band . this also holds for the @xmath128 kev and @xmath129 kev . while the two profiles in the @xmath130 kev band are very similar , the lines are separate and distinct between @xmath131 , when accounting for statistical uncertainty . with these intensity profiles we confirm the detection of the north spur and eastern limb at energies as high as @xmath132 kev . we repeated this process with the north spur , orienting the red line along the north / south axis . lastly , we also confirmed the detection of the eastern limb and north spur in residual maps from fitting the pwn , as mentioned in section 4.1 . the residual maps had areas of faint excess , indicating that there exists emission aside from the pwn . however , the broad _ nustar _ psf prevents us from studying any further morphology of the faint emission . we therefore turn to image deconvolution in an attempt to remove the effects of the psf . we applied an iterative deconvolution technique to the _ nustar _ images using * arestore * ( ciao v4.4 ) and the effective psfs described in section 4 . the iterative image deconvolution can produce artificial features if the process is over - iterated and/or the background region is deconvolved . great care was taken to ensure that the deconvolved image was both stable and reproducible . we deconvolved the @xmath120 kev mosaiced _ nustar _ image with several iterations , from 20 up to 200 in increments of 20 . each of these images exhibited the same features of the north spur and eastern limb that are characterized by areas of brightened emission to the north and northeast of the pwn . additionally , we confirmed that the features visible through deconvolution were not dependent on telescope rotation , detector module , or observation . the data sets were first grouped by _ nustar _ module , summing all the fpma images before deconvolving , and likewise for all the fpmb images . the deconvolved images have identical features . we also grouped the data by epoch , summing the 2012 and 2013 data separately . as before , the two deconvolved images are very similar . as a final verification of our deconvolution process , we compared the output images to the @xmath120 kev image from _ chandra_. figure [ fig : decon ] ( a ) shows the _ chandra_image at the top left corner . one can see the eastern limb and north spur , which is identified as ` knot ' in the image . the north spur is visible as the excess of emission @xmath133 north of the pwn . the eastern limb is an arced feature that begins at the southern edge and wraps clockwise around to the northern edge , located approximately @xmath134 from the center of the plerion . for reference , the outer green circle has a radius of @xmath84 . figure [ fig : decon ] ( b ) shows the final deconvolved @xmath120 kev _ nustar _ image , created with 20 iterations . the _ nustar _ features correspond well to those seen by _ chandra _ , allowing us to confidently extend this deconvolution technique to higher energies . once we established that the _ nustar _ results reproduce the _ chandra _ images stably for a range of lucy - richardson iterations , we applied the same deconvolution to the higher energy images . figures [ fig : decon ] ( b - f ) show the _ nustar _ deconvolved images at @xmath120 kev , @xmath128 kev , @xmath129 kev , @xmath130 kev , and @xmath135 kev , respectively . one can visually identify the eastern limb and north spur at energies as high as @xmath136 kev , as seen in figure [ fig : decon ] ( d ) . this is not surprising , since all of the previously reported spectral fitting of these two regions have required a non - thermal component . this is , however , the first direct measurement showing that these features have emission above 10 kev . at energies above @xmath136 kev the limb and spur become very faint . one can still see emission along the eastern edge of the limb in figure [ fig : decon ] ( e ) . we can , however , verify the existence of these features above @xmath136 kev by producing intensity profiles of the deconvolved images at various azimuthal angles . applying the same 1-d profile method as in section 4.2.1 , we obtained profiles along three different radial lines , each bisecting 21 at angles in increments of @xmath137 as shown in the left image of figure [ fig : deconprof ] . the arrows overlaid on the lines indicate the direction along which the profiles were obtained . the intensity profiles themselves for the @xmath120 kev image are shown on the right in figure [ fig : deconprof ] . each profile corresponds to the line of matching color overlaid on the decononvolved image on the left . one can clearly see a sharp increase corresponding to the eastern limb in the blue and yellow profiles , while the red excess indicates the existence of the north spur . similarly , the eastern limb and north spur are clearly visible in the profiles taken from the images up to energies of @xmath136 kev ( see figure [ fig : deconprofall ] ) . the intensity profiles from the @xmath130 kev image also show statistically significant photon excess from the limb and spur , confirming the existence of these features to energies as high as @xmath132 kev . the flux in the @xmath138 kev band from the wind nebula generated by psr j1833@xmath891034 completely dominates the pulsar itself . the pulsar is barely resolved from the pwn at arcsecond resolution by _ chandra _ and completely swamped by the pwn emission for other x - ray telescopes . all previous searches for x - ray pulsation have been unsuccessful despite extensive x - ray data sets collected with sufficient timing resolution . given the evidence of synchrotron burn - off above @xmath139 kev presented in section 4.4 , the extended energy band of _ nustar _ presents a new opportunity to further isolate the pulsar signal from the pwn . to search for the signal from psr j1833@xmath891034 , we initially selected photons in the energy range @xmath140 kev from a small source extraction aperture ( @xmath141 ) ( see figure [ fig : rve ] ) . we searched these photons for significant power around the expected period using the two derivative radio timing ephemeris presented in ( * ? ? ? * epoch 2009 ) extrapolated to the _ nustar _ observation epoch . this solution is preferred over the slightly updated five derivative model of @xcite whose extrapolated behavior is not predictive . lacking a coeval ephemeris , it is not possible to maintain phase unambiguously over the 206 day gap between the two _ nustar_observations . instead , we separately search data collected in 2012 and 2013 which span @xmath142 d and @xmath143 d , respectively . taking into account the increased uncertainty in the timing parameters for the extrapolated ephemeris , we searched for a significant signal over a frequency range of @xmath144 , where @xmath145 is the uncertainty in the frequency measurement , oversampled by three times the fourier resolution . we evaluate the power at each frequency using the @xmath146 test statistic for @xmath147 , to be sensitive to both broad and narrow pulse profiles , possibly single or double peaked . the most significant signal in this search range was @xmath148 and @xmath149 , using 6.4 kcts and 13.2 kcts , respectively , for the 2012 and 2013 observations . this corresponds to a significance of @xmath150 and @xmath151 after taking into account the number of search trials , @xmath132 and @xmath152 , for the two observations . we repeated our search for a additional combination of energy ranges @xmath153 kev , @xmath154 kev and aperture size with radius @xmath155 but find no signal with a greater significance than the initial @xmath156 result . we conclude that no pulsed x - ray signal in detected from psr j1833@xmath891034 in the optimal _ nustar _ band and place an upper limit at the @xmath157 ( @xmath158 ) confidence level on a sinusoidal signal pulse fraction of @xmath159 and @xmath160 ( including unknown pwn emission ) for the two observations , respectively . _ nustar _ is not able to independently measure the spectrum of the pulsar . however , @xcite were able to isolate and fit the spectrum of a @xmath161 region located at the cite of the radio pulsar . using their best - fit non - thermal spectrum we can approximate the contributions of the pulsar and pwn in the _ nustar _ region . this increases the pulse fraction to @xmath162 and @xmath163 , respectively . _ chandra _ observations of 21 show spectral softening over the pwn , with the photon spectral index ranging from @xmath164 at the inner ( radius @xmath165 ) region to @xmath166 at the outer @xmath167 radius annulus . _ nustar _ observations confirm this by showing spectral softening below @xmath168 kev , as shown in table [ tab : specfit ] . the broad _ nustar _ psf causes some mixing of the spectra in different annuli , and as a result the variation of @xmath169 with radius obtained by _ _ is less pronounced than that seen by _ chandra_. _ _ finds a spectral break at @xmath170 kev in the integrated pwn emission . the spectral break is also observed in the inner regions with radius @xmath171 and in an annulus from @xmath167 . the break is statistically significant at radii @xmath172 . above @xmath44 kev the photon spectral index remains constant . there are several possible origins for the spectral break seen by _ nustar_. one possibility is that the _ nustar _ psf mixes the radially dependent power - law indices seen by _ chandra _ @xcite , and softens the spectrum in such a way as to cause a sharp break . however , simulations that fold _ chandra _ maps through the _ nustar _ response indicate that this is not the case . the effects of the large ism extinction @xcite again would not cause such a sharp , defined energy break . similarly , the loss effect of dust scattering at lower energies is negligible ( below @xmath173)in the _ nustar _ band because of the relatively low column absorption . contributions from the pulsar are also not likely to be responsible . most pulsars have a harder spectrum than the pwn they power . while this could in fact cause a sharp transition between photon indices , it would cause spectral hardening with energy , not the softening seen by _ nustar_. finally , spectral breaks are often attributed to synchrotron cooling , as proposed by @xcite . however , the break energy of 9 kev would require an unreasonably low magnetic field strength of @xmath174 @xmath175 , compared to the @xmath176 @xmath177 g derived from equipartition arguments @xcite . the prominent spectral steepening between radio and x - rays , if interpreted with simple cooling models , yields magnetic field strengths ranging from @xmath178 @xcite to @xmath179 @xcite . it is likely that the spectral break results from physical effects , either due to a break in the injected electron energy spectrum or due to energy losses due to particle transport in the pwn . pulsars emit pairs of relativistic electrons and positrons , which are accelerated at a termination shock near the pulsar itself . downstream of the termination shock the accelerated electrons interact with the magnetic field , also produced by the pulsar , and subsequently emit synchrotron radiation from the radio through gamma energy bands . the injection spectrum can therefore shape and influence the spectrum of the synchrotron radiation , as noted by @xcite . these authors propagate a broken power - law electron spectrum through a time - dependent model that includes energy losses due to synchrotron radiation , inverse compton scattering , and adiabatic cooling . _ provides confirmation that the relatively simple injection spectrum used in @xcite is not adequate to fit the observed x - ray data , as seen in figure [ fig : sed ] . the model , noted as the solid black line , has a steep negative slope in the x - ray band and does not fit the x - ray spectra obtained from _ chandra _ , _ xmm - newton _ , _ integral_/ibis , and now _ nustar_. it is therefore reasonable to explore whether a more complex model can explain the 9 kev spectral break . @xcite extend the models with the addition of diffusive losses as well as a broken injection spectrum with a discontinuity at the break energy . however , both the aforementioned sed models do not include the spatial dependence of parameters such as the magnetic field of the pwn , which provides additional complexity and can perhaps better fit the x - ray data . we explore these effects in the following sections . particle transport in pwne has long been a matter of discussion and study . unfortunately , few celestial objects are both bright enough and large enough to allow distinguishing radially - dependant features to fit to the various existing models . the crab nebula , 21 and 3c 58 are such pwne . they have been frequently observed and analyzed , and have provided valuable insights into the physics of magnetohydrodynamic ( mhd ) flows and relativistic shocks that govern the appearance of pwne ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . extending the energy range of analysis will allow for more detailed probing of radially - dependant physics . this information is important both to understand pwne and to isolate the properties of the relativistic shock at the pwn inner edge from the subsequent spectral evolution downstream . we shall ask : can the observable parameters of 21 from radio to x - ray be reproduced assuming the shock injects only a single structureless power - law particle spectrum ? one such parameter used as a spectral fingerprint of various models is the radial dependency of the power - law photon index . numerous observations have confirmed softening of the 21 pwn spectrum with increasing projected radius @xcite . this is shown at higher energies for the _ nustar _ observations in figure [ fig : rve ] . the softening is associated with synchrotron cooling of the electrons , and the corresponding decrease in the maximum emitted x - ray energy in the bulk velocity flow downstream of the termination shock . an alternative approach is to characterize the variation of source size with photon energy using a characterization of the ` cooling length ' @xmath125 , such as the source fwhm . since only the energy - dependence of this length is important for modeling the spectral steepening , its precise definition is not important . this scale depends only on mapping the total number of photons as a function of radius and energy to determine the scale length @xmath125 . this is more straightforward than spectral modeling because counting statistics are almost always limited , and length scale measurements do not require determining a spectrum at each radius . we therefore use the cooling length scale , with fwhm as its surrogate , as the fundamental diagnostic for extracting information about physical conditions . we shall also make use of the observed steepening from radio to x - rays of 21 through the parameter @xmath180 . most discussions of pwn physics ( e.g. * ? ? ? * ) simply take this as an intrinsic property , but we shall attempt to explain it through evolutionary effects . there exist two prominent mechanisms that have been invoked to explain pwne particle transport : advection and diffusion . beginning with @xcite and @xcite , diffusion has long been investigated as a cause for the characteristics of a pwn . however , the early models of particles propagating outwards from a central source purely by diffusion , applied exclusively to the crab nebula , have been unable to account for the detailed x - ray properties of the crab , such as its change of size with frequency @xcite . pure advection models have also been proposed , such as @xcite . the canonical theory involving pure advection as the method of particle transport was presented by ( * ? ? ? * hereafter kc84 ) . when kc84 was used to predict the radial behavior of the spectrum , the resultant spectral photon index has little to no variation from the center of the nebula outwards , and begins to vary only towards the pwn periphery ( * ? ? ? * hereafter tc12 , figure 2 ) . this also does not match the observed behavior of a slowly - steepening x - ray spectrum @xcite . tc12 provide a nuanced approach to particle transport by diffusion and present two updated models . they claim that the magnetic field is not predominately toroidal far from the termination shock , as is often assumed in pwn theory , but has a more complex geometry with cross - field transport best described by diffusion . their first model consequently incorporates pure diffusion only , with both the magnetic field and diffusion coefficient constant with radius , and synchrotron emission as the only loss of energy . tc12 argue that such a model better explains the radial dependence of @xmath169 as seen in the crab , 3c58 , and 21 , with proper adjustment of the diffusion coefficient . involving complexities such as an energy - dependent diffusion coefficient might be more physically reasonable , but the data are not sufficiently constraining to distinguish these cases from a simple diffusion model . while the pure diffusion model of tc12 appears to describe observations of the crab and 3c 58 relatively well , we argue that it is less appropriate for 21 . it does provide a good description of @xmath181 for the crab and 3c 58 ( @xmath182 ) , however the fit of this model to the 21 @xmath183 kev data is poor ( @xmath184 ) , as seen in figure 6 in tc12 . in addition , the ratio of advective to diffusive timescales that determines what process dominates is radially dependant . using @xmath185 downstream of the pwn termination shock ( kc84 ) , and the know relationships @xmath186 and @xmath187 , we obtain @xmath188 . 21 is very compact , with a size more than two times smaller than that of the other two pwne , and thus advection is likely to dominate . finally , the advective model provides a good fit to the energy dependent cooling length scale ( see below ) . tc12 presented the radially - dependant spectral index @xmath181 rather than using the cooling length scale @xmath125 as characteristic of their model , which makes fitting the model to higher energy _ data too difficult . the second model presented by tc12 involves a monte carlo simulation that includes both diffusion and advection transport close to and farther away from the pulsar , respectively . this allows a more complex treatment than previous analytical models @xcite . while likely a more appropriate approach for 21 , tc12 only applied this simulation to the crab and 3c 58 . kc84 provide a complete advective solution for a steady , spherically symmetric wind terminated by a magnetohydrodynamic ( mhd ) shock . although kc84 is routinely applied to determine quantities such as the mean downstream magnetic field , its range of applicability is in fact more narrow . kc84 represents an idealized theory , suited to the case of constant injection of electrons in a spherical geometry , transported outward by pure advection in an ideal mhd flow with an ordered , toroidal magnetic field . the model does not attempt to reproduce the crab spectrum from radio to x - rays , but only the optical to x - ray portion , with an injection spectral index of optically emitting electrons of @xmath189 . the predicted value of steepening of @xmath190 of 0.51 is roughly appropriate ( so their cooling break is at uv wavelengths ) . this value is fortuitously close to the value of 0.5 for a stationary , homogeneous source . thus this model also can not reproduce radio - to - x - ray seds of many other pwne , including 21 , which have a larger @xmath191 . as mentioned above , the @xmath181 relationship predicted by the kc84 model is flatter in the pwn interior and softens towards the edges , as shown by @xcite , while generally a gradual variation in @xmath181 is observed . this has motivated several generalizations of kc84 . the values we obtain from the _ nustar _ data for @xmath125 and @xmath192 are clearly inconsistent with the model presented in kc84 . the gradients implied by the assumptions of kc84 , predict @xmath193 , independent of spectral index , and @xmath194 for the physically important inner flow region . if we attempt to apply the kc84 formalism to describe the radio - to - x - ray spectrum of g21.5 - 0.9 with @xmath195 and @xmath196 , we fail on both counts . first , we find @xmath4 ( @xmath197 ) . second , we find the observed spectral indices in the radio and x - ray bands produce @xmath198 , instead of 5/9 from a kc model using the injected spectral index of @xmath199 . * hereafter r09 ) noted that there are a number of physical effects not accounted for in kc84 that could produce a steeper spectral break than @xmath200 . for example , the magnetic field may have a significant radial or turbulent component , or it may not satisfy mass conservation due to cloud evaporation , or it may not have magnetic flux conservation ( e.g. due to magnetic reconnection or turbulent amplification ) . r09 constructed a simple model that includes these effects . the model involves generating simple scaling relations for the downstream magnetic field @xmath201 , fluid velocity field @xmath202 , and fluid density @xmath203 in terms of the dimensionless length scale @xmath204 where @xmath205 is the inner injection radius . the non - spherical geometrical effects could also be parameterized in terms of jet width @xmath206 where @xmath207 corresponds to a conical jet or a section of spherical outflow ; a confined jet has @xmath208 , while a flaring jet would have @xmath209 . by assuming @xmath201 , @xmath202 and @xmath203 all vary as power - laws in @xmath210 with indices @xmath211 , @xmath212 and @xmath213 , r09 obtains a series of general consistency relations that these indices must satisfy with each other and with observable parameters . these variables of @xmath214 , @xmath215 and @xmath191 , the energy scaling of the cooling length , particle injection index and spectral index break , respectively , provide constraints on the allowed values of @xmath216 , @xmath217 , @xmath218 , and @xmath219 . the results of r09 can be rewritten in terms of the functional form of @xmath214 , and assuming @xmath220 as observed for g21.5 - 0.9 : @xmath221 independent of @xmath218 . since we observe @xmath222 and @xmath223 , this gives @xmath224 . the possible solutions are thus very restrictive for our measured value of @xmath191 and @xmath214 . additionally , if the flow conserves mass ( disallowing , for instance , mass loading by evaporation of thermal material ) , then the density index @xmath219 is linked to the velocity index @xmath218 . in this case , some solutions are unphysical , such as those with @xmath225 , corresponding to an accelerating downstream fluid . with some judicious rejection of such solutions we can draw some interesting general conclusions , based on the possible values of the observables , which may guide further investigations . for a conical or spherical flow , @xmath226 , so either density or magnetic - field strength , or both , must _ rise _ with radius ( since either @xmath219 or @xmath217 is positive , or both ) . mass conservation links @xmath219 to @xmath218 , but if that assumption is abandoned , there is no relation and no constraint on the velocity profile only the density profile matters for observable quantities . a steeply decelerating flow can produce @xmath227 with mass conservation , but mass loading can also do this . similarly , @xmath228 can be a reasonable outcome of flux non - conservation through processes such as reconnection . our observations require one or both of these effects : addition of mass to the flow through some kind of evaporation , and increase in magnetic - field strength beyond flux freezing . a strongly confined jet ( @xmath208 ) can relax some constraints ; for @xmath229 , we only require @xmath230 . however , even here either mass or flux nonconservation is necessary . one can construct constant density solutions ( @xmath231 ) but such a geometry is disfavored due to the high symmetry of 21 as evidenced in both the broad axial symmetry observed in the radio @xcite and in the soft x - ray band @xcite . while @xcite expands on the treatment of kc84 , both fail to reproduce the steadily steepening spectrum with radius shown by g21.50.9 and other pwne . this shortcoming is characteristic of models in which particles are transported outwardly purely by advection , so that all particles at a given radius have similar ages . to produce steady spectral steepening probably requires a mixture of particles of different ages at each radius . this could be caused by more complex fluid flow such as the back flows found in simulations by @xcite , or by particle diffusion . we conclude that a model describing both the radio - to - x - ray spectrum of 21 and the size shrinkage with x - ray energy we observe can be accommodated in a pure advection model requiring the injection of only a straight power - law spectrum of electrons , @xmath232 . however , as with all pure advection models , the gradual rather than sudden steepening of the spectrum with radius is not reproduced . the viability of this explanation for the observed properties of 21 will depend on whether the addition of diffusion can reproduce the gradual steepening while preserving the successes of the advection model . _ nustar _ has detected , for the first time , the north spur and eastern limb above 10 kev . three main theories have been proposed to explain the nature of these features : they are extensions of the pwn itself , they are limb - brightened shock fronts propagating into surrounding ism and accelerating cosmic rays , or they result from an interaction of ejecta with the envelope of the progenitor snr , presumably a type iip supernova @xcite . since the north spur and eastern limb have different spectral and spatial properties , we discuss them separately . a multi - wavelength analysis is required to get a full understanding of the north spur . this feature was observed in the radio and the soft x - ray , most recently by @xcite and @xcite , respectively . @xcite reported a radio detection of the north spur , with a 1.43 ghz flux density of @xmath233 mjy , and a fwhm size of @xmath234 . this is notable because it is the only feature , other than the pwn itself that is detected in the radio band . the eastern limb has no radio emission detected to date . @xcite obtained @xmath235 ks of _ chandra _ data , and found that the north spur has a spectrum comprised of non - thermal and thermal components . the thermal component , represented by the * pshock * model , is best - fit to temperature of @xmath236 kev and contributes only @xmath237 to the overall @xmath238 kev flux . the non - thermal component , however , is equally well - fit by either the * srcut * or * powerlaw * model . the model * srcut * describes the synchrotron emission from a homogeneous source consisting of a power - law energy distribution of electrons with an exponential cutoff , radiating in a uniform magnetic field . the emitted spectrum is a power - law that steepens slowly above the photon energy corresponding to the electron cutoff energy . this slow curvature can mimic a steeper power - law in a limited energy band . however , in a broader energy band the * srcut * model can fall well below the extrapolation of a power - law with the same slope at lower energies . since x - ray emission is visible from the north spur at energies as high as @xmath132 kev , the correct spectral model must provide a photon flux from @xmath130 kev that is statistically higher than the background . due to signal - to - noise limitations we can not use _ nustar _ data to spectrally fit the spur , however we can use the deconvolved images to distinguish between the * srcut * and * powerlaw * models . we simulated spectra with these models using the _ fakeit _ command in xspec using the model parameters reported by @xcite . a photon spectral index of @xmath239 was applied to the * powerlaw * model , while the * srcut * model used a radio index of @xmath240 and a rolloff frequency of @xmath241 hz . the _ nustar _ response files were based on a point source extraction with @xmath242 . we were thus able to obtain fluxes from each respective model from @xmath129 kev , @xmath130 kev , and from @xmath135 kev . both models have count rates higher than that of the background within the @xmath129 kev energy band . in the @xmath130 kev band , the * powerlaw * model has a count rate 7 times higher than the background , while the * srcut * model is only 3 times higher . finally , within the @xmath130 kev band , the count rates are 4 and 1.5 higher than the background for the * powerlaw * and * srcut * models , respectively . this implies that the * srcut * model should be marginally visible up to @xmath132 kev , and above @xmath132 kev should have a count rate equal to that of the background . this matches well with what is seen in the _ nustar _ images . the * powerlaw * model , however , should be detectable at energies as high as @xmath243 kev . if the north spur were an extension of the pwn , it would have a spectral photon index similar to that of the pwn itself . while this is true , the analysis above indicates that the north spur is not described by a * powerlaw * model extending to higher energies . if it were , the north spur would easily be detectable at energies as high as @xmath243 kev . however , the _ nustar _ images with a combined exposure of 281 ksec do not detect any statistical emission above @xmath132 kev , indicating that * srcut * is the more plausible spectral model for the north spur . it is possible that the north spur results from the interaction of the inner sn ejecta with the h - envelope of the progenitor @xcite . this is supported by the thermal component in the spectral fit of @xcite , a * pshock * model with solar abundances , low temperature of @xmath244 kev , and low ionization timescales . this is also supported by the morphology of the north spur itself . with the inclusion of projection effects , the north spur is located between @xmath245 and the edge of the sn shell at @xmath246 . x - ray emission from the shell of 21 is clearly visible in the _ nustar _ image . our results confirm the existence of this shell , revealed by _ chandra _ @xcite in both their image and in an extracted shell spectrum ( up to @xmath174 kev ) , but also hinted at in earlier _ xmm - newton _ data @xcite . the _ xmm - newton _ observations reveal evidence of the shell in a @xmath247 kev energy band image , after careful subtraction of a modeled dust scattering component below @xmath248 kev . the _ nustar _ detection extends up to much a higher energy of @xmath1 kev . the morphology of the _ nustar _ emission is striking in its similarity to the _ chandra _ and especially the _ xmm - newton _ image . emission is detected from pa @xmath249 to @xmath250 in the @xmath128 kev image ( pa = 0 at n , positive clockwise ) , with the extent shrinking as the energy increases until it is visible mainly in the east and north at the highest energies . this is consistent with the intensity distribution with position angle seen in the lower energy image . a question unresolved by previous x - ray observations is whether the shell emission is thermal or non - thermal . the extended energy response of _ nustar _ can be exploited to answer this question . @xcite found that the spectrum of the eastern limb could be characterized equally by four distinct models . a thermal fit to the * pshock * model provided a temperature of @xmath251 kev while a non - thermal * powerlaw * model obtained a spectral photon index of @xmath252 . two * srcut * models were also well - fit to the eastern limb spectrum . ideally * srcut * requires both a radio flux density and radio spectral index for the shell , but the eastern limb has not been detected in the radio . therefore , two values of the radio index @xmath215 that are reasonable for a sn shell were chosen ( @xmath253 ) while all other parameters were left free to vary . care was taken to ensure the best - fit radio flux was below the upper limit reported by @xcite . as with the north spur , we extrapolated the four _ chandra _ spectra for the eastern limb into the _ nustar _ band . we created an effective area file for an extended source shaped like the eastern limb , then used the _ fakeit _ command in xspec to simulate spectral data . the thermal * pshock * model predicted a shell flux which would not produce the x - ray emission seen by _ nustar _ at @xmath254 kev . in contrast , the three non - thermal models produced x - ray fluxes consistent with imaging of the eastern limb by _ nustar _ , although the * srcut * and * powerlaw * models could not be distinguished from each other . nonetheless , the _ nustar _ observations firmly establish the non - thermal nature of the shell x - ray emission . the detection of a non - thermal shell up to quite high x - ray energy in 21 is interesting , and is in contrast to observations of other crab - like supernova remnants . recently shells have been detected in 3c58 @xcite and g54.1 + 0.3 @xcite , and a clear , detached shell of emission in kes 75 @xcite . however the 3c58 shell is clearly thermal , with no sign of a non - thermal component . the shell of @xmath255 can be fit with both thermal and non - thermal models , however a thermally emitting shell seems much more likely . assuming thermal emission , @xcite were able to use pwn - snr evolutionary models to obtain an age consistent with the characteristic age derived from pulsar observations , obtain the proper supernova - pwn radius ratio , and predict that the reverse shock has not encountered the pwn yet , consistent with other observations . the ages of 3c58 and g54.1 + 0.3 are @xmath256 years and @xmath257 years , respectively . 21 is thus unique in that it is much younger ( @xmath258 years ) , and potentially has higher forward shock speed , both of which could lead to the detectable non - thermal shell . the crab nebula itself , of comparable age to 21 but much closer , still shows no non - thermal shell , presumably due to a very low interstellar medium density . the detection of this 21 non - thermal shell at quite high x - ray energy is thus likely due to its younger age compared to these other crab - like snrs . we have presented an analysis of a 281 ks _ nustar _ observation of 21 to probe the spatial and spectral characteristics revealed by high - energy x - ray emission . _ nustar _ s broad energy band makes it uniquely suited to analyze not only the pwn but observe the characteristics of the snr shell as well . the spectrum taken from the entire remnant is described by a best - fit broken power - law with a break energy of @xmath2 kev . this is the first instance where a single instrument was able to capture this break . the addition of the _ nustar _ spectrum to sed models from @xcite produces a poor fit to the x - ray data . this suggests that further modeling is required : more complex electron injection spectra , additional loss processes such as diffusion , or radial dependence of the pwn parameters . spectra extracted from various radial annuli were also fit with both an absorbed power - law and an absorbed broken power - law model . the two regions @xmath259 and @xmath260 are statistically better fit with a broken power law with @xmath261 kev , while the regions with radii @xmath262 are best - fit with a single power law . we observe spectral softening of the spectral index below the spectral break , while the spectral index above the break is constant within uncertainties . image analysis allows us to measure the energy - dependent cooling length scale and fit the relationship with a power - law model of @xmath124 . this yields an index of @xmath126 . incorporating this with the spectral indices in both the radio and x - ray bands , we are able to systematically approach the equations of @xcite and inspect the parameter space for physically consistent solutions . we found that , for a conical jet or spherical outflow , the most reasonable solutions do not conserve magnetic flux but do conserve mass , indicating turbulent magnetic field amplification . the bulk velocity decelerates steeper than that predicted by kc84 . we detect the eastern limb and north spur at energies above 10 kev for the first time . a deconvolution method provids clear evidence of emission from the north spur up to @xmath132 kev . extrapolation of the spectral fits obtained by _ chandra _ show that the * srcut * model is favored over the * powerlaw * model . this further solidifies the assumption that the north spur is an interaction of the sn ejecta with the remnant . we also detect the eastern limb up to @xmath132 kev . we have confirmed the existence of non - thermal emission from the limb , and conclude this faint feature is the sn shell of 21 . we are unable to distinguish between the non - thermal models fit to the eastern limb by _ chandra_. this work was supported under nasa contract no . nng08fd60c , and made use of data from the _ nustar _ mission , a project led by the california institute of technology , managed by the jet propulsion laboratory , and funded by the national aeronautics and space administration . we thank the _ nustar _ operations , software and calibration teams for support with the execution and analysis of these observations . this research has made use of the _ nustar _ data analysis software ( nustardas ) jointly developed by the asi science data center ( asdc , italy ) and the california institute of technology ( usa ) . , o. c. , ferreira , s. e. s. , & djannati - ata , a. 2008 , in american institute of physics conference series , vol . 1085 , american institute of physics conference series , ed . f. a. aharonian , w. hofmann , & f. rieger , 199202 , a. , mcdowell , j. c. , allen , g. e. , et al . 2006 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 6270 , society of photo - optical instrumentation engineers ( spie ) conference series

the design , optimisation and construction of an anti - coincidence veto detector to complement the zeplin iii direct dark matter search instrument is described . one tonne of plastic scintillator is arranged into 52 bars individually read out by photomultipliers and coupled to a gadolinium - loaded passive polypropylene shield . particular attention has been paid to radiological content . the overall aim has been to achieve a veto detector of low threshold and high efficiency without the creation of additional background in zeplin iii , all at a reasonable cost . extensive experimental measurements of the components have been made , including radioactivity levels and performance characteristics . these have been used to inform a complete end - to - end monte carlo simulation that has then been used to calculate the expected performance of the new instrument , both operating alone and as an anti - coincidence detector for zeplin iii . the veto device will be capable of rejecting over 65% of coincident nuclear recoil events from neutron background in the energy range of interest in zeplin iii . this will reduce the background in zeplin iii from.4 to.14 events per year in the wimp acceptance region , a significant factor in the event of a non - zero observation . furthermore , in addition to providing valuable diagnostic capabilities , the veto is capable of tagging over 15% for-ray rejection , all whilst contributing no significant additional background . in conjunction with the replacement of the internal zeplin iii photomultiplier array , the new veto is expected to improve significantly the sensitivity of the zeplin iii instrument to dark matter , allowing spin independent wimp - nucleon cross sections below 10 pb to be probed .

1004.4207

zeplin iii @xcite is a two - phase ( liquid / gas ) xenon detector developed to observe low energy nuclear recoils resulting from the elastic scattering of galactic weakly interacting massive particles ( wimps ) @xcite . a key feature of zeplin iii is the ability to discriminate between incident particle species for each event by recording both vacuum ultraviolet scintillation light and electroluminescence from ionisation of the xenon target @xcite . the ratio in signal strength from these channels differs for electron and for nuclear recoil interactions , allowing the efficient rejection of most background events . however , in the absence of an electromagnetic interaction , weak and strong elastic scattering events appear identical , making the rejection of single scatter background neutron events difficult @xcite . additionally , even with good discrimination , for large exposures a small fraction of electron - recoil events can be misidentified as nuclear recoils . with predicted wimp event rates of less than 0.1 events / kg / day , and with energy depositions of the order of 10 kev , dark matter detectors such as zeplin iii must reduce the number of background events , especially due to neutrons , by a very large factor . typically this is achieved through the use of high purity components of low radionuclide content , hydrocarbon and lead shielding external to the detector to attenuate and moderate neutrons and @xmath1-rays originating in the local environment , as well as the operation of the detector at a site with large rock overburden to reduce the cosmic muon flux . having been manufactured from low background components such as oxygen free copper , shielded with 30 cm of high density polypropylene and lead , and situated at the boulby underground laboratory , uk , the background event rate of zeplin iii is already greatly reduced . sensitivity to spin - independent wimp - nucleon cross sections as low as 8.1@xmath310@xmath2 pb for a wimp mass of 55 gev / c@xmath4 has been demonstrated @xcite , as well as a spin - dependent wimp - neutron cross section as low as 1.8@xmath310@xmath5 pb @xcite . however , to improve the sensitivity of the device further , zeplin iii is being retrofitted with an active veto detector . since wimps are highly unlikely to interact within a detector system twice , any event doing so may be discarded . following this upgrade , the predicted combined background event rate in zeplin iii translates into much less than one count within the wimp acceptance region for a one year dataset . the implementation of this veto system , in conjunction with new custom - built ultra - low background photomultiplier tubes ( pmts ) , will allow zeplin iii to probe spin independent wimp - nucleon cross sections below the 10@xmath2 pb level . experimental measurements have been made of the individual performance and radiological content of all components . these results have been fed into an end - to - end monte carlo simulation of the zeplin iii instrument @xcite , coupled with the entire veto detector array located in the boulby mine , to evaluate the overall performance . an iterative process has been followed in which , with the geometry of the scintillator defined , alternative additional components and light collection efficiency were evaluated , informed by further experimental data where necessary . in the following sections we first describe the general design of the veto . we then provide details of the measurements of the radioactivity content of the components , the performance of components , and the optimisation process . finally we provide results from the evaluation of the predicted overall performance of the veto . figures [ vetofig1 ] and [ cad ] depict schematically the veto system , while table [ table ] presents key parameters of the overall system and its components . it consists of 52 individual bars of plastic scintillator arranged so as to provide @xmath63@xmath7 sr coverage around zeplin iii . thirty - two bars are 100 cm long , 15 cm thick parallelepipeds with trapezoidal cross section , that form a closed ring of outer diameter 160 cm ( the ` barrel ' ) . these are arranged to stand on a continuous 30 cm thick base piece of passive polypropylene shielding . a single 3-inch pmt is optically coupled to one end of each bar . pmts are a likely significant source of background radioactivity . as such they are positioned on the lower face of the scintillator bars , in recessed holes in the base piece , thus keeping them farther from the zeplin iii xenon target . three dimensional rendering of the zeplin iii veto detector , showing only the plastic scintillator bars with attached pmts . the diameter of the barrel is 160 cm , width=325 ] cross - sectional view of the zeplin iii experiment in its second science run configuration . in the centre is the zeplin iii detector ( labelled a ) showing the copper vacuum vessel enclosing the xenon target and the ln2 vessel below it . forming a barrel around zeplin iii are the 32 gd - loaded polypropylene pieces and roof plug ( labelled b and hatched ) . surrounding these are the active scintillator modules ( c ) with pmts housed in cups and recessed into the lower polypropylene structure . the 20 roof modules of scintillator ( d ) rest on the roof plug . the lower polypropylene structure ( e ) contains no gd and rests on a copper and a lead base ( f ) . finally , the lead castle ( g ) envelopes the entire assembly.,width=362 ] .[table]key parameters of the zeplin iii veto . [ cols="<,^ " , ] between the scintillator and the zeplin iii instrument are positioned 32 pieces of 15 cm thick gadolinium ( gd ) loaded polypropylene forming a barrel of outer diameter 130 cm . the scintillator and polypropylene barrels are rotated with respect to one another to ensure no line - of - sight gaps to the detector . polypropylene has been chosen because of its high hydrogen content , machinebility , and low radioactivity @xcite . for ease of manufacture the gadolinium is added to the polypropylene in the form of 10 @xmath8 m gd@xmath9o@xmath10 powder mixed in to rutherford type 71 epoxy @xcite . this mixture is set within 2 mm wide , 10 mm pitch vertical slots throughout the polypropylene . monte carlo simulations and experimental tests indicate that for neutron capture there is negligible difference between a uniform distribution and a discrete pattern , provided that the pitch of the slots does not exceed @xmath1115 mm ( see section [ pitchmc ] below ) . the polypropylene and the scintillator parallelepiped bars are secured together to provide mechanical stability . monte carlo simulations have also been employed to confirm that mm - sized voids between or within attached standing scintillator modules result in neither a significant loss in neutron detection efficiency nor increased exposure to background for zeplin iii ( see section [ voidsmc ] below ) . a 15 cm thick , 115 cm diameter disc of gadolinium - loaded polypropylene forms a roof piece , with the remaining 20 plastic scintillator bars resting on it to form a roof that extends over the full diameter of the barrel modules , as shown in figure [ vetofig1 ] . a single pmt is optically coupled to the outer end of each roof bar , again maintaining maximum distance between veto pmts and the xenon target . the combined thickness of the passive and active hydrocarbon shielding is 30 cm , the same used in the first zeplin iii science run @xcite . the veto system is itself enclosed within the existing 20 cm thick pb shielding of the zeplin iii instrument ( see figure [ cad ] ) . the geometry described above provides active rejection of background as follows . firstly , neutrons entering the hydrocarbon shielding are efficiently moderated to thermal energies , mostly via h scattering , and undergo radiative capture predominantly on @xmath12gd ( natural abundance 15.7% ) which has an extraordinarily high capture cross section of 2.4@xmath310@xmath13 barns @xcite . the level of gadolinium loading required was explored and is described in section [ pitchmc ] , where a value of at least 0.5@xmath14 by weight , averaged over the polypropylene , is determined . the neutron capture is accompanied with the emission of typically 34 @xmath1-rays of energy totalling @xmath118 mev , and at a loading of 0.5@xmath14 these are emitted with a mean delay of 35 @xmath8s from the time the neutron enters the loaded material . this is advantageous over a pure hydrocarbon system where radiative capture would occur after much longer periods of time and would lead to the release of a single 2.2 mev @xmath1-ray . thus , with the gadolinium - loaded components located within the plastic scintillators , there is a very high probability of recording at least one significant energy deposition . such a design exposes the scintillator to an increased rate of neutron background from the environment relative to zeplin iii , thereby reducing diagnostic capability , but greatly increases the efficiency as an anti - coincidence detector for wimp - like neutron recoils in zeplin iii the primary function of the device . in the default mode of operation , data acquisition for the veto will be triggered predominantly by energy deposition in zeplin iii , with event - by - event veto waveforms recording for 20 @xmath8s before and 300 @xmath8s afterwards , allowing for offline searching of coincidences . similarly , @xmath1-ray emission , directly from components within zeplin iii or otherwise , can lead to energy deposition in the scintillator . this allows coincident background events within zeplin iii , predominantly from compton scattering , to be identified . in addition to operating as an active veto , this device will provide significant diagnostic capability for zeplin iii , using independent triggering and providing a measurement of cosmic - ray muon fluxes , enhanced calibration capabilities and , by being a large volume low threshold device , permitting independent measurements of the neutron and @xmath1-ray background environment in the laboratory around zeplin iii . the veto design has several further advantages . unlike with a liquid scintillator , where there is a possibility of leaks developing , the plastic veto scintillator bars are less of a chemical and fire hazard . additionally , the modular design structure means that individual modules can easily be removed and repaired or modified if required without disturbing the remaining detectors or systems . as a diagnostic device , since signals generated in the veto are highly unlikely to appear in more than a few modules , a degree of directional reconstruction is possible , especially for cosmic - rays . thirdly , with the scintillator on the outside of the polypropylene , the veto has the ability to run in several modes simultaneously , with , for example , continuous recording of background as well as timelines where zeplin iii has triggered , regardless of whether the veto has detected an event . finally , the response of the entire system can be modelled accurately by incorporating the measured performance of each unit into a simulation , as opposed to the use of global parameters applied across a large volume and mass . at a vertical depth of 1070 m ( 2850 m water - equivalent shielding ) , the cosmic ray muon flux is reduced by a factor of @xmath1110@xmath15 to a level of ( [email protected])@xmath310@xmath2 muons cm@xmath5 s@xmath18 @xcite . the resultant neutrons from cosmic ray muon spallation and secondary cascades lead to a neutron scattering rate in zeplin iii of @xmath191 event / year for a nuclear recoil energy above 10 kev within the central 8 kg of xenon @xcite . thus , the most important background that must be considered comes from the local environment of the laboratory and the instruments themselves . consequently , all components proposed for use in the veto have been assayed for their radiological content . the main contributions come from radioisotopes along the u and th decay chains , which produce background neutrons via ( @xmath20,n ) reactions on other materials and from spontaneous fission , as well as numerous @xmath1-rays . the @xmath21-decay of @xmath22k , which contaminates natural potassium at 100 ppm concentrations , generates 1.461 mev @xmath1-rays . to assess the overall neutron and @xmath1-ray environment that the instruments will be exposed to , additional measurements of many other components have also been considered , for example the zeplin iii instrument pmts and the cavern rock . two principal methods have been employed to determine the radioactivity content of components . l c c c c component & mass , kg & + & & u ( ppb ) & th ( ppb ) & k ( ppm ) + + _ hpge measurements _ & & & & + plastic scintillator & 1057.0 & [email protected] & [email protected] & [email protected] + ptfe inner wrap & 8.9 & [email protected] & [email protected] & [email protected] + silicone & 0.1 & [email protected] & [email protected] & [email protected] + ptfe tape & 3.1 & [email protected] & [email protected] & [email protected] + veto pmts & 6.2 & [email protected] & [email protected] & [email protected] + pmt preamplifiers & 0.7 & [email protected] & [email protected] & [email protected] + pmt base & 5.5 & [email protected] & [email protected] & [email protected] + epoxy & 70.0 & [email protected] & [email protected] & [email protected] + gd oxide & 8.0 & [email protected] & [email protected] & [email protected] + + _ icp ms / oes _ & & & & + copper tape & 26.0 & [email protected] & [email protected] & [email protected] + ptfe inner wrap & 8.8 & [email protected] & [email protected] & @xmath194 + veto pmts & 6.2 & [email protected] & [email protected] & [email protected] + pmt preamplifiers & 0.7 & [email protected] & [email protected] & [email protected] + pmt base & 5.5 & [email protected] & [email protected] & [email protected] + polypropylene & 510 & @xmath191 & @xmath191 & @xmath195 + pmt mounting & 15.8 & [email protected] & @xmath1910 & @xmath1910 + cabling & 30.2 & [email protected] & [email protected] & [email protected] + connectors & 2.1 & @xmath1910 & @xmath1910 & @xmath194 + optical gel & 0.3 & @xmath191 & @xmath191 & @xmath191 + gd oxide & 8.0 & [email protected] & [email protected] & @xmath194 + direct measurements of the @xmath1-ray emission from candidate components has been made using a high purity germanium ( hpge ) detector located in a dedicated low background counting facility at the boulby mine . the hpge detector head was encased in an inner copper and outer lead castle leaving a ( 30 cm)@xmath23 test cell . the ambient @xmath1-ray flux experienced by the hpge detector in the absence of a test sample was then assessed through a week - long background run . a sample of material was placed close to the head and further data accrued over a period of several days . an example for a background - subtracted sample spectrum is shown in figure [ gespec ] , revealing photopeaks at energies corresponding to known @xmath1-ray emissions from u and th chains and from k as well as other dominant lines of interest . to assess the quantity of an isotope that must have been present in order to generate the measured excess in the background - subtracted spectrum , secular equilibrium was assumed and a geant4 @xcite monte carlo simulation of the low background counting setup performed . each simulation was individually tailored to the geometry of the sample being tested . sensitivities at the level of 0.1 ppb for u and th , and 0.1 ppm for k were achieved using this technique . the conversion between @xmath1-rays emitted from an isotope and the source contamination level is given using the following @xmath1-ray yields per kg@xmath24day : 2310 @xmath1-rays / ppb u , 958 @xmath1-rays / ppb th and 285 @xmath1-rays / ppm k. a summary of the activity levels found is presented in table [ radtable ] , together with the total mass of each component used in the construction . illustrative spectra from the underground hpge detector . in this case the sample being measured was a collection of three pmts , positioned close to the hpge detector head . shown is a background - subtracted exposure of 160 hours duration . the energies of @xmath1-ray lines of interest are indicated.,width=359 ] to supplement some of the @xmath1-ray measurements , and to provide a cross check of their accuracy , a subset of components was measured using different techniques at the geosciences advisory unit ( gau radioanalytical ) at the university of southampton . samples were dissolved and subsequently diluted in 2% hno@xmath10 . activity due to @xmath25th , @xmath26u and @xmath27u was measured using an inductively - coupled plasma mass spectrometer ( icp ms ) , while potassium content was assessed using an inductively - coupled plasma optical emission spectrometer ( icp oes ) . a calibration @xmath28u recovery tracer was added . the results from this process are also presented in table [ radtable ] . differences arise due to the assumption of secular equilibrium used to infer contamination levels in the direct @xmath1-ray measurements , as opposed to the direct measurement of atoms using mass - spectroscopy . in all cases , upper limits from either set of measurements ( hpge or icp ms / oes ) have been used as inputs for simulations , described in section [ mc ] , performed to assess the impact of radiological contamination in the veto components on zeplin iii as well as in the veto itself . the overall capability of the veto relies on the performance of individual components as well as on the geometry and methodology in which they are used . in this section we report on the technical performance of key components : the pmts , the plastic scintillator , the front - end electronics and the data acquisition system . the pmts used are model 9302 kb from et enterprises ltd ( etel ) . these are 78 mm diameter window devices with bialkali photocathode and 9 dynode stages . their envelopes are manufactured from low background glass , with manufacturer - quoted contamination levels of 30 ppb u , 30 ppb th and 60 ppm k. as described in section [ gemeasurements ] , experimentally measured values using a hpge detector in a low background environment confirm these levels . voltage divider networks ( etl c647bfn2 - 01 ) were fitted to low background b14a bases supplied by etel , with u , th and k contamination levels measured to be consistent with , or lower than , the manufacturer stated values of 13 ppb , 19 ppb and 21 ppm , respectively . to amplify the small signals from low energy events and to impedance - match the output of the pmt to the data acquisition digitisers , pre - amplifiers have been designed and manufactured that give a gain of @xmath1110 . the preamplifiers are tooled onto surface mount printed circuit boards made from light - weight low - background single - sided resin bonded paper . the pmts and their bases ( including the voltage divider network and the preamplifier board ) are secured within a pvc cylinder which is itself chemically bonded directly to the scintillator . springs mounted in the cylinder cap provide support for the pmt , pushing it into contact with the scintillator surface via bc-630 optical coupling grease . the pmts are operated with negative bias supplied by lecroy 1443nf boards within a lecroy 1440 hv high voltage system . all power supply and signal cabling for the photocathode - anode ( k - a ) hv , pre - amplifier power , and signal output , has been manufactured to order with minimal radiological contamination . single photoelectron ( spe ) integrated pulse height spectra from dark thermionic emission have been obtained for each pmt , an example of which is shown in figure [ spespec ] . the average dark count rate is found to be approximately 300 hz . by design , the 10 mhz sampling frequency of the data acquisition is sufficient for good recovery of spe signals , allowing these to be used routinely for relative gain characterisation . these spectra were taken with no scintillator or other light sources present . for these measurements a single reference pmt was used assuring environmental and electrical uniformity across tests , with both the reference and test pmt housed in a light - tight box . the bias voltage of the test pmt was varied until the spectral peak position observed from single photoelectron emission from the photocathode was matched to that of the reference pmt . this was repeated for all pmts . a typical single photoelectron spectrum ( solid line ) , with a partial gaussian fit to the peak ( dashed line ) , obtained from a pmt in the absence of a light source and with no plastic scintillator attached.,width=325 ] with pmt bias voltages adjusted to yield normalised spe spectra , the quantum efficiency ( qe ) and gain of each pmt was calibrated . this was performed by illuminating the test pmt photocathode with a blue led to stimulate emission from the photocathode . the number of photoelectrons observed was then calculated using two different methods . in the first method , a statistical approach is adopted where the number of photoelectrons detected is determined from the width of the distribution generated . the width of the peak seen from the led is dominated by the poissonian statistics of the photoelectron emission , _ i.e. _ @xmath29 where @xmath30 is the measured standard deviation , @xmath31 is the centroid of the peak , and @xmath32 is the mean number of photoelectrons emitted for events in the peak . this will underestimate the total number of photoelectrons from a pmt , since it does not account for additional contributions to @xmath30 , the most important being the width of the spe spectrum itself . however , as was the case here , this can be mitigated somewhat by generating a large number of photoelectrons such that the width of the distribution is much greater than the width of the spe spectrum . the second and more traditional method is to calculate the number of detected photoelectrons by simply dividing the peak position of the led spectrum by that obtained from the single photoelectron response from dark spontaneous emission . the measured response was then normalised to absolute qe when 3 out of the 52 pmts were absolutely calibrated at etel . the distribution of qes , as plotted in figure [ qe ] , shows good agreement between both methods used for estimating the number of photoelectrons . the gain normalised measurements confirmed the relative qes between pmts for sensitivities of 50 a / lm and 200 a / lm as quoted by the manufacturer . the mean qe at 360 nm is 30.1% and at 420 nm is 27.1% . the quantum efficiencies of the pmts used in the zeplin iii veto . two methods have been used giving broadly consistent results . three pmts spanning the range of performance were assessed for absolute quantum efficiency externally ; these are indicated by the filled symbols . a linear fit through the data points is drawn to highlight the bias between the two methods used.,width=325 ] the 1057 kg of scintillator material is polystyrene based ups923a ( p - terphenyl 2% , popop 0.02% ) produced by amcrys h , kharkov , ukraine @xcite . this material emits scintillation light with a peak intensity at 420 nm and with a rise time of 0.9 ns and decay time of 3.3 ns . the density of ups923a is 1.06 g @xmath33 and its refractive index is 1.52 . the nominal light output for the material is stated as 55% of anthracene . the transparency of the scintillator material is determined by the so - called bulk attenuation length ( bal ) which is the length that reduces the initial light intensity by factor e according to the buger-lambert law , and is expected to be greater than 1 m ( see later for measured values ) . a previous study by the manufacturers suggests this material to have long - term stability , such that there is no natural degradation reported over a 12 year period @xcite . to minimise loss of light within a scintillator bar each piece has been wrapped along its length with ptfe sheet giving it a high diffuse reflectivity . experimental tests were performed comparing a number of diffuse and specular reflective wrapping materials including white paper , aluminised mylar , vituki 3 m @xcite and other ptfe - based sheeting . three layers of 76 @xmath34 thick ptfe film @xcite was found to provide the most cost effective option , increasing light collection by up to a factor of 2 at the pmt . at the far end from the pmt a specular reflective aluminised mylar foil is placed that acts as a mirror to improve the uniformity and increase the effective bal of the unit . the pmt is optically coupled to the scintillator with bc630 silicone - based optical couplant and is housed in a pvc cylindrical tube chemically bonded to the scintillator . the entire assembly has then been wrapped in black opaque pvc sheet @xcite to provide light tightness . figures [ unwrapped ] and [ wrapped ] show unwrapped and wrapped plastic scintillator barrel bars . photograph of an unwrapped barrel bar of plastic scintillator with the apparatus used to wrap bars with ptfe to increase light collection.,width=325 ] photograph of a wrapped plastic scintillator module . around the bare plastic is a layer of ptfe , a mirror at the opposite end from the pmt , and black wrapping to light tighten the module . the pmt itself is housed in a light tight cup bonded to the scintillator.,width=325 ] the performance of each of the 52 pieces has been assessed individually as follows . for each fully wrapped plastic scintillator module , spectra of the pmt response to a @xmath35na calibration source were recorded for each of six positions along the length . to ensure consistency , the same pmt , voltage divider network , preamplifier and operating bias were used in these measurements . additionally , the use of optical grease was omitted to reproduce conditions as exactly as possible across scintillators . it was found that without the optical grease the number of photoelectrons detected is approximately halved , and this is confirmed by monte carlo simulations @xcite . the measurements were performed in a surface laboratory of predominantly concrete construction , and thus , for plastic scintillators of this size , a high rate of @xmath1-rays from the local environment was observed . to reduce this contribution , events seen in the plastic scintillators were required to be in coincidence with events in a 3 `` diameter , 3 '' long nai scintillation detector placed on the other side of the radioactive source from the plastic scintillator . since @xmath35na decays by @xmath36 emission this greatly enhanced the fraction of events caused by the 511 kev annihilation and 1275 kev de - excitation @xmath1-rays . however , since @xmath1-rays produce largely compton electrons in organic scintillators at these energies , the trigger signal of the nai detector can come either from another @xmath1-ray , or from the compton - scattered @xmath1-rays detected in the plastic scintillator . in the latter , the shape of the energy spectrum may be effected since compton electrons in the scintillator deposit lower energy and yet the scattered @xmath1-ray retains sufficient energy to trigger the nai . monte carlo simulations of the experimental setup confirmed that , with the geometry used for these measurements , the rate of such events relative to true coincidence triggers from separate gamma - rays was negligible . a typical compton spectrum is shown in fig . [ balfig ] , revealing features corresponding to the 511 kev @xmath1-rays , as well as a weaker peak corresponding to the 1275 kev @xmath1-ray emission following @xmath36 decay of @xmath35na , and a background well described by an exponential trend . the calibration between observed compton peak position and number of photoelectrons has been used to allow this spectrum to be plotted in terms of numbers of photoelectrons . a two - partial - gaussian - plus - exponential fit was made to the data and the movement of the centroid of the partial gaussians and the compton edge as a function of the position of the source along the length was used as the indicator of the technical attenuation length ( tal ) . the bal describes the attenuation of photons in a beam due to the scintillator only , whereas the tal parameter is an auxiliary parameter . the difference between the and bal comes from the fact that the real light path from the vertex of the interaction to the pmt is longer than a straight line path , since most of of the light reaching the pmt has been reflected at the surface of the scintillator bar through total internal reflection . the measured tal combines geometry effects ( light path , reflection ) , light attenuation , wavelength shift ( due to wavelength dependent absorption and re - emission ) and wavelength dependent quantum efficiency of the photo cathode . in the case of the modules measured here , a highly - reflective specular mirror surface has also been placed at the opposite end to the pmt . consequently , one expects , to first order , the response as a function of distance from the pmt to vary as the sum of two exponentials , one describing the reduction due to the separation of pmt and source position , and one which describes the reduction due to the distance between the pmt and the image of the source in the mirror : @xmath37 where @xmath38 is the centroid of the measured photopeak , @xmath39 is the distance from the pmt face , @xmath40 is the length of the scintillator module ( the distance between the pmt and the mirror ) , @xmath41 is the tal and @xmath42 an arbitrary scaling constant . in principle , one could use a different scaling for the term describing the response due to reflections from the mirror , effectively accounting for imperfect reflection from it , but the quality of the data available did not justify an additional free parameter . the measurements of tal have been complemented with cosmic ray muon measurements in which the nai coincidence detector was replaced by two small plastic scintillator detectors placed above and below the test piece . in this case the spectra obtained are well described and fitted by a landau distribution , and again the response as a function of measurement position determined . consistent results were found for the tal measured either with the @xmath35na source or with cosmic rays . typical spectrum ( solid histogram ) showing the response of a single barrel module of plastic scintillator to irradiation with a @xmath35na @xmath1-ray source . see text for further details of the source geometry . the solid line is the result of a fit to the data assuming partial gaussians for the peaks attributed to the 511 and 1275 kev @xmath1-rays and an exponential to describe the background . also shown ( dashed histogram ) is the result of a monte carlo simulation of this exposure . , width=325 ] to assess whether the measured tals are consistent with the expected bal of the material , geant4 monte carlo simulations of these tests have been performed ( see section [ mc ] ) . these included the geometry of the plastic scintillator , laboratory and source , as well as physical parameters such as the light output of the plastic , gain and quantum efficiency of the pmt ( see section [ pmts ] ) . the simulations include full tracking of scintillation photons from @xmath1-ray energy deposition through to incidence on pmt photocathode and stochastic generation of photoelectrons , allowing realistic spectra to be generated . the lighter histogram in figure [ balfig ] shows the result of the simulation tailored to that particular measurement s geometry , illustrating that the key features are very well reproduced . the simulations are repeated for different values of bal , each time generating spectra for different source target positions , and these spectra evaluated in the same manner as was done for real data to extract a tal . figure [ responsevspos ] shows the position of the centroid of the peak attributed to the 511 kev @xmath1-ray as a function of the source location both for real data from one of the barrel modules , and evaluated from monte carlo simulations of those exposures with an appropriately chosen value of bal . this bal is then taken as corresponding to the experimental value of tal indicted by the data . lines have been added to guide the eye in this figure and in subsequent figures where appropriate . as expected , for each scintillator bar geometry a monotonic dependence between bal and tal is found , with tals having smaller values than bals , and the most uniform geometry bars having the highest tals . these results are summarised in figure [ baltal ] . we find a large dispersion in bal , ranging from @xmath1160 cm to greater than 2.5 m , with significant variations even between bars polymerised in the same batch . the measurements allow the matching of individual bars of plastic scintillator to particular pmts , matching lower bal with higher qe to provide a more uniform performance across the entire veto and informing the placement of scintillator modules within the array . this also provides input for the monte carlo simulations described in section [ ubersim ] in which the overall performance of the whole veto system is evaluated . variation of the peak position from the 511 kev @xmath1-ray with source location both for data from one of the 1 m long barrel modules and evaluated from monte carlo simulations of those tests . lines have been drawn to guide the eye.,width=325 ] measured technical attenuation length ( tal ) and monte carlo predicted bulk attenuation length ( bal ) for each bar of plastic scintillator.,width=325 ] although many processes contribute to the scintillation light emission @xcite , the absolute light output of the plastic scintillator can be calculated by matching the measured and simulated spectra . a consistent value of @xmath115500 ph / mev was required for all bars , regardless of which polymerisation batch they came from . this is lower than expected , as 55% of anthracene would correspond to nearer 8500 ph / mev , necessitating the use of reflective wrapping around the scintillators . during normal operation of the veto a blue led unit will continuously flash 52 separate optic fibres coupled directly to the ends of the plastic scintillator modules furthest from the pmt . this will allow any degradation of the scintillator performance over time to be monitored . the outputs from the pmt preamplifiers are fed to caen v1724 adcs housed in a vme8011 crate with caen v2718 pci bridge communicating via optical link to a dedicated linux - based data acquisition computer . the digitisers have @xmath162.25 v input range with 14-bit resolution and 40 mhz bandwidth with a sampling rate of up to 100 ms / s simultaneously on each channel , with internal buffering to ensure zero deadtime performance at expected data rates . custom made run - control software provides a graphical user interface for beginning and terminating data acquisition runs with easily changeable run parameters and toggles over various operation modes . the software also provides remote capabilities to fully automate instrument operation . a screenshot of this interface is shown in figure [ screenshot ] . amongst the most important operation modes are the veto ` slave ' mode and calibration / diagnostic ` master ' mode . in _ slave _ mode , all modules take an external trigger sourced from the zeplin iii instrument . as such , the veto is guaranteed to record data when zeplin iii itself has recorded an event . besides simplicity of design , this minimises data volume and obviates the need for a separate hardware - enforced trigger and associated efficiency loss at low energies . additionally , spontaneous photoelectron emission from the pmts , present in the timelines recorded where pulses are absent , can be used for equalising gains , calibrating and monitoring stability of the pmts without acquiring dedicated data , where conditions between data runs may vary . finally , since even with high energy depositions in the veto only a few of the scintillator modules are involved , the remaining modules provide diagnostic information at the time of the event . monte carlo simulations indicate background @xmath1-ray events depositing @xmath6200 kev distribute @xmath690% of the energy in 6 modules or fewer ( see section [ ubersim ] ) . graphical user interface of veto data acquisition.,width=325 ] the expected time delay between signals in zeplin iii and in the veto defines the requirements for the duration of the waveforms to be be recorded . coincident @xmath1-ray events will be essentially simultaneous . for single scatter neutron recoils within zeplin iii , where the veto signal is provided by energy deposition of @xmath1-rays following neutron capture on gadolinium , the coincidence is expected to have a mean delay of @xmath1135 @xmath8s ( see section [ gdloading ] ) . however , for small energy deposits in zeplin iii , the trigger is in fact provided by the s2 electroluminescence , which is itself delayed by up to 17 @xmath8s ( due to the time it takes electrons to drift through the xenon ) . it is therefore necessary that a period of pre - trigger data is also recorded : values of 20 @xmath8s pre - trigger and 300 @xmath8s post - trigger have been chosen . an example of a waveform recorded in the veto data acquisition is presented in figure [ waveforms ] . this event corresponded to irradiation by a @xmath43co source located approximately half way along the length of one of the barrel modules . by design , the gain of the system is such that @xmath1-rays of about this energy populate the mid range of the 14-bit adc . the baseline noise is observed to be between typically 2 mv , or about 30 adc channels . in addition to the @xmath1-ray event at 60 @xmath8s , a typical single photoelectron can be seen at 230 @xmath8s , clearly visible above the noise level . an example waveform from a barrel module plastic scintillator in response to a @xmath1-ray from a @xmath43co source , with the pulse starting at 60 @xmath8s . also clearly visible is a single photoelectron at 230 @xmath8s.,width=362 ] the _ master _ diagnostic / calibration mode takes either a forced trigger to allow the veto modules to record effectively continuously without imposing a signal threshold for triggering , or can be set to record data when a defined number of scintillator modules detect a signal above a defined threshold . these are necessary for _ in situ _ calibration of the veto device and for recording background data in all modules . as such , when operating in this master mode , the veto system can operate as a detector independently of zeplin iii if required . the default operating mode will be a combination of these triggering techniques , concurrently recording a level of background without hindering the coincidence detection capabilities : the veto will trigger whenever zeplin iii observes a signal , regardless of the presence of a signal in the veto , but it will also trigger whenever the veto observes a signal in two or more modules above some threshold set to exclude noise and maintain a low count rate . since the acquisition can run with zero deadtime by storing events in its internal buffer , there is no possibility of missing a zeplin iii trigger . to maintain accurate synchronisation between zeplin iii and the veto data acquisitions , a dedicated hardware unit produces a digital time stamp which is sent to both acquisitions so that events can be correlated with 100 ns accuracy . finally , taking advantage of the segmented nature of the veto , pmt outputs from the 20 roof modules are fed into a dedicated triggering unit that passes signals through to the adc inputs but also sums them and shapes this summed signal . the shaping parameters are set such that if the unit finds a pulse in any module that is characteristic of a cosmic ray muon , all 52 veto channels are triggered and the event recorded , regardless of a coincident event in zeplin iii . this is a valuable diagnostic allowing measurement of the cosmic ray flux through the laboratory , as well as providing a measure of muon induced neutron background in the local environment . monte carlo simulations have been used extensively in this work , both to characterise individual components , for example correlating the bal and tal responses of individual plastic scintillator modules , and as a tool for optimising and understanding the overall performance of the veto . the requirements for this monte carlo simulation are relatively demanding . the programs should be capable of modelling the transport and interactions of particles ( essentially neutrons and @xmath1-rays ) internal and external to zeplin iii and the veto through to the production of electron and nuclear recoils . the program must then simulate the physical processes involved in the generation of the optical response to scintillation ( and electroluminescence in zeplin iii ) , and finally generate the response for all channels in order to produce realistic datasets . the existing zeplin iii monte carlo simulation package @xcite , and the veto simulations developed here , have utilised the geant4 toolkit @xcite , which is capable of fulfilling all of these requirements . in this section we first present the general features of the simulation , leading to calculation of the expected ability of the veto to reject both neutron and @xmath1-ray events that , based on zeplin iii data alone , might be misidentified as wimps . a key feature of the veto design is the inclusion of gadolinium to enhance the response to neutrons . we have used the monte carlo methodology to explore the dependence of the design to the amount and location of gadolinium used , and this is presented . we then bring together the measurements of radiological content and performance of all the veto components to estimate the absolute event rate that the veto will experience , and its impact on the rejection performance of the instrument . finally , we utilise the monte carlo simulations to explore effects such as misalignment between modules introduced during assembly . the monte carlo simulation has also been used to inform numerous other choices not mentioned , for example the geometry of the active and passive modules . the geant4v9.1 monte carlo toolkit has been used . all simulations included accurate geometry of the plastic scintillators , their wrapping including the 95% specular reflective mirror at the far end , the pmts including envelope , photocathode , internal elements and pvc support structure and optical grease ( in terms of changes to refractive indices ) . also incorporated are the geometry and materials of the surrounding area ; for simulations of the whole veto this includes an extensive description of the boulby laboratory with full details of zeplin iii itself , the gd and all related processes , the local polypropylene and lead shielding , the building , and rock walls . by varying the depth of rock included in the simulation it was found that a depth of 3 m for neutrons and 0.25 m for @xmath1-rays is sufficient to accurately simulate the flux emanating from these surfaces ( @xmath191% contribution to uncertainty ) . for simulations of measurements made in surface laboratories , the details of the walls , tables and other large objects in the vicinity were included as necessary . the light yield and bal of the scintillator and the single photoelectron response and the qe of the pmts are taken from measurements described earlier . simulations have been performed for radiation originating from the cavern walls , from zeplin iii components , from the veto components and from calibration sources and , in each case , measured values of activity have been used . in the case of u and th , the sources4c @xcite code has been used to generate the emission spectra for neutrons . all particles and their secondaries are tracked . the location , size and event time of all energy depositions are recorded . a nuclear recoil energy deposition generates much less electronic excitation ( primary scintillation and ionisation ) than an electron recoil event of similar energy due to ` quenching ' . a quenching coefficient is defined as the ratio of the amount of light induced by a recoil nucleus to the amount of light induced by an electron of the same kinetic energy . a ( conservative ) quenching coefficient of 0.1 is assumed for neutron energy depositions occurring in the plastic scintillator @xcite . since the efficiency for rejecting coincident events with zeplin iii comes from detection of @xmath1-rays , a large variation in quenching factor of the plastic scintillator makes negligible difference to the results presented . each energy deposition occurring within an active volume is converted to a number of photoelectrons seen by the pmt . this has been performed in two ways . firstly , an option in the monte carlo simulation allows each of these energy depositions to be converted to scintillation photons , and for these scintillation photons to be tracked to the photocathode , with losses occurring at points of reflection and due to the bulk attenuation . rayleigh scattering is included . poisson statistics are then applied to determine how many photoelectrons are generated . since this is effectively a further complete simulation for each energy deposition , this option is slow and has been used only where necessary , principally to correlate measured tal values with intrinsic bal of the scintillators , and to allow the light emission of the scintillator to be determined . the second much faster option is to simply apply the expected loss due to the measured tal given the distance from the end of the scintillator . since the measured and simulated tals agree , a further refinement to the simulation has been to use the monte carlo to estimate the fully 3d - dependent tal throughout the scintillator , rather than a single mean value that describes the unit as a whole . the central function of the veto detector is to facilitate rejection of events that , on the basis of zeplin iii data alone , could be mistaken for wimps . the principal characteristics of candidate events in zeplin iii are that they have an electroluminescence - to - scintillation ratio characteristic of nuclear recoils , occur within the fiducial volume , and cause an energy deposition of @xmath4450 kevnr@xmath45 . the complete end - to - end simulation has been performed with all energy depositions in zeplin iii and the veto recorded . while for the first science run the main contribution to background was the activity of the pmts of the zeplin iii device itself , after the upgrade it is expected that no single source will dominate overwhelmingly - the contributions from pmts , components and from the rock walls being similar . differences arise due to the activity levels particular to each case , and due to the propagation of the radiation from point of origin through to detection . thus , the efficiency of the veto in rejecting background neutrons has been estimated by considering each of these contributions separately , and then adding the contributions appropriately . in its simplest mode , an event that satisfies the zeplin iii criteria for a wimp can be rejected by the veto if any one of the scintillator modules registers , within an appropriate time window ( @xmath4620 to + 300 @xmath8s ) , a signal significantly above background . these are dubbed ` tagged ' events , and a tagging efficiency can be calculated as a function of the veto signal size ( see figure [ effvsphe ] ) . a reasonably weak dependence on veto threshold is seen , with a maximum tagging efficiency at zero threshold of 80% possible : about 20% of neutron background events that satisfy the wimp - search criteria in zeplin iii have no interaction in any of the veto modules . in reality , a threshold of 6 photoelectrons in any single scintilator is sufficient to provide a minimal accidental coincidence contribution to overall event rate , and therefore this has been used as the conservative hard threshold within any module in simulations of the veto . lower limit efficiencies derived from this threshold determine the minimum neutron and gamma - ray background reduction in zeplin iii . for the barrel modules , 6 photoelectrons corresponds to a typical energy deposition of 135 kev at the far end and it is predicted that @xmath665@xmath14 of background neutrons that could otherwise have been misidentified as wimps will be rejected by the veto with this threshold . however , as stated , since there is no hard threshold trigger for the veto when triggering in its slave mode , a higher tagging efficiency for coincident events ( up to a maximum of 80% ) is achievable . this represents a substantial improvement on previous anti - coincidence systems such as the veto used by the zeplin ii experiment @xcite , which had a neutron tagging efficiency of approximately 50% , with a similar @xmath1-ray tagging efficiency . tagging efficiency of the veto for neutron events in zeplin iii as a function of the signal size seen in at least one module of the veto , in terms of photoelectrons.,width=325 ] figure [ effvsphe ] shows the neutron background rejection power of the veto based on signals above threshold in single scintillator modules only . in principle , it could be possible to increase the tagging efficiency further by searching for smaller energy depositions occurring in multiple modules simultaneously . figure [ tagfrac ] extends figure [ effvsphe ] to explore the summed energy deposition in multiple scintillator modules . since there exists no energy threshold for coincidence veto events when in slave mode of operation it is possible to apply not just single module thresholds , but a threshold on the sum energy of a veto event . in this case a veto event would be defined as the sum of all veto modules with signals above a minimal threshold as low as 1 photoelectron and occuring in coincidence with one another within a short time window of a few @xmath8s duration . the former condition takes advantage of the mutliple @xmath1-rays emitted following gd capture causing interactions in multiple modules to maximise the tagging efficiency of the veto system . the latter condition restricts the number of random coincidence events with zeplin iii by demanding all @xmath1-ray signals seen in multiple modules belong to the same event constrained within the timing window . since 320 @xmath8s timelines are recorded for each module a number of such events may be observed . figure [ tagfrac ] shows the neutron tagging efficiency of the veto as a function of the sum of photoelectrons detected from each modules registering at least one photoelectron . it is seen that the tagging efficiency has a much stronger dependence on the number of scintillator modules than on the signal size . with a summed signal of 6 photoelectrons the nominal 65% tagging efficiency is acieved from any single module registering 6 or more photoelectrons . however , this efficiency is supplemented by events where the 6 photoelectrons are distributed , in any hit pattern , across multiple modules and the efficiency is increased to almost 70% . the scope for improvement is limited to a maximum of approximately 80% since , as seen in figure [ effvsphe ] , some 20% of candidate events result in no energy deposition in the veto at all . the efficiency for the veto to tag background neutron events that , based on zeplin iii data alone , could have been misidentified as wimps , as a function of the multiplivity of the event and the summed number of photoelectrons across all modules of the veto . the multiplicity is defined as the number of modules contributing to the summed signal with at least one photoelectron.,width=325 ] zeplin iii has very good ability to discriminate between electron recoils , principally from @xmath1-rays , and nuclear recoils from neutrons or candidate wimps . however , with sufficient exposure some @xmath1-rays can still leak into the nuclear recoil acceptance region . some fraction of these are likely to have also deposited energy in the veto , and thus could be rejected . simulations have been performed in which @xmath1-rays were emitted from the known activities of the rock walls , laboratory , shielding , and from the detector components . instances in which energy was deposited in both zeplin iii ( 216 kevee , single scatter in the fiducial volume ) and in one or more of the veto plastic scintillators were searched for . the resulting efficiency for the veto to reject @xmath1-rays that otherwise might have been misidentified as wimps , has been evaluated as a function of the number of photoelectrons seen in at least one module of the veto , and is presented in figure [ geffvsphe ] . it is predicted that with a threshold of 6 photoelectrons being seen in a single module , @xmath1115% of background @xmath1-rays depositing 216 kevee through single scatters in the zeplin iii fiducial volume can be rejected by the veto . @xmath1-ray tagging efficiency of the veto as a function of the number of photoelectrons seen in at least one plastic scintillator.,width=325 ] [ pitchmc ] the adopted design relies heavily on the inclusion of gadolinium loaded into the polypropylene pieces of the veto . the very high neutron capture cross section of @xmath12gd , combined with the subsequent emission of up to 8 mev of @xmath1-rays , is the principle reason for the high neutron tagging efficiency . removing the gd entirely results in the neutron tagging efficiency for this geometry to drop to below 45% for a 6 phe threshold . the results presented so far are for a loading of 0.5% by weight , which was the value eventually adopted ; here we explore the effect of different concentrations and distributions of gadolinium . the incorporation of gd into the polypropylene has two main effects . firstly , increasing the fraction of gadolinium increases the fraction of internal neutrons that are captured by the gadolinium isotopes , as opposed ( mainly ) to capture on the hydrogen content of the polypropylene and plastic scintillator . since the @xmath12gd capture in particular is followed by emission of several energetic @xmath1-rays , this also tends to increase the tagging efficiency . this is illustrated in the upper panel of figure [ gdloadingfig ] in which an exposure to neutrons from known background sources of the entire zeplin iii veto system has been simulated . for events that satisfy the wimp search criteria ( fiducial volume , @xmath4450 kevnr ) the fraction that result in 6 or more photoelectrons in a single module of the veto is shown , as a function of the gadolinium loading . a gradual increase in tagging efficiency is seen up to about 0.5% fraction by weight . above this little further gain is made , indicating that at this level of loading , essentially all neutrons that thermalise in the polypropylene are captured by the gadolinium . the second effect of including gadolinium is that the mean time before neutron capture is reduced . in terms of experimental infrastructure , a short capture time is desirable because it allows a larger fraction of coincidences to be identified for a given coincidence time window , or alternatively , for a shorter window to be used reducing data volume and the false - coincidence rate . the mean time between tagged signals in zeplin iii and the veto , as a function of gadolinium concentration , is shown in the lower panel of figure [ gdloadingfig ] . here we see that above about a 1% loading , further increase ceases to reduce the capture time , indicating that capture is occurring very soon after thermalisation . simulations have also shown that neutrons are effectively moderated to thermal energies whether they originate externally to the veto ( such as from the rock around the laboratory ) or from within the internal volumes enclosed by the veto ( such as from zeplin iii itself ) . external neutrons from the rock predominantly undergo radiative capture on h within the scintillator since very few are able to penetrate the 15 cm into the gd - loaded polypropylene . the mean neutron energy at capture is 0.035 ev . on the other hand , internal neutrons are captured on gd and , at 0.5% gd concentration , the mean energy before capture is 0.045 ev ( a slightly higher value due to the enhanced capture cross section relative to h ) . in both cases , neutrons entering the veto are effectively moderated and captured within the veto acquisition time window of 300 @xmath8s , with a mean time of 35 @xmath8s for the internal neutrons . monte carlo simulations of the zeplin veto system for various concentrations of gadolinium loading ( by weight ) . upper panel : the efficiency for the veto to tag events that based on zeplin iii data alone could have been misidentified as wimps . lower panel : for tagged events , the mean time delay in microseconds between signals seen in zeplin iii and the veto.,width=325 ] the gadolinium is suspended in epoxy that is located in 2 mm wide , 10 mm pitch slots that cut through the full depth of the polypropylene . this choice has partly been driven by mechanical considerations , as narrower slots with a higher pitch would significantly increase the cost of machining . in figures [ slots1 ] and [ slots2 ] the monte carlo simulation has been used to estimate the tagging efficiency as a function of slot width and slot pitch for a fixed average loading throughout the polypropylene of 0.5% . clearly , if the slot pitch increases beyond about 10 mm , the efficiency of the veto in tagging coincident events within zeplin iii begins to drop significantly . this is because , at this scale , neutrons moderating and thermalising within the polypropylene have a trajectory that sometimes never crosses a slot , and thus never provides an opportunity for capture on the gadolinium loading . the width of the slots has less effect , and thus a value amenable to construction can be chosen that minimises cost and is optimal for the epoxy to hold the gadolinium in suspension . monte carlo simulation of the efficiency of the veto in rejecting events that , based on zeplin iii data alone , could have been identified as wimps , as a function of the pitch of the gd - loaded epoxy filled slots . the width of the slots , 2 mm , and the overall loading fraction of gd , 0.5% by weight , is maintained in all cases.,width=325 ] monte carlo simulation of the efficiency of the veto in rejecting events that , based on zeplin iii data alone , could have been identified as wimps , as a function of the width of the gd - loaded epoxy filled slots . the pitch of the slots , 10 mm , and the overall loading fraction of gd , 0.5% by weight , is maintained in all cases.,width=325 ] the large volume of plastic scintillator used in the veto , combined with a low threshold , has the potential to create an extremely high event rate compatible neither with the data acquisition hardware nor the coincidence methodology to be used . the thorough approach to control and evaluate the radioactivity content of items used , coupled to the development of a detailed monte carlo simulation benchmarked against known performance , allows the absolute event rates to be estimated . the dominant rate comes from @xmath1-rays , although it is the neutron background generated by the veto that is of fundamental importance to zeplin iii . excluding any contributions from the veto , the net nuclear recoil background in the second science run of the zeplin iii experiment is expected to be @xmath00.4 events per year . this is reduced to @xmath00.14 events per year for a 65% veto tagging efficiency . for each contribution of background from the veto the complete monte carlo was performed , with the neutrons and @xmath1-rays distributed in energy and location as appropriate . energy depositions in the veto and in zeplin iii were recorded . the total expected event rate could then be calculated by summing the contributions . the net single scatter contribution to the background in the zeplin iii acceptance region , assuming no rejection through vetoing the events , is 0.02 nuclear recoils per year . the nuclear recoil rate drops to 0.007 events per year when a conservative self - vetoing efficiency of 65% is applied . the veto is predicted to contribute @xmath191000 electron recoils per year from @xmath1-ray emission within the zeplin iii fiducial volume , in the energy range of 216 kevee . considering the veto as a stand - alone instrument , figure [ exposure ] shows the raw event rate that is expected in the veto as a function of the number of photoelectrons observed in a single plastic scintillator module . this shows , for example , that if a veto trigger requirement were imposed such that events were recorded if any single module recorded six or more photoelectrons , then a total event rate of 70 hz would result . predicted event rate of energy depositions in the veto , as a function of signal size , originating from @xmath1-ray emission of the measured activities of components . statistical errors , although present , are too small to be seen clearly on this scale . also shown is the probability for events seen in the veto following @xmath1-ray emission from components being misidentified as coincident events with zeplin iii.,width=325 ] we may consider how the background radioactivities will impact the veto when operating in coincidence with zeplin iii by calculating the ` accidental coincidences rate ' . this is the probability of any recorded timeline having an uncorrelated background - radioactivity caused signal within it . for the usual mode of slave operation , with the veto triggered to record for 320 @xmath8s by events from zeplin iii at an estimated rate of 1 hz , the accidental coincidences rate is indicated by the right hand axis in figure [ exposure ] . for nominal operational parameters the rate is less than 2% . the veto instrument consists of many modular pieces mechanically held together . as described in section [ overalldesign ] , the plastic scintillators are deliberately offset relative to the polypropylene modules to prevent line - of - sight gaps that would compromise the shielding . there is , however , still a significant chance of introducing small gaps during construction and assembly . the monte carlo simulation has been employed to estimate the impact of such gaps occurring between the active modules , both in terms of degrading the shielding from external sources of background , and in terms of lowering the ability of the veto to tag coincident nuclear recoil events . the simulation was performed for two distinct cases : when gaps existed between both active and passive modules , and when gaps existed between only active modules ( in which case the passive modules were assumed to be completely closed ) . in the former case a direct path is formed for radiation that has penetrated the external lead shield to reach zeplin simulation of a three year exposure that included gaps of 1 mm width , in excess of engineering constraints , showed no statistically significant increase in the zeplin iii active volume single scatter rate . this is consistent with previous simulations performed to explore the impact of shielding due to the existing holes needed for pipework and cabling . moreover , the veto tagging efficiency remains unaffected until the gaps exceed a width of 1 cm . this is because a neutron , having originally entered the shielding or been internally generated , is unlikely to interact with zeplin iii and exit the shielding through a gap again . in the latter sets of simulations , in which gaps between adjacent active scintillator modules were introduced assuming a continuous inner gd loaded polypropylene barrel , it was found that the efficiency for tagging coincident veto iii nuclear recoil events begins to be significantly reduced for gaps of widths greater than 2 cm , essentially due to less coverage for @xmath1-ray detection following neutron capture . the neutron flux seen by zeplin iii from external background is unaffected until the gap exceeds 1 cm between scintillator modules . all veto components have been shipped underground to the boulby laboratory with active and passive modules coupled in preparation for installation . the full veto array , complete with data acquisition and synchronisation hardware , led calibration systems and data reduction software has been successfully assembled within the pb castle . this was done in the absence of zeplin iii to test systems prior to final installation . neutron and @xmath1-ray source exposures as well as background runs have been performed and data recorded for _ in situ _ calibration and characterisation of the detector . the veto has since been dismantled to allow access to zeplin iii . it will be installed following the completion of the upgrade to the zeplin iii photomultiplier array . performance results of the veto will be presented following commencement of the second science run of the zeplin iii experiment . the second science run of the zeplin iii project will feature two main upgrades : lower background photomultiplier tubes and the use of an active veto . here , the design of the veto has been presented , together with details of the radiological content and performance of the components to be used . detailed geant4 monte carlo simulations have been used to aid in the characterisation of the veto , and to estimate its overall performance . in addition to providing valuable diagnostic information , it is expected that the veto will be able to reject over 65% of neutrons , and over 15% of @xmath1-rays from background radioactivities , whilst contributing negligibly to the zeplin iii acceptance region background , expected to be @xmath00.4 events per year . the veto will reduce this background to less than @xmath00.14 events per year . this is a significant factor in the event of a non - zero observation . the veto has been fully assembled and systems integrated , and will be installed around zeplin iii shortly for the commencement of the second science run as this instrument probes yet greater sensitivity . the uk groups acknowledge the support of the science & technology facilities council ( stfc ) for the zeplin iii project and for maintenance and operation of the underground palmer laboratory which is hosted by cleveland potash ltd ( cpl ) at boulby mine , near whitby on the north - east coast of england . the project would not be possible without the co - operation of the management and staff of cpl . we also acknowledge support from a joint international project award , held at itep and icl , from the russian foundation of basic research ( 08 - 02 - 91851 ko a ) and the royal society . lip coimbra acknowledges financial support from fundao para a cincia e tecnologia ( fct ) through the project - grants cern / fp/83501/2008 and cern / fp/109320/2009 , as well as the postdoctoral grants sfrh / bpd/27054/2006 and sfrh / bpd/47320/2008 . this work was supported in part by sc rosatom ; by russian grant ss-1329.2008.2 and by the russian ministry of education and science contract 02.740.11.0239 . the university of edinburgh is a charitable body , registered in scotland , with the registration number sc005336 . _ , sources4c : a code for calculating ( @xmath20,n ) spontaneous fission and delayed neutron sources and spectra , american nuclei society / radiation protection and shielding division , 2002 .

we present continued radio observations of the tidal disruption event swiftj164449.3 + 573451 extending to days after discovery . the data were obtained with the evla , ami large array , carma , the sma , and the vlba+effelsberg as part of a long - term program to monitor the expansion and energy scale of the relativistic outflow , and to trace the parsec - scale environment around a previously - dormant supermassive black hole ( smbh ) . the new observations reveal a significant change in the radio evolution starting at month , with a brightening at all frequencies that requires an increase in the energy by about an order of magnitude , and an overall density profile around the smbh of ( pc ) with a significant flattening at pc . the increase in energy can not be explained with continuous injection from an tail , which is observed in the x - rays . instead , we conclude that the relativistic jet was launched with a wide range of lorentz factors , obeying . the similar ratio of duration to dynamical timescale for and grbs suggests that this result may be applicable to grb jets as well . the radial density profile may be indicative of bondi accretion , with the inferred flattening at pc in good agreement with the bondi radius for a m black hole . the density at pc is about a factor of 30 times lower than inferred for the milky way galactic center , potentially due to a smaller number of mass - shedding massive stars . from our latest observations ( d ) we find that the jet energy is erg ( erg for ) , the radius is pc , the lorentz factor is , the ambient density is , and the projected angular size isas , below the resolution of the vlba+effelsberg . assuming no future changes in the observed evolution and a final integrated total energy of erg , we predict that the radio emission from should be detectable with the evla for several decades , and will be resolvable with vlbi in a few years .

1112.1697

the discovery of the unusual @xmath22-ray / x - ray transient swiftj164449.3 + 573451 ( hereafter , ) , which coincided with the nucleus of an inactive galaxy at @xmath23 , has opened a new window into high - energy transient phenomena , with potential implications to our understanding of relativistic outflows in systems such as gamma - ray bursts ( grbs ) and active galactic nuclei ( agn ) . the prevailing interpretation for this event is the tidal disruption of a star by a dormant supermassive black hole ( smbh ) with a mass of @xmath24 m@xmath9 ( @xcite ; but see @xcite for alternative explanations ) . the argument for a tidal disruption origin is based on : ( i ) a positional coincidence ( @xmath25 kpc ) with the host galaxy nucleus ; ( ii ) rapid time variability in @xmath22-rays and x - rays ( @xmath26 s ) , which requires a compact source of @xmath27 au , a few times the schwarzschild radius of a @xmath28 m@xmath9 black hole ; ( iii ) high @xmath22-ray and x - ray luminosity of @xmath29 erg s@xmath30 , which exceeds the eddington limit of a @xmath28 m@xmath9 black hole by @xmath31 orders of magnitude ; ( iv ) a lack of previous radio to @xmath22-ray activity from this source to much deeper limits than the observed outburst , pointing to a rapid onset ; and ( v ) long - term x - ray luminosity evolution following @xmath32 , as expected from the fallback of tidally disrupted material ( e.g. , @xcite ) . equally important , was accompanied by bright radio synchrotron emission , with an initial peak in the millimeter band ( @xmath33 mjy ) and a steep spectral slope at lower frequencies indicative of self - absorption ( @xcite ; hereafter , zbs11 ) . the properties of the radio emission established the existence of a relativistic outflow with a lorentz factor of @xmath34 ( zbs11 , @xcite ) . the spectral energy distribution also demonstrated that the lack of detected optical variability required significant rest - frame extinction ( @xmath35 mag ; zbs11 , @xcite ) , and that the x - rays were produced by a distinct emission component , rather than inverse compton scattering by the radio - emitting relativistic electrons ( zbs11 ) . finally , the evolution of the radio emission on a timescale of @xmath36 d pointed to an ambient density with a radial profile of roughly @xmath37 , as well as a mild increase in the energy of the outflow ( zbs11 ) . the formation of a relativistic jet with dominant x - ray and radio emission were not predicted in standard tidal disruption models ( e.g. , @xcite ) , which instead focused on the thermal optical / uv emission from the long - term accretion of the stellar debris . a signature of the latter process is a mass accretion rate that evolves as @xmath38 , presumably leading to emission with the same temporal dependence ( e.g. , @xcite ) . shortly before the discovery of , @xcite investigated the potential signature of a putative relativistic outflow , and concluded that the interaction of the outflow with the ambient medium will lead to radio emission on a timescale of @xmath39 yr ( for typical off - axis observers ) . while the mechanism for the radio emission from is interaction with an external medium , the actual light curves differ from the off - axis prediction . to address this issue , in a follow - up paper @xcite ( hereafter , mgm11 ) reconsidered the model for a relativistic jet interacting with an ambient medium . they draw on the inferences from the early radio emission described in zbs11 to infer the properties of the environment and the jet kinetic energy , and use this information to predict the future evolution of the radio emission . this long - term radio evolution is of great interest because it can provide several critical insights : * the integrated energy release in the relativistic outflow , including the anticipated injection from on - going accretion . * the density profile around a previously - dormant smbh on @xmath40 pc scales , which can not be otherwise probed in agn . * the potential to spatially resolve the outflow with very long baseline interferometry ( vlbi ) , and hence to measure the dynamical evolution ( expansion and potentially spreading ) of a relativistic jet . * predictions for the radio emission from tidal disruption jets as viewed by off - axis observers on timescales of months to years to decades . the energy scale and jet dynamics are of particular importance since the total energy input and the structure of the jet may also have implications for relativistic jets in grbs and agn . the ability to trace the environment on parsec scales provides a unique probe of gas inflow or outflow around an inactive smbh on scales that can not be probed outside of the milky way . finally , the long - term radio emission from will inform future radio searches for tidal disruption events ( tdes ) that can overcome the low detection rate in @xmath22-rays / x - rays ( due to beaming ) , and obscuration due to extinction in the optical / uv ( as in the case of ) . to extract these critical properties we are undertaking long - term monitoring of the radio emission from using a wide range of centimeter- and millimeter - band facilities . here we present radio observations of that extend to @xmath11 d , and use these observations to determine the evolution of the total energy and ambient density . we find that the evolution of both quantities deviates from the behavior at @xmath41 month ( presented in zbs11 ) , thereby providing crucial insight into the structure of the relativistic outflow and the ambient medium . this paper is the first in a series that will investigate the long - term radio evolution of and the implications for relativistic jets and parsec - scale environments around supermassive black holes , including efforts to resolve the source with vlbi and to measure polarization . the current paper is organized as follows . we describe the radio observations in [ sec : obs ] , and summarize the radio evolution at @xmath42 d in [ sec : prem ] . in [ sec : model ] we present our modeling of the radio emission , which utilizes the formulation of mgm11 . the implications for the energy scale and ambient density are discussed in [ sec : energy ] and [ sec : density ] , respectively , and we finally consider the implications for relativistic jets and the parsec - scale environments of smbhs in [ sec : implic ] . although first triggered the _ swift _ burst alert telescope on 2011 march 28.55 ut , discernible @xmath22-ray emission was detected starting on 2011 march 25 ut @xcite . we therefore consider 2011 march 25.5 ut to be the actual initial time for the event and the associated relativistic outflow that powers the observed radio emission . our radio observations of commenced on 2011 march 29.36 ut ( @xmath43 hr after the _ swift _ trigger and about @xmath44 d after the initial @xmath22-ray detection ) . observations extending to @xmath45 d were presented in zbs11 . here we report observations extending to @xmath11 d. throughout the paper we use the standard cosmological constants with @xmath46 km s@xmath30 mpc@xmath30 , @xmath47 and @xmath48 . we observed with the evla using the new wideband interferometric digital architecture ( widar ) correlator to obtain up to 2 ghz of bandwidth at several frequencies . at all frequencies , we used 3c286 for bandpass and flux calibration . at 1.4 ghz , we used j1634 + 6245 for phase calibration . for phase calibration at all other frequencies , we used j1638 + 5720 , and also included a third calibrator , j1639 + 5357 , at 5.8 ghz . the data were reduced and imaged with the astronomical image processing system ( aips ) software package . the observations are summarized in table [ tab : data ] . minor changes to data values with respect to those reported in zbs11 are due to additional flagging of the data . the errors reported in table [ tab : data ] are statistical uncertainties only ; the overall systematic uncertainty in the flux calibration is @xmath49 . we observed with the ami large array at 15.4 ghz with a bandwidth of 3.75 ghz . the maximum baseline is about 110 m , with a resulting angular resolution of 25 arcsec . observations ranged in duration from 45 min to 11 hr . observations of the compact source j1638 + 5720 were interleaved at intervals of 10 min as a phase reference , and the flux density scale was established by regular observations of the calibrators 3c48 and 3c286 . the telescope measures linearly - polarized signals ( stokes i+q ) . the observations are summarized in table [ tab : data ] . we observed with carma at frequencies of 87.3 and 93.6 ghz with a total bandwidth ranging between 6.8 and 7.8 ghz . we used neptune as our primary flux calibrator , and j1824 + 568 and j1638 + 573 as bandpass and phase calibrators . the overall uncertainty in the absolute flux calibration is @xmath50 . data calibration and imaging were done with the miriad and aips software packages . the observations are summarized in table [ tab : data ] . we note some changes with respect to the flux densities reported in zbs11 , especially those epochs where the total integration time on source was less than @xmath51 min ( @xmath52 d ) . we observed with the sma using at least seven of the eight antennas , in a wide range of weather conditions , with @xmath53 ranging from 0.04 to 0.3 . in each observation we combined the two sidebands , each with a bandwidth of 4 ghz separated by 10 ghz , to increase the signal - to - noise ratio . the data were calibrated using the mir software package developed at caltech and modified for the sma , while for imaging and analysis we used miriad . gain calibration was performed using j1642 + 689 , 3c345 , and j1849 + 670 . absolute flux calibration was performed using real - time measurements of the system temperatures , with observations of neptune to set the overall scale . bandpass calibration was done using 3c454.3 , j1924 - 292 , and 3c279 . the observations are summarized in table [ tab : data ] . we observed with the nrao very long baseline array ( vlba ) and the 100-m effelsberg telescope in six epochs between 2011 april 2 and september 17 ut at 8.4 and 22 ghz . the observations were performed with eight frequency bands of 8 mhz bandwidth each in dual circular polarization , resulting in a total data rate of 512 mbps . icrfj1638 + 5720 @xcite , located only 0.92@xmath54 from was used for phase - referencing at both frequencies . at 22 ghz , we switched between the target and calibrator every 40 s , while at 8.4 ghz we spent 40 s on the calibrator and 90 s on the target . this resulted in a total integration time on of about 93 min at 8.4 ghz and 78 min at 22 ghz in epochs 1 , 2 , 5 , and 6 . epochs 3 and 4 were 1 hour shorter , resulting in integration times on source of 75 min at 8.4 ghz and 60 min at 22 ghz . a second calibrator , j1657 + 5705 @xcite , was observed for 6 min at each frequency to check the overall data calibration . at 22 ghz , we also performed @xmath55-min blocks of geodetic observations to perform atmospheric calibration ( for details see @xcite ) . the results from epoch 1 were presented in zbs11 . here , we only discuss the 22 ghz data ; the full vlbi data set will be presented together with future vlbi observations in an upcoming paper . the data were correlated at the vlba array operations center in socorro , new mexico and calibrated using aips and parseltongue @xcite . we applied the latest values of the earth orientation parameters and performed zenith delay corrections based on the results of the geodetic block observations . total electron content maps of the ionosphere were used to correct for ionospheric phase changes . amplitude calibration used system temperature measurements and standard gain curves . we performed a `` manual phase - calibration '' using the data from one scan of j1638 + 5720 to remove instrumental phase offsets among the frequency bands . we then fringe - fitted the data from j1638 + 5720 . since j1638 + 5720 has extended structure , we performed phase self - calibration , and later amplitude and phase self - calibration on j1638 + 5720 to construct robust models of j1638 + 5720 at both frequencies . these models were then used to fringe - fit the data again . finally , the calibration was transferred to and j1657 + 5705 . the data were imaged in aips using robust weighting ( with robust=0 ) . a linear fit to the positions of from epochs 26 gives a proper motion of @xmath56 @xmath20as yr@xmath30 in right ascension and @xmath57 @xmath20as yr@xmath30 in declination , with an additional systematic uncertainty of about 40 @xmath20as yr@xmath30 . this is consistent with no detectable motion . the corresponding @xmath58 upper limit on the motion at the redshift of is @xmath59 pc . since the first epoch was correlated at a different position , it was not included in this fit to avoid second order systematic errors . to obtain an accurate estimate of the angular size in our latest epoch ( 2011 september 17 ut ) , we fit gaussian models directly to the visibility data . we first self - calibrated the data in phase using solution intervals of 20 min , and then fitted a circular gaussian model using a weighted least - squares fit . the source is not resolved in our observation . for unresolved sources the fitted size can be strongly correlated with the antenna amplitude gains , which are only imprecisely known . our uncertainty on the fitted fwhm size , and consequently our upper limit , was estimated from a monte - carlo simulation in which we randomly varied the amplitude gains of the antennas by conservative factors of @xmath60% ( for a more elaborate discussion of the uncertainties in estimating angular source sizes from a similar procedure see @xcite ) . the @xmath58 upper limit on the fwhm source size is @xmath61 mas , corresponding to a projected physical diameter of @xmath62 pc . for spherical expansion this corresponds to an upper limit on the apparent expansion velocity of @xmath63 . for a collimated relativistic source with an opening angle @xmath64 the projected diameter is @xmath65 , indicating that for @xmath14 the limit on the size of corresponds to @xmath66 at @xmath67 d. we compare this result with estimates of the source size from modeling of the radio emission in [ sec : energy ] . the light curves at observed frequencies of 4.9 , 6.7 , 15.4 , 19 , 24 , 43 , 90 , and 230 ghz are presented in figure [ fig : lcs ] . as described in zbs11 , the light curves at 4.9 to 24 ghz exhibit an initial rapid increase , with a flux density of @xmath68 , followed by a shallower increase , with @xmath69 . in the millimeter band , the light curves peak on a timescale of a few days and then decline as @xmath70 . this evolution is due to a synchrotron spectrum that is optically thin in the millimeter band and optically thick in the centimeter band , with a peak frequency and flux density that decline as a function of time as @xmath71 and @xmath72 ( zbs11 ) . the properties of the spectral energy distribution , along with measurements of interstellar scintillation , established that the outflow is relativistic ( @xmath73 ) , that the density profile of the circumnuclear medium ( cnm ) is roughly @xmath74 , and that the post - shock energy increases as @xmath75 ( zbs11 ) . the most striking result from our new observations is that the millimeter flux on a timescale of @xmath76 d is significantly brighter than expected from an extrapolation of the early decline . indeed , at 90 ghz the brightness is comparable to the initial peak at @xmath77 d. a similar effect is observed at lower frequencies , where the light curves exhibit an upturn starting at @xmath78 d , with @xmath79 . the 90 ghz light curve is already in a declining ( i.e. , optically thin ) phase at @xmath80 d , while the light curves at 4.9 to 24 ghz peak at @xmath81 d ( 19 and 24 ghz ) , @xmath82 d ( 15.4 ghz ) , @xmath83 d ( 8.4 ghz ) , and @xmath84 d ( 4.9 and 6.7 ghz ) , with peak flux densities of @xmath85 mjy , @xmath86 mjy , @xmath87 mjy , and @xmath88 mjy , respectively . these observations are at odds with the light curve evolution predicted by mgm11 based on our radio data at @xmath89 d , using a model that assumes a constant energy and a steady @xmath90 cnm profile ( figure [ fig : lcs ] ) . the observed change in the light curve evolution at @xmath91 month requires an increase in the outflow kinetic energy and/or the cnm density . the evolution of the peak frequency , roughly @xmath92 at @xmath93 d , is slower than expected for the characteristic synchrotron frequency , @xmath94 ( see [ sec : model ] ) . since @xmath95 only depends on the kinetic energy , @xmath96 @xcite , the observed evolution requires an increase in the energy , with @xmath97 . at the same time , the similar light curve evolution at 4.9 and 24 ghz , which straddle the self - absorption frequency ( see [ sec : model ] ) , indicates that the density must also change . in particular , at 4.9 ghz we expect @xmath98 , while at 24 ghz we expect @xmath99 ; here @xmath100 is a fiducial density that parametrizes the cnm density , with @xmath101 . for the observed light curve evolution of @xmath79 at both frequencies , and using the inferred evolution of @xmath102 , we find that @xmath103 . thus , both the energy and density normalization appear to increase at @xmath104 d. in the next section we model the radio emission in detail to extract the temporal evolution of the energy and density . to determine the temporal evolution of the relativistic outflow and the radial density profile we use the formulation of mgm11 , which draws on the grb afterglow model of @xcite ( hereafter , gs02 ) . the mgm11 model is largely motivated by the basic properties of the outflow that were determined in our initial study of the radio emission ( zbs11 ) , as well as by the observed evolution of the x - ray emission @xcite . the model assumes that the outflow is collimated , with an opening angle @xmath64 , and has a lorentz factor of @xmath105 ; in zbs11 we demonstrated that @xmath106 ( see also @xcite ) . the kinetic luminosity of the outflow , @xmath107 , is assumed to be constant for a timescale @xmath108 , followed by a decline as @xmath109 at @xmath110 . this is expected for fallback accretion in a tde , and is observationally motivated by the evolution of the x - ray light curve , which also indicates that @xmath111 s @xcite . as a result of the long - term injection of energy , the expectation is that the kinetic energy will gradually increase from an initial value of @xmath112 to a final level of @xmath113 . from the initial x - ray luminosity , @xmath114 erg s@xmath30 , the expected total energy is @xmath115 erg , where @xmath116 is the efficiency of the jet in producing x - rays ( mgm11 ) , while the predicted beaming - corrected energy is @xmath117 erg . the interaction of the outflow with the cnm leads to synchrotron emission due to the acceleration of electrons and amplification of magnetic fields . based on the early radio observations , zbs11 demonstrated that the cnm density roughly follows @xmath74 ( hereafter , wind medium ) , which is the profile adopted for the analysis in mgm11 . the lorentz factor of the fluid behind the forward shock is @xmath118 , where @xmath119 is the circumnuclear density and @xmath120 is the density of the ejecta ( see mgm11 ) ; this relation is appropriate for the case of a newtonian reverse shock , i.e. , @xmath121 @xcite , which as we find below is appropriate for ( table [ tab : params ] ) . a key difference between the mgm11 model and the gs02 grb afterglow model is that the former includes a suppression of the flux density by a factor of @xmath122 due to the finite extent of the collimated outflow ( i.e. , @xmath123 ) . thus , the synchrotron emission from the relativistic outflow is determined by @xmath107 , @xmath119 , and the fractions of post - shock energy in the radiating electrons ( @xmath124 ) and the magnetic field ( @xmath125 ) . we use this model ( equations 711 of mgm11 ) , along with the smoothing formulation of gs02 , to fit snap - shot broad - band spectra of on a timescale of @xmath42 d. in each epoch we fix @xmath126 and @xmath127 , and we then determine the best - fit values of @xmath107 and @xmath119 ; here @xmath128 is the power law index of the electron lorentz factor distribution , @xmath129 . for our choice of @xmath128 value the synchrotron frequencies ( self - absorption : @xmath130 ; peak : @xmath95 ) and flux normalization [ @xmath131 for @xmath132 are given by ( mgm11 , gs02 ) : @xmath133&=&345\ , \epsilon_{e,-1}\ , l_{\rm j , iso,48}^{1.5}\ , t_{j,6}^{2}\ , n_{18}^{-1.5}\ , \theta_{j,-1}^2\ , \nu_{a,10}^2\ , \left(\frac{t}{t_j}\right)^{0.5}\,\ , { \rm mjy},\,\label{eqn : fnua}\end{aligned}\ ] ] where we use the notation @xmath134 , and @xmath135 is the cnm density ( @xmath119 ) at a fiducial radius of @xmath136 cm . at @xmath137 we use the same equations but with the time dependencies modified to @xmath138 , @xmath139 , and @xmath140 ( mgm11 ) . the synchrotron spectrum is given by ( c.f . , spectrum 1 of @xcite ) : @xmath141^{-1/s_1}\times \nonumber \\ \left[1+\left(\frac{\nu}{\nu_m}\right)^{s_2(\beta_2- \beta_3)}\right]^{-1/s_2 } , \label{eqn : spec1}\end{aligned}\ ] ] where @xmath142 and @xmath143 are smoothing parameters ( gs02 ) , and @xmath144 , @xmath145 , and @xmath146 are the power law indices for each segment of the synchrotron spectrum . equations [ eqn : nua ] and [ eqn : spec1 ] are appropriate when @xmath147 , and have to be modified when @xmath148 as follows ( gs02 , mgm11 ) : @xmath149 @xmath150^{-1/s_1}\times \nonumber \\ \left[1+\left(\frac{\nu}{\nu_a}\right)^{s_2(\beta_2- \beta_3)}\right]^{-1/s_2 } , \label{eqn : spec2}\end{aligned}\ ] ] where the spectrum is now normalized at @xmath151 , @xmath142 and @xmath143 , take on different values ( gs02 ) and @xmath144 , @xmath152 , and @xmath146 . to smoothly connect the evolution in the two phases we use a weighted average of equations [ eqn : spec1 ] and [ eqn : spec2 ] , with the weighting determined by the time difference relative to the transition time , @xmath153 , defined as @xmath154 . we find that @xmath155 d , so this transition does not affect the data presented in this paper . instead of imposing a specific temporal evolution on @xmath107 and @xmath119 , we model each broad - band radio spectrum independently to extract the evolution of these quantities , and in turn the time evolution of the emission radius ( @xmath156 ) and @xmath157 ( and hence @xmath105 ) , as well as the radial density profile . the results of the individual fits are shown in figure [ fig : specs ] and the relevant extracted parameters are plotted in figure [ fig : params ] and summarized in table [ tab : params ] . the broad - band seds reveal a complex evolution . at @xmath158 d both the peak frequency and peak flux density decrease with time ( as noted by zbs11 ) , but starting at @xmath159 d the peak flux density begins to increase , reaching a maximum at @xmath160 d , and subsequently declining again . the vlbi size limit of @xmath161 pc at @xmath162 d can be used to set an independent upper bound on the ratio @xmath163 . in particular , on this timescale we expect @xmath164 pc , indicating that @xmath165 ( for @xmath14 ) . from the results of our radio modeling we find on a similar timescale that @xmath166 and @xmath167 ( table [ tab : params ] ) , and hence @xmath168 is in agreement with the vlbi size limit . on the other hand , if we assume @xmath169 the resulting value of @xmath170 increases by about a factor of 15 , while @xmath135 decreases by about a factor of 2 , leading to @xmath171 , in violation of the vlbi size limit we therefore conclude from the vlbi results that @xmath172 . a detailed analysis of the evolution of @xmath95 ( shown in figure [ fig : nuanum ] ) reveals a complex behavior . while it continuously declines as a function of time , the decline rate is always shallower than the expected @xmath174 for constant energy . since @xmath95 only depends on @xmath107 and the equipartition fractions ( which are not expected to change with time ) , the shallower than expected evolution directly implies that @xmath107 continuously increases with time . in particular , at @xmath89 d and @xmath175 d , the observed evolution of @xmath176 indicates that @xmath177 , while in the intermediate phase ( @xmath178 ) d , @xmath179 points to a more rapid increase in the energy scale , with @xmath180 . the time evolution of @xmath107 is shown in figure [ fig : ljiso ] . we find that at @xmath181 d @xmath182 erg s@xmath30 ( or @xmath183 erg for @xmath184 s ) . changing the values of the equipartition fractions from our assumed values @xmath126 , we find that @xmath107 increases by about a factor of 3 if @xmath185 or by a factor of 15 if @xmath169 ( @xmath186 is ruled out by the vlbi limits on the source size ) . these scalings hold for the overall time evolution of @xmath107 . also shown in figure [ fig : ljiso ] is the expected evolution for a model with @xmath187 at @xmath188 and @xmath189 at @xmath190 , as inferred from the x - ray light curve ( @xcite , mgm11 ) . in this model , we expect the integrated value of @xmath107 ( i.e. , the total isotropic - equivalent energy ) to approach @xmath191 as @xmath192 , and to roughly double within @xmath193 d. this seems to be the case based on the radio data at @xmath194 d , but the subsequent rapid increase in energy by about an order of magnitude at @xmath195 d can not be explained with continued injection from a @xmath196 tail . on the other hand , the x - ray light curve of ( figure [ fig : ljiso ] ) agrees well with this simple luminosity evolution , with @xmath197 erg s@xmath30 at @xmath198 d. a comparison to the inferred value of @xmath199 indicates a high efficiency of producing x - rays of @xmath200 . since continuous energy injection from an @xmath5 tail can not explain the inferred rise in integrated kinetic energy , the radio data require a different energy injection mechanism . one possibility is that the outflow has a distribution of lorentz factors , with increasing energy as a function of decreasing lorentz factor , @xmath201 @xcite . in this scenario the energy will increase with time as material with a lower lorentz factor catches up with decelerated ejecta that initially had a higher lorentz factor . the continuous injection of energy will also lead to a more rapid increase in radius , with @xmath202 @xcite ; here @xmath203 is the power law index describing the cnm density profile ( @xmath204 ) . in figure [ fig : params ] we plot the inferred radius as a function of time , and find that it follows @xmath205 , faster than expected in a simple wind model ( @xmath206 ) . we also find from the density profile that @xmath207 ( see [ sec : density ] ) , thereby leading to @xmath208 . thus , we expect in this scenario a fairly steep profile for the energy as a function of lorentz factor of @xmath6 . with this relation , and the inferred time evolution of the jet lorentz factor ( figure [ fig : params ] ) we can infer the expected increase in energy . we find @xmath209 at @xmath210 d and @xmath211 at @xmath212 d , indicating an expected increase in energy by about a factor of 10 for our inferred profile . this is in excellent agreement with the data . thus , a jet with a distribution of lorentz factors of @xmath213 naturally explains the evolution of source size and the substantial increase in energy beyond the on - going input from a @xmath196 tail . in addition to the substantial increase in energy , we also find that the normalization of the density profile , @xmath214 , changes as a function of time ( figure [ fig : params ] ) . this indicates that the radial density structure is not a simple wind profile , for which @xmath135 is by definition constant . in particular , we find that @xmath135 decreases at @xmath215 d , followed by a steady increase at @xmath216 d is about a factor of 8 . the resulting radial density profile is shown in figure [ fig : nr ] , and clearly reflects the complex evolution of @xmath135 . the density profile at @xmath217 pc is @xmath218 , followed by a uniform density at @xmath4 pc , and a slow transition back to @xmath219 by @xmath15 pc . the inferred density is @xmath220 @xmath18 at @xmath221 pc , respectively . these values scale with the choice of equipartition fractions : for @xmath185 the density is about 3 times larger , while for @xmath169 it is about 1.7 times lower . thus , the inferred density is fairly robust to large changes in the equipartition fractions . the flattening at @xmath4 pc coincides with the rapid increase in energy at @xmath222 d. this can be understood in the context of an outflow with a distribution of lorentz factors since the relative increase in density compared to the previous @xmath223 profile leads to enhanced contribution from lower lorentz factor ejecta . a similar effect has been observed in the radio emission from some core - collapse supernovae , which have a steep ejecta profile with @xmath224 ( e.g. , @xcite ) . also shown in figure [ fig : nr ] are the inferred densities in the central parsec of the milky way galactic center from x - ray measurements @xcite . the galactic center density is similar that inferred for at @xmath225 pc , but is about 30 times larger at @xmath226 pc . if the bulk of the gas in the central parsec is due to mass loss from massive stars ( e.g. , @xcite ) , the lower density inferred here may indicate a much smaller number of massive stars in the central parsec of the host galaxy of . although we can not investigate star formation on scales smaller than @xmath227 kpc in the host of , the overall star formation rate in this galaxy is indeed a factor of several times lower than in the milky way @xcite . regardless of the reason for the difference in density , it is quite remarkable that the radio emission from offers as detailed a view ( or better ) of the density profile in the inner parsec around an inactive smbh at @xmath23 as available for the galactic center at a distance of only 8.5 kpc . using the radio modeling we can also predict the optical and near - ir emission from the jet . in figure [ fig : oir ] we plot the optical @xmath156-band upper limits from @xcite , as well as their near - ir @xmath228-band measurements . the latter include the total flux from and its host galaxy since transient - free templates are not presently available . we estimate the host contribution to be about @xmath229 @xmath20jy ( @xmath230 ab mag ) by requiring that the overall shape of the @xmath228-band light curve match our predicted light curve . we note that this only affects the light curve shape at @xmath231 d since at earlier times the observed flux is dominated by the transient itself . a simple extrapolation of our model over - predicts the @xmath228-band flux density by about an order of magnitude , indicating the presence of an additional break in the spectrum between the radio and optical / near - ir band . such a break is indicative of the synchrotron cooling frequency , @xmath232 . to explain the @xmath228-band flux density requires @xmath233 hz . even if we include this break , the observed @xmath156-band limits are still a factor of about 70 times fainter than the model prediction . to explain this discrepancy requires host galaxy extinction of @xmath234 mag ( see also @xcite , zbs11 ) . with the addition of a cooling break and host galaxy extinction we find an excellent match between our model and the near - ir fluxes . since these data were not used in the fitting , the agreement between our model and the data provides an independent confirmation of the model . our on - going radio observations have uncovered two unexpected and critical results regarding the nature of the relativistic outflow from and the parsec - scale environment around a previously - dormant smbh . we find that the relativistic outflow is not dominated by a single lorentz factor , as typically assumed in grb jet models . instead , the outflow has a lorentz factor distribution with @xmath235 , at least over the range of lorentz factors probed so far , @xmath236 . this scenario is reminiscent of supernova ejecta , in which the energy is a strong function of velocity , @xmath224 @xcite . in the sn case , this coupling reflects shock acceleration through the steep density gradient in the outer envelope of the star . in the case of we do not expect such a density gradient , suggesting instead that the ejecta structure may be an intrinsic property of relativistic jet launching , potentially through the blandford - znajek mechanism @xcite . in this case , the same structure may also apply to the relativistic jets of grbs , which are generally assumed to have a single lorentz factor . to assess this possibility we compare the ratio of event duration to the accretion disk dynamical timescale , @xmath237 , where @xmath238 is the black hole schwarzschild radius . for similar ratios of @xmath239 , since @xmath240 , the dynamical timescale in is about @xmath241 times longer than in grbs . the ratio of durations is similar , @xmath28 s for and @xmath242 s for grbs . therefore , the ratio of duration to dynamical timescale is similar , and the jet launching mechanism may imprint a similar profile in the case of and grbss . of course , in long grbs the jet still has to propagate through the stellar envelope , which may modify the jet structure . we next address the potential implications of the radial density profile and the flattening at @xmath4 pc . one possibility is that this scale represents the bondi radius for spherical accretion from the cnm , @xmath243 ; here @xmath244 is the sound speed in the cnm and @xmath245 is the schwarzschild radius of the smbh . for a cnm gas temperature of @xmath246 k @xcite , we find @xmath247 pc for the smbh mass of @xmath248 m@xmath9 inferred for . thus , the radius at which we infer a density enhancement is in reasonable agreement with the expected bondi radius . in addition , we note that the expected density profile inside the bondi radius , @xmath2 , matches the inferred profile for on a scale of @xmath249 pc . in the standard scenario , the medium outside the bondi radius is assumed to have a uniform density , while here we infer a continued decline in the density at @xmath250 pc . this may reflect the complex conditions in the inner few parsecs around the smbh in which the interstellar density profile may be mainly influenced by mass loss from stars ( as in the case of the milky way galactic center ; e.g. , @xcite ) . indeed , from the inferred density of @xmath251 @xmath18 at @xmath252 pc , the required mass loss rate is @xmath253 m@xmath9 yr@xmath30 for a wind velocity of @xmath254 km s@xmath30 . in particular , in the case of the galactic center it has long been believed that the gas in the central parsec is supplied by mass loss from massive stars , with the bulk of the gas ( @xmath255 ) being thermally expelled in a wind ( @xcite , but see @xcite for an alternative interpretation of the x - ray emission ) . using a spherically symmetric hydrodynamic simulation of the gas supplied by stellar winds , @xcite showed that beyond about 0.4 pc gas is mainly expelled with a resulting wind profile ( @xmath37 ) on larger scales , while on scales of @xmath256 pc the profile is roughly as expected for bondi accretion ( @xmath2 ) ; see figure [ fig : nr ] . the profile we find here is somewhat different from this model , but this may be due to a different distribution of massive stars , or to the simplifying assumption of spherical symmetry in the model . we presented radio observations of extending to @xmath0 d and spanning a wide range of frequencies . the evolution of the radio emission changes dramatically at @xmath257 month , requiring an increase in the total energy by about an order of magnitude , a density profile of @xmath2 ( @xmath3 pc ) , and a flattening at @xmath4 pc . a comparison of the model to optical limits and near - ir detections indicates a cooling break at @xmath233 hz and host galaxy extinction of @xmath258 mag . the increase in energy can not be explained by injection from an @xmath5 tail that is expected in tidal disruption events and which matches the evolution of the x - ray emission . we conclude that a natural explanation is a structured outflow with @xmath6 . the inferred density profile and the radial scale of the density enhancement are in rough agreement with the expectation for bondi accretion from a circumnuclear medium . the jet energetics and structure , as well as the detailed density profile on @xmath259 pc scale are a testament to the important insight that can be gained from continued radio observations of . in particular , the radial density profile is traced in greater detail than even the inner parsec of the milky way . continued radio observations will probe the environment to a scale of @xmath260 pc in the coming decade . using the results of our analysis we can predict the future evolution of the radio emission ( modulo any future unpredictable changes in energy and/or density as we have found here ) . we use the evolution of @xmath107 and @xmath119 as inferred from the data at @xmath261 d , and assume that the density will continue to evolve as @xmath218 and that the energy will increase to a maximum beaming - corrected value of @xmath262 $ ] with @xmath263 erg . the resulting light curves at 6 and 22 ghz are shown in figure [ fig : predlc ] . the long - term evolution is marked by a break when @xmath102 achieves its maximum value , corresponding to about @xmath264 , @xmath265 , and @xmath266 yr for our three choices of maximum energy . using the @xmath267 sensitivity of the evla in an observation of a few hours , we find that the emission at 22 and 6 ghz should be detectable for at least @xmath268 yr and @xmath269 yr , respectively . indeed , any significant upgrades to the evla or the construction of more sensitive radio facilities in the coming decades may extend the range of detectability to centuries . the same is true if the total energy scale is @xmath270 erg . an equally important question is whether the jet will be resolvable with vlbi in the future . the projected radius is @xmath271 , as long as the jet maintains its collimation . in figure [ fig : predrad ] we plot the predicted future evolution of @xmath156 using the prescription described above . we find that for @xmath14 and a best - case vlbi angular resolution of @xmath272 mas ( fwhm ) , the source should become resolvable at @xmath273 yr . on this timescale the 22 ghz flux density is expected to be only @xmath274 mjy ( figure [ fig : predlc ] ) , still accessible with vlbi . while the flux density at 6 ghz is expected to be larger by about a factor of @xmath275 , the angular resolution at this frequency is poorer by about a factor of 3.7 , making it less competitive than 22 ghz . thus , we conclude that the radio emission from may be marginally resolved in a few years . on the other hand , if the jet undergoes significant spreading on the timescale at which it becomes non - relativistic ( as expected for grb jets : e.g. , @xcite ) it is possible that it will become resolvable at @xmath276 yr when the expected 22 ghz flux density is still @xmath260 mjy . we are undertaking continued multi - frequency radio monitoring of to follow the long - term evolution of the relativistic outflow and the radial profile of the ambient medium . even in the absence of any future dramatic changes relative to the current evolution , we expect that in the next few years we may be able to determine the total energy of the relativistic outflow , measure the spreading of the jet , and study the radial density profile to a scale of @xmath260 pc . future papers in this series will detail these results . we thank ramesh narayan , ryan chornock , and nicholas stone for helpful discussions , and glen petitpas for assistance with the sma data reduction . acknowledges support from swift ao6 grant nnx10ai24 g and from the national science foundation through grant ast-1107973 . a.b . is supported by a marie curie outgoing international fellowship ( fp7 ) of the european union ( project number 275596 ) . this work is partially based on observations with the 100-m telescope of the mpifr ( max - planck - institut f@xmath277r radioastronomie ) at effelsberg . the submillimeter array is a joint project between the smithsonian astrophysical observatory and the academia sinica institute of astronomy and astrophysics , and is funded by the smithsonian institution and the academia sinica . support for carma construction was derived from the gordon and betty moore foundation , the kenneth t. and eileen l. norris foundation , the james s. mcdonnell foundation , the associates of the california institute of technology , the university of chicago , the states of california , illinois , and maryland , and the national science foundation . ongoing carma development and operations are supported by the national science foundation under a cooperative agreement , and by the carma partner universities . the ami arrays are supported by the university of cambridge and the stfc . this work made use of data supplied by the uk swift science data centre at the university of leicester . , m. , van langevelde , h. j. , reynolds , c. , & cotton , b. 2006 , in astronomical society of the pacific conference series , vol . 351 , astronomical data analysis software and systems xv , ed . c. gabriel , c. arviset , d. ponz , & s. enrique , 497 rccr 0.15 in 6.79 & evla & 1.4 & @xmath278 + 126.59 & evla & 1.4 & @xmath279 + 174.47 & evla & 1.4 & @xmath280 + 197.41 & evla & 1.4 & @xmath281 + 3.87 & evla & 4.9 & @xmath282 + 4.76 & evla & 4.9 & @xmath283 + 5.00 & evla & 4.9 & @xmath283 + 5.79 & evla & 4.9 & @xmath284 + 6.78 & evla & 4.9 & @xmath285 + 7.77 & evla & 4.9 & @xmath286 + 9.79 & evla & 4.9 & @xmath287 + 14.98 & evla & 4.9 & @xmath288 + 22.78 & evla & 4.9 & @xmath289 + 35.86 & evla & 4.9 & @xmath290 + 50.65 & evla & 4.9 & @xmath291 + 67.61 & evla & 4.9 & @xmath292 + 94.64 & evla & 4.9 & @xmath293 + 111.62 & evla & 4.9 & @xmath294 + 126.51 & evla & 4.9 & @xmath294 + 143.62 & evla & 4.9 & @xmath295 + 164.38 & evla & 4.9 & @xmath296 + 174.47 & evla & 4.9 & @xmath297 + 197.41 & evla & 4.9 & @xmath298 + 213.32 & evla & 4.9 & @xmath299 + 3.87 & evla & 6.7 & @xmath300 + 4.76 & evla & 6.7 & @xmath301 + 5.00 & evla & 6.7 & @xmath302 + 5.79 & evla & 6.7 & @xmath303 + 6.79 & evla & 6.7 & @xmath287 + 7.77 & evla & 6.7 & @xmath304 + 9.79 & evla & 6.7 & @xmath305 + 14.98 & evla & 6.7 & @xmath306 + 22.78 & evla & 6.7 & @xmath307 + 35.86 & evla & 6.7 & @xmath308 + 50.65 & evla & 6.7 & @xmath309 + 67.61 & evla & 6.7 & @xmath310 + 94.64 & evla & 6.7 & @xmath311 + 111.62 & evla & 6.7 & @xmath312 + 126.51 & evla & 6.7 & @xmath313 + 143.62 & evla & 6.7 & @xmath314 + 164.38 & evla & 6.7 & @xmath315 + 174.47 & evla & 6.7 & @xmath316 + 197.41 & evla & 6.7 & @xmath317 + 213.32 & evla & 6.7 & @xmath318 + 14.97 & evla & 8.4 & @xmath319 + 127.69 & evla & 8.4 & @xmath320 + 159.77 & evla & 8.4 & @xmath321 + 174.47 & evla & 8.4 & @xmath322 + 177.50 & evla & 8.4 & @xmath323 + 197.41 & evla & 8.4 & @xmath324 + 213.32 & evla & 8.4 & @xmath325 + 219.22 & evla & 8.4 & @xmath326 + 5.81 & ami - la & 15.4 & @xmath327 + 6.64 & ami - la & 15.4 & @xmath328 + 7.62 & ami - la & 15.4 & @xmath329 + 8.55 & ami - la & 15.4 & @xmath330 + 9.56 & ami - la & 15.4 & @xmath331 + 10.78 & ami - la & 15.4 & @xmath332 + 11.56 & ami - la & 15.4 & @xmath333 + 13.64 & ami - la & 15.4 & @xmath334 + 14.57 & ami - la & 15.4 & @xmath335 + 16.47 & ami - la & 15.4 & @xmath336 + 18.65 & ami - la & 15.4 & @xmath337 + 19.73 & ami - la & 15.4 & @xmath338 + 21.78 & ami - la & 15.4 & @xmath339 + 22.76 & ami - la & 15.4 & @xmath340 + 25.38 & ami - la & 15.4 & @xmath341 + 26.74 & ami - la & 15.4 & @xmath342 + 31.45 & ami - la & 15.4 & @xmath343 + 33.61 & ami - la & 15.4 & @xmath344 + 34.79 & ami - la & 15.4 & @xmath345 + 35.71 & ami - la & 15.4 & @xmath346 + 37.49 & ami - la & 15.4 & @xmath347 + 38.60 & ami - la & 15.4 & @xmath348 + 39.78 & ami - la & 15.4 & @xmath349 + 40.67 & ami - la & 15.4 & @xmath350 + 41.69 & ami - la & 15.4 & @xmath351 + 43.63 & ami - la & 15.4 & @xmath352 + 45.74 & ami - la & 15.4 & @xmath353 + 48.69 & ami - la & 15.4 & @xmath354 + 49.64 & ami - la & 15.4 & @xmath355 + 50.54 & ami - la & 15.4 & @xmath356 + 53.39 & ami - la & 15.4 & @xmath357 + 55.62 & ami - la & 15.4 & @xmath358 + 56.52 & ami - la & 15.4 & @xmath359 + 60.73 & ami - la & 15.4 & @xmath360 + 62.54 & ami - la & 15.4 & @xmath361 + 63.45 & ami - la & 15.4 & @xmath362 + 65.51 & ami - la & 15.4 & @xmath363 + 68.60 & ami - la & 15.4 & @xmath364 + 70.69 & ami - la & 15.4 & @xmath365 + 71.38 & ami - la & 15.4 & @xmath366 + 73.48 & ami - la & 15.4 & @xmath367 + 74.48 & ami - la & 15.4 & @xmath368 + 76.65 & ami - la & 15.4 & @xmath369 + 77.47 & ami - la & 15.4 & @xmath370 + 78.49 & ami - la & 15.4 & @xmath371 + 79.49 & ami - la & 15.4 & @xmath372 + 80.47 & ami - la & 15.4 & @xmath373 + 81.48 & ami - la & 15.4 & @xmath374 + 83.42 & ami - la & 15.4 & @xmath375 + 86.65 & ami - la & 15.4 & @xmath376 + 88.37 & ami - la & 15.4 & @xmath377 + 89.64 & ami - la & 15.4 & @xmath378 + 91.58 & ami - la & 15.4 & @xmath379 + 92.58 & ami - la & 15.4 & @xmath380 + 95.39 & ami - la & 15.4 & @xmath381 + 98.42 & ami - la & 15.4 & @xmath382 + 100.61 & ami - la & 15.4 & @xmath383 + 101.60 & ami - la & 15.4 & @xmath384 + 102.60 & ami - la & 15.4 & @xmath385 + 105.59 & ami - la & 15.4 & @xmath386 + 107.38 & ami - la & 15.4 & @xmath387 + 108.32 & ami - la & 15.4 & @xmath388 + 110.34 & ami - la & 15.4 & @xmath389 + 112.50 & ami - la & 15.4 & @xmath390 + 113.50 & ami - la & 15.4 & @xmath391 + 115.49 & ami - la & 15.4 & @xmath392 + 118.31 & ami - la & 15.4 & @xmath393 + 119.48 & ami - la & 15.4 & @xmath394 + 120.55 & ami - la & 15.4 & @xmath395 + 121.55 & ami - la & 15.4 & @xmath396 + 122.18 & ami - la & 15.4 & @xmath397 + 123.45 & ami - la & 15.4 & @xmath398 + 124.38 & ami - la & 15.4 & @xmath399 + 125.51 & ami - la & 15.4 & @xmath400 + 127.38 & ami - la & 15.4 & @xmath401 + 128.35 & ami - la & 15.4 & @xmath402 + 130.38 & ami - la & 15.4 & @xmath403 + 131.41 & ami - la & 15.4 & @xmath404 + 132.49 & ami - la & 15.4 & @xmath405 + 133.35 & ami - la & 15.4 & @xmath406 + 134.36 & ami - la & 15.4 & @xmath407 + 135.32 & ami - la & 15.4 & @xmath408 + 136.46 & ami - la & 15.4 & @xmath409 + 137.51 & ami - la & 15.4 & @xmath410 + 138.51 & ami - la & 15.4 & @xmath411 + 139.51 & ami - la & 15.4 & @xmath412 + 140.50 & ami - la & 15.4 & @xmath413 + 141.13 & ami - la & 15.4 & @xmath414 + 142.30 & ami - la & 15.4 & @xmath415 + 143.26 & ami - la & 15.4 & @xmath416 + 144.42 & ami - la & 15.4 & @xmath417 + 146.48 & ami - la & 15.4 & @xmath418 + 147.48 & ami - la & 15.4 & @xmath419 + 148.27 & ami - la & 15.4 & @xmath420 + 149.27 & ami - la & 15.4 & @xmath421 + 150.47 & ami - la & 15.4 & @xmath422 + 151.22 & ami - la & 15.4 & @xmath423 + 152.27 & ami - la & 15.4 & @xmath424 + 153.46 & ami - la & 15.4 & @xmath425 + 155.45 & ami - la & 15.4 & @xmath426 + 156.35 & ami - la & 15.4 & @xmath427 + 157.45 & ami - la & 15.4 & @xmath428 + 158.29 & ami - la & 15.4 & @xmath429 + 159.28 & ami - la & 15.4 & @xmath430 + 160.27 & ami - la & 15.4 & @xmath431 + 161.20 & ami - la & 15.4 & @xmath432 + 162.21 & ami - la & 15.4 & @xmath433 + 163.28 & ami - la & 15.4 & @xmath434 + 166.00 & ami - la & 15.4 & @xmath435 + 168.13 & ami - la & 15.4 & @xmath436 + 169.13 & ami - la & 15.4 & @xmath437 + 172.25 & ami - la & 15.4 & @xmath438 + 174.33 & ami - la & 15.4 & @xmath439 + 177.32 & ami - la & 15.4 & @xmath440 + 181.30 & ami - la & 15.4 & @xmath441 + 184.03 & ami - la & 15.4 & @xmath442 + 187.19 & ami - la & 15.4 & @xmath443 + 187.91 & ami - la & 15.4 & @xmath444 + 189.37 & ami - la & 15.4 & @xmath445 + 193.19 & ami - la & 15.4 & @xmath446 + 194.01 & ami - la & 15.4 & @xmath447 + 197.02 & ami - la & 15.4 & @xmath448 + 200.01 & ami - la & 15.4 & @xmath449 + 203.01 & ami - la & 15.4 & @xmath450 + 207.08 & ami - la & 15.4 & @xmath451 + 210.19 & ami - la & 15.4 & @xmath452 + 214.17 & ami - la & 15.4 & @xmath453 + 217.16 & ami - la & 15.4 & @xmath454 + 220.20 & ami - la & 15.4 & @xmath455 + 221.22 & ami - la & 15.4 & @xmath456 + 225.91 & ami - la & 15.4 & @xmath457 + 232.87 & ami - la & 15.4 & @xmath458 + 235.01 & ami - la & 15.4 & @xmath459 + 237.85 & ami - la & 15.4 & @xmath460 + 4.79 & evla & 19.1 & @xmath461 + 6.75 & evla & 19.1 & @xmath462 + 7.77 & evla & 19.1 & @xmath463 + 8.87 & evla & 19.1 & @xmath464 + 9.78 & evla & 19.1 & @xmath465 + 21.89 & evla & 19.1 & @xmath466 + 31.74 & evla & 19.1 & @xmath467 + 35.86 & evla & 19.1 & @xmath468 + 50.65 & evla & 19.1 & @xmath469 + 67.61 & evla & 19.1 & @xmath470 + 94.64 & evla & 19.1 & @xmath471 + 111.62 & evla & 19.1 & @xmath472 + 127.83 & evla & 19.1 & @xmath473 + 142.62 & evla & 19.1 & @xmath474 + 159.77 & evla & 19.1 & @xmath475 + 177.50 & evla & 19.1 & @xmath476 + 198.22 & evla & 19.1 & @xmath477 + 219.22 & evla & 19.1 & @xmath478 + 4.79 & evla & 24.4 & @xmath479 + 6.75 & evla & 24.4 & @xmath480 + 7.77 & evla & 24.4 & @xmath481 + 8.87 & evla & 24.4 & @xmath482 + 9.78 & evla & 24.4 & @xmath483 + 21.89 & evla & 24.4 & @xmath484 + 31.74 & evla & 24.4 & @xmath485 + 35.86 & evla & 24.4 & @xmath486 + 50.65 & evla & 24.4 & @xmath487 + 67.61 & evla & 24.4 & @xmath488 + 94.64 & evla & 24.4 & @xmath489 + 111.62 & evla & 24.4 & @xmath490 + 127.83 & evla & 24.4 & @xmath491 + 142.62 & evla & 24.4 & @xmath492 + 159.77 & evla & 24.4 & @xmath493 + 177.50 & evla & 24.4 & @xmath494 + 198.22 & evla & 24.4 & @xmath495 + 219.22 & evla & 24.4 & @xmath496 + 5.75 & evla & 43.6 & @xmath497 + 6.75 & evla & 43.6 & @xmath498 + 8.87 & evla & 43.6 & @xmath499 + 111.62 & evla & 43.6 & @xmath500 + 127.83 & evla & 43.6 & @xmath501 + 159.77 & evla & 43.6 & @xmath502 + 177.50 & evla & 43.6 & @xmath503 + 198.22 & evla & 43.6 & @xmath504 + 219.22 & evla & 43.6 & @xmath505 + 4.90 & carma & 87 & @xmath506 + 8.19 & carma & 87 & @xmath507 + 9.14 & carma & 87 & @xmath508 + 10.23 & carma & 87 & @xmath509 + 12.14 & carma & 87 & @xmath510 + 17.66 & carma & 87 & @xmath511 + 22.11 & carma & 87 & @xmath512 + 25.12 & carma & 87 & @xmath513 + 99.75 & carma & 87 & @xmath514 + 131.52 & carma & 87 & @xmath515 + 148.66 & carma & 87 & @xmath516 + 175.56 & carma & 87 & @xmath517 + 19.25 & sma & 200 & @xmath518 + 24.32 & sma & 200 & @xmath519 + 10.30 & sma & 230 & @xmath520 + 11.13 & sma & 230 & @xmath521 + 17.23 & sma & 230 & @xmath522 + 18.25 & sma & 230 & @xmath523 + 20.24 & sma & 230 & @xmath524 + 21.25 & sma & 230 & @xmath525 + 125.05 & sma & 230 & @xmath526 + 5.13 & sma & 345 & @xmath527 ccccccccccc 0.1 in @xmath528 & @xmath529 & @xmath530 & @xmath531 & @xmath532 & @xmath533 & @xmath534 & @xmath535 & @xmath536 & @xmath537 & @xmath538 + @xmath539 & @xmath540 & @xmath541 & @xmath542 & @xmath543 & @xmath544 & @xmath545 & @xmath546 & @xmath547 & @xmath548 & @xmath549 + @xmath550 & @xmath551 & @xmath552 & @xmath553 & @xmath554 & @xmath555 & @xmath533 & @xmath535 & @xmath556 & @xmath557 & @xmath558 + @xmath559 & @xmath560 & @xmath561 & @xmath562 & @xmath563 & @xmath564 & @xmath565 & @xmath566 & @xmath567 & @xmath568 & @xmath569 + @xmath570 & @xmath571 & @xmath572 & @xmath573 & @xmath574 & @xmath575 & @xmath576 & @xmath577 & @xmath578 & @xmath579 & @xmath580 + @xmath581 & @xmath560 & @xmath582 & @xmath583 & @xmath584 & @xmath585 & @xmath586 & @xmath587 & @xmath588 & @xmath589 & @xmath590 + @xmath591 & @xmath592 & @xmath593 & @xmath594 & @xmath595 & @xmath596 & @xmath575 & @xmath597 & @xmath536 & @xmath598 & @xmath599 + @xmath600 & @xmath571 & @xmath601 & @xmath602 & @xmath603 & @xmath604 & @xmath605 & @xmath606 & @xmath607 & @xmath608 & @xmath609 + @xmath610 & @xmath611 & @xmath612 & @xmath613 & @xmath614 & @xmath615 & @xmath596 & @xmath616 & @xmath617 & @xmath618 & @xmath619 + @xmath620 & @xmath621 & @xmath622 & @xmath623 & @xmath624 & @xmath615 & @xmath604 & @xmath625 & @xmath626 & @xmath627 & @xmath628 + @xmath629 & @xmath630 & @xmath631 & @xmath632 & @xmath633 & @xmath634 & @xmath635 & @xmath636 & @xmath624 & @xmath627 & @xmath637 + @xmath638 & @xmath639 & @xmath592 & @xmath640 & @xmath641 & @xmath642 & @xmath643 & @xmath644 & @xmath645 & @xmath646 & @xmath647

we describe the astrometric calibration of the gemini nici planet - finding campaign . the campaign requires a relative astrometric accuracy of mas across multi - year timescales in order to distinguish true companions from background stars by verifying common proper motion and parallax with their parent stars . the calibration consists of a correction for instrumental optical image distortion , plus on - sky imaging of astrometric fields to determine the pixel scale and image orientation . we achieve an accuracy of mas between the center and edge of the 18 nici field , meeting the 20 mas requirement . most of the campaign data in the gemini science archive are accurate to this level but we identify a number of anomalies and present methods to correct the errors .

1411.0641

the gemini nici planet - finding campaign was a direct - imaging survey of about 250 nearby stars for substellar and planetary - mass companions conducted at gemini south observatory between 2008 and 2012 @xcite . the campaign used the near - infrared coronagraphic imager ( nici , * ? ? ? * ) , which combines adaptive optics , coronagraphy , angular differential imaging @xcite , and dual - channel methane - band infrared imaging to achieve an h - band contrast detection limit of 14.4 magnitudes at 1@xmath2 radius @xcite . the principal scientific results have been published by @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite ; the pipeline processing algorithm is described by @xcite . the multi - epoch imaging data acquired for the campaign require accurate astrometric calibration in order to detect common proper motion and parallax between the host star and any candidate companions and therefore distinguish true companions from background stars . in this paper , we describe the process used to establish the alignment of the nici science detectors relative to the celestial coordinate system and calibrate the world coordinate system ( wcs ) contained in the data headers . nici consists of two parts : an adaptive optics ( ao ) system and a dual - channel science camera @xcite . the ao - corrected wavefront entering the science camera first passes through a focal plane coronagraph mask and a pupil / spider mask , then is divided between the two science channels by the _ dichroic wheel _ ( dw ) . light transmitted through the dw enters the _ red _ channel while reflected light enters the _ blue_. the dw can be set to one of the following elements : _ h-50/50 _ beamsplitter , which directs 50% of h - band ( 1.65 ) light to each science channel ; _ h / k dichroic _ , which divides light between the h and k ( 2.2 ) bands ; _ mirror _ , which reflects all light to the blue channel ; or _ open _ , which passes all light to the red channel . after the dw , each science channel contains a filter wheel , reimaging optics , and an aladdin ii 1024@xmath31024 insb array detector . the campaign most frequently used 4% bandpass filters centered in and out of the @xmath4 m methane absorption feature ( named ch4-h4%l and ch4-h4%s , respectively ) to search for methane - bearing planets very close to the primary star , and a broadband h filter for deeper searches at larger separations . the on - sky pixel size is @xmath5 milli - arcseconds ( mas ) and the field size is @xmath6 . the mapping of the sky onto each of the two science detectors is different and changes with the dw element and filter . nici data are processed by the gemini data - handling system ( dhs ) and are written to multi - extension fits ( mef ) files which contain a primary header followed by two extensions , one for each of the two science channels . the header for each extension contains its own wcs specifying the astrometric parameters for that channel . the calibration of the wcs for each channel is a principal subject of this work . details on the data format , wcs , and other header data are specified in the appendix , and more information on nici is available at http://www.gemini.edu/sciops/instruments/nici . the astrometric calibration requirements for the nici planet - finding campaign arise from three aspects of the observing strategy . first , most campaign data were taken in a mode called _ angular differential imaging _ ( adi , * ? ? ? * ; * ? ? ? * ) , in which the cassegrain rotator is held fixed and the sky allowed to rotate on the science detectors so that astronomical objects surrounding the primary star under observation move relative to the fixed speckles in the nici point - spread function ( psf ) . the individual @xmath7 minute exposures taken over a range of field angles ( in some cases @xmath8 ) were derotated and registered before being co - added by the pipeline - processing software . to achieve accurate alignment over the entire field , image distortion must be corrected before derotation . second , targets were imaged simultaneously in the two science channels in and out of the @xmath4 m methane absorption feature , in a mode known as _ angular spectral differential imaging _ ( asdi ) , to suppress the stellar continuum further and enhance potential methane - bearing planets . the distortion , rotation and scale difference between the two channels must be corrected before the pipeline can accurately register and subtract the two images , especially when they are added in order to find non - methane - bearing planets . in practice , sufficient accuracy can be achieved because these differences are static properties of the instrument that can be measured once and applied to all data . third , the campaign strategy required measuring the relative proper motion ( pm ) between a target star and any candidate companions detected in the surrounding field in order to distinguish true companions ( with common pm ) from background stars . because only the relative pm is required , we are not concerned with absolute astrometry ( i.e. the absolute celestial coordinate zero points ) , but only with the image distortion , field rotation , and pixel scale . in order to assess proper motion and parallax robustly , this calibration had to be maintained over multi - year timescales , during which time nici was dismounted and remounted on the telescope , sometimes on different instrument ports with different numbers of reflections in the optical path to the instrument . therefore , the astrometric calibration required observations of astrometric fields during each instrument mounting . in addition , because a field observed at multiple epochs was not typically observed at the same hour angle each time , the field rotation was often significantly different , again requiring correction of the image distortion before the two datasets could be compared . the required astrometric accuracy is dictated by the pm of the target stars . in the campaign s target list of nearby stars , the lowest pms are @xmath0 mas yr@xmath9 , or @xmath10 nici pixel yr@xmath9 ; most are at least a few times larger . an accuracy of 1 pixel from the center of the nici field ( the usual location of a primary star ) to the edge therefore permits most candidate companions to be checked for common pm within one year , especially since accuracy near the field center will be considerably better . for the pixel scale , 1 pixel out of 512 is 0.2% , while a rotation error of @xmath11 corresponds to 1 pixel across half the field . these constraints will serve as guidelines throughout this paper . our astrometric calibration process has three principal components : 1 . correcting the image distortion introduced by the instrument optics ; 2 . measuring the field rotation and setting the instrument alignment angle ; 3 . constructing the wcs . the two nici channels suffer from image distortion arising from the off - axis reflective optics in both the ao system and science camera . to measure the distortion , we imaged a pinhole grid mask mounted in the _ fiber optic calibration source _ ( focs ) , which can be deployed into the telescope focal plane at the entrance to the ao system . the focs mask distortion at the science detectors represents the combined distortions of the ao system and science camera optics . we preferred to use the mask rather than an astronomical field due to the grid spots very high signal - to - noise , their uniform size and distribution over almost the entire field , and the freedom from seeing and anisoplanatism effects . focs grid mask images are shown in figure [ fig - gridmask ] . the grid is rotated by @xmath12 relative to the detector . the blue channel image is reflected left - right relative to the red due to the extra reflection at the dw optic in the path to the blue channel . to determine the distortion , a grid mask image is first background - subtracted and the centroid of each spot measured . a synthetic rectilinear grid is then fitted to the spot positions within 256 pixels of the image center , the region of lowest distortion , by minimizing the @xmath13 statistic @xmath14|^2}{\sigma_i^2},\ ] ] where @xmath15 is the measured centroid of spot @xmath16 , @xmath17 $ ] are the spot positions in the synthetic grid , and @xmath18 is the uncertainty of the centroid measurement . there are four free parameters : @xmath19 and @xmath20 represent the overall grid position , @xmath21 is the spot separation in pixels ( assumed to be the same in x and y ) , and @xmath22 is the grid rotation . the @xmath13 value is minimized using the nelder - mead simplex method @xcite as implemented in the idl built - in routine amoeba ( exelis visual information solutions , boulder , colorado ) . the simplex algorithm begins by computing @xmath13 for a set of trial parameters , then finds the trajectory through parameter space that steadily reduces @xmath13 until a minimum is reached . the uncertainty @xmath23 is set to 1 for all points because we are only interested in the best - fit rectangular grid at this first stage , not the positional uncertainties . we used the idl routine poly_2d to transform the images geometrically to correct the distortion . the transformation is defined by polynomials of degree @xmath24 : @xmath25 @xmath26 where @xmath27 and @xmath28 are the initial pixel coordinates and @xmath29 and @xmath30 are the transformed coordinates . @xmath31 and @xmath32 are coefficients determined from a least - squares fit performed by the companion routine poly_warp , based on the measured grid mask ( @xmath33 ) positions and the rotated rectilinear grid ( @xmath34 ) positions . after computing ( @xmath34 ) for all the pixels , poly_2d generates a transformed image using cubic convolution . the residual errors of the distortion correction were computed by applying the correction to the grid mask images , then measuring the difference between the rectilinear points and the mask spots . we derived distortion corrections for six pairs of images taken in five different instrument configurations as shown in table [ tab - dcparams ] . by experimentation , we determined that a fifth - order ( @xmath35 ) fit had a residual error of 0.5 0.6 pixels root - mean - square ( rms ) for both channels , and that higher - order polynomials did not reduce the error . the initial distortion is as high as 12 pixels in the array corners . given the high signal - to - noise images of the spots , the uncertainty of the individual centroid measurements is much smaller than 0.5 pixels , so the quality of the fits is limited by some source of systematic error . ( in statistical terms , if @xmath23 is set to 0.1 pixel , the resultant @xmath13 per degree of freedom of the perfect grid relative to the distortion - corrected images is @xmath36 , much higher than the value of 1 expected when random errors dominate . ) the main source of systematic error is most likely the use of a polynomial to approximate the distortion ; an optical raytrace model may be required to improve the fit . the accuracy of the polynomial fit is within our 1 pixel goal , so we accept it as satisfactory for the campaign . we fitted six different images to evaluate the consistency of the distortion across multiple instrument configurations . the consistency is a concern because the grid position on the blue channel array varies by up to a few dozen pixels ( table [ tab - spotpos ] ) due to the slightly different mounting angles of the optical elements in the dichroic wheel , and in some cases the failure of dw to seat consistently in its detent . the blue channel variations are much larger than the red because the reflected beam from dw is deviated more by optical element tilts than the transmitted beam . for the h-50/50 beamsplitter , the blue channel ( @xmath37 ) positions vary by up to 25 pixels between different positionings ( two extremes `` min '' and `` max '' are listed in table [ tab - spotpos ] ) ; evidently the dw does not fully settle into its detent at the h-50/50 position . such large variations are not detected for the other dw elements . across the six datasets , there are only small variations in each of the parameters for scale and rotation ( see table [ tab - dcparams ] ) . the spot separations vary by 0.05% for the red channel and 0.10% for the blue , while the rotations are consistent to @xmath38 for the red channel and @xmath39 for the blue . if the open position is omitted , a reasonable action given that dw substrates probably introduce aberrations into the transmitted beam that would be absent for the open position , the red channel rotation range is only @xmath40 . the blue channel exhibits dw - dependent rotations which appear to be correlated with the largest position offsets listed in table [ tab - spotpos ] , indicating that the irreproducible positions of the dw are causing both effects . the rotations are smaller than our @xmath41 precision goal , and not much larger than the rms error of the fit , but nevertheless appear to be systematic . the fitting results do reveal significant differences in scale and rotation between the red and blue channels . how these differences are managed is described in the wcs section below . to evaluate distortion variations between the instrument configurations , we applied the correction for the h-50/50 + ch4-h4% images to the other four datasets . table [ tab - dch5050 ] shows that in all cases the errors are @xmath42 pixel rms , or @xmath43 . given the acceptable repeatability of the h-50/50 + ch4-h4correction across all the instrument configurations , we adopted it as the standard correction for all campaign data . the distortion correction coefficients for this mode are listed in table [ tab - dccoeffs ] , and maps of the distortion correction vectors and the post - correction residual errors are displayed in figure [ fig - distmap ] . more than half of the campaign data were taken in the dual - channel asdi mode with this configuration . applications which require the highest possible accuracy , however , may wish to use more detailed calibration datasets taken in a particular configuration . the telescope control system ( tcs ) uses a simple formula to control the cassegrain rotator ( cr ) so that the desired position angle on sky , @xmath44 , the _ instrument position angle _ , is oriented parallel to the detector columns toward the top of the image . a quantity @xmath45 , the _ instrument alignment angle _ , is defined to represent the rotation between the cr and science detector reference frames . the cr angle @xmath46 is related to the other angles by @xmath47 where @xmath48 is the parallactic angle of the target under observation . for example , if a detector is mounted exactly aligned with the cr frame , @xmath49 . to achieve @xmath50 ( north up on detector ) on the meridian where @xmath51 , @xmath46 would be 180@xmath52 . note that @xmath45 is dependent on the mechanical orientation of the instrument on the telescope ; small rotations or other changes in the mounting may in turn require a different @xmath45 setting in order to achieve an accurate @xmath44 . observations may be defined with cr in two modes : _ follow _ mode , in which case the tcs sets and continuously updates the cr to keep the specified @xmath44 vertical on the detector , and _ fixed _ mode where the tcs sets the cr to a fixed value , allowing the sky ( and therefore @xmath44 ) to rotate on the detector during the observation . follow mode is used for most observations with other science instruments , but fixed is used for nici adi and asdi observations . for nici , @xmath45 can represent only one of the two science channels . the natural choice is the red channel , due to its smaller dw - dependent systematic variations in scale and rotation than the blue s . with this calibration , at @xmath50 , north is exactly vertical on the red channel , while it is offset by @xmath53 on the blue channel . setting @xmath45 requires observing an astrometric standard target or field , an exercise which must be performed each time nici is mounted on the telescope . table [ tab - iaa ] lists the individual nici mountings between 2008 and 2012 , the astrometric standard observed , and the @xmath45 and pixel scale results . over the course of the campaign , our calibration technique grew more sophisticated as we added distortion corrections and switched from binary stars to a more precise astrometric field . for the campaign s first two years , from 2008 august to 2010 october , we used two binary stars for astrometric calibration . an ideal binary would have a separation between 5@xmath2 and 10@xmath2 ( small enough to fit into nici s 18@xmath2 field yet large enough to provide a sufficient lever arm for an accurate measurement ) , a separation precision @xmath54 a few mas , and a position angle precision @xmath55 . such high precision requires modern speckle measurements ; orbits based on older historical data are not sufficiently precise . several well - known systems with precise orbits were rejected because their stars were too bright , saturating on nici even with the shortest possible exposure times . we adopted ( stf 2272ab , wds 18055 + 0230 ) , which has an accurate speckle - based orbit and 55 separation in mid-2008 , as our initial astrometric standard . the @xmath56 primary component was bright enough to provide sufficient s / n when placed under the coronagraph mask . the off - mask image of the secondary component , with @xmath57 , saturated slightly in the psf core , but its position could still be measured to @xmath58 pixel based on the unsaturated halo . in this way , @xmath45 for mounting 1 was measured ( see table [ tab - iaa ] ) , with an estimated uncertainty of @xmath59 . when nici changed iss ports ( from side - looking to up - looking ) in late 2009 , was not accessible , so we observed the binary ( ccdm j22398 - 1942 b , wds 22398 - 1942 ) instead . was observed again for mounting 3 in 2010 march . at this time we had not developed the distortion - correction algorithm , so these data were reduced and @xmath45 set with no distortion correction . follow - up checks and corrections of these early binary star calibrations are described in [ sec - earlycheck ] . at the start of mounting 4 in 2010 october , we observed the trapezium cluster in orion as an astrometric field . we used the _ hst_-derived coordinates from @xcite , applied the grid mask distortion correction procedure described in [ sec - distortion ] , and fitted the measured position centroids to the celestial coordinates as described for the lmc field in [ sec - lmc ] . although the technique appeared to be superior to the binary star calibrations , the fitting errors were unacceptably large : @xmath60 pixels rms , considerably larger than what we eventually achieved for the lmc field using the same observing and reduction techniques . the large errors may be caused by systematic errors in the _ hst _ optical astrometry or the underlying two micron all sky survey ( 2mass ) catalog @xcite absolute astrometry against which the _ hst _ data were registered . the magnitude of these errors led us to develop the lmc - based calibration described in the next section . note that during this period in 2010 october - december , while new software was being developed to apply the distortion correction , measure the alignment , and compute @xmath45 changes , a number of errors were made computing the direction of the @xmath45 corrections and the wcs parameters in the headers , as indicated in table [ tab - iaa ] . methods for correcting these errors are described in [ sec - wcs ] . in 2010 november we began to observe a field in the large magellanic cloud ( lmc ) that was the subject of a detailed _ hst_-based astrometric calibration in support of _ jwst _ @xcite . for _ jwst _ , the field was chosen to be relatively free of bright stars , but fortunately it contains three @xmath61 mag stars bright enough to serve as natural guide stars for nici s ao system . the data file containing high - precision coordinates was supplied to us by j.anderson . the subfield with the brightest guide star , named `` f606w_11.341 '' in the stsci documentation but which we refer to as `` lmc-11mag , '' contains 12 stars with @xmath62 which can be detected by nici in about 5 minutes observing time . for consistency , the field was observed in cass rotator follow mode at @xmath44 = 0 , always with the clear focal - plane mask ( no coronagraph ) , the h-50/50 beamsplitter , and the ch4-h4%l and s filters . image pairs taken with a 4@xmath2 dither were subtracted to cancel the background emission ( fig . [ fig - lmc11mag ] ) , then the standard distortion correction was applied and the stars centroid positions in the red channel measured . the centroid position probable errors were estimated by synthesizing images with gaussian psf s over a range of s / n combined with backgrounds with random variations . after the centroid for a given s / n psf was measured for each of the random backgrounds , the standard deviation @xmath23 could be computed directly , and a lookup table of s / n vs. @xmath23 constructed . the probable error of each star in the real image was then estimated by measuring its s / n and finding the corresponding @xmath23 in the lookup table . the centroid positions were fitted with a @xmath13 minimization technique similar to the grid mask fitting described by eqn.[eq - chisq ] to determine the pixel scale and the @xmath45 correction , using the probable errors @xmath23 estimated for each star . table [ tab - lmc ] displays the results of an lmc-11mag astrometric fit taken for mounting 10 on 2012 december 22 utc . the @xmath63 mag guide star consistently shows a large 5 - 6 pixel error in all images , so it is not included in the fit . ( the exact reason for the error is unknown ; the star is certainly a foreground star , but the error appears to be too large to be explained by proper motion , and no proper motion was detected between the different nici astrometric images over a period of 3 years . ) as with the pinhole - grid images , the value of @xmath13 per degree of freedom @xmath64 indicates that the errors are dominated by systematic effects . in addition to the distortion correction errors described in [ sec - fitproc ] , additional sources of error on a star field are errors in the provided celestial coordinates , proper motions , atmospheric refraction , and anisoplanicity effects which cause stars far from the center of the field to become elongated . the residual error between the expected and measured positions of all the lmc field stars is 0.68 pixels rms , with no errors @xmath65 pixel , better than our goal of @xmath42 pixel across half the field . this 2012 december dataset was taken in good seeing and is one of our best results , but in all our calibrations the residual errors are @xmath66 pixels across the field , or @xmath67 and 0046 in rotation . this corresponds to a center - to - edge accuracy of @xmath68 pixels or 7 mas , better than our 20 mas requirement . the errors are similar for both up - looking and side - looking port mountings , even with the extra reflection on the side port . after 2010 the lmc-11mag field became the primary nici astrometric reference , observed regularly after most instrument mountings and port changes and periodically during observing semesters . we find that @xmath45 varies by a few tenths degree between mountings , but periodic checks within a mounting are consistent to @xmath69 , exceeding our calibration requirements . we calibrated a secondary field , , relative to lmc-11mag for use during the southern winter months when the lmc is inaccessible ; our derived astrometric data for this field are listed in table [ tab - hip62403 ] . the lmc fits also yield the image pixel scale which , unlike @xmath45 , should not vary between mountings . over 12 separate lmc measurements , we find a red channel scale of [email protected] mas / pixel . the relative probable error is 0.12% , or about 1 pixel across the detector , again within our requirements . after developing the distortion correction and lmc - based calibration in late 2010 , we checked the earlier binary star calibrations in two ways . first , the binary star centroids were remeasured after applying the grid mask distortion correction . we found that the mounting 1 and 2 calibrations were unaffected by the distortion correction to a level of @xmath71 , because both binary components were placed within the low - distortion central region . for mounting 3 , however , 70 oph b was placed in the lower left corner of the field and suffered a distortion of @xmath72 pixels , causing a rotation error of @xmath73 . as a second check method , we utilized several campaign target fields with multiple background stars that had been observed both with binary star and later lmc - based astrometric calibration . the results are summarized in table [ tab - astromcorr ] . this method verified the accuracy of the mounting 1 and 2 calibrations and the error with mounting 3 . two good comparison fields for mounting 3 indicated an error of @xmath74 and a true @xmath45 = 112.2 , which we judge to be more reliable than the 70 oph derived value due to saturation of 70 oph b and its position in the corner of the field . a detailed procedure for correcting the @xmath45 error is presented in [ sec - wcs ] ; these corrections were applied to the campaign data reductions , but _ not _ to the raw data stored in the gemini science archive . these checks revealed one range of dates , 2009 april 26 - 27 utc , for which the field rotation is in error by @xmath74 , an unusually large amount . and data taken on these dates both exhibit the anomaly . we suspect that an error in the cass rotator position or datum sometime between april 8 and 26 caused this error . after april 27 , nici operations were suspended for the winter , so we do not have futher data until 2009 august , by which time the cass rotator position is correct . the wcs in each nici mef extension specifies the mapping between sky and detector for that channel , including the field rotation , pixel scale , and reflection between sky and detector . it has limitations , however , in that it describes the absolute astrometric position only to low accuracy and does not represent the image distortion . in addition , during nici s operations there were many instances when @xmath45 was set incorrectly and therefore the wcs does not represent the true astrometry . therefore , in this section we will review how the wcs is constructed , its limitations and errors , and how the errors can be corrected . the gemini algorithm that builds the wcs for each mef file begins with a mapping file that contains a series of @xmath37 positions in detector coordinates along with their corresponding on - sky angular offsets from the field center in the telescope coordinate system . the algorithm fits a transformation to these points , then applies the current telescope pointing and cassegrain rotator angles to compute the on - sky wcs . the original intent was to populate the mapping file with points measured by offsetting a star to different positions on the detector and recording both the detector and celestial positions . for nici s very narrow field of view and small pixels , however , this technique is insufficiently precise . ( it relies on the mapping accuracy of the peripheral wavefront sensor probe arm , which could contain errors of several tenths of an arcsec . ) therefore , we developed a simple script to generate synthetic mapping files based on the pixel scale and @xmath45 rotation angle determined from the calibrations . this technique produces a wcs limited only by the quality of the astrometric field measurements as described in preceeding sections . the red channel mapping file is generated directly from the measured red channel scale and @xmath45 . the blue channel file is generated using the more precise scale and rotation angle relative to the red channel determined from the grid mask images . for the lmc calibrations after 2010 november , the blue wcs rotation angle is set @xmath53 from the red angle , and the scale is 1/0.9980 = 1.0020 @xmath3 the red scale ( with the scale , in mas / pixel , being inversely proportional to the magnification ) . gemini s wcs standard does not include distortion parameters such as the sip system , which defines fits header distortion coefficients to supplement the wcs @xcite . therefore , for nici files after 2010 november , the wcs represents the coordinate system _ after _ the standard distortion correction ( for h-50/50 + ch4-h4% ) described in [ sec - distortion ] has been applied . in addition , note that the wcs represents the orientation at the _ beginning _ of the exposure . if the cr mode is follow , then this orientation is correct throughout the exposure . however , if the cr mode is fixed , as is the case for adi observations , a rotation must be applied to correct the change in field angle between the start and mid - point of the exposure . as we previously described in [ sec - binstars ] and [ sec - trapezium ] , the @xmath45 and pixel scale values measured during several calibrations contain errors . these errors are also manifested in the wcs . known issues are : 1 . the wcs values for each mounting are based on the pixel scale derived from a particular calibration dataset . for highest accuracy and consistency , we recommend using the mean lmc - based scales of 17.958 mas / pixel for the red channel and 17.994 for blue for all epochs . 2 . the values of @xmath45 and wcs may be incorrect for data taken immediately after each mounting , until astrometric calibrations could be taken and analyzed . the affected dates are listed in table [ tab - iaa ] . mounting 4 is the most severe example : data from 2010 october and november contain several errors due to problems with the distortion - correction and astrometric fitting software used at that time for the trapezium and lmc fields . 3 . before 2010 - 12 - 14 , the rotation of the blue wcs relative to the red did not use the correct value of @xmath53 . 4 . for mounting 3 from 2010 march to october , the @xmath45 value of 111.70 was too low by @xmath74 . data taken on 2009 april 26 - 27 have a rotation error of @xmath74 , apparently caused by a short - term error in the cass rotator . ( see last paragraph of [ sec - earlycheck ] the wcs rotation correction for the red channel is listed in table [ tab - iaa ] . in addition , we have developed an idl script that automatically corrects a mef file s wcs according to the date of observation ( see appendix ) . note that the script does not modify the image data ; only the wcs is updated to represent the true astrometry of the data . the updated file with the corrected wcs can then be pipeline - processed as usual . several dozen of the campaign s target fields contained candidate companions which required follow - up checks for common proper motion and parallax . the vast majority of these candidates were determined to be background stars . we can use these multi - epoch observations to evaluate the consistency of the astrometric calibration over the course of the campaign . our technique is simply to compare the relative positions of the primary target and background stars at the available epochs , after correcting the primary s relative parallax and proper motion to the first epoch . differences in the corrected positions between epochs indicate errors in the position measurements themselves , the proper motion correction , the astrometric calibrations , or a combination of all three effects . some dense fields contained multiple background stars , permitting more detailed checks independent of the proper motion correction , but because many fields just contained the primary and one background star we will discuss only the primary - to - background position differences here . figure [ fig - offsetdistrib ] shows the offset distributions for a total of 218 pairs of observations of 205 background stars in 113 fields . the offsets are indicated parallel and perpendicular to the radius vector from the primary to the background star , which is appropriate given the large derotations applied to the adi frames before stacking and helps to identify systematic errors in scale and rotation . the left panel of figure [ fig - offsetdistrib ] shows the separation @xmath75 and perpendicular @xmath76 offsets vs. @xmath75 itself . the distribution of the @xmath75 offsets shows a possible increase with radius , suggesting scale errors on the order of 0.1% . the @xmath21 offsets , however , maintain a consistent width from small to large @xmath75 , indicating that the primary source of error is in the individual position measurements of the stars . in other words , if the dominant source of errors were in the rotation calibration , the distribution would be expected to increase proportional to @xmath75 , which is not observed . the right panel shows the distribution histograms in the same pixel units . the distributions have a central core with @xmath77 pixels in @xmath75 and 1.20 in @xmath21 ; values that are only slightly larger than the rms errors of the distortion correction and the lmc field astrometric fits . above 3.5 pixels total offset lie 9 _ outlier _ points , all of which have @xmath78 . the large errors in these cases are caused by inaccurate distortion correction near the edges of the field and very different positions of the stars on the detector between the two epochs due to either different field angles , or to the two observations being taken on different telescope ports with an odd and even number of reflections , which reversed the image and caused the same object to appear on opposite sides of the detector . that the primary source of the @xmath21 offset errors appears to be due to individual position errors does not have an obvious explanation . because the psf core fwhm is usually 3 - 4 pixels , centroid measurement errors as large as 3 pixels would typically be several @xmath23 . it should be noted that the astrometry is measured from final images which are coadded from many derotated adi frames , which tends to broaden the individual stars . possible causes of systematic error are small - scale irregularities in the distortion correction , shifts in the position of the primary star which is dimmed by more than 6 magnitudes by the partially transparent occulting mask , or other irregularities in the mask . we have developed an astrometric calibration for the nici planet - finding campaign based on a grid mask distortion correction and an lmc astrometric field . the accuracy achieved by the calibration is @xmath79 @xmath80 in rotation , corresponding to @xmath81 mas center - to - edge . the calibration for each channel is represented by the mef header wcs . before 2010 november the calibrations and wcs contain known errors ; we provide tables and an idl routine to correct archival nici data to the final calibration . this work was supported in part by nsf grants ast-0713881 and ast-0709484 . the authors thank the nici team members j. hinds and c. lockhart , and gemini engineers r. galvez , g. gausachs , j. luhrs , g. perez , r. rogers , r. rojas , l. solis , g. trancho , and c. urrutia for their expert support of nici during commissioning and operations . f. rantakyro , e. christensen , and e. artigau provided valuable operational and software support . we thank j. anderson for supplying the _ hst _ data for the lmc fields , and the anonymous referee for providing valuable suggestions . this research has made use of the simbad database , operated at cds , strasbourg , france ; the washington double star catalog maintained at the u.s . naval observatory ; and nasa s astrophysics data system . _ facilities : _ the two nici science channels are red ( transmitted light through dw ) and blue ( reflected by dw ) . the red and blue channels are also referred to as `` holmes '' and `` watson , '' respectively . the mef files contain a primary header unit ( phu ) followed by two extensions : 1 . primary header 2 . extension 1 : red ( holmes ) channel header and data 3 . extension 2 : blue ( watson ) channel header and data selected astrometry - related keywords in an example nici mef header . .... # primary header unit object = ' f606w_11.341 ' / object name frame = ' fk5 ' / target coordinate system equinox = 2000 . / equinox of coordinate system epoch = 2000 . / target coordinate epoch ra = 80.48629583 / target right ascension dec = -69.44843889 / target declination ha = ' -00:52:22.05 ' / telescope hour angle pa = 0 . / sky position angle at start of exposure ( ipa ) inport = 1 / number of iss port where instrument is located crpa = -49.5685362028393 / current cass rotator position angle crmode = ' follow ' / cass rotator mode iaa = 247.5 / instrument alignment angle focs = ' open ' / fiber optic calib . fpmw = ' clear_g5710 ' / nici focal plane mask wheel dichroic= ' h-50/50_g5701 ' / nici dichroic wheel # extension 1 header channel = ' red ' / nici science camera channel filter_r= ' ch4-h4\%l_g0740 ' / nici red filter wheel radecsys= ' fk5 ' / r.a./dec . coordinate system reference ctype1 = ' ra --- tan ' / r.a . in tangent plane projection ctype2 = ' dec -- tan ' / dec . in tangent plane projection crval1 = 80.4883844 / ra at ref pix in decimal degrees crval2 = -69.4482734 / dec at ref pix in decimal degrees crpix1 = 413.999522 / ref pix of axis 1 crpix2 = 594.997314 / ref pix of axis 2 cd1\_1 = 5.01646534e-06 / wcs matrix element 1 1 cd1\_2 = 8.26096075e-11 / wcs matrix element 1 2 cd2\_1 = -3.29874171e-10 / wcs matrix element 2 1 cd2\_2 = -5.01703714e-06 / wcs matrix element 2 2 # extension 2 header channel = ' blue ' / nici science camera channel filter_b= ' ch4-h4\%s_g0743 ' / nici blue filter wheel radecsys= ' fk5 ' / r.a./dec . coordinate system reference ctype1 = ' ra --- tan ' / r.a . in tangent plane projection ctype2 = ' dec -- tan ' / dec . in tangent plane projection crval1 = 80.4883844 / ra at ref pix in decimal degrees crval2 = -69.4482734 / dec at ref pix in decimal degrees crpix1 = 570.000515 / ref pix of axis 1 crpix2 = 650.997321 / ref pix of axis 2 cd1\_1 = -5.0159887e-06 / wcs matrix element 1 1 cd1\_2 = -6.99793144e-08 / wcs matrix element 1 2 cd2\_1 = 7.03997325e-08 / wcs matrix element 2 1 cd2\_2 = -5.01655469e-06 / wcs matrix element 2 2 .... an idl routine nici_fixwcs is available at http://www.gemini.edu/sciops/instruments/nici/data-format-and-reduction to correct known errors in the nici wcs . it automatically applies corrections stored in lookup tables according to the observation date , using the following algorithm : 1 . extract the iss port ( inport ) and @xmath82 ( iaa ) from the primary header unit and the red wcs from extension 1 . 2 . compute the wcs rotation angle @xmath83 . 3 . determine @xmath84 from a lookup table according to the observation date . 4 . compute @xmath85 . 5 . compute the new red wcs angle @xmath86 . compute the new blue wcs angle @xmath87 . compute the new red and blue wcs s with a left - handed coordinate system using the mean pixel scales and new rotation angles . . reflect the red wcs left - right for the side - looking port or the blue wcs left - right for the up - looking port . 9 . write the new wcs s into the mef extension headers . these wcs computations can also be performed with standard wcs processing routines such as the idl astronomy library astrometry routines getrot and hrotate . table [ tab - wcscorr ] lists the date - specific @xmath84 values used by nici_fixwcs . a typical data - reduction strategy is outlined below . additional details , up - to - date information , and the distortion correction files are available at http://www.gemini.edu/sciops/instruments/nici/data-format-and-reduction . 1 . apply recommended corrections to the red and blue channel wcs s . 2 . apply the standard distortion corrections to the red and blue data . 3 . for cassrot fixed data , apply a rotation to the wcs to correct it to the mid - point of the exposure . the amount of correction is proportional to the rate of change of the parallactic angle during the exposure , which may vary significantly as targets rise , transit , and set . at this point the wcs should accurately represent the astrometric properties of the data . 4 . reflect both channels up - down so that pixel ( 1,1 ) appears in the lower - left corner of standard image display tools , using a routines such the idl astronomy library s hrotate that also reflects the wcs . depending on the iss port , reflect either the red or blue channel left - right to produce a left - handed coordinate system . apply further processing ( adi pipeline , derotation , etc . ) using the wcs to correct the rotation and scale of individual exposures . lccccccccccc h-50/50 ( max ) & 4%l & 29.574 & 22.313 & 0.54 & & 4%s & 29.514 & -21.210 & 0.53 & 0.9980 & -1.103 + h-50/50 ( min ) & 4%l & 29.574 & 22.311 & 0.54 & & 4%s & 29.507 & -21.148 & 0.53 & 0.9977 & -1.163 + mirror & & & & & & 4%s & 29.495 & -21.125 & 0.56 & & + open & 4%l & 29.568 & 22.382 & 0.57 & & & & & & & + h / k & 4%l & 29.560 & 22.286 & 0.57 & & 4%s & 29.487 & -21.213 & 1.01 & 0.9974 & -1.073 + h / k & ks & 29.562 & 22.293 & 0.57 & & h & 29.484 & -21.206 & 0.59 & 0.9974 & -1.087 lccccccc h-50/50 ( max ) & 4%l & 530 & 514 & & 4%s & 450 & 566 + h-50/50 ( min ) & 4%l & 530 & 514 & & 4%s & 451 & 542 + mirror & & & & & 4%s & 455 & 580 + open & 4%l & 529 & 513 & & & & + h / k & 4%l & 531 & 508 & & 4%s & 465 & 593 + h / k & ks & 531 & 508 & & h & 465 & 593 rrrrrrr j=0 & 3.30330e+00 & -3.24809e-02 & 1.71647e-04 & -4.15214e-07 & 4.42279e-10 & -1.64635e-13 + 1 & 9.56919e-01 & 6.51225e-04 & -4.02601e-06 & 1.04369e-08 & -1.13916e-11 & 4.35036e-15 + 2 & 2.56604e-04 & -3.82735e-06 & 2.39133e-08 & -6.28410e-11 & 6.92881e-14 & -2.67375e-17 + 3 & -6.51553e-07 & 9.23475e-09 & -5.73017e-11 & 1.50925e-13 & -1.66960e-16 & 6.47106e-20 + 4 & 7.14946e-10 & -9.69040e-12 & 5.96375e-14 & -1.56947e-16 & 1.73795e-19 & -6.75174e-23 + 5 & -2.79985e-13 & 3.67357e-15 & -2.24523e-17 & 5.89872e-20 & -6.53292e-23 & 2.54179e-26 + j=0 & 1.42582e+01 & 9.41488e-01 & 2.18931e-04 & -5.22009e-07 & 6.23837e-10 & -2.70063e-13 + 1 & -5.40761e-02 & 6.07982e-04 & -3.77042e-06 & 9.95836e-09 & -1.15149e-11 & 4.81094e-15 + 2 & 0.000191206 & -3.30812e-06 & 2.08871e-08 & -5.49720e-11 & 6.29836e-14 & -2.60687e-17 + 3 & -3.77292e-07 & 7.52945e-09 & -4.79591e-11 & 1.26167e-13 & -1.44296e-16 & 5.96523e-20 + 4 & 3.80158e-10 & -7.67197e-12 & 4.89736e-14 & -1.28818e-16 & 1.47450e-19 & -6.10716e-23 + 5 & -1.42414e-13 & 2.88797e-15 & -1.84127e-17 & 4.84259e-20 & -5.55290e-23 & 2.30676e-26 + j=0 & -1.49018e+00 & 1.07896e-03 & 1.57155e-05 & -4.13557e-08 & 3.27833e-11 & -8.19736e-15 + 1 & 9.93247e-01 & 6.31284e-05 & -4.91790e-07 & 1.18236e-09 & -1.03838e-12 & 2.99391e-16 + 2 & 8.41072e-05 & -5.02536e-07 & 3.61158e-09 & -8.97672e-12 & 8.16102e-15 & -2.40618e-18 + 3 & -2.42011e-07 & 1.44899e-09 & -1.03877e-11 & 2.65441e-14 & -2.48127e-17 & 7.50551e-21 + 4 & 2.74100e-10 & -1.79337e-12 & 1.29831e-14 & -3.36450e-17 & 3.20101e-20 & -9.88390e-24 + 5 & -1.10542e-13 & 8.03376e-16 & -5.82388e-18 & 1.51399e-20 & -1.45466e-23 & 4.55593e-27 + j=0 & 1.17133e+01 & 9.88950e-01 & -5.97344e-05 & 1.98457e-07 & -2.06889e-10 & 7.90131e-14 + 1 & -6.68061e-03 & -3.01784e-04 & 1.63966e-06 & -3.76719e-09 & 3.90648e-12 & -1.50652e-15 + 2 & -6.09404e-05 & 1.72898e-06 & -9.59428e-09 & 2.21625e-11 & -2.30184e-14 & 8.88496e-18 + 3 & 1.92937e-07 & -4.21968e-09 & 2.39915e-11 & -5.58094e-14 & 5.78653e-17 & -2.22094e-20 + 4 & -2.04690e-10 & 4.55594e-12 & -2.63308e-14 & 6.15381e-17 & -6.36298e-20 & 2.42601e-23 + 5 & 7.95641e-14 & -1.77333e-15 & 1.03499e-17 & -2.42680e-20 & 2.50371e-23 & -9.49402e-27 rlcclcccccl 1&2008 - 07 - 27&5&10 - 07&70 oph & 112.30 & 17.900&17.900&@xmath88 & & gs - nici - comm1 - 318 + 2&2009 - 10 - 24&1&10 - 29&hdo 171 b - c&247.50 & 18.060&18.060 & & & eng files 20101024 1 - 11 + 3&2010 - 03 - 20&5&05 - 09&70 oph & * 111.70*&17.800&17.800 & & @xmath89&gs - cal20100509 - 2 + 4&2010 - 10 - 20&5&10 - 28&trapezium & * 111.32*&17.910&17.910 & & @xmath90&gs - cal20101031 + & & 5&11 - 02&trapezium & * 112.20*&17.960&17.960 & & @xmath90 & + & & 5&11 - 10 & & & & & & @xmath91 & + & & 5&11 - 20 & & & & & @xmath921.137 & @xmath91 & + & & 5&12 - 14&lmc-11mag & 112.61 & 17.932&17.970&@xmath921.1&@xmath93 & gs - cal20101225 - 2 + 5&2011 - 01 - 14&1&02 - 16&lmc-11mag & 247.50 & 17.973&18.009 & & & gs - cal20110117 - 2 + 6&2011 - 03 - 11&1&03 - 14&lmc-11mag & 247.50 & 17.973&18.009 & & & gs - cal20110314 - 2 + 7&2011 - 04 - 27&1&05 - 12&lmc-11mag & 247.50 & 17.973&18.009 & & & gs - cal20110512 - 1 + 8&2011 - 06 - 08&5&06 - 24&hip 62403 & 112.82 & 17.978&18.014 & & & gs - cal20110626 - 3 + 9&2012 - 02 - 15&5&02 - 21&lmc-11mag & 112.64 & 17.950&17.986 & & & gs - cal20120331 - 1 + 10&2012 - 07 - 13&5&08 - 27&lmc-11mag & 112.42 & 17.936&17.973 & & & gs - cal20121222 - 5 + 11&2013 - 05 - 07&5&05 - 12&hip 62403 & 112.25 & 17.936&17.973 & & & gs - cal20130512 - 1 + 12&2013 - 06 - 19&1&06 - 21&hip 62403 & 247.50 & 17.936&17.973 & & & gs - cal20130621 - 1 rrccccccccccccccc 0 & 46787 & 80.486295&-69.448440 & 538.2 & 495.8 & & 3500 & 0.01 & & 535.7 & 499.7 & 4.58 & & 535.7 & 499.7 & 4.58 + 1 & 46775 & 80.490369&-69.450562 & 251.4 & 70.4 & & 71 & 0.09 & & 251.3 & 70.5 & 0.19 & & 251.2 & 70.6 & 0.28 + 2 & 46767 & 80.488462&-69.450358 & 385.7 & 111.3 & & 86 & 0.08 & & 385.7 & 110.3 & 0.99 & & 385.6 & 110.3 & 0.96 + 3 & 3014 & 80.488284&-69.447263 & 398.2 & 731.8 & & 15 & 0.28 & & 397.8 & 732.7 & 0.95 & & 397.8 & 732.7 & 0.95 + 4 & 46799 & 80.487740&-69.448175 & 436.5 & 549.0 & & 82 & 0.08 & & 436.7 & 549.6 & 0.67 & & 436.7 & 549.6 & 0.69 + 5 & 3421 & 80.487645&-69.448639 & 443.1 & 457.7 & & 14 & 0.29 & & 443.5 & 457.5 & 0.43 & & 443.5 & 457.5 & 0.42 + 6 & 2771 & 80.486258&-69.447291 & 540.8 & 726.2 & & 23 & 0.20 & & 541.4 & 726.3 & 0.64 & & 541.4 & 726.3 & 0.68 + 7 & 46768 & 80.484459&-69.448820 & 667.4 & 419.7 & & 34 & 0.16 & & 667.0 & 418.9 & 0.84 & & 667.0 & 418.9 & 0.87 + 8 & 3694 & 80.484371&-69.450446 & 673.6 & 93.6 & & 14 & 0.29 & & 674.2 & 94.4 & 0.99 & & 674.1 & 94.4 & 0.92 + 9 & 2518 & 80.482859&-69.447614 & 780.0 & 661.4 & & 19 & 0.23 & & 779.7 & 661.1 & 0.47 & & 779.7 & 661.1 & 0.48 + 10 & 2053 & 80.481952&-69.446614 & 843.9 & 862.3 & & 12 & 0.32 & & 843.5 & 862.1 & 0.44 & & 843.6 & 862.0 & 0.42 + 11 & 46772 & 80.481336&-69.447447 & 887.2 & 694.9 & & 88 & 0.08 & & 886.9 & 695.1 & 0.36 & & 886.9 & 695.0 & 0.30 + & & & & & 0.69 & & & & 0.68 rrrrr 62403 & 191.8289838 & -66.2374484 & 507.5 & 487.5 + 1 & 191.8346683 & -66.2355670 & 49.0 & 864.0 + 2 & 191.8333292 & -66.2348526 & 157.0 & 1007.0 + 3 & 191.8305338 & -66.2394045 & 382.5 & 96.0 + 4 & 191.8283639 & -66.2360693 & 557.5 & 763.5 + 5 & 191.8277686 & -66.2394120 & 605.5 & 94.5 + 6 & 191.8250159 & -66.2385176 & 827.5 & 273.5 + 7 & 191.8247250 & -66.2357544 & 851.0 & 826.5 + 8 & 191.8245947 & -66.2363066 & 861.5 & 716.0 + lccccclclcccccclccc + hd 61005 & 2 & 2009 - 01 - 13 & b & 50/50 & 4%s & 112.30 & & 2011 - 04 - 25 & b & mir & h & 6 & 1 & 247.50 & lmc & 0.215 & 0.23 + rxj 1224.8 & 2 & 2009 - 02 - 06 & r & 50/50 & 4%l & 112.30 & & 2012 - 04 - 06 & r & h / k & ks & 9 & 5 & 112.64 & lmc & 0.040 & 0.92 + gamma oph & 7 & 2009 - 04 - 08 & b & mir & h & 112.30 & & 2010 - 04 - 08 & b & mir & h & 3 & 5 & 112.20 & & 0.150 & 1.54 & + prox cen & 8 & 2009 - 04 - 26 & r & 50/50 & 4%l & 112.30 & & 2009 - 04 - 08 & r & 50/50 & 4%l & 1 & 5 & 112.30 & 70 oph & 0.478 & 0.39 & + hd 196544 & 2 & 2009 - 04 - 26 & r & 50/50 & 4%l & 112.30 & & 2011 - 04 - 25 & b & mir & h & 6 & 1 & 247.50 & lmc & 0.700 & & + hd 196544 & 2 & 2009 - 04 - 27 & r & 50/50 & 4%l & 112.30 & & 2011 - 04 - 25 & b & mir & h & 6 & 1 & 247.50 & lmc & 0.400 + + + hip 41373 & & 2010 - 01 - 10 & b & 50/50 & 4%s & 247.50 & & 2011 - 02 - 05 & b & mir & h & 5 & 1 & 247.50 & lmc & 0.274 & 0.93 + hip 62403 & 4 & 2010 - 02 - 28 & r & 50/50 & 4%l & 247.50 & & 2011 - 04 - 15 & r & 50/50 & 4%l & 6 & 1 & 247.50 & lmc & 0.077 & 2.04 & + + + 70 oph & & 2010 - 04 - 07 & r&b & 50/50 & 4%l , s + hd 168443 & 6 & 2010 - 04 - 08 & b & 50/50 & 4%l & 112.15 & & 2012 - 04 - 06 & r & 50/50 & 4%l & 9 & 5 & 112.64 & lmc & 0.046 & & + 70 oph & & 2010 - 04 - 09 & r&b & 50/50 & 4%l , s & 112.15 & & & & & & & & & & & & + 70 oph & & 2010 - 05 - 09 & r&b & 50/50 & 4%l , s & 111.70 & & & & & & & & & & & & + hip 59315 & 2 & 2010 - 05 - 09 & r & 50/50 & 4%l & 111.70 & & 2011 - 05 - 12 & r & h / k & ks & 7 & 1 & 247.50 & lmc & -0.604 & & + & & & & & & 112.20 & & & & & & & & & & 0.02 & 0.92 & + rccc 1 & 2008 - 07 - 27 & 5 & 112.30 + 2 & 2009 - 10 - 24 & 1 & 247.50 + 3 & 2010 - 03 - 20 & 5 & 112.20 + 4 & 2010 - 10 - 20 & 5 & 112.61 + 5 & 2011 - 01 - 14 & 1 & 247.50 + 6 & 2011 - 03 - 11 & 1 & 247.50 + 7 & 2011 - 04 - 27 & 1 & 247.50 + 8 & 2011 - 06 - 08 & 5 & 112.82 + 9 & 2012 - 02 - 15 & 5 & 112.64 + 10 & 2012 - 07 - 13 & 5 & 112.42 + 11 & 2013 - 05 - 07 & 5 & 112.25 + 12 & 2013 - 06 - 19 & 1 & 247.50

_ a priori _ dimension reduction is a widely adopted technique for reducing the computational complexity of stationary inverse problems . in this setting , the solution of an inverse problem is parameterized by a low - dimensional basis that is often obtained from the truncated karhunen - love expansion of the prior distribution . for high - dimensional inverse problems equipped with smoothing priors , this technique can lead to drastic reductions in parameter dimension and significant computational savings . in this paper , we extend the concept of _ a priori _ dimension reduction to non - stationary inverse problems , in which the goal is to sequentially infer the state of a dynamical system . our approach proceeds in an _ offline - online _ fashion . we first identify a low - dimensional subspace in the state space before solving the inverse problem ( the offline phase ) , using either the method of `` snapshots '' or regularized covariance estimation . then this subspace is used to reduce the computational complexity of various filtering algorithms including the kalman filter , extended kalman filter , and ensemble kalman filter within a novel subspace - constrained bayesian prediction - and - update procedure ( the online phase ) . we demonstrate the performance of our new dimension reduction approach on various numerical examples . in some test cases , our approach reduces the dimensionality of the original problem by orders of magnitude and yields up to two orders of magnitude in computational savings . _ keywords _ : state estimation , bayesian filtering , kalman filter , ensemble kalman filter , dimension reduction .

1508.06452

an inverse problem converts noisy , incomplete , and possibly indirect observations to characterizations of some unknown parameters or states of a physical system . these unknowns are often functions defined on a spatial domain and linked to the observables via a forward model . high dimensionality due to the numerical discretization of the unknown functions is often viewed as one of the grand challenges in designing scalable inference methods . this challenge motivates the development of dimension reduction approaches , which exploit the possibly low - dimensional intrinsic structure of the inverse problem , to alleviate the effect of the `` curse of dimensionality . '' a typical inverse problem is ill - posed ; the unknowns are not uniquely identified by the observations . this is a joint effect of noisy , incomplete observations and the smoothing properties of the forward model . in the bayesian inference framework @xcite , ill - posedness is addressed by employing a suitable prior distribution and characterizing the posterior distribution of the unknowns conditioned on the observations . in this setting , the priors often encode structural information about the unknowns , such as spatial smoothness properties . this _ a priori _ structural information opens up the possibility of prior - based dimension reduction , especially in cases where the variation of the high - dimensional unknowns can be explained by a small number of basis functions . for instance , the truncated karhunen - love expansion @xcite of the prior distribution is employed in @xcite for identifying such a _ a priori _ low - dimensional basis in _ static _ inverse problems . computational cost can be greatly reduced by projecting the original high - dimensional unknowns onto the subspace spanned by the resulting low - dimensional basis . in this paper , we extend this concept of _ a priori _ dimension reduction to non - stationary inverse problems , in which the goal is to sequentially infer the state of a dynamical system . such problems can be solved efficiently using filtering methods , where the posterior prediction from the previous time step is used as the prior for the current state , and the new posterior is obtained by conditioning the current state on data observed at the current time . the computational difficulty of applying filtering methods to high - dimensional problems stems both from propagating the distribution of the high - dimensional state forward in time and from solving the high - dimensional inference problem when the new data is observed . to reduce the computational complexity of filtering methods , our proposed dimension reduction method is applied in an _ offline - online _ fashion . in the _ offline _ phase , we identify a low - dimensional subspace of the state space before solving the inverse problem , using either the method of snapshots or regularized covariance estimation . in the _ online _ phase , the computational complexity of various ( gaussian ) filtering algorithms including the kalman filter , extended kalman filter , and ensemble kalman filter is reduced by constraining the update and the prediction steps within the resulting subspace , in a unified subspace - constrained bayesian framework . the success of the proposed approach naturally requires that the unknown states can be captured by a low - dimensional basis . this is the case , for instance , if either the model states are sufficiently smooth or the states can be captured by a low - dimensional attractor . we show numerical examples where the reduction provides significant computational savings in different ways by reducing the dimension of the linear systems involved in the prediction and update steps , but also , in ensemble filtering approaches , by reducing the number of ensemble members required . we discuss the limitations of the proposed approach in cases where the state can not be represented efficiently in a fixed low - dimensional subspace . we also discuss several different ways to construct the subspace based on existing `` snapshots '' of the model states . the idea of reducing the dimension of filtering algorithms has been investigated before . the closest existing algorithm to our approach is the reduced - order kalman filter ( rokf ) @xcite , where dimension reduction is obtained by projecting the model dynamics onto a fixed low - dimensional subspace . while the rokf shares some similarities with the present approach , fundamental differences remain : our algorithms do not explicitly project the model dynamics onto the subspace , but rather constrain the inference ( update ) step using the subspace . we show that the latter strategy yields more appropriate prior distributions for each inference problem . moreover , we extend the discussion of dimension reduction to ensemble filtering methods . differences between the present approach and the rokf are analyzed in detail later in the paper . another related approach is the reduced - rank kalman filter ( rrkf ) of @xcite . in this approach , one propagates prediction uncertainties only along directions in the state space where the variance of the states grows most quickly ( the so - called hessian singular vectors ) . the difference with our approach is that the subspace of rrkf is re - computed at each filtering step through the solution of an eigenvalue problem , whereas in our approach the basis is fixed and computed offline . thus , rrkf is a local - in - time approach , based on local linearization , whereas our approach tries to find a low - dimensional subspace for the state based on an analysis of global - in - time dynamics . our approach is less computationally intensive , but its applicability may be restricted to cases where a global low - dimensional representation for model states exists . our numerical examples , however , demonstrate that this simple strategy can yield significant computational savings in a range of filtering algorithms . in @xcite , dimension reduction is sought not in the gaussian filtering context , but for a sequential monte carlo ( particle filtering ) method . again , dimension reduction is performed locally in time . filtering is restricted to coordinates spanning the most unstable modes around the current nominal trajectory ; interestingly , for spatially distributed systems , these unstable modes are often the low - wavenumber components of the state . dimension reduction approaches for filtering problems can be beneficial in many ways . for instance , they reduce memory requirements , which are prohibitively large for high - dimensional problems when standard kalman filters are applied . indeed , memory constraints have motivated the development of various approximate filtering methods @xcite . the dimension reduction approach presented here reduces memory requirements , but also offers speedups that may be beneficial for ( even smaller scale ) real - time estimation and control problems , e.g. , chemical process tomography @xcite . as noted above , speedups also extend to ensemble filtering methods : constraining the inference step onto a subspace implicitly regularizes the problem , and thus reduces the number of ensemble members required to achieve a given accuracy . the paper is organized as follows . in section [ sec : redu ] , we review prior - based dimension reduction for static inverse problems and develop the linear - gaussian filtering equations for subspace coordinates . in section [ sec : nonlin ] , we discuss how the method applies to extended kalman filtering and ensemble methods . section [ sec : subspace ] discusses techniques for constructing the low - dimensional subspace . section [ sec : rokf ] analyzes the differences between our approach and the rokf . in sections [ sec : numex][sec : numex2 ] , we study the behavior of the dimension reduction approach via linear and nonlinear examples . section [ sec : conc ] offers some concluding remarks . our starting point is the prior - based dimension reduction technique for static inverse problems , which we briefly review here . the unknown function @xmath0 , @xmath1 , is defined in some spatial domain @xmath2 . discretizing @xmath0 on a grid defined by a set of nodes @xmath3 and some basis functions yields a @xmath4-dimensional vector @xmath5^\top \in \mathbb{r}^d$ ] . the discretized unknown vector @xmath6 is related to observations @xmath7 via the model @xmath8 where @xmath9 and @xmath10 is a ( possibly nonlinear ) mapping from the unknown @xmath6 to the observable output . moreover , let us assume that we have a gaussian prior @xmath11 . then , the posterior density for @xmath6 is @xmath12 where @xmath13 denotes the quadratic form @xmath14 . the idea in prior - based dimension reduction is to constrain the problem onto a subspace that contains most of the variability allowed by the prior ; see , for instance @xcite . this can be done by computing the singular value decomposition ( svd ) of the prior covariance matrix @xmath15 , where @xmath16 contains the singular vectors @xmath17 ( as colums ) and @xmath18 has the singular vectors in the diagonal . dimension is reduced by representing the unknown as a linear combination of the @xmath19 leading ( scaled ) singular vectors : @xmath20.\ ] ] inserting this parameterization into the problem leads to the following posterior density for subspace coordinates @xmath21 : @xmath22 it is easy to verify that the prior term simplifies to @xmath23 , where @xmath24 is the @xmath25 identity matrix . the posterior can be thus written simply as @xmath26 and the dimension of the inverse problem has been reduced from @xmath4 to @xmath19 . this can be helpful , for instance , when mcmc samplers , which are challenging to apply in high - dimensional problems , are used to quantify uncertainty in the parameters . if the model is linear , @xmath27 , the posterior is gaussian @xmath28 with mean and covariance matrix given by @xmath29 thus , one needs to apply the model to the @xmath19 columns of @xmath30 and solve @xmath19-dimensional linear system , which is computationally much easier than solving the full problem if @xmath31 . here , we discuss how dimension reduction can be implemented for dynamical state estimation problems . let us begin with the following linear gaussian state space model : @xmath32 in the above system , @xmath33 is the forward model that evolves the state in time and @xmath34 is the observation model that maps the state to the observations . the model and observation errors are assumed to be zero mean gaussians : @xmath35 and @xmath36 with known covariance matrices @xmath37 and @xmath38 . the linear gaussian problem can be solved with the kalman filter , which proceeds sequentially as follows . assume that , at time step @xmath39 , the marginal posterior is the following gaussian : @xmath40 the _ prediction step _ involves propagating this gaussian forward with the model @xmath41 , which yields the gaussian @xmath42 where @xmath43 and @xmath44 . throughout the paper , we follow the notation commonly used in weather forecasting and data assimilation literature : we use the superscript @xmath45 to refer to the `` analysis '' ( posterior ) estimate updated with the most recent observations , and the superscript @xmath10 to refer to the `` forecast '' ( prior ) estimate . in the _ update step _ , the predicted gaussian is updated with the new observations that become available . the resulting posterior density is @xmath46 which is , again , gaussian with known mean and covariance matrix given by the standard kalman filter formulas , which we choose not to rewrite here ( see any standard textbook on the subject , e.g. , @xcite ) . the most straightforward application of the prior - based dimension reduction technique in dynamical problems would be to define @xmath30 separately for each filter step via the leading singular values and vectors of the prior covariance matrix @xmath47 . this approach has a few potential problems . first , computing the local leading singular vectors at each time step can be a computationally challenging task . second , by truncating the prior covariance , we might discard directions that seem less important ( i.e. , that have low prior variance ) locally in time , but that become relevant at a later time step . in our experiments , this approach led to inconsistent behavior of the filter ; good filtering accuracy was obtained for some cases , but in other cases the filter performed poorly or even diverged . here , we examine an alternative , simpler strategy , where a global subspace is constructed _ a priori _ ( before the filtering is started ) and is then fixed for the filtering . this approach is motivated by the fact that the state of a dynamical system often lives in a subspace of much smaller dimension than the full state space ; the state vector often has some properties ( e.g. , smoothness ) that enable it to be effectively described in a low - dimensional subspace . if we can capture the subspace where the essential dynamics of the system happen , we can potentially reduce the whole filtering procedure onto this subspace . this idea is discussed here . now , we parameterize the unknown as @xmath48 , where @xmath49 is a fixed reduction operator that does not change in time and @xmath50 is the predicted ( prior ) mean . for now , we assume that such a representation exists ; discussion about how to construct @xmath30 is reserved for section [ sec : subspace ] . to derive the filtering equations for the subspace coordinates @xmath51 , assume that the marginal posterior distribution for @xmath52 at time @xmath53 is @xmath54 transforming this gaussian distribution to the original coordinates using @xmath55 yields the following gaussian distribution in the full state space : @xmath56 by propagating this gaussian distribution forward with @xmath41 , we obtain the mean and the covariance matrix of the predictive distribution @xmath57 in the original coordinates : @xmath58 applying this as the prior , and inserting the parameterization @xmath59 into ( [ fullpost ] ) yields the following marginal posterior density for @xmath51 : @xmath60 this is equivalent to a linear problem with gaussian likelihood @xmath61 and zero mean gaussian prior @xmath62 . the resulting posterior is thus @xmath63 where @xmath64 and @xmath65 is the prediction residual . note that now @xmath30 does not whiten the prior , in contrast with the prior - based dimension reduction discussed in section [ sec : static ] for static problems , and thus the term @xmath66 is not equal to the identity . moreover , the matrix @xmath47 can not be formed explicitly in high - dimensional problems . to efficiently evaluate @xmath67 , we recall that here @xmath68 where @xmath69 is the matrix square root @xmath70 . to shorten the notation , let us denote @xmath71 . now , applying the sherman - morrison - woodbury matrix inversion formula yields @xmath72 where @xmath24 is the @xmath25 identity matrix . now , the matrix @xmath73 that needs to be inverted is only @xmath25 . thus , a product @xmath74 can be efficiently computed as long as @xmath75 is easy to compute . this condition must hold in order for this technique to work . in practice , @xmath76 is usually a simple ( e.g. , diagonal ) matrix postulated by the user . to sum up , a single step of the reduced kalman filter with a fixed basis is given as an algorithm below . assume that we have the previous estimate @xmath77 and its covariance matrix @xmath78 available . then the algorithm reads as follows : + + + + computationally , this version is much lighter than the standard kalman filter if @xmath31 . in the standard kalman filter , the prediction covariance matrix is computed as @xmath79 ; that is , we need to compute products of @xmath80 matrices ( or to apply the forward model to the @xmath4 columns of @xmath81 ) . moreover , when updating the prior covariance , one needs to operate with @xmath82 and solve a system of @xmath83 linear equations . in the approach described here , one needs to work with @xmath84 matrices @xmath85 , solve a system of @xmath19 linear equations , and do one inversion of a @xmath25 matrix . also , here the basis vectors are fixed , so we avoid solving local eigenvalue problems , which are needed , for instance , in the reduced rank kalman filter of @xcite . here , the parameterization is centered at the predicted mean , @xmath86 . an alternative would be to use a fixed mean , @xmath87 , where @xmath88 is some fixed offset . the former looks for a correction to the predicted mean in the subspace , whereas the latter attempts to describe the state vector itself in a fixed subspace . in our experiments , the former yields much better filter accuracy , especially with small @xmath19 . here , we discuss two ways to extend the dimension reduction idea to problems where the evolution and/or observation models are nonlinear . we still assume additive gaussian model and observation errors , so our state space model now reads as @xmath89 where @xmath90 and @xmath91 are the nonlinear forward and observation models . we start by discussing the extended kalman filter , which requires linearizations of the forward and observation models . then , we discuss ensemble filtering techniques where linearizations are not needed . the extended kalman filter ( ekf ) replaces the model and observation matrices in the kf with their linearized versions . thus , the algorithm is the same as algorithm 1 in section [ sec : redu ] , but the mean is propagated with the nonlinear forward model and the prediction residual is calculated with the nonlinear observation model . elsewhere @xmath41 and @xmath92 are replaced with @xmath93 note that for large scale problems , computing the above matrices explicitly is not feasible . instead , one often derives the linearized model analytically ( as an operator ) , which makes it possible to propagate vectors forward with the linear model , which is equivalent to computing products @xmath94 where @xmath95 . here , these _ tangent linear codes _ need to be applied in steps ( iii ) and ( v ) of algorithm 1 . in step ( iii ) , we compute @xmath96 , which can be done by applying the linearized forward model to the @xmath19 columns of @xmath97 . in step ( v ) , we need @xmath98 , which can be computed by applying the linearized observation model to the @xmath19 columns of @xmath30 . in both cases , we only need to propagate @xmath19 vectors through the linearized models , instead of @xmath4 vectors as in the standard ekf . ensemble filters have become popular for solving very high - dimensional dynamical state estimation problems arising in geophysical applications such as numerical weather prediction . the development started started from the ensemble kalman filter ( enkf , @xcite ) in the 1990s , and different variants are under active development . the idea of the enkf is to represent the state and its uncertainty with samples ( an `` ensemble '' of states ) , and , roughly speaking , to replace the covariances in the filtering formulas with their empirical estimates calculated from the samples . for high - dimensional problems that involve complex physical models , the ensemble size is necessarily much smaller than the dimension of the problem . as a result , the obtained covariance estimates are rank - deficient and can suffer from `` spurious correlations '' ( unphysical correlations appearing randomly due to small sample size ) ; see , e.g. , @xcite for discussion . to overcome these issues , various _ localization _ techniques have been proposed , where the empirical covariance estimates are regularised by , for instance , explicitly removing unrealistic distant correlations from the covariance matrices ; see @xcite . recently , adaptive localization techniques have also been developed , where the localization mechanism is tuned on - line in the filter @xcite . localization is one of the key techniques to make enkfs work for small ensemble sizes . in addition to rank deficiency , the classical enkf suffers from sampling error , as the observation and model errors are accounted for by randomly perturbing the observations and predicted ensemble members during the estimation . to avoid this additional variance in the resulting estimates , so - called square root enkfs have been developed , which are deterministic schemes where no such random perturbations are used for the observations @xcite . accounting for model error still remains a difficulty , although some techniques have been recently proposed ; see , e.g. , @xcite and the references therein . in the method discussed here , the state is constrained onto a subspace , which heavily regularises the estimation problem ; for instance , if the state is smooth , the rough features are explicitly removed from the estimation problem , and all of the information in the samples can be used for inferring the smooth features . as a result , the need for localization is diminished , as demonstrated in the numerical examples in section [ sec : numex ] . moreover , model error can be can be easily included in the approach , provided that we are able to apply the model error covariance matrix to a vector efficiently . here , we present one way of extending the dimension reduction idea to ensemble filtering . let us now consider the state space model where the forward model is nonlinear but the observation model is linear ( the nonlinear observation model case is discussed later ) : @xmath99 in ensemble filtering , we represent the distribution of @xmath51 with samples . let us assume that at time step @xmath53 we have @xmath100 posterior samples @xmath101 available , sampled from the gaussian posterior @xmath102 . these obviously correspond to samples @xmath103 in the full state space . in the prediction step , we move the posterior samples forward with the dynamical model : @xmath104 . then , we compute the empirical covariance matrix of the predicted samples and add the model error covariance to obtain the prediction error covariance : @xmath105 where @xmath106 / \sqrt{n_{\mathrm{ens } } } \in \mathbb{r}^{d \times n_{\mathrm{ens}}}. \label{xk}\ ] ] thus , @xmath107 is the empirical covariance estimate computed from the prediction ensemble . the mean @xmath108 is taken to be the posterior mean from the previous step propagated via the evolution model , @xmath109 , instead of the empirical mean computed from the prediction ensemble , which is why we divide by @xmath110 instead of @xmath111 , see the remarks below for more discussion about this choice . if the observation model is linear , the reduced dimension ensemble filtering algorithm stays almost the same as the reduced kf algorithm ( algorithm 1 in section [ sec : redu ] ) . the only difference is the prior covariance matrix @xmath47 , which is now defined via the prediction ensemble . when @xmath112 , operating with @xmath113 , needed in step ( v ) of algorithm 1 , can still be done efficiently via the sherman - morrison - woodbury inversion formula : @xmath114 note that now , when applying @xmath113 to a vector , we are left with the inversion of an @xmath115 matrix instead of an @xmath25 matrix . to summarise , one step of the ensemble kalman filter with dimension reduction is given below . + + + + the computational cost of the ensemble algorithms is dictated by both the number of basis vectors @xmath19 and the number of ensemble members @xmath100 . the computational cost is similar to that of the standard enkf . what makes dimension reduction attractive from ensemble filtering point of view is that the number of samples needed to capture the distribution of the @xmath19-dimensional variable @xmath51 can be much smaller than the number of samples needed to get accurate filtering results for the @xmath4-dimensional variable @xmath116 in the full space . thus , similar performance can be obtained with fewer ensemble members , as can be observed in the numerical examples in sections [ sec : numex][sec : numex2 ] . the ensemble filter presented above differs from the classical ensemble kalman filter ( enkf ) developed in @xcite . the classical enkf is a non - gaussian filter ; it applies a linear update with perturbed observations to non - gaussian prediction samples to get the posterior ensemble . the version presented here is a gaussian filter in the sense that the prior is assumed to be a gaussian whose covariance matrix is estimated from the prediction ensemble . another difference between the proposed method and many other ensemble filters is that the prior mean here is @xmath117 instead of the empirical ensemble mean used , for instance , in the classical enkf . technically , one could easily choose the ensemble mean as @xmath108 as well . however , we have noticed that the choice @xmath117 for propagating the mean , analogous to ekf and variational ( 3d - var and 4d - var ) methods , works better for many problems . one reason might be that the @xmath108 obtained this way lies close to the attractor of the forward model , whereas the sample mean obtained from a small number of ensemble members might be further away from it and thus represent an `` unphysical '' state . a similar approach was taken in some recently developed filtering algorithms ; see , for instance , @xcite for some discussion . if the observation model is nonlinear , the posterior distribution for @xmath51 is no longer gaussian . then , one way forward is to apply the linearized observation operator as in the ekf to obtain a gaussian approximation of the posterior , and to sample new members from the gaussian . another way is to sample new posterior samples for @xmath51 directly from the non - gaussian posterior @xmath118 using , for instance , markov chain monte carlo ( mcmc ) techniques , which should be feasible if @xmath19 is not too large and @xmath91 is relatively simple ( note that evaluating the posterior density does not require the forward model @xmath90 ) . for instance , novel optimization - based sampling techniques like @xcite ( which requires gaussian priors as above ) can potentially be used to generate posterior samples efficiently , as discussed in @xcite in connection with high - dimensional filtering problems . the subspace representation also opens up a way to implement other sample - based filtering techniques for high - dimensional problems , such as the popular unscented kalman filter ( see , e.g. , @xcite ) , which , in the subspace version , would require @xmath119 samples to propagate the covariance forward instead of @xmath120 as in the full state space version . even particle filtering in the subspace might be possible with a reasonable number of particles . these ideas are left for future study and not pursued further here . the reduced subspace basis @xmath30 for representing the unknown state @xmath116 is constructed in a similar way as for static problems discussed in section [ sec : static ] . consider the covariance matrix @xmath121 that represents the covariance structure of states , and in particular its eigendecomposition @xmath122 . we compute the basis @xmath30 from the @xmath19 leading eigenvectors @xmath123 and square roots of the corresponding eigenvalues @xmath124 : @xmath125 if the eigenvalues of the covariance matrix @xmath121 decay quickly , the variation of states can be captured by a low - dimensional subspace spanned by the leading eigenvectors . we note that scaling the basis by the eigenvalues in ( [ eq : repr ] ) is not necessary , but can unify the scales of the different state variables . of course , this method requires access to @xmath121 . in the rest of this section , we present several ways to estimate this covariance matrix . principal component analysis ( pca ) can be applied to `` snapshots''which are possible model states obtained from existing model simulations for constructing low - dimensional subspaces of high - dimensional dynamical systems . depending on the field of application , this procedure is also named empirical orthogonal functions @xcite in meteorology , or proper orthogonal decomposition @xcite in model reduction . the reduced basis obtained from pca can then be used in either model reduction @xcite or filtering ( e.g. , the rokf method @xcite or our approach as presented in section [ sec : dyn ] ) . in non - stationary inverse problems , our main interest is dynamical systems without steady states ( e.g. , chaotic models ) . in this setting , we use trajectories obtained from either a sufficiently long free model simulation or multiple model simulations with randomized initial conditions . given a sufficient number of snapshots @xmath126 , the subspace basis is computed using ( [ eq : repr ] ) via the eigendecomposition of empirical state covariance @xmath127 where @xmath128 is the empirical state mean . for high - dimensional dynamical systems , it is not feasible to form the empirical covariance directly and apply dense matrix eigendecomposition methods . in this case , either krylov subspace methods @xcite or randomized methods @xcite should be applied together with matrix - free operations the matrix vector product with @xmath121to compute the eigendecomposition . the number of snapshots naturally can affect the quality of the basis @xmath30 , and a sufficiently large @xmath129 should be chosen to capture the essential behavior of the model . above , the basis is constructed from un - regularised empirical covariance estimates , without assuming any particular form for the covariance . if the number of snapshots available is limited , we can instead employ regularised covariance estimation techniques , where certain assumptions about the covariance structure are introduced to regularise the estimation problem . these techniques are discussed below . another viable route for state covariance estimation is to infer the state correlation structure from snapshots and _ a priori _ information such as smoothness assumptions . here , we discuss how gaussian processes ( gps ) can be used for the task . we consider two types of gps : a stationary gp modeled by a kernel function @xcite , or a ( possibly ) non - stationary gp modeled by a a differential operator @xcite . we first discuss the kernel approach , where each element of the covariance matrix is given by a kernel function @xmath39 of the form @xmath130 where @xmath131 are spatial locations used to discretize the states , and all the information about smoothness , correlation length , and variability can be encoded in the parameter @xmath132 . for example , one commonly used kernel function is the squared exponential kernel @xmath133 where @xmath134 and @xmath135 control the variability and correlation length , respectively , and @xmath136 is a distance between points @xmath137 and @xmath138 . with this kernel function , the eigenvalues decay quickly ( exponentially ) , and it is easy to capture the kernel with a finite basis . given the mean of the gp , @xmath88 , empirically estimated from the snapshots , we estimate the parameters @xmath132 in a bayesian framework . the likelihood function of @xmath132 , given a snapshot collection @xmath126 , takes the form @xmath139 combining the likelihood with a prior distribution @xmath140 , we obtain the _ maximum a posteriori _ estimate @xmath141 of the kernel parameters . the reduced subspace basis can then be computed from the eigendecomposition of the state covariance @xmath142 . here the smoothness assumption is used to `` fill the gap '' between the unknown high - dimensional correlations of the state and the information provided by a limited number of snapshots . for a gp defined by stationary kernels , this assumption can be enforced by using a smooth kernel such as the squared exponential kernel . the stationary assumption we used in the above - mentioned kernel method may not be suitable for dynamical systems where the states have heterogeneous spatial correlations . furthermore , operations with the dense covariance matrix @xmath143 in ( [ eq : like ] ) can be computationally challenging for high - dimensional states , because the covariance matrix can be singular and computational costs of dense matrix operations especially factorization and inversion scale poorly with dimension . this motivates us to model the gp using the inverse of the covariance , i.e. , the precision matrix @xmath144 , so that gaussian markov random field ( gmrf ) models @xcite can be used to construct the precision as a sparse matrix . we particularly mention the work of @xcite , in which the sparse precision matrix is constructed from the finite element discretization of the following stochastic partial differential equation ( spde ) : @xmath145 where @xmath146 is a white noise process in space . for spatially constant @xmath147 and @xmath148 , it can be shown that the solution @xmath149 of the spde ( [ eq : spde ] ) defines a gaussian process with the matrn family of correlation functions ; see @xcite and references therein . the functions @xmath150 and @xmath151 together control the correlation length and variability of the gp , and the scalar @xmath152 controls the smoothness , and therefore a nonstationary gp can be defined by prescribing spatially heterogeneous @xmath150 and @xmath151 . for a positive integer @xmath152 , discretizing the spde ( [ eq : spde ] ) yields a sparse precision matrix @xmath153 . here , we discretize ( [ eq : spde ] ) using the finite element method with linear basis functions , @xmath154 where @xmath155 , is a set of nodal points and @xmath156 , is a set of linear basis functions associated with the nodal points . here the nodes are defined to coincide with the locations of the states in the filtering problem . the precision matrix @xmath153 can be constructed given the mass matrix and the stiffness matrix of the finite element discretization , which are defined as @xmath157 \phi_j(s ) ds,\ ] ] where @xmath158 . the functions @xmath150 and @xmath151 can also be discretized by linear basis functions , which are given as @xmath159 respectively . this yields the local mass and stiffness matrices @xmath160 \phi_j(s ) ds,\end{aligned}\ ] ] such that the overall mass and stiffness matrices can be written as @xmath161 where @xmath162^\top$ ] and @xmath163^\top$ ] . here , the local mass and stiffness matrices , @xmath164 and @xmath165 , can be precomputed for a given set of finite element basis functions . following the recursive definition given in @xcite , the precision matrix @xmath166 parameterized by a scalar @xmath152 and vectors @xmath167 and @xmath168 is @xmath169 in the present work , we will pre - select the order of the differential operator by choosing an @xmath152 value , @xmath170 , where @xmath171 . a larger exponent , e.g. , @xmath172 , is not recommended as it will lead to a high computational cost in the following parameter estimation problem . for a fixed @xmath173 , the parameters @xmath167 and @xmath168 can be estimated in a bayesian manner . the likelihood function for this estimation problem takes the form @xmath174 choosing spatially varying parameters @xmath167 and @xmath168 provides the flexibility needed to model a non - stationary covariance structure . however , estimating @xmath167 and @xmath168 from a limited number of snapshots is itself an ill - posed inverse problem , so priors must be assigned to these parameters to remove the ill - posedness . to limit the degrees of freedom in the estimation , we prescribe the function @xmath150 to be a scalar , i.e. , @xmath147 , and use only a spatially varying @xmath151 to control the nonstationarity of the resulting gp . we use the mass lumping technique to approximate the mass matrix rather than dealing with the computationally prohibitive inversion @xmath175 . for the case @xmath147 , the lumped mass matrix is given as @xmath176 we use an exponential prior to enforce the positivity of @xmath177 . furthermore , a smooth lognormal process is prescribed as the prior for @xmath151 to enforce the positive semi - definiteness of the stiffness matrix @xmath178 . by defining @xmath179 , the posterior distribution of the bayesian inference problem can be written as @xmath180 where @xmath181 define the mean and precision of the lognormal prior for @xmath182 . the precision matrix @xmath183 in ( [ eq : post ] ) is given as @xmath184 we can then obtain the _ maximum a posteriori _ estimate of the gmrf parameters @xmath185 by maximizing the logarithm of the posterior density , which yields @xmath186 the optimization problem ( [ eq : mrf_post ] ) is continuous , and the gradient and hessian of the objective with respect to @xmath177 and @xmath187 can be analytically derived ( not reported here for brevity ) ; hence , gradient - based nonlinear programing tools can be used to obtain an optimum . in particular , we employ the subspace trust region method of @xcite with inexact newton steps . note that in the inexact newton solve , we use hessian - vector products rather than explicitly forming the full hessian , to ensure that computational costs and memory requirements scale favorably with parameter dimension . given the solution of ( [ eq : mrf_post ] ) , the low - dimensional subspace for our reduced filters can be computed from the eigendecomposition of the estimated state covariance @xmath188 using the matrix - free methods discussed in the pca case ( section [ sec : pca ] ) . a subtle issue in constructing the precision matrix is the choice of boundary condition for the spde ( [ eq : spde ] ) . for a gp specified with scalar @xmath177 and @xmath189 , prescribing a dirichlet boundary condition will enforce zero variability on the boundary , and prescribing a no flux boundary condition which is a common choice in the literature will roughly double the variability at the boundary compare to the variability in the interior . clearly both choices will result a gp that has nonstationary behavior near the boundary , even though the gp with scalar @xmath177 and @xmath189 defined on an infinite domain should be stationary in theory . in our numerical examples , boundary conditions do not present any difficulties , since we use periodic boundary conditions that are inherited from the structure of the corresponding data assimilation problem . in a more general setting , we recommend to use a zero - flux boundary condition and to let the data determine the nonstationary correlation structure through the solution of ( [ eq : mrf_post ] ) . this way , artifacts created by the boundary condition can be potentially compensated for via the inhomogeneous @xmath151 field . the reduced - order kalman filter ( rokf ) , developed in @xcite and discussed further in @xcite is , in principle , very similar to the dimension reduction approaches presented in this paper ; like our approach , rokf uses a fixed low - dimensional subspace to represent the state vector . therefore , it is useful to discuss the differences between the techniques in more detail . to see the difference between the methods , it is instructive to compare what kind of priors ( predictive distributions ) the methods induce for the subspace coordinates . the prior mean is propagated in the same way , but the covariance is handled differently . in the rokf , the prediction covariance in the reduced space , denoted here by @xmath190 , is computed by evolving @xmath191 as follows : @xmath192 where @xmath193 is the posterior covariance of the previous time step propagated forward with @xmath41 . this has two interpretations : ( _ a _ ) projecting the forward dynamics @xmath41 onto the subspace spanned by the columns of @xmath30 and using it to propagate the reduced covariance forward , and ( _ b _ ) propagating the posterior covariance of the previous time step with the full model and projecting the resulting prediction covariance onto the subspace . on the other hand , the prediction covariance in our approach ( see section [ sec : dyn ] ) is @xmath194 that is , while rokf projects the predicted _ covariance _ matrix @xmath47 , we project the predicted _ precision _ matrix @xmath113 . this difference has a nice geometric interpretation ; projecting the covariance matrix is equivalent to _ marginalizing _ the prior onto the subspace , whereas projecting the precision matrix amounts to taking the _ conditional _ of the prior in the subspace . to visually see the difference , consider the following simple example . assume that the full dimension of the state space is @xmath195 and that the subspace ( of dimension @xmath196 ) is the x - axis : @xmath197^\top$ ] . the model @xmath41 is taken to be a @xmath198 matrix with random entries sampled from @xmath199 . we start with zero mean and covariance @xmath191 in the subspace ( x - axis ) , and propagate it forward with both the rokf and our approach . the results are shown in figure [ fig : rokf_demo ] . one can clearly see that the prior induced by rokf can be significantly different than the prior induced by our approach . specifically , the prior induced by the rokf is always wider than the conditional prior of our approach . it is clear that marginalizing the prior can yield posterior estimates which are _ outside the essential support of the prior . _ here , we demonstrate the proposed dimension reduction algorithms with two synthetic filtering problems : a 240-dimensional version of the lorenz model and a 1600-dimensional example using the quasi - geostrophic model . as the reference methods , we use the standard extended kalman filter ( ekf ) and the standard ensemble kalman filter ( enkf ) ; see @xcite . the purpose of the experiments is to highlight some of the properties of the proposed approach , such as its behaviour with small sample sizes in ensemble filtering , rather than to draw conclusions about the performance of the approach relative to all the recent developments in the ensemble filtering literature . for this reason , and to keep the comparisons simple , we choose the well - known standard enkf as the reference method instead of one of the many variants developed recently . a thorough performance comparison with all the recent developments in filtering methods is a challenging task ( e.g. , handling all the tuning issues of the various filters ) and left for future research . for the enkf , we implement a simple and widely used localization scheme , obtained by _ tapering _ ( setting the covariance between distant points to zero ) the prediction covariance matrix using the 5th order piecewise rational function @xcite . the correlation cut - off length was chosen experimentally so that roughly optimal filter performance was obtained . as a small scale nonlinear example , we consider a generalized version of the lorenz 96 model , the model ii described in @xcite . the evolution model is given by an ode system of @xmath129 equations , each defined as @xmath200 where @xmath201 and @xmath202 is a chosen _ odd _ integer and @xmath203 . the variables are periodic : @xmath204 and @xmath205 for @xmath206 . with @xmath207 , the system reduces to the standard lorenz 96 model introduced in @xcite . in our experiments we use a range of values for @xmath202 , and choose the forcing @xmath208 so that the model attains chaotic behaviour ( verified experimentally ) . in the prediction model used in the estimation , we use values @xmath209 and the corresponding forcing values @xmath210 for all @xmath211 . increasing @xmath202 introduces stronger spatial dependence between neighbouring variables , and yields spatially smoother solutions ; example solutions with @xmath212 and varying @xmath202 are given in figure [ fig : lorex ] . controlling the smoothness allows us to demonstrate how the dimension reduction works in different cases : the smoother the unknown , the fewer basis vectors we need to describe it accurately . this is demonstrated in figure [ fig : energy ] , where we plot the fraction of the energy of the empirical covariance matrix of the model trajectories as a function of the number of basis vectors used in the representation . more precisely , we plot @xmath213 as a function of @xmath19 , where @xmath214 is the @xmath215th largest eigenvalue of the empirical covariance matrix computed from model simulation output . one can see that with large @xmath202 , most of the variability of the model state can be captured in a low - dimensional subspace , whereas with small @xmath202 more basis vectors are needed . in the experiments , we use values @xmath209 and the corresponding forcing values @xmath210 for all @xmath211 . we generate data for the estimation by simulating the model and adding 1% normally distributed random perturbations to the forcing values to introduce error into the prediction model . the observation frequency is @xmath216 time units ( twice the time step used to solve the ode ) and every @xmath217th state variable is observed ( @xmath218 ) , which yields altogether 24 measurements per observation time step . data is generated for @xmath219 time units , which yields altogether 400 observation times . the ode was solved using the 4th order runge - kutta method . the model error covariance matrix used in the experiments is simply @xmath220 for all @xmath39 , where @xmath221 is chosen experimentally from the interval @xmath222 $ ] so that roughly optimal tracking performance is obtained ( rms error between the estimates and the truth is minimised ) for each filter . the candidate subspaces were constructed by the pca , gp , and gmrf techniques discussed in section [ sec : subspace ] . for pca , we used the trajectory obtained by simulating the model for 1200 steps ( that is , the covariance was estimated using 1200 samples ) . gp and gmrf parameters were fitted using 32 snapshots of the model state . for the gp approach , the squared exponential covariance function was used , see equation ( [ squared - exp ] ) , and the obtained estimates for the variance and correlation length parameters were @xmath223 and @xmath224 . for the gmrf approach , we use an exponent of @xmath225 . first , we compare the reduced ekf described in sections [ sec : redu ] and [ sec : ekf ] to the full ekf for the @xmath226 case , where the model state is spatially rather smooth and thus well described in a low - dimensional subspace ; see figures [ fig : lorex ] and [ fig : energy ] . the first 16 basis vectors obtained via pca , gp , and gmrf are given in figure [ fig : basis ] ; the first vectors represent large scale smooth features and the later ones describe finer scale features . all methods were started from an all - zero initial state , @xmath227 and identity covariance @xmath228 . in figure [ fig : lor_res1 ] , we compare the rms error of the ekf and reduced ekf with varying numbers of basis vectors @xmath19 , three ways for constructing the subspace ( pca , gp , and gmrf ) , and two ways to parameterize the unknown ( solid lines : centered at the prior mean @xmath86 , dashed lines : fixed offset @xmath87 ) . as the fixed offset @xmath88 , we use the empirical mean of the simulated model trajectory for pca and a constant @xmath88 for gp and gmrf . we observe that centering the parameterization at the prior mean improves the results dramatically compared to fixed mean . with small @xmath19 , the pca basis works slightly better than gp and gmrf . with the fixed mean parameterization , pca and gmrf work roughly equally well , and gp a little worse . but all in all , the three different ways of constructing the subspace all yield similar results . in this example , we are able to obtain a reasonably accurate filter even with @xmath229 ( ! ) , whereas the fixed offset parameterization requires roughly @xmath230 for similar accuracy . thus , we are able to reduce the dimension and the computational complexity almost two orders of magnitude compared to the full state dimension @xmath212 . a summary of the results for all cases @xmath209 using the pca basis is given in figure [ fig : lor_res_allk ] . we plot the relative difference between the rms error obtained with the full ekf and with the reduced ekf for all cases with varying @xmath19 , using only the parameterization @xmath86 . we observe that with high @xmath202 ( when the trajectories are smooth ) , a small @xmath19 is enough to capture the system ; for instance , with @xmath231 and @xmath232 , only roughly @xmath233 vectors are needed to get an accurate filter . with small @xmath202 , however , the system contains more fine scale features and larger @xmath19 is needed to obtain good filtering performance ; for example , @xmath234 requires roughly @xmath235 for similar accuracy . this example illustrates how the efficiency of the proposed approach depends on the smoothness properties of the system . and @xmath19 . ] here , we compare the reduced enkf described in section [ sec : enkf ] to the standard enkf . the results for the @xmath226 case with varying @xmath19 are given in figure [ fig : lor_res3 ] for different ensemble sizes . we observe that the reduced enkf works much better than the standard enkfs with small ensemble sizes ; with the reduced enkf , we are able to obtain a convergent filter with an ensemble size as small as @xmath236 , whereas the standard enkf has problems converging with @xmath237 . for example , the reduced enkf with @xmath238 and @xmath236 yields similar performance as full enkf with @xmath239 . again , with sufficiently high @xmath100 ( here @xmath240 ) the enkf performance starts to catch up . figure [ fig : lor_res3 ] also illustrates the interesting connection between @xmath19 ( number of basis vectors used ) and @xmath100 . with smaller @xmath19 , the performance that can be obtained is poorer , but , on the other hand , a smaller @xmath100 is needed to achieve that performance . with larger @xmath19 , better performance can be obtained , but only if @xmath100 is set high enough . that is , for each @xmath100 , there seems to be an @xmath19 that provides an optimal compromise between representation error ( due to small @xmath19 ) and sampling error ( due to small @xmath100 ) . we conclude that in ensemble filtering , the dimension reduction approach , when feasible , can offer a way to develop a reasonably accurate filter with fewer ensemble members . restricting the inference to a subspace can reduce the need for sample covariance matrix regularization via localization techniques , which otherwise are needed when high - dimensional filtering problems are solved with small ensemble sizes @xcite . next , we test the subspace filtering algorithms using the two - layer quasi - geostrophic model ( qg model ) @xcite , which is often used as a benchmark system for data assimilation studies for numerical weather prediction ( nwp ) . the model provides a reasonably good analogue of large - scale mid - latitude chaotic dynamics , while being relatively cheap computationally @xcite . next , we briefly describe the model equations and our estimation setup . for more details about the model as we use it , refer to @xcite . the two - layer quasi - geostrophic model simulates atmospheric flow for the geostrophic ( slow ) wind motions . the geometrical domain of the model is specified by a cylindrical surface vertically divided into two `` atmospheric '' layers . the model also accounts for an orographic component that defines the surface irregularities affecting the bottom layer of the model . the latitudinal boundary conditions are periodic , whereas the values on the top and the bottom of the cylindrical domain are user - supplied constant values . the geometrical layout of the two - layer qg model mapped onto a plane is illustrated in figure [ qg_geometry ] . in the figure , parameters @xmath241 and @xmath242 denote mean zonal flows in the top and the bottom atmospheric layers , respectively . the model formulation we use is dimensionless , where the non - dimensionalization is defined by the length scale @xmath243 , velocity scale @xmath244 , and the layer depths @xmath245 and @xmath246 . the model operates with variables called _ potential vorticity _ and _ stream function _ , where the latter is analogous to pressure . the model is formulated as a coupled system of pdes ( [ qg_conservation ] ) describing a conservation law for potential vorticity . the conservation law is given as @xmath247 where @xmath248 denotes the substantial derivatives for latitudinal wind @xmath249 and longitudinal wind @xmath250 , defined as @xmath251 ; @xmath252 denote the potential vorticity functions ; index @xmath215 specifies the top atmospheric layer ( @xmath253 ) and the bottom layer ( @xmath254 ) . interaction between the layers , as well as relation between the potential vorticity @xmath252 and the stream function @xmath255 , is modeled by the following system of pdes : @xmath256 @xmath257 here @xmath258 and @xmath221 denote dimensionless orography component and the northward gradient of the coriolis parameter , which we hereafter denote as @xmath259 . the relations between the physical attributes and dimensionless parameters that appear in ( [ pv_eq_1])([pv_eq_2 ] ) are as follows : @xmath260 where @xmath261 defines the potential temperature change across the layer interface , @xmath262 is the mean potential temperature , @xmath263 is acceleration of gravity , @xmath264 is the rossby number associated with the defined system , and @xmath265 and @xmath266 are dimensional representations of @xmath267 and @xmath221 , respectively . the system of ( [ qg_conservation])([pv_eq_2 ] ) defines the two - layer quasi - geostrophic model . the state of the model , and thus the target of estimation , is the stream function @xmath255 . for the numerical solution of the system , we consider potential vorticity functions @xmath268 and @xmath269 to be known , and invert the spatial equations ( [ pv_eq_1 ] ) and ( [ pv_eq_2 ] ) for @xmath255 . more precisely , we apply @xmath270 to equation ( [ pv_eq_1 ] ) and subtract @xmath271 times ( [ pv_eq_2 ] ) and @xmath272 times ( [ pv_eq_1 ] ) from the result , which yields the following equation : @xmath273-\left(f_1+f_2\right)\left[\nabla^2\psi_1\right ] = \nonumber\\ & \nabla^2 q_1-f_2\left(q_1-\beta y\right)-f_1\left(q_2-\beta y - r_s\right ) . \label{qg_helmholtz_eq}\end{aligned}\ ] ] equation ( [ qg_helmholtz_eq ] ) can be treated as a non - homogeneous helmholtz equation with negative parameter @xmath274 and unknown @xmath275 . once @xmath275 is solved , the stream function for the top atmospheric layer is determined by a poisson equation . the stream function for the bottom layer can be found by plugging the obtained value for @xmath276 into ( [ pv_eq_1 ] ) , ( [ pv_eq_2 ] ) and solving the equations for @xmath277 . the potential vorticity functions @xmath252 are evolved over the time by a numerical advection procedure which models the conservation equations ( [ qg_conservation ] ) . we run the qg model with @xmath278 grid in each layer , and the dimension of the state vector is thus 1600 . to generate data , we run the model with 1 hour time step using layer depths @xmath279 and @xmath280 . data is generated at every 6th step ( filter step is thus 6 hours ) by adding random noise for 100 randomly chosen grid points with standard deviation @xmath281 . for the estimation , bias is introduced to the forward model by using wrong layer depths , @xmath282 and @xmath283 . the model error covariance matrix was @xmath284 , where @xmath285 was chosen experimentally so that roughly optimal ekf performance was obtained ( here , the tracking performance was quite insensitive to @xmath221 ) . a snapshot of the model simulation at one time step and the measurement locations are illustrated in figure [ fig : qg_setup ] . in figure [ qg_res ] we compare the mean rms errors ( computed over 400 filter steps ) of the full ekf and the reduced ekf with two different parameterizations and different numbers of basis vectors @xmath19 used . only the pca basis is considered here . we observe , as in the other example , that the parameterization centered at the predicted mean works much better , especially with a small @xmath19 . we are able to obtain reasonably accurate filtering results using only @xmath230 basis vectors , which reduces the cpu time ( and memory requirements ) by almost two orders of magnitude compared to full ekf . interestingly , with a sufficiently high @xmath19 , the average rms error is actually _ lower _ than with the full ekf . this can be explained by the additional prior information brought into the problem by restricting the inference onto a subspace . when @xmath19 approaches the full dimension of the problem @xmath4 , all methods agree . next , we run the reduced enkf using the parameterization centered at the predicted mean , @xmath86 . in figure [ qg_res_enkf ] , we plot the mean rms error computed over 400 filter steps for varying ensemble sizes and varying number of basis vectors @xmath19 . the results qualitatively follow the same pattern as with the lorenz model in section [ sec : numex ] ; with large @xmath19 , a larger ensemble size is needed to get an accurate filter , and with small @xmath19 , a smaller ensemble size is sufficient to get close to the optimal performance that is achievable with that @xmath19 . for instance , with @xmath286 , ensemble size @xmath287 yields better tracking performance than @xmath288 with @xmath289 . the results are much better than what could be obtained with the standard enkf ; here , @xmath290 would be required even to get the standard enkf to converge ; see the results of @xcite . localization methods dramatically improve enkf performance ; for comparison , in figure [ qg_res_enkf ] we show the results for a simple localization , where we taper the prediction covariance matrix again using the 5th order piecewise rational function @xcite , experimentally tuning the cutoff length in the localization to achieve roughly optimal performance . however , the subspace algorithms still yield better results , as in the lorenz example . restricting the filtering onto a subspace regularizes the problem enough so that the need for localization is diminished . figure [ qg_res_enkf ] contains some results with ensemble size @xmath291 . here , zero ensemble size means that no samples were used to propagate the uncertainty ( only the posterior mean was propagated ) , and the prediction covariance was taken to be the model error directly : @xmath292 . for small @xmath19 , this simple 3d - var type of strategy with a fixed prior was enough to get a convergent filter . when the uncertainty propagation via samples was added and the sample size was increased , the filter accuracy was improved , as expected . the larger the value of @xmath19 , the more crucial the uncertainty propagation . this behavior can be explained by the fact that restricting the inference onto a subspace already heavily regularizes the problem , and thus propagating the covariance accurately is less important . for instance , using a small @xmath19 restricts the inference to a subspace spanned by spatially smooth basis vectors . in the full space algorithms , such smoothness information would be obtained by propagating the covariance forward in time . in the subspace method with small @xmath19 , non - smooth directions are explicitly removed from the estimation problem , and even a simple , fixed prior can yield reasonably accurate results . note also that here the number of observations is 100 , which is , in many cases , larger than the dimension of the subspace , making the estimation problems numerically well posed and the role of the prior less important . used . ensemble size 0 means that a fixed prior covariance was used , without error propagation : @xmath292 . ] in this paper , we presented an effective and simple - to - implement dimension reduction strategy for solving non - stationary inverse problems in a bayesian framework . by identifying a global reduced subspace that captures the essential features of the state vectors , we provided a new subspace - constrained bayesian estimation technique for reducing the computational cost of filtering algorithms . our approach is first applied to the kalman filter for linear gaussian models , and then generalized to nonlinear problems via the extended and ensemble kalman filters . in the kalman filter and extended kalman filter cases , the computational savings of our subspace - constrained technique is due to two sources : ( _ a _ ) the number of forward model simulations required in each prediction step is only equal to the reduced subspace dimensions , as the error is only propagated along the coordinates of subspace basis ; and ( _ b _ ) the update step can be formulated efficiently on the subspace coordinates . this way we also avoid handling matrices in the full dimension of the state space . in the ensemble version , computational savings stem from the fact that when the inference is constrained into a low - dimensional subspace , fewer ensemble members are needed for covariance estimation compared to the full space approach . also , the need for covariance localization techniques to regularize the predicted covariance is diminished . two approaches for constructing the reduced subspace are discussed . the first idea widely used in model reduction community is to obtain snapshots of typical model states ( e.g. , by performing a sufficiently long free model simulation ) and to compute the leading eigenvectors of the resulting empirical state covariance matrix . the second idea is to infer the state covariance matrix from a limited number of snapshots using a gaussian process hypothesis . this choice `` fills in '' the missing information about model states ( due to a limited number of snapshots ) using the correlation structure encoded in the particular choice of gp . we discussed gp constructions using either stationary kernels that directly specify the covariance matrix , or non - stationary differential operators that correspond to sparse precision matrices . the gp construction also opens the door to other possible state covariance reconstruction approaches ; for instance , one could infer the state covariance from previous data sets . we will investigate this extension in future work . we demonstrated the performance of our approach using two numerical examples . the first one is a 240-dimensional lorenz system , where the smoothness of the model states can be controlled with a tuning parameter . this rather low - dimensional example is used to demonstrate the performance of our dimension reduction approach in various regimes . for smooth settings , the dimension can be reduced dramatically ( to less than 10 ) while still obtaining filtering accuracy for both extended kalman filtering and ensemble filtering comparable to the full space algorithms . on the other hand , for non - smooth settings with `` rough '' features , the level of dimension reduction that maintains filter accuracy becomes less dramatic . the second example is a 1600-dimensional two - layer quasi - geostrophic model , where the state dimension can be a reduced to about 30 without losing filtering accuracy for both the extended and ensemble filters , compared to their full space counterparts . a two order of magnitude reduction in computing time is achieved for extended kalman filtering in this case . in ensemble filtering , our subspace approach yields accurate filtering results with smaller ensemble sizes than the standard enkf with localization . we thank alexander bibov for providing the quasi - geostrophic model implementation . a. solonen and j. hakkarainen acknowledge support from the academy of finland ( project numbers 284715 and 267442 ) . t. cui and y. marzouk also acknowledge support from the us department of energy , office of advanced scientific computing research ( ascr ) under grant number de - sc0003908 . 10 auvinen h. , bardsley j.m . , haario h. and kauranne t. , 2009 . the variational kalman filter and an efficient implementation using limited memory bfgs , international journal on numerical methods in fluids , 64(3 ) , pp . 314335 . bardsley j. , solonen a. , haario h. and laine m. , 2014 . randomize - then - optimize : a method for sampling from posterior distributions in nonlinear inverse problems , siam j. sci . comput . , 36(4 ) , a1895a1910 . cane , m.a . , kaplan , a. , miller , r.n . , tang , b. , hackert , e.c . , busalacchi , a.j . , 1996 . mapping tropical pacific sea level : data assimilation via a reduced state kalman filter . j. geophys . 101 , 599617 . liberty e. , woolfe f. , martinsson p. , rokhlin v. , and m. tygert , 2007 . randomized algorithms for the low - rank approximation of matrices . proceedings of the national academy of sciences , 104(51 ) , 2016720172 . lindgren f. , rue h. , and j. lindstrm , 2011 . an explicit link between gaussian fields and gaussian markov random fields : the stochastic partial differential equation approach , journal of the royal statistical society : series b , 73 , 423498 . e. ott , b. r. hunt , i. szunyogh , a. v. zimin , e. j. kostelich , m. corazza , e. kalnay , d. patil , and j. a. yorke , a local ensemble kalman filter for atmospheric data assimilation , tellus a , 56 ( 2004 ) , pp .

we consider a locally trivial fiber bundle over a compact oriented two - dimensional manifold , and a section of this bundle defined over , where is a discrete subset of . we call the set the _ set of singularities of the section _ . we assume that the behavior of the section at the singularities is controlled in the following way : coincides with the interior part of a surface with boundary , and is . for such sections we define an index of at a point of , which generalizes in the natural way the index of zero of a vector field , and then prove that the sum of this indices at the points of can be expressed as integral over of a-form constructed via a connection in . then we show that the classical hopf - poincar - gauss - bonnet formula is a partial case of our result , and consider some other applications .

1510.01395

let @xmath1 be a @xmath11-dimensional compact oriented manifold , and @xmath13 be a vector field on @xmath1 . the classical hopf - poincar theorem affirms that the sum of indices of zeros of @xmath13 is equal to the euler characteristic of @xmath1 : @xmath14 where @xmath15 is the set of zeros of @xmath13 and @xmath16 is the index of @xmath13 at @xmath17 ( see for example @xcite ) . if , in addition , the manifold @xmath1 is endowed by a riemannian metric @xmath18 , then the gauss - bonnet theorem says that the euler characteristic is expressed in terms of the curvature @xmath19 of this metric : @xmath20 where @xmath21 is the area form of @xmath18 . therefore , we have that @xmath22 then this formula can be generalized for sections of a unit sphere subbundle of the tangent bundle of a riemannian manifold , thus we arrive at the euler class ( see , e. g. @xcite , appendix 20 : gauss - bonnet theorem ) and it also can be generalized to arbitrary sphere bundles @xcite . in this paper we find a generalization of this formula to arbitrary locally trivial bundles with compact fibers over two - dimensional closed manifolds . namely , we consider a locally trivial fiber bundle @xmath0 over a compact oriented two - dimensional manifold @xmath1 , and a section @xmath2 of this bundle defined over @xmath3 , where @xmath4 is a discrete subset of @xmath1 . we call the set @xmath4 the _ set of singularities of the section _ @xmath5 . we assume that the behavior of the section @xmath2 at the singularities is controlled in the following way : @xmath6 coincides with the interior part of a surface @xmath7 with boundary @xmath8 , and @xmath8 is @xmath9 ( see details in subsection [ subsec:2_1 ] ) . it is worth noting that this idea was used by s .- s.chern in @xcite . for such sections @xmath2 we define an index of @xmath2 at a point of @xmath4 , which generalizes in the natural way the index of zero of a vector field ( subsection [ subsec:2_2 ] ) , and then prove that the sum of these indices at the points of @xmath4 can be expressed as integral over @xmath10 of a @xmath11-form constructed via a connection in @xmath12 . thus in theorem [ thm:2 ] we obtain a generalization of the formula . in section [ sec:4 ] we show that the classical fact that the index of a vector field on a closed two - dimensional riemann manifold is the integral of the curvature of the levi - civita connection can be obtained from our generalized version . also in this section we consider the projective bundles and projective connections and find a relation between the index of a section of a projective bundle and the curvature of a projective connection in this bundle . in section [ sec:5 ] our results are applied to the theory of @xmath23-structures with singularities . if @xmath24 is a @xmath25-principal bundle , which is reduced to a @xmath23-principal bundle over @xmath3 , where @xmath4 is a submanifold of @xmath1 , we say that @xmath26 is a @xmath23-principal bundle with singular set @xmath4 . if the @xmath25-principal bundle @xmath26 is a subbundle of the linear frame bundle @xmath27 of a manifold @xmath1 , we say that @xmath26 is a @xmath23-structure with singularities . almost all singularities which appear in classical differential geometry define @xmath23-structures with singularities . some examples are : a vector field on a riemannian @xmath28-dimensional manifold defines @xmath29-structure with singularities , a metric on an @xmath28-dimensional manifold @xmath1 which degenerates along a submanifold , determines a @xmath30-structure with singularities on @xmath1 ( examples of such metrics were considered for example in @xcite ) , @xmath31-webs with singularities on a two - dimensional manifolds , which in particular appear in considerations of algebraic ordinary differential equations on manifolds ( see , e.g. , @xcite ) define a @xmath23-structure with singularities , where @xmath32 is the lie subgroup of diagonal matrices @xcite , also sub - riemannian manifolds with singularities determine the corresponding @xmath23-structure with singularities @xcite . a @xmath23-principal bundle @xmath26 with singularities determines a section @xmath33 , which is a section with singularities of the bundle @xmath34 . for the case @xmath35 , we apply the results obtained in section [ sec:2 ] to this situation and obtain the relation between the index of this section and the connection in the bundle @xmath24 ( see section [ sec:5 ] for details ) . let @xmath1 be a @xmath11-dimensional oriented closed manifold , and @xmath36 a locally trivial fiber bundle with typical fiber @xmath37 and structure group @xmath23 . we will assume that @xmath23 is a connected lie group . for each @xmath38 , by @xmath39 we will denote the fiber of @xmath40 over @xmath41 , and by @xmath42 the inclusion . let @xmath43 be a discrete subset . let us consider a section @xmath5 , we will call @xmath4 the _ singular set _ of the section @xmath2 . assume that _ there exist a @xmath11-dimensional oriented compact manifold @xmath10 with boundary @xmath8 and a map @xmath44 such that the map @xmath45 restricted to @xmath46 is a diffeomorphism between @xmath47 and @xmath6_. we will call the map @xmath45 a _ resolution of the singularities of the section @xmath2_. _ a ) _ @xmath48 is a diffeomorphism . _ b ) _ the map @xmath49 is surjective . _ c ) _ @xmath50 . _ d ) _ let @xmath51 , where @xmath52 are the boundary components . then for any @xmath53 , there exist one and only one boundary component @xmath52 such that @xmath54 . _ e ) _ for any @xmath53 exists a neighborhood @xmath55 of @xmath41 such that 1 . @xmath55 is diffeomorphic to a disk ; 2 . there exists a closed set @xmath56 diffeomorphic to a ring such that the boundary of @xmath13 consists of two connected components @xmath57 and @xmath58 , where the connected component @xmath57 is a closed curve in @xmath47 and the other connected component @xmath58 is the connected component @xmath52 of the boundary @xmath8 which corresponds to @xmath41 by d ) , and @xmath59 [ thm:1 ] _ in what follows we will label each point @xmath53 such that @xmath54 by the corresponding index @xmath60_. [ rem:1 ] \a ) @xmath61 is a diffeomorphism and @xmath62 is the diffeomorphism inverse to @xmath2 . therefore @xmath48 is a diffeomorphism \b ) for each point @xmath63 , the point @xmath64 has the property that @xmath65 . let us take a point @xmath53 and a sequence of points @xmath66 converging to @xmath41 . the sequence of points @xmath67 has a convergent subsequence @xmath68 because @xmath10 is compact . if @xmath69 , then @xmath70 \c ) as @xmath71 is surjective and @xmath72 because @xmath73 , we have that @xmath50 . \d ) and e ) for a point @xmath53 let us take a neighborhood @xmath55 diffeomorphic to an open disk such that @xmath74 . the closure @xmath75 , where @xmath76 is the boundary of @xmath55 . let @xmath77 . then @xmath78 because @xmath1 and @xmath10 are compact . as @xmath48 is a diffeomorphism , and @xmath79 , we have a curve @xmath80 such that @xmath81 . this means that @xmath57 is a part of the boundary @xmath82 which lies in the interior part of @xmath10 . now , by b ) we know that @xmath83 , @xmath84 . finally we have that @xmath85 , and we know that @xmath86 is diffeomorphic to @xmath87 , and @xmath88 is diffeomorphic to @xmath76 , that is @xmath57 is diffeomorphic to a circle @xmath89 . thus @xmath86 is an open set in @xmath10 diffeomorphic to an open ring , and the boundary of @xmath86 consists of a closed curve in the interior part of @xmath10 and some components of the boundary of @xmath10 . however , this means that the boundary of @xmath86 contains one and only one component @xmath52 of @xmath8 , otherwise the fundamental group of @xmath90 . thus we get the affirmation d ) and the decomposition @xmath91 with the properties given by . therefore , we have proved e ) . from the proof of items d ) and e ) of theorem [ thm:1 ] it follows that , for each boundary component @xmath52 , we have a closed neighborhood @xmath92 of @xmath52 and a diffeomorphism of manifolds with boundary @xmath93 \to v_i$ ] such that @xmath94 , @xmath95)$ ] is a closed neighborhood of @xmath96 diffeomorphic to a disk , @xmath97 is the boundary of @xmath55 , and @xmath98 ( see figure [ fig:1 ] ) . ] [ rem:2 ] let @xmath40 be a unit vector bundle of @xmath99 , @xmath100 , and consider a section @xmath101 which takes value @xmath102 with respect to the polar coordinate system on @xmath103 . the point @xmath104 is the singular point of this section . if we take the coordinate system @xmath105 on @xmath106 , then @xmath107 . now the map @xmath108 \to e$ ] is given with respect to coordinates @xmath109 on @xmath110 by @xmath111 . therefore @xmath112 is a curve consisting of helices , @xmath113 is the fiber @xmath114 represents the image of @xmath115 . ] ] [ ex:1 ] first let us first observe that @xmath0 is trivial on the homotopy level . let @xmath116 be a chart of the atlas of @xmath40 . let @xmath117 where @xmath118 is the canonical projection onto @xmath37 . for each @xmath119 the map @xmath120 restricted to @xmath121 induces a diffeomorphism @xmath122 . note that if we take another chart @xmath123 , and @xmath124 is the corresponding map , then on @xmath125 we have that @xmath126 where @xmath127 is the gluing map of the charts . now , for any @xmath128 , we have @xmath129 , and , as @xmath23 is connected , @xmath130 is homotopic to the identity map . this means that for any @xmath38 we have well defined isomorphisms of the homotopy and ( co)homology groups : @xmath131 h_*(\eta_x ) : h_*(f_x ) \to h_*(f ) , \quad h^*(\eta_x ) : h^*(f ) \to h^*(f_x ) , \end{array } \label{eq:8}\ ] ] which do not depend on the chart . in particular , the following diagram is commutative : @xmath132_{\langle \cdot , \cdot \rangle}\ar[rr]^{\pi_1(\eta_x)\times h^1(\eta_x^{-1 } ) } & & \pi_1(f ) \times h^1(f ) \ar[dl]^{\langle \cdot , \cdot \rangle } \\ & \mathbb{r } & } \label{eq:9}\ ] ] where @xmath133 is the natural pairing through the integration . this means that , for @xmath134 \in \pi_1(f_x)$ ] and @xmath135 \in h^1(f)$ ] , we have @xmath136 , h^1(\eta_x)[\alpha ] \right\rangle = \int_\gamma \eta_x^ * \alpha = \int_{\eta_x \gamma } \alpha = \left\langle \pi_1(\eta_x)[\gamma ] , [ \alpha ] \right\rangle , \label{eq:10}\ ] ] and this equality neither depends on the chart , nor on the choice of representatives @xmath137 or @xmath138 of their equivalence classes . for @xmath96 the map @xmath139 \to e$ ] restricted to @xmath140 gives a map @xmath141 , and therefore an element @xmath142 via the isomorphism @xmath143 ( see ) . this element @xmath144 will be called the _ index of the section @xmath2 at the point @xmath96_. [ df:1 ] the index does not depend on the choice of the maps @xmath145 . [ rem:3 ] the index of the section @xmath2 at the point @xmath104 in example [ ex:1 ] is equal to @xmath146 , where @xmath137 is a generator of the group @xmath147 . [ ex:2 ] for @xmath148 we define the _ index @xmath149 of the section @xmath2 at the point @xmath96 with respect to @xmath150 _ as @xmath151 where @xmath152 is the curve representing the element @xmath142 and @xmath153 is a closed form representing the element @xmath148 . _ the index of a section @xmath2 with respect to @xmath150 _ is the number @xmath154 [ df:2 ] by , @xmath155 where @xmath135 = a$ ] and @xmath156 is defined in definition [ df:1 ] . [ rem:4 ] let @xmath157 \in h^1(\mathbb{s}^1 ; \mathbb{r})$ ] , where @xmath158 is the angular form on @xmath89 , then @xmath150 is a base of @xmath159 . then index of the section @xmath2 at the point @xmath104 in example [ ex:1 ] with respect to @xmath150 is equal to @xmath28 . let @xmath160 be a connection in the fiber bundle @xmath12 . let us denote by @xmath13 the _ vertical distribution _ , that is the distribution of the tangent spaces to the fibers , then we have @xmath161 . let @xmath148 and @xmath160 be a connection in @xmath12 . there exists a @xmath162-form @xmath163 such that 1 . @xmath164 ; 2 . for each @xmath38 , @xmath165 and @xmath166 = h^1(\eta_x ) a$ ] . [ stat:1 ] let us take an atlas @xmath167 of the bundle @xmath40 , and fix a @xmath162-form @xmath168 such that @xmath169 = a$ ] . let @xmath170 be the partition of unity subordinate to the covering @xmath171 , and @xmath172 ( see ) . then take the collection of @xmath162-forms @xmath173 , and then construct @xmath174 . then @xmath175 therefore , @xmath176 and @xmath177 = \sum_i \rho_i(x)\ , [ i_x^ * \eta_i^ * \bar\alpha ] = a , \label{eq:16}\ ] ] because @xmath178 . we have the decomposition @xmath161 , and therefore we can define the desired @xmath163 by @xmath164 and @xmath179 . let @xmath181 be a distribution on a manifold @xmath182 . a _ partial @xmath181-connection _ in a vector bundle @xmath183 is given by a covariant derivative : @xmath184 such that @xmath185 . [ subsubsec:3_2_1 ] the connection @xmath160 induces a partial linear @xmath160-connection @xmath186 in the vector bundle @xmath180 : @xmath187 ) , \quad x \in \mathfrak{x}^h(e ) , y \in \mathfrak{x}^v(e ) \label{eq:18}\ ] ] in the following way . one can easily check that the properties hold true . the geometrical sense of the partial @xmath160-connection is as follows . the connection @xmath160 induces a parallel translation of the fibers of the bundle @xmath12 along the paths on the base @xmath1 : for @xmath188 \to m$ ] , we have the diffeomorphism @xmath189 . therefore , if @xmath190 is a horizontal lift of @xmath137 , we have the parallel translation @xmath191 along the curve @xmath190 . as @xmath192 is linear , from this parallel translation we get a partial linear connection @xmath186 which is exactly the connection . let us fix a linear partial @xmath13-connection @xmath193 in the bundle @xmath180 , then the connection @xmath193 induces a liner connection in each fiber . we will assume that this connection does not have torsion . then let us consider the linear connection @xmath194 in the bundle @xmath180 : @xmath195 where @xmath196 corresponds to the decomposition @xmath161 , and @xmath197 . let @xmath198 be a vector bundle of rank @xmath199 over a manifold @xmath182 , and @xmath200 be a linear connection in this bundle . then we can extend the exterior differential @xmath201 to a differential operator @xmath202 , where @xmath203 is the module of sections of @xmath40 so that @xmath204 [ stat:2 ] let @xmath205 be a local frame field of @xmath40 over @xmath206 . then any @xmath207 restricted to @xmath55 can be written in the form @xmath208 . if we change the frame field to @xmath209 , then @xmath210 and @xmath211 , where @xmath212 are functions on @xmath55 . then , @xmath213 because @xmath214 . @xmath215 , where @xmath216 is the curvature tensor of the linear connection @xmath200 . [ rem:5 ] for @xmath217 , and @xmath218 , @xmath219 ) . \label{eq:22}\ ] ] [ stat:3 ] let @xmath220 be a local frame field of @xmath40 , and @xmath208 . then @xmath221 therefore @xmath222 ) = \\ x(\omega^\alpha(y ) ) e_\alpha + \omega^\alpha(y ) \nabla_x e_\alpha - y(\omega^\alpha(x ) ) e_\alpha - \omega^\alpha(x ) \nabla_y e_\alpha - \omega^\alpha([x , y ] ) e_\alpha = \\ d\omega^\alpha(x , y ) e_\alpha - \omega^\alpha \wedge \nabla e_\alpha ( x , y ) = d\omega ( x , y ) . \label{eq:24}\end{gathered}\ ] ] now let us assume that we have chosen the linear connection @xmath200 in the vector bundle @xmath180 given by , and therefore we have the corresponding differential operator @xmath223 , where @xmath224 is the lie algebra of vertical vector fields . let @xmath225 be the connection form of the connection @xmath160 . then @xmath226 where @xmath227 is horizontal . [ stat:4 ] it is sufficient to prove that @xmath228 if @xmath197 . we will apply . if @xmath229 , then @xmath230 ) = \nabla_x y - \omega ( [ x , y ] ) = 0 \label{eq:26}\ ] ] by the definition of @xmath200 ( see and ) . if @xmath231 , then @xmath230 ) = 0 \label{eq:27}\ ] ] again by the definition of @xmath200 and because the connection @xmath193 is torsion free . the tensor @xmath232 is called the _ curvature form of the connection @xmath160_. the curvature form @xmath233 vanishes if and only if @xmath160 is integrable . [ cor:1 ] indeed , by statement [ stat:3 ] , we have that , for any horizontal vector fields @xmath234 and @xmath235 , @xmath236 ) , \label{eq:28}\ ] ] therefore @xmath237 , for any @xmath238 and @xmath239 if and only if the lie bracket of horizontal vector fields is again a horizontal vector field , this is equivalent to integrability of @xmath160 . the decomposition @xmath161 induces the bigrading on the space of differential forms on @xmath12 : @xmath240 , and @xmath241 because the distribution @xmath13 is integrable . let @xmath242 and @xmath243 is a @xmath162-form such that @xmath244 ) = a$ ] . in statement [ stat:1 ] we have constructed a @xmath162-form @xmath245 such that @xmath246 vanishes on each fiber . note that the form @xmath247 satisfies the conditions ( 1 ) and ( 2 ) of statement [ stat:1 ] . let us calculate @xmath246 . as @xmath248 , we have @xmath249 ) = x \widetilde{\alpha}(\omega(y ) ) - y \widetilde{\alpha}(\omega(x ) ) - \widetilde{\alpha}(\omega([x , y ] ) = \\ \nabla_x \widetilde{\alpha}(\omega(y ) ) + \widetilde{\alpha } ( \nabla_x(\omega(y ) ) - \nabla_y \widetilde{\alpha}(\omega(x ) ) + \widetilde{\alpha } ( \nabla_y(\omega(x ) ) - \widetilde{\alpha}(\omega([x , y ] ) = \\ \nabla_x \widetilde{\alpha}(\omega(y ) ) - \nabla_y \widetilde{\alpha}(\omega(x ) ) + \widetilde{\alpha}(d\omega ) . \label{eq:29}\end{gathered}\ ] ] here we use statement [ stat:3 ] . let us consider the @xmath11-form @xmath250 if @xmath251 , then @xmath252 . now if @xmath253 , then @xmath254 . therefore , the form @xmath255 , and for @xmath229 and @xmath197 , by , we get that @xmath256 ) = l_x \alpha . \label{eq:31}\ ] ] now , using statement [ stat:3 ] , we arrive at the following statement . the form @xmath138 constructed in statement [ stat:1 ] lies in @xmath257 and @xmath258 , where @xmath259 and @xmath260 , and @xmath261 [ stat:5 ] note that the forms @xmath262 and @xmath263 in depends only on the connection @xmath160 and the form @xmath138 . now we apply the stokes theorem to the surface @xmath10 and arrive at a generalization of the gauss - bonnet theorem : for @xmath148 , let @xmath163 be the corresponding vertical form constructed in statement [ stat:1 ] . also let @xmath264 as in statement [ stat:5 ] . then @xmath265 [ thm:2 ] by the stokes theorem , @xmath266 we have that @xmath267 therefore , @xmath268 on the other hand , by statement [ stat:5 ] , @xmath269 and thus we get the required statement . let @xmath270 be a compact oriented two - dimensional riemannian manifold , and @xmath271 the bundle of unit vectors . then @xmath12 is a local trivial bundle with structure group @xmath272 and typical fiber @xmath89 . any @xmath273 defines a section @xmath274 , where @xmath4 is the set of zeros of @xmath13 . assume that the set @xmath4 is discrete . we will consider trivializations of the bundle @xmath12 given by local orthonormal frames @xmath275 . for each @xmath96 let us consider a orientation preserving diffeomorphism @xmath276 , where @xmath277 is a disk centered at the origin @xmath278 . let @xmath279 be an orthonormal frame field over @xmath280 which determines a diffeomorphism @xmath281 . now let consider a map @xmath282 \to u_i , \quad ( \phi , \rho ) \mapsto c_i ( ( 1-\rho)\cos\phi , ( 1-\rho)\sin\phi ) ) , \label{eq:38}\ ] ] and let @xmath283 be the local representation of the section @xmath2 with respect to the given trivialization , this means that @xmath284 now @xmath285 induces a diffeomorphism @xmath286 and therefore we get a map @xmath287 . now assume that the map @xmath288 can be prolonged to a map @xmath289 \to \mathbb{s}^1 $ ] . under this assumption we can construct the desired compact @xmath11-dimensional manifold @xmath10 and the map @xmath44 with properties ( 1 ) and ( 2 ) ( see section [ sec:2 ] ) . in geometric terms , our assumption means that the angle @xmath290 of the unit vector field @xmath2 with the vector field @xmath291 ( see ) has limit when we approach to the point @xmath292 by some curve , but the limit depends on the curve . [ rem:6 ] in order to construct @xmath10 , first we take @xmath293 , this is a submanifold with boundary which consists of the curves @xmath294 , then consider the disjoint union @xmath295 of @xmath296 copies of the cylinder @xmath297 \cong \partial d \times [ 0 , 1]$ ] , and attach @xmath295 to @xmath298 by the map @xmath299 determined by the maps @xmath300 thus we get the manifold @xmath301 . now the map @xmath44 is defined as follows : for each @xmath302 , we set @xmath303 , and for each @xmath304 $ ] , if @xmath305 belongs to @xmath60-th copy of @xmath297 \cong \partial d \times [ 0 , 1]$ ] , we set @xmath306 . from the construction it is clear that @xmath45 is well defined on @xmath307 . let @xmath308 be the class represented by the angle form @xmath309 . the riemannian connection @xmath200 of the metric @xmath18 defines a connection @xmath160 in the bundle @xmath310 . let @xmath163 be the @xmath162-form constructed in statement [ stat:1 ] . in terms of the trivialization @xmath311 where @xmath312 and @xmath275 is an orthonormal frame field , the form @xmath138 is written as follows : @xmath313 where @xmath314 and @xmath315 , and @xmath316 , @xmath317 are the connection coefficients of @xmath23 with respect to the orthonormal frame @xmath275 . if @xmath318 , then @xmath319 , therefore @xmath320 is parallel if and only if @xmath321 . this means that , in terms of the trivialization , the connection form is exactly the form @xmath138 given by . therefore @xmath138 vanishes over @xmath160 ( condition ( 1 ) in statement [ stat:1 ] ) . at the same @xmath138 restricted to a fiber is @xmath309 , this means satisfies condition ( 2 ) in statement [ stat:1 ] . [ rem:7 ] denote by @xmath322 the map @xmath323 \to \mathbb{s}^1 $ ] restricted to @xmath324 . then @xmath325 is homotopic to @xmath326 , therefore @xmath327 . however , @xmath328 restricted to @xmath329 and its degree coincides with the degree of @xmath330 restricted to @xmath331 , therefore the _ @xmath332 is equal to the classical index @xmath333 of the vector field @xmath2 at the point @xmath292_. therefore , @xmath334 this means that _ the classical index @xmath333 of vector field @xmath2 at the singular point @xmath292 coincides with the index @xmath335_. on the other hand @xmath336 , where @xmath19 is the curvature of the metric @xmath18 . therefore , in statement [ stat:5 ] @xmath337 and @xmath338 . thus , gives us @xmath339 at the same time , by the construction , @xmath48 is a diffeomorphism , @xmath4 is a set of measure zero , and @xmath340 is a @xmath11-form defined globally on @xmath1 , therefore @xmath341 and we get the classical result @xmath342 a _ projective bundle _ is a locally trivial bundle @xmath343 with typical fiber @xmath344 and structure group @xmath345 . the number @xmath28 is called the _ rank _ of @xmath346 . [ df:3 ] let @xmath310 be a vector bundle . then the bundle @xmath347 of @xmath162-dimensional subspaces of the fibers of @xmath12 is called the _ projectivization _ of @xmath12 and it is projective bundle . [ ex:4 ] if we have a projective bundle @xmath343 , then can we find a vector bundle @xmath12 such that @xmath348 ? note that for @xmath349 , we have @xmath350 . now for an even @xmath28 , the group @xmath351 , @xmath352 , is isomorphic to @xmath353 because for each equivalence class @xmath354 \in pgl(n)$ ] one can find a representative @xmath355 $ ] such that @xmath356 . this @xmath182 is unique because for @xmath357 $ ] such that @xmath358 , we have that @xmath359 and @xmath360 , this implies that @xmath361 . therefore in this case @xmath362 . for an odd @xmath28 , the group @xmath363 , where @xmath364 \mid \det a > 0 \right\ } , \quad pgl_{-}(n ) = \left\ { [ a ] \mid \det a < 0 \right\}. \label{eq:47}\ ] ] by the same arguments as for case of even @xmath28 we get that @xmath365 . note that @xmath366 is diffeomorphic to @xmath367 . if the structure group @xmath345 of a projective bundle @xmath346 reduces to @xmath367 , then @xmath346 is called _ orientable _ , on the contrary @xmath346 is called _ no orientable_. if a vector bundle @xmath12 is no orientable , then @xmath347 is no orientable . [ ex:5 ] now let consider the following situation . let @xmath368 be a lie group , @xmath369 be the center of @xmath368 , and @xmath370 . so we have the exact sequence @xmath371 & z \ar[r]^i & \widetilde{g } \ar[r]^p & g \ar[r ] & 0 . } \label{eq:48}\ ] ] let @xmath372 be a @xmath23-principal bundle , @xmath373 be a good covering of a manifold @xmath1 , and @xmath374 be the transition functions of @xmath1 . as the covering @xmath375 is good , we can lift @xmath376 up to @xmath377 of @xmath376 , this means that the following diagram is commutative @xmath378_p \\ u_{ij } \ar[ur]^{\widetilde{g}_{ij } } \ar[r]_{g_{ij } } & g% } \label{eq:49}\ ] ] consider the cech cochain @xmath379 over the covering @xmath375 , where @xmath380 as @xmath381 , the functions @xmath382 take values in @xmath383 , therefore in fact @xmath384 . 1 . the ech cochain @xmath17 is a normalized cocycle . 2 . the ech cohomology class @xmath385 \in \check{h}^2(m ; z)$ ] is trivial if and only if there exists a @xmath368-principal bundle @xmath386 such that @xmath387 . [ stat:6 ] let us prove that @xmath388 is a hech cocycle on @xmath375 . first of all note that we have @xmath389 then @xmath390 therefore @xmath391 . at the same time , @xmath392 hence follows that @xmath393 . moreover @xmath394 if at least two indices coincide . thus @xmath395 is a normalized cochain . now we have @xmath396 and @xmath397 thus @xmath398 , this implies @xmath399 if @xmath400 , where @xmath401 , then @xmath402 therefore we can construct the @xmath368-principal bundle @xmath403 with the transition functions @xmath404 because @xmath405 now we have that @xmath406 and from the following lemma it follows that there exists a principal bundle morphism @xmath407 written with respect to the trivializations as @xmath408 . let @xmath409 , @xmath410 be lie groups . let @xmath411 , @xmath410 be a @xmath409-principal bundle and @xmath412 be a lie group homomorphism . assume that @xmath373 is a trivializing covering for @xmath413 and @xmath414 at the same time . if the transition functions @xmath415 with respect to the covering @xmath375 are related by @xmath416 , then there exists a principal fiber bundle homomorphism @xmath417 corresponding to @xmath418 , i. e. @xmath419 . [ lm:1 ] we define @xmath420 by the commutative diagram @xmath421^{f_i } \ar[d]_{\psi_i^1 } & \pi_2^{-1}(u_i ) \ar[d]^{\psi_i^2 } \\ u_i \times g_1 \ar[r]^ { 1 \times f } & u_i \times g_2% } \label{eq:59}\ ] ] where @xmath422 are the trivializations which are principal bundle isomorphisms , therefore @xmath423 , where the action of @xmath409 on @xmath424 is given by @xmath425 . from this follows that @xmath426 . on the other hand , for @xmath427 we have @xmath428 . indeed , we have the commutative diagram for @xmath410 : @xmath429^{(x , g_a ) \to ( x , g^a_{ji } g_a ) } & & u_{ij } \times g_a \\ & \pi_1^{-1}(u_{ij } ) \ar[lu]^{\psi^a_i } \ar[ru]_{\psi^a_j } & % } \label{eq:60}\ ] ] let @xmath427 and @xmath430 , then @xmath431 . also @xmath432 , hence @xmath433 . therefore , @xmath434 thus @xmath435 on @xmath436 and therefore @xmath437 determine a homomorphism @xmath417 . from the construction of the bundle @xmath386 it follows that @xmath438 and the principal fiber bundle homomorphism constructed in lemma [ lm:2 ] is exactly the projection @xmath439 . let @xmath346 be a projective bundle of rank @xmath28 , @xmath28 is odd . assume that @xmath346 is orientable . let @xmath171 be a good covering of @xmath1 , @xmath440 , and @xmath441 the corresponding cocycle of transition functions . as @xmath442 are contractible , we can choose the maps @xmath443 such that @xmath444 , where @xmath445 is the canonical projection . then @xmath446 , therefore @xmath447 takes values in @xmath448 . note that @xmath448 is the center of the group @xmath353 . if @xmath343 is a projective bundle of odd rank @xmath28 and is oriented , then the obstruction to existence of a vector bundle @xmath12 such that @xmath348 is an element in @xmath449 . [ cor:2 ] let @xmath343 be a projective bundle of rank @xmath28 . the principal bundle @xmath450 with the group @xmath345 adjoint to @xmath346 is called the _ bundle of projective frames_. [ df:4 ] on the total space of a vector bundle @xmath310 the group @xmath451 acts freely by scaling , and @xmath348 is the quotient of @xmath12 with respect to this action . therefore , we have the @xmath452-principal bundle @xmath453 . also we have the induced action of @xmath452 on the total space of the linear frame bundle @xmath454 , and the quotient space is @xmath450 . the quotient map @xmath455 is @xmath452-principal bundle . if @xmath12 is an orientable vector bundle of rank @xmath456 , where @xmath28 is odd , then the structure group @xmath457 of @xmath454 reduces to @xmath353 , and in this case we have the double covering @xmath458 . for an orientable vector bundle @xmath12 of rank @xmath456 , one can take a metric @xmath18 on @xmath12 , therefore the @xmath457-principal bundle @xmath454 of linear frames reduces to a @xmath459-principal bundle of positively oriented orthonormal frames @xmath460 . therefore , for @xmath348 , the bundle @xmath450 reduces to the subgroup @xmath461 . [ ex:6 ] let @xmath343 be a projective bundle . a projective connection is a connection in @xmath450 . any linear connection in @xmath12 induces a projective connection in @xmath347 . indeed , the parallel transport preserves one - dimensional subspaces in fibers of @xmath12 . at the same time the canonical projection @xmath462 is a principal bundle morphism corresponding to the morphism of groups @xmath463 . therefore , any connection in @xmath454 determines a unique connection in @xmath464 @xcite(chapter ii , proposition 1 ) . two linear connections @xmath200 and @xmath465 in the vector bundle @xmath12 induce the same projective connection in @xmath347 if and only if the deformation tensor @xmath466 has the form @xmath467 [ stat:7 ] if @xmath348 , where @xmath12 is an oriented vector bundle , then for any projective connection in @xmath346 we can find an equi - affine connection @xmath200 in @xmath12 with respect to a given volume form @xmath468 , i.e. @xmath469 , such that @xmath200 induces the given connection in @xmath346 . indeed in this case @xmath470 covers @xmath471 , @xmath472 , and any connection in @xmath473 lifts to a connection in @xmath470 . let @xmath12 be an oriented vector bundle , and @xmath468 be a volume form in @xmath12 . then there is one - to - one correspondence between equi - affine connections on @xmath12 with respect to @xmath309 and the projective connections in @xmath347 . [ stat:8 ] in terms of parallel translation this correspondence can be explained as follows . let @xmath76 be a projective connection in @xmath348 induced by a linear connection @xmath200 in @xmath12 . in terms of the parallel translation this means that a section @xmath474 of @xmath346 along @xmath137 is parallel with respect to the connection @xmath76 if there exists a section @xmath2 of @xmath12 along @xmath137 which is parallel with respect to @xmath200 and generates @xmath474 . this can be expressed also in terms of covariant derivative : @xmath475 $ ] is parallel with respect to @xmath76 if and only if @xmath476 . now any projective frame field @xmath477 in @xmath348 determines a frame field @xmath478 in @xmath12 such that @xmath479 $ ] , and @xmath480 $ ] , where @xmath481 , up to a scaling . if in addition we have a volume form @xmath309 , the projective frame field determines the linear frame field @xmath482 up to a sign . then @xmath483 and @xmath484 , hence follows that @xmath485 . if we assume that the linear connection @xmath200 is equi - affine with respect to a volume form @xmath309 , then @xmath469 and the function @xmath486 . [ rem:8 ] let @xmath346 be a projective bundle of rank @xmath162 . let @xmath487 be the subgroup of affine transformations of line . then @xmath488 and a section of projective bundle @xmath346 is equivalent to the reduction of the bundle @xmath450 to the subgroup @xmath489 . [ stat:9 ] this follows from the fact that the group @xmath490 acts transitively on @xmath491 , and the isotropy subgroup of a point in @xmath491 is the group @xmath489 . let @xmath310 be an oriented vector bundle and @xmath492 = 2 . let @xmath348 be the corresponding projective bundle . also let us take a projective connection @xmath76 in @xmath346 . the typical fiber of @xmath346 is @xmath491 . let us consider the canonical projection @xmath493 , @xmath494 $ ] , and the corresponding double covering @xmath495 . let us consider the form @xmath496 then @xmath497 is the angular form on @xmath89 and there exists @xmath498 such that @xmath499 . then , as @xmath500 , we have that @xmath501 . therefore , @xmath502 is a generator of @xmath503 . now let us take a projective connection @xmath160 is @xmath346 which is determined by an equi - affine connection @xmath504 in @xmath12 . now we can construct the form @xmath138 in @xmath505 such that @xmath164 , @xmath506 , and @xmath507 represents the class @xmath508 \in h^1(\mathbb{r}p^1)$ ] . let us take a metric @xmath18 in @xmath12 , the volume form @xmath309 associated with @xmath18 , and let @xmath509 be the connection form on @xmath12 , that is the projection onto the vertical subspace parallel to @xmath504 . now let us consider the @xmath162-form @xmath510 where @xmath511 , @xmath512 , and @xmath278 is the zero section of @xmath12 . this form @xmath513 vanishes on @xmath160 , and the metric @xmath18 allows us to choose the trivialization of the bundle @xmath12 such that the maps @xmath514 are isomorphisms of the euclidean vector spaces @xmath515 and @xmath516 with the standard metric . therefore by @xmath514 the form @xmath513 restricted to the fibers is sent to the form . now let us consider local coordinate system @xmath517 on @xmath12 such that @xmath518 , then @xmath519 . consider the local coframe @xmath520 , @xmath521 . @xmath522 where @xmath523 is the curvature tensor of the linear connection in @xmath12 . [ lm:2 ] we have @xmath524 the form @xmath525 is written with respect to this frame as follows : @xmath526 now let us assume that the connection is adapted to the metric @xmath18 in @xmath12 , in this case @xmath527 . then we have @xmath528 because @xmath529 also we have that @xmath530 y^1 d dy^2 = \gamma_i{}^2_1 y^1 dy^1 \wedge dx^i + r_{ij } { } ^2_1 ( y^1)^2 dx^i \wedge dx^j , \\[5pt ] dy^2 \wedge dy^1 = dy^2 \wedge dy^1 - \gamma_i{}^2_1 y^1 dx^i \wedge dy^1 , \\[5pt ] y^2 d dy^1 = \gamma_i{}^1_2 y^2 dy^2 \wedge dx^i + r_{ij } { } ^1_2 ( y^2)^2 dx^i \wedge dx^j , \end{array } \label{eq:70}\ ] ] therefore , @xmath531 thus , @xmath532 = \\ - \frac{1}{((y^1)^2 + ( y^2)^2)^2 } \left ( y^1 dy^1 + y^2 dy^2 \right ) \wedge \left ( -y^2 dy^1 + y^1 dy^2 \right ) + \\ + \frac{1}{(y^1)^2 + ( y^2)^2 } dy^1 \wedge dy^2 + \frac{1}{2 } r_{ij}{}^2_1 dx^i \wedge dx^j \\ = \frac{1}{2 } r_{ij}{}^2_1 dx^i \wedge dx^j . \label{eq:72}\end{gathered}\ ] ] note that @xmath533 is the curvature form of the connection @xmath160 restricted to @xmath534 , where @xmath55 is the coordinate neighborhood . let @xmath310 be a vector bundle , @xmath348 , and @xmath535 . 1 . there exist a @xmath162-form @xmath536 such that @xmath537 . 2 . @xmath164 3 . @xmath506 , and @xmath507 represents the class @xmath508 \in h^1(\mathbb{r}p^1)$ ] . 4 . @xmath538 . [ thm:3 ] let @xmath310 be a vector bundle and @xmath348 . assume that @xmath18 is a metric in @xmath12 . let @xmath233 be the curvature form of the connection in @xmath346 induced by a connection in @xmath12 adapted to @xmath18 . let @xmath539 is represented by the form @xmath540 . if @xmath541 is a section with singularities , @xmath542 [ cor:3 ] let @xmath543 be a lie group , and @xmath544 be a lie subgroup of @xmath23 . let @xmath1 be a manifold , and @xmath4 be a submanifold of @xmath1 . a principal @xmath543-bundle @xmath545 is called a _ principal @xmath23-bundle with singularities @xmath4 _ if the structure group @xmath543 of @xmath546 reduces to @xmath23 . [ df:5 ] let @xmath234 be a vector field on a manifold @xmath1 with a discrete set of zeros @xmath43 . then the frame bundle @xmath27 is a principal @xmath547-bundle with singularities on @xmath1 . it is clear that on @xmath3 the bundle @xmath27 reduces to the principal @xmath547-subbundle consisting of frames with the first vector equal to the value of @xmath234 at the corresponding point . [ ex:7 ] if @xmath1 is a two - dimensional oriented riemannian manifold , and @xmath234 is a vector field on @xmath1 with a discrete set of zeros @xmath43 , then the orthonormal frame bundle @xmath548 is a principal @xmath549-bundle with singularities where the set of singularities @xmath4 because the vector field induced trivialization of @xmath548 over @xmath3 . [ ex:8 ] let us recall that there is an important relation between @xmath160-reductions of a principal @xmath23-bundle @xmath372 and sections of the fiber bundle @xmath552 which is given by the following statement . let @xmath553 be a @xmath23-principal bundle and let @xmath160 be a lie subgroup of @xmath23 . then @xmath23 has a reduction to a principal @xmath160-bundle if and only if the associated bundle @xmath554 admits a section.in particular , if @xmath555 then @xmath553 has a section . [ stat:10 ] in what follows by @xmath558 we denote the lie algebra of @xmath543 , and by @xmath559 the lie algebra of @xmath23 . moreover , we suppose that the lie group @xmath543 is compact and connected . then @xmath23 is also compact and the quotient @xmath560 is reductive , i.e , @xmath558 has an @xmath561-invariant decomposition @xmath562 . let @xmath1 be a @xmath11-dimensional compact connected oriented manifold . let @xmath563 be a principal @xmath23-bundle with singularities , where the set of singularities @xmath4 consists of isolated points . consider the fiber bundle @xmath564 . recall that the canonical projection @xmath565 is a principal @xmath23-bundle @xcite(chapter i , proposition 5.5 ) , and @xmath566 is a fiber bundle morphism : @xmath567^{\pi_g } \ar[rd]_{\pi } & & e = \bar{p}/g \ar[ld]^{\pi_e } \\ & m & % } \label{eq:74}\ ] ] assume that @xmath568 is endowed with a connection form @xmath569 , and let @xmath570 be the horizontal distribution of the connection , and @xmath571 the vertical distribution on @xmath568 . note that @xmath572 and @xmath573 are vector subbundles of @xmath574 , and @xmath575 . \a ) there exists a operator field @xmath598 on @xmath12 such that , for each @xmath588 , the following diagram is commutative : @xmath599^{\widetilde{\omega } } \ar[d]_{d\pi_g } & \bar{v}_{\bar{p } } \ar[d]^{d\pi_g } \\ t_{\pi_g(p ) } e \ar[r]^{\omega_e } & ve_{\pi_g(p ) } } \label{eq:76}\ ] ] \a ) if @xmath606 , then @xmath607 , and @xmath608 , where @xmath609 and @xmath610 with @xmath611 . by lemma [ lm:3 ] , we have @xmath612 then , @xmath613 this proves that @xmath598 in the diagram is well defined . \b ) from the diagram it follows that @xmath614 . for any @xmath615 , take @xmath616 such that @xmath617 . then @xmath618 , this means that @xmath619 , where @xmath611 . then @xmath620 , therefore @xmath621 and @xmath622 . thus @xmath623 and therefore @xmath624 . the bundle @xmath310 is the bundle associated with the principal @xmath543-bundle @xmath629 with the fiber @xmath560 with respect to the left action of @xmath543 on @xmath560 . the distribution @xmath630 is the connection induced on the associated bundle by the connection @xmath572 on @xmath568 . [ rem:9 ] the vertical subbundle @xmath600 is isomorphic to the bundle associated with the principal @xmath23-bundle @xmath632 with the fiber @xmath633 with respect to the left action of @xmath23 on @xmath633 : @xmath634 . [ lm:4 ] the isomorphism @xmath635 follows from the fact that @xmath636 @xcite(chapter iv , section 18.18 ) and that @xmath637 @xcite(chapter 4 , section 5 ) . in the exact form the isomorphism is written as follows : @xmath638 \mapsto d\pi_g(\sigma_{\bar{p } } ( a ) ) . \label{eq:79}\ ] ] let @xmath629 be a principal @xmath639-bundle with singularities , where @xmath543 is a compact lie group . let us take a connection @xmath572 in @xmath568 with the corresponding connection form @xmath509 . let @xmath640 and @xmath160 be the connection in @xmath12 induced by @xmath572 with the connection form @xmath598 ( see statement [ stat:11 ] and remark [ rem:9 ] ) . in this case one can construct the form @xmath138 explicitly . as @xmath560 is a homogeneous space of a compact lie group , we can take an invariant form @xmath643 such that @xmath644 = a$ ] ( see , e.g. , @xcite , ch.1 , 1 ) . note that @xmath40 is determined by its value @xmath645 at @xmath646 \in \bar{g}/g$ ] . then from lemma [ lm:4 ] it follows that @xmath40 defines a form @xmath647 : @xmath648 ) = \xi_0(a ) . \label{eq:80}\ ] ] recall that the trivializations of @xmath654 are constructed in the following way . for a good covering @xmath280 of @xmath1 we take sections @xmath655 , then we define the charts @xmath656 ) \to [ s_i(x ) \bar{g } ] = \pi_g(s_i(x ) \bar{g } ) . \label{eq:82}\ ] ] therefore , for @xmath657 , @xmath658 , @xmath659 = \pi_g(s_i(x)\bar{g})$ ] . now let us fix @xmath660 and let @xmath661 . for any @xmath662 and @xmath663 , we have that @xmath664 = \eta_x^{-1}(\bar{g})$ ] and @xmath665 = d\eta_x^{-1 } ( dl_{\bar{g } } a)$ ] , where @xmath666 , and @xmath667 , @xmath668 . note that here we use the isomorphism constructed in lemma [ lm:4 ] . then @xmath669 therefore @xmath670 , thus @xmath671 , then @xmath672 . the curvature form @xmath675 of the connection @xmath598 ( see section [ sec:3 ] , equation ) can be represented by the following equality : @xmath676 , \label{eq:84}\ ] ] where @xmath677 : \omega^{k}(e , te)\times \omega^{l}(e , te)\rightarrow \omega^{k+l}(e , te)$ ] is the frolicher - nijenhuis bracket @xcite(chapter iv , section 16.3 ) . in particular , for @xmath162-forms @xmath678 and @xmath679 this bracket is given by the following equation @xmath680(x , y)&=[\omega_1(x ) , \omega_2(y)]-[\omega_1(y ) , \omega_2(x)]-\omega_2([\omega_1(x ) , y])\-\\ & \omega_1([\omega_2(x ) , y])-[\omega_1(y),x]-[\omega_2(y ) , x]+(\omega_1 \circ \omega_2+\omega_2 \circ \omega_1)[x , y].\end{aligned}\ ] ] we repeat the arguments of subsection [ subsec:3_4 ] and get that @xmath681 ) . \label{eq:85}\ ] ] thus , by assuming the existence of the @xmath11-dimensional oriented compact manifold @xmath10 and the map @xmath682 satisfying the properties given in section [ sec:1 ] , by theorem [ thm:2 ] , we obtain the following equality @xmath683))=\sum_{i=\overline{1 , k } } index_{x_i}(s , a ) , \label{eq:86}\ ] ] where @xmath684 is the section corresponding to the reduction of @xmath546 to @xmath23 . the result given by the equation will be called _ the hopf - gauss - bonnet theorem for principal @xmath23-bundle with singularities_. if @xmath234 is a vector field on @xmath1 , and @xmath4 is the set of zeros of @xmath234 , then @xmath685 defines a section of the unit tangent bundle @xmath686 over @xmath687 . since @xmath688 is the associated bundle to the bundle of orthonormal frames @xmath689 with respect of standard action of @xmath272 on @xmath690 , then @xmath691 reduces to the trivial subgroup @xmath692 of @xmath693 now , @xmath694 , and @xmath272 is diffeomorphic to @xmath690 , then @xmath695 is @xmath162-dimensional vector space generated by the angle form on @xmath690 . therefore , we can choose this generator to construct the form @xmath696 , and the connection form on @xmath548 induced by the levi - civita connection . this connection form is given by equation . thus , if @xmath10 and @xmath697 are the manifold and the map , respectively , constructed in section [ sec:4 ] , the equation reduces to the equation . hence , hopf - gauss - bonnet theorem in principal @xmath23-bundle with singularities is a generalization of the classical gauss - bonnet theorem . [ rem:10 ] let @xmath698 be @xmath296 vector bundles of rank @xmath11 over a compact connected riemannian two - dimensional manifold @xmath1 , and for each @xmath60 , let @xmath699 be the zero section of @xmath700 . let @xmath701 . for all @xmath60 let us take sections @xmath702 . then @xmath703 is a section of @xmath704 with singularities , whose set of singularities is the set @xmath705 of zeros of @xmath703 . thus , if we consider the product bundle @xmath706 over @xmath687 , where @xmath707 , then the map @xmath708 is a section with singularities of @xmath709 . in addition , suppose that the section @xmath710 admits a resolution of singularities ( see section [ sec:1 ] ) . this means that there exists a @xmath11-dimensional oriented compact manifold @xmath10 with boundary @xmath711 , and a map @xmath712 such that @xmath45 restricted to @xmath46 is a diffeomorphism between @xmath47 and @xmath6 . let us take connection forms @xmath713 on @xmath714 , for @xmath715 , then @xmath716 is a connection form on @xmath12 . the curvature form of @xmath509 is @xmath717 , where @xmath718 is the curvature form of @xmath719 . since the fiber of the bundle @xmath714 is @xmath720 , and the cohomology class @xmath721 $ ] represented by the angular form on @xmath722 is a generator of @xmath723 , then the @xmath162-form @xmath724 represents a nonzero cohomology class @xmath725 , where @xmath726 is the fiber of the bundle @xmath12 . note that @xmath37 retracts by deformation to the @xmath296-torus @xmath727 . with this in mind , we can choose the @xmath162-form @xmath728 satisfying properties ( 1 ) and ( 2 ) of statement [ stat:1 ] as follows : @xmath729 where @xmath730 is the connection form of @xmath704 ( compare with ) . let @xmath1 be a compact connected two - dimensional manifold , let @xmath738 , are @xmath296 vector bundles of rank @xmath11 over @xmath1 , let @xmath739 , where @xmath740 is the zero section of @xmath700 , and let @xmath702 be a section of @xmath700 for each @xmath60 . if @xmath741 are connections on @xmath742 respectively , @xmath743 are the curvatures of these connections , and @xmath744 , then @xmath745 ) , \label{eq:89}\ ] ] where @xmath746 is the angular form of @xmath747 , and @xmath309 is the area form on @xmath1 . [ thm:5 ] let @xmath10 be an oriented @xmath11-manifold with boundary , and let @xmath748 a map which satisfies the condition given in section [ sec:1 ] , that is , @xmath749 restricted to @xmath46 is a diffeomorphism between @xmath47 and @xmath750 . now , because @xmath751 , @xmath45 has the form @xmath752 , and for each @xmath60 , the map @xmath753 also satisfies this condition . this fact follows from the commutative diagram @xmath754^{id } \ar[d]_{s } & m\setminus \sigma \ar[d]^{s_i}\\ \overset{\circ}{s}\ar[r]^{q } & s(m\setminus \sigma)\ar[r]^{\tilde{p}_i } & s_i(m\setminus \sigma ) } \label{eq:90}\ ] ] since the vertical arrows are diffeomorphisms , we have that @xmath749 is a diffeomorphism . therefore , for each @xmath755 , the map @xmath756 is also a diffeomorphism . let us take the cohomology class @xmath757 $ ] of the fiber of @xmath709 , where @xmath758 is the angular form on the fiber @xmath722 of @xmath700 , then @xmath759 + \cdots + [ d\phi_k]$ ] . thus , the expression at the right side of equation can be written as follows @xmath760 ) . \label{eq:91}\ ] ] furthermore , @xmath761 , and @xmath762 . therefore , the expression at the left side of equation reduces to the following equality @xmath763 hence , we obtain the following equation @xmath764 ) . \label{eq:93}\ ] ] note that if @xmath765 is non empty , and @xmath766 , then the indexes @xmath767 and @xmath768 could be different , still in the case where the vector bundles @xmath700 and @xmath769 are the same . [ rem:11 ] jos ricardo arteaga b , mikhail malakhaltsev , and alexander haimer trejos serna . isometry group and geodesics of the wagner lift of a riemannian metric on two - dimensional manifold . , 33(4):293311 , oct 2012 .

we consider a financial market model driven by an-valued gaussian process with stationary increments which is different from brownian motion . this driving noise process consists of independent components , and each component has memory described by two parameters . for this market model , we explicitly solve optimal investment problems . these include ( i ) merton s portfolio optimization problem ; ( ii ) the maximization of growth rate of expected utility of wealth over the infinite horizon ; ( iii ) the maximization of the large deviation probability that the wealth grows at a higher rate than a given benchmark . the estimation of paremeters is also considered .

math0506621

in this paper we study optimal investment problems for a financial market model with memory . this market model @xmath2 consists of @xmath1 risky and one riskless assets . the price of the riskless asset is denoted by @xmath3 and that of the @xmath4th risky asset by @xmath5 . we put @xmath6 , where @xmath7 denotes the transpose of a matrix @xmath8 . the dynamics of the @xmath0-valued process @xmath9 are described by the stochastic differential equation @xmath10 , \quad t\ge 0 , \\ \quad s_i(0)&=s_i,\qquad\qquad\qquad\qquad\qquad\qquad\qquad i=1,\dots , n , \end{split } \label{eq:1.1}\ ] ] while those of @xmath3 by the ordinary differential equation @xmath11 where the coefficients @xmath12 , @xmath13 , and @xmath14 are continuous deterministic functions on @xmath15 and the initial prices @xmath16 are positive constants . we assume that the @xmath17 volatility matrix @xmath18 is nonsingular for @xmath19 . the major feature of the model @xmath2 is the @xmath0-valued driving noise process @xmath20 which has memory . we define the @xmath21th component @xmath22 by the autoregressive type equation @xmath23 where @xmath24 , @xmath25 , is an @xmath0-valued standard brownian motion defined on a complete probability space @xmath26 , the derivatives @xmath27 and @xmath28 are in the random distribution sense , and @xmath29 s and @xmath30 s are constants such that @xmath31 ( cf . anh and inoue @xcite ) . equivalently , we may define @xmath22 by the moving - average type representation @xmath32 ds,\quad t\in\mathbf{r } \label{eq:1.5}\ ] ] ( see ( * ? ? ? * examples 2.12 and 2.14 ) ) . the components @xmath22 , @xmath33 , are gaussian processes with stationary increments that are independent of each other . each @xmath22 has short memory that is described by the two parameters @xmath29 and @xmath30 . in the special case @xmath34 , @xmath22 reduces to the brownian motion @xmath35 . driving noise processes with short or long memory of this kind are considered in @xcite , anh et al . @xcite and inoue et al . @xcite , for the case @xmath36 . we define @xmath37 where @xmath38 is the @xmath39-null subsets of @xmath40 . this filtration @xmath41 is the underlying information structure of the market model @xmath2 . from ( [ eq:1.5 ] ) , we can easily show that @xmath42 is a semimartingale with respect to @xmath43 ( cf . * section 3 ) ) . in particular , we can interpret the stochastic differential equation ( [ eq:1.1 ] ) in the usual sense . in actual calculations , however , we need explicit semimartingale representations of @xmath44 . it should be noticed that ( [ eq:1.5 ] ) is not a semimartingale representation of @xmath44 ( except in the special case @xmath34 ) . for , @xmath35 involves the information of @xmath45 with @xmath46 and vice versa . the following two kinds of semimartingale representations of @xmath44 are obtained in ( * ? ? ? * example 5.3 ) and ( * ? ? ? * theorem 2.1 ) , respectively : @xmath47ds , \quad t\ge 0,\quad j=1,\dots , n , \label{eq:1.6 } \\ y_j(t)&=b_j(t)-\int_0^t\left[\int_0^s l_j(s , u)db_j(u)\right]ds , \quad t\ge 0,\quad j=1,\dots , n , \label{eq:1.7}\end{aligned}\ ] ] where , for @xmath33 , @xmath48 is the so - called _ innovation process _ , i.e. , an @xmath49-valued standard brownian motion such that @xmath50 notice that @xmath51 s are independent of each other . the point of and is that the deterministic kernels @xmath52 and @xmath53 are given explicitly by @xmath54 with @xmath55 , \quad s\ge 0 . \label{eq:1.10}\ ] ] we have the equalities @xmath56 many authors consider financial market models in which the standard driving noise , that is , brownian motion , is replaced by a different one , such as fractional brownian motion , so that the model can capture _ memory effect_. to name some related contributions , let us mention here comte and renault @xcite , rogers @xcite , heyde @xcite , willinger et al . @xcite , barndorff - nielsen and shephard @xcite , barndorff - nielsen et al . @xcite , hu and ksendal @xcite , hu et al . @xcite , elliott and van der hoek @xcite , and heyde and leonenko @xcite . in most of these references , driving noise processes are assumed to have stationary increments since this is a natural requirement of simplicity . among such models , the above model @xmath2 driven by @xmath44 which is a gaussian process with _ stationary increments _ is possibly the simplest one . one advantage of @xmath2 is that , by the semimartingale representations ( [ eq:1.6 ] ) and ( [ eq:1.7 ] ) of @xmath44 , it admits _ explicit calculations _ in problems such as those considered in this paper . another advantageous feature of the model @xmath2 is that , assuming @xmath57 , real constants , we can easily estimate the characteristic parameters @xmath29 , @xmath30 and @xmath58 from stock price data . we consider this parameter estimation in appendix c. for the market model @xmath2 , we consider an agent who has initial endowment @xmath59 and invests @xmath60 dollars in the @xmath4th risky asset for @xmath61 and @xmath62x^{x,\pi}(t)$ ] dollars in the riskless asset at each time @xmath63 , where @xmath64 denotes the agent s wealth at time @xmath63 . the wealth process @xmath64 is governed by the stochastic differential equation @xmath65\frac{ds_0(t)}{s_0(t ) } + \sum\nolimits_{i=1}^{n}\pi_i(t)\frac{ds_i(t)}{s_i(t ) } , \quad x^{x,\pi}(0)=x . \label{eq:1.12}\ ] ] here , we choose the self - financing strategy @xmath66 from the admissible class @xmath67 for the finite time horizon of length @xmath68 , where @xmath69 denotes the euclidean norm of @xmath0 . if the time horizon is infinite , we choose @xmath70 from the class @xmath71 let @xmath72 and @xmath73 . in this paper , we consider the following three optimal investment problems for the model @xmath2 : @xmath74 , \tag{\textbf{p1}}\\ & j(\alpha):=\sup_{\pi\in\mathcal{a } } \limsup_{t\to\infty } \frac{1}{\alpha t}\log e\left[(x^{x,\pi}(t))^{\alpha}\right ] , \tag{\textbf{p2}}\\ & i(c):=\sup_{\pi\in\mathcal{a } } \limsup_{t\to\infty}\frac{1}{t } \log p\left[x^{x,\pi}(t)\ge e^{ct}\right ] . \tag{\textbf{p3}}\end{aligned}\ ] ] the goal of problem p1 is to maximize the expected utility of wealth at the end of the finite horizon . this classical optimal investment problem dates back to merton @xcite . we refer to karatzas and shreve @xcite and references therein for work on this and related problems . in hu et al . @xcite , this problem is solved for a black scholes type model driven by fractional brownian motion . in section [ sec:2 ] , assuming @xmath75 for @xmath33 , we explicitly solve this problem for the model @xmath2 . our approach is based on a cameron martin type formula which we prove in appendix a. this formula holds under the assumption that a relevant riccati type equation has a solution , and the key step of our arguments is to show the existence of such a solution ( lemma [ lem:2.1 ] ) . the aim of problem p2 is to maximize the growth rate of expected utility of wealth over the infinite horizon . this problem is studied by bielecki and pliska @xcite , and subsequently by other authors under various settings , including fleming and sheu @xcite , kuroda and nagai @xcite , pham @xcite , nagai and peng @xcite , hata and iida @xcite , and hata and sekine @xcite . in section [ sec:3 ] , we solve problem p2 for the model @xmath2 by verifying that a candidate of optimal strategy suggested by the solution to problem p1 is actually optimal . in so doing , existence results on solutions to riccati type equations ( lemmas [ lem:2.1 ] and [ lem:3.5 ] ) play a key role as in problem p1 . the result of nagai and peng @xcite on the asymptotic behavior of solutions to riccati equations , which we review in appendix b , is also an essential ingredient in our arguments . the purpose of problem p3 is to maximize the large deviation probability that the wealth grows at a higher rate than the given benchmark @xmath76 . this problem is studied by pham @xcite , in which a significant result , that is , a duality relation between problems p2 and p3 , is established . subsequently , this problem is studied by hata and iida @xcite and hata and sekine @xcite under different settings . in section [ sec:4 ] , we solve problem p3 for the market model @xmath2 . in the approach of @xcite , one needs an explicit expression of @xmath77 . since our solution to problem p2 is explicit , we can solve problem p3 for @xmath2 using this approach . as in @xcite , our solution to problem 3 is given in the form of a sequence of nearly optimal strategies . for @xmath78 with certain constant @xmath79 , an optimal strategy , rather than such a nearly optimal sequence , is obtained by ergodic arguments . in this section , we consider the finite horizon optimization problem p1 for the market model @xmath2 . throughout this section , we assume @xmath72 and @xmath80 thus @xmath75 rather than @xmath81 ( see remark [ rem:2.6 ] below ) . let @xmath20 and @xmath82 be the driving noise and innovation processes , respectively , described in section [ sec:1 ] . we define an @xmath0-valued deterministic function @xmath83 by @xmath84 , \quad t\ge 0 , \label{eq:2.2}\ ] ] where @xmath85 . for the kernels @xmath52 s in , we put @xmath86 we denote by @xmath87 the @xmath88-valued process @xmath89 , i.e. , @xmath90 by , , , and , the wealth process @xmath64 evolves according to @xmath91dt + \pi'(t)\sigma(t)db(t),\quad t\ge 0,\ ] ] whence , by the it formula , we have , for @xmath19 , @xmath92 . \end{split } \label{eq:2.4}\ ] ] we define an @xmath49-valued process @xmath93 by @xmath94 , \quad t\ge 0.\ ] ] since @xmath95 is a continuous gaussian process , the process @xmath93 is a @xmath39-martingale ( see , e.g. , example 3(a ) in liptser and shiryayev ( * ? ? ? * section 6.2 ) ) . we define the @xmath49-valued process @xmath96 by @xmath97 , \quad 0\le t\le t,\ ] ] where @xmath98 is the conjugate exponent of @xmath99 , i.e. , @xmath100 notice that @xmath101 ( resp . @xmath102 ) if @xmath103 ( resp . @xmath104 ) . in view of theorem 7.6 in karatzas and shreve ( * chapter 3 ) , to solve problem p1 , we only have to derive a stochastic integral representation for @xmath105 . we define an @xmath49-valued @xmath39-martingale @xmath106 by @xmath107 , \quad t\ge 0.\ ] ] then , by bayes rule , we have @xmath108 \\ & = k(t)\bar{e}\left[\left.\exp\left\{-\frac{1}{2}\beta(1-\beta ) \int_0^t \vert\lambda(s)-\xi(s)\vert^2ds\right\}\right| \mathcal{f}_t\right ] \end{split}\ ] ] for @xmath109 $ ] , where @xmath110 stands for the expectation with respect to the probability measure @xmath111 on @xmath112 such that @xmath113 . thus @xmath114 . \end{split } \label{eq:2.5}\ ] ] we are to apply theorem [ thm : a1 ] in appendix a to . by , the dynamics of @xmath115 are described by the @xmath1-dimensional stochastic differential equation @xmath116 where @xmath117 , @xmath118 , and @xmath119 with @xmath120 s as in ( [ eq:1.10 ] ) . write @xmath121ds$ ] for @xmath122 $ ] . then @xmath123 is an @xmath0-valued standard brownian motion under @xmath111 . by , the process @xmath115 evolves according to @xmath124dt+l(t)d\bar{b}(t ) , \quad t\ge 0 , \label{eq:2.7}\ ] ] where @xmath125 , @xmath126 with @xmath127 by theorem [ thm : a1 ] in appendix a , we are led to consider the following one - dimensional backward riccati equations : for @xmath33 @xmath128 the following lemma , especially ( iii ) , is crucial in our arguments . [ lem:2.1]let @xmath129 . 1 . if @xmath34 , then has a unique solution @xmath130 . 2 . if @xmath103 , then has a unique nonnegative solution @xmath131 . 3 . if @xmath132 and @xmath104 , then ( [ eq:2.10 ] ) has a unique solution @xmath130 such that @xmath133 for @xmath109 $ ] . ( i ) if @xmath34 , then ( [ eq:2.10 ] ) is linear , whence it has a unique solution . ( ii ) if @xmath103 , then @xmath134 , so that , by the well - known result on riccati equations ( see , e.g. , fleming and rishel ( * ? ? ? * theorem 5.2 ) and liptser and shiryayev ( * ? ? ? * theorem 10.2 ) ) , ( [ eq:2.10 ] ) has a unique nonnegative solution . ( iii ) when @xmath132 and @xmath104 , write @xmath135 then the equation for @xmath136 $ ] becomes @xmath137 where @xmath138.\ ] ] since @xmath139 and @xmath140 , we see that @xmath141 = \frac{2(p_j+q_j)-\beta l_j(t)}{l_j(t)^3}\cdot\frac{dl_j}{dt}(t)>0.\ ] ] we write @xmath142 as @xmath143 + [ ( p_j+q_j)-l_j(t)]^2,\ ] ] which is positive since @xmath144 . thus @xmath145 , so that ( [ eq:2.12 ] ) has a unique nonnegative solution @xmath146 . the desired solution to ( [ eq:2.10 ] ) is given by @xmath147 $ ] . in what follows , we write @xmath130 for the unique solution to ( [ eq:2.10 ] ) in the sense of lemma [ lem:2.1 ] . then @xmath148 satisfies the backward matrix riccati equation @xmath149 where @xmath150 denotes the @xmath17 unit matrix . for @xmath33 , let @xmath151 be the solution to the following one - dimensional linear equation : @xmath152v_j(t)+ \beta(1-\beta)\lambda_j(t)-r_j(t;t)\rho_j(t)=0,\\ & \quad0\le t\le t,\quad v_j(t)=0 . \end{split } \label{eq:2.14}\ ] ] then @xmath153 satisfies the matrix equation @xmath154v(t)+ \beta(1-\beta)\lambda(t)-r(t;t)\rho(t)=0,\\ & \quad0\le t\le t,\quad v(t)=0 . \end{split } \label{eq:2.15}\ ] ] we put , for @xmath33 and @xmath155 , @xmath156 where @xmath157 we are now ready to give the desired representation for @xmath105 . [ prop:2.2 ] write @xmath158 , \quad 0\le t\le t. \label{eq:2.18}\ ] ] then , for @xmath109 $ ] , we have @xmath159 with @xmath160 . \label{eq:2.19}\ ] ] it follows from , , , and theorem [ thm : a1 ] that @xmath161 . \label{eq:2.20}\ ] ] the equality ( [ eq:2.19 ] ) follows from this . a straightforward calculation based on , and the it formula gives @xmath162 , where @xmath163 is as in . thus the proposition follows . recall that we have assumed @xmath72 and . here is the solution to problem p1 . [ thm:2.3 ] for @xmath68 , the strategy @xmath164 defined by @xmath165 \label{eq:2.21}\ ] ] is the unique optimal strategy for problem p1 . the value function @xmath166 in ( p1 ) is given by @xmath167^{\alpha } \exp\left[\frac{(1-\alpha)}{2 } \sum\nolimits_{j=1}^{n}\int_0^t g_j(t;t)dt\right ] . \label{eq:2.22}\ ] ] by theorem 7.6 in karatzas and shreve ( * ? ? ? * chapter 3 ) , the unique optimal strategy @xmath168 for problem p1 is given by @xmath169 , \quad 0\le t\le t,\ ] ] which , by ( [ eq:2.18 ] ) , is equal to @xmath170 . thus the first assertion follows . by the same theorem in @xcite , @xmath171^{\alpha}\gamma^{1-\alpha}(0)$ ] . this and ( [ eq:2.19 ] ) give ( [ eq:2.22 ] ) . [ rem:2.4 ] we can regard @xmath172 , which is the only random term on the right - hand side of , as representing the memory effect . to illustrate this point , suppose that @xmath173 is a constant matrix . then , by in appendix c , we can express @xmath44 , whence @xmath115 , in terms of the past prices @xmath174 , @xmath175 $ ] , of the risky assets . [ rem:2.5 ] from ( * ? ? ? * theorem 7.6 ) , we also find that @xmath176 [ rem:2.6 ] regarding , we assume this to ensure the existence of solution to for @xmath33 . under the weaker assumption , we could show by a different argument that , for @xmath33 , has a solution if @xmath177 , where @xmath178 $ ] is defined by @xmath179 from this , we see that the same result as theorem [ thm:2.3 ] holds under if @xmath180 , @xmath181 , where @xmath182 . however , we did not succeed in extending the result to the most general case @xmath183 , @xmath181 . such an extension , if possible , would lead us to the solution of problem p3 under ( see remark [ rem:3.8 ] ) . in this section , we consider the infinite horizon optimization problem p2 for the financial market model @xmath2 . throughout this section , we assume and the following two conditions : @xmath184 here recall @xmath83 from ( [ eq:2.2 ] ) . in the main result of this section ( theorem [ thm:3.4 ] ) , we will also assume @xmath185 , @xmath181 , where @xmath186 with @xmath187 notice that @xmath188 . to give the solution to problem p2 , we take the following steps : 1 . for the value function @xmath166 in ( p1 ) , we calculate the following limit explicitly : @xmath189 . \label{eq:3.5}\ ] ] 2 . for @xmath190 in below , we calculate the growth rate @xmath191 , \label{eq:3.6}\ ] ] and verify that @xmath192 . 3 . since the definition of @xmath193 implies @xmath194 \le \tilde{j}(\alpha),\quad \forall\pi\in\mathcal{a } , \label{eq:3.7}\ ] ] we conclude that @xmath195 is an optimal strategy for problem p2 and that the optimal growth rate @xmath77 in ( p2 ) is given by @xmath196 . let @xmath72 and @xmath98 be its conjugate exponent as in section [ sec:2 ] . for @xmath33 , recall @xmath197 from ( [ eq:2.9 ] ) . we have @xmath198 , where @xmath199 notice that @xmath200 . we consider the equation @xmath201 when @xmath34 , we write @xmath202 for the unique solution @xmath203 of this linear equation . if @xmath132 , then @xmath204+q_j^2\ge q_j^2>0,\ ] ] so that we may write @xmath202 for the larger solution to the quadratic equation let @xmath205 then @xmath206 . as in section [ sec:2 ] , we write @xmath130 for the unique solution to ( [ eq:2.10 ] ) in the sense of lemma [ lem:2.1 ] . recall @xmath207 from ( [ eq:2.17 ] ) . the next proposition provides the necessary results on the asymptotic behavior of @xmath208 . [ prop:3.1 ] let @xmath183 , @xmath181 , and @xmath129 . then 1 . @xmath208 is bounded in @xmath207 . 2 . @xmath209 . 3 . for @xmath210 such that @xmath211 , @xmath212 if @xmath34 , then @xmath213 and @xmath214 for @xmath19 , so that the assertions follow from theorem [ thm : b3 ] in appendix b. we assume @xmath132 . since @xmath215 the function @xmath120 converges to @xmath29 exponentially fast as @xmath216 . hence the coefficients of the equation ( [ eq:2.10 ] ) converge to their counterparts in ( [ eq:3.8 ] ) exponentially fast , too . if @xmath103 , then the desired assertions follow from theorem [ thm : b1 ] in appendix b ( due to nagai and peng @xcite ) . suppose @xmath104 . let @xmath217 , @xmath218 and @xmath219 be as in ( [ eq:2.11 ] ) . since @xmath220 and @xmath221 is bounded from below in @xmath207 , so is @xmath208 . to show that @xmath208 is bounded from above in @xmath207 , we consider the solution @xmath222 to the linear equation @xmath223m_j(t)+a_3+a_1(t)\bar{r}^2_j=0 , \quad 0\le t\le t,\quad m_j(t)=0.\ ] ] since @xmath224 and @xmath225 + 2[a_2(t)-\bar{r}_ja_1(t)][m_j(t)-r_j(t ) ] = -a_1(t)\left[r_j(t)-\bar r_j\right]^2\le 0,\ ] ] we have @xmath226 in @xmath207 . however , @xmath227 as @xmath216 , so that @xmath228 is bounded from above in @xmath207 , whence so is @xmath208 . the desired assertions now follow from theorem [ thm : b2 ] in appendix b. let @xmath129 . for @xmath229 in ( [ eq:2.8 ] ) , we have @xmath230 , where @xmath231 let @xmath151 be the solution to as in section [ sec:2 ] . define @xmath232 by @xmath233 [ prop:3.2 ] let @xmath183 , @xmath181 , and @xmath129 . then 1 . @xmath234 is bounded in @xmath207 . 2 . @xmath235 . 3 . for @xmath210 such that @xmath211 , @xmath236 the coefficients of ( [ eq:2.14 ] ) converge to their counterparts in ( [ eq:3.9 ] ) . also , @xmath237 = \bar{b}_j - p_j^2\bar{r}_j =- k_j<0.\ ] ] thus the proposition follows from theorem [ thm : b3 ] in appendix b. for @xmath33 and @xmath183 , @xmath181 , we put @xmath238 } , \label{eq:3.10}\\ \begin{split } & g_j(\alpha)\\ & \ \ : = ( p_j+q_j)-q_j\alpha-(1-\alpha)^{1/2 } \left[(1-\alpha)(p_j+q_j)^2+\alpha p_j(p_j+2q_j)\right]^{1/2}. \end{split } \label{eq:3.11}\end{gathered}\ ] ] recall the value function @xmath166 from ( p1 ) and its representation ( [ eq:2.22 ] ) . in the next proposition , we compute @xmath239 in . [ prop:3.3 ] let @xmath183 , @xmath181 . then the limit @xmath239 in exists and is given by @xmath240 where @xmath241 more explicitly , @xmath242 recall @xmath243 from ( [ eq:2.16 ] ) . by propositions [ prop:3.1 ] and [ prop:3.2 ] , @xmath244 from this and ( [ eq:2.22 ] ) , @xmath245 & = \frac{\log x}{t } + \frac{1}{t}\int_0^t r(t)dt + \frac{1-\alpha}{2\alpha}\sum_{j=1}^n\frac{1}{t}\int_0^t g_j(t;t)dt\\ & \to \bar{r}+\frac{1-\alpha}{2\alpha}\sum_{j=1}^{n}\bar{g}_j \quad \mbox{as $ t\to\infty$ , } \end{split}\ ] ] which implies . we have @xmath246 . also , @xmath247 and @xmath248 . thus @xmath249\\ & = \frac{\beta\bar{\lambda}_j^2}{k_j^2 } \left[\{\beta p_j^2(p_j\bar{r}_j)^2 - 2\beta p_j^2\bar{r}_jk_j\ } + 2\beta(1-\beta)p_j(p_j^2\bar{r}_j - k_j)\right.\\ & \left.\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\quad + \beta p_j^2(1-\beta)^2-(1-\beta)k_j^2\right]\\ & = \frac{\beta(1-\beta)\bar{\lambda}_j^2}{k_j^2 } \left[-\beta^2p_j^2 + 2\beta p_j\bar{b}_j+\beta(1-\beta)p_j^2 -\{\bar{b}_j^2+\beta(1-\beta)p_j^2\}\right]\\ & = -\frac{\beta(1-\beta)\bar{\lambda}_j^2}{k_j^2}[\bar{b}_j-\beta p_j]^2 = \frac{\alpha}{(1-\alpha)^2}\frac{(p_j+q_j)^2}{k_j^2}\bar{\lambda}_j^2 . \end{split}\ ] ] this and @xmath250 imply @xmath251 since @xmath252 , it follows that @xmath253=-p_j+(\alpha-1)q_j = q_j\alpha-(p_j+q_j).\ ] ] also , @xmath254 combining , we obtain . recall @xmath115 from . taking into account ( [ eq:2.21 ] ) , we consider @xmath255 defined by @xmath256 , \quad t\ge 0 , \label{eq:3.14}\ ] ] where @xmath117 as in section [ sec:2 ] , and @xmath257 , @xmath258 . recall that we have assumed , and . recall also @xmath259 from with . here is the solution to problem p2 . [ thm:3.4 ] let @xmath185 , @xmath181 . then @xmath195 is an optimal strategy for problem p2 with limit in . the optimal growth rate @xmath77 in ( p2 ) is given by @xmath260 where @xmath261 s and @xmath262 s are as in and , respectively . for simplicity , we put @xmath263 . for @xmath239 in and @xmath264 in , we claim @xmath192 , that is , @xmath265=\tilde{j}(\alpha ) . \label{eq:3.16}\ ] ] as mentioned before , and imply that @xmath195 is an optimizer for problem p2 . the equality follows from this and we complete the proof of the theorem by proving . _ step _ 1 . we calculate @xmath266 $ ] . define the @xmath49-valued martingale @xmath267 by @xmath268 , \quad t\ge 0.\ ] ] from ( [ eq:2.4 ] ) , we have @xmath269^{\alpha}l(t)\exp[\int_{0}^{t}n(s)ds]$ ] for @xmath19 , where @xmath270\\ & = \frac{\alpha(1-\alpha)}{2}\{\sigma'(t)\hat{\pi}(t)\ } ' \left [ ( 1-\beta)\{\lambda(t)-\xi(t)\}+p\bar{r}\xi(t)-p\bar{v } \right]\\ & = -\frac{\beta}{2 } \left[\lambda(t)-\xi(t)-(1-\alpha)\{p\bar{r}\xi(t)-p\bar{v}\}\right]'\\ & \qquad\qquad\qquad\qquad \cdot \left[\lambda(t)-\xi(t)+(1-\alpha)\{p\bar{r}\xi(t)-p\bar{v}\}\right]\\ & = -\frac{\beta}{2 } \left [ \{\lambda(t)-\xi(t)\}'\{\lambda(t)-\xi(t)\ } -(1-\alpha)^2\{p\bar{r}\xi(t)-p\bar{v}\}'\{p\bar{r}\xi(t)-p\bar{v}\ } \right ] . \end{split}\ ] ] notice that we have used @xmath252 , @xmath271 . we write @xmath272 where @xmath273,\quad t\ge 0,\quad j=1,\dots , n,\ ] ] and @xmath274 , @xmath275 with @xmath276,\quad t\ge 0,\quad j=1,\dots , n,\\ & h_j(t):=\alpha(\alpha-1)p_j^2\bar{r}_j\bar{v}_j-\beta\lambda_j(t ) , \quad t\ge 0,\quad j=1,\dots , n.\end{aligned}\ ] ] therefore , @xmath277 & = [ xs_0(t)]^{\alpha } \exp\left[\frac{1}{2}\sum\nolimits_{j=1}^n \int_0^t u_j(t)dt\right]\\ & \quad\times\bar{e}\left[\exp\left\{-\int_0^t \left(\frac{1}{2}\xi'(t)q\xi(t)+h'(t)\xi(t)\right)dt \right\}\right ] , \end{split } \label{eq:3.17}\ ] ] where @xmath110 denotes the expectation with respect to the probability measure @xmath111 on @xmath112 such that @xmath278 . _ step _ 2 . we continue the calculation of @xmath266 $ ] . we are about to apply theorem [ thm : a1 ] in appendix a to . write @xmath279 for @xmath19 . then @xmath123 is an @xmath0-valued standard brownian motion under @xmath111 . by , the process @xmath115 evolves according to the @xmath1-dimensional stochastic differential equation @xmath280dt + l(t)d\bar{b}(t),\quad t\ge 0 , \label{eq:3.18}\ ] ] where @xmath281 , @xmath282 with @xmath283 for @xmath33 , let @xmath284 be the unique solution to the one - dimensional backward riccati equation @xmath285 in the sense of lemma [ lem:3.5 ] below , and let @xmath286 be the solution to the one - dimensional linear equation @xmath287m_j(t ) -h_j(t)-u_j(t;t)\gamma_j(t)=0\\ & \quad 0\le t\le t , \quad m_j(t)=0 . \end{split } \label{eq:3.20}\ ] ] then , from and theorem [ thm : a1 ] , we obtain @xmath288 = [ xs_0(t)]^{\alpha } \exp\left[\frac{1}{2}\sum\nolimits_{j=1}^{n}\int_0^t f_j(t;t)dt\right ] , \label{eq:3.21}\ ] ] where , for @xmath289 and @xmath33 , @xmath290 _ step _ 3 . we compute the limit @xmath264 in . let @xmath129 . write @xmath291 then @xmath292 converges to @xmath293 , as @xmath216 , exponentially fast . now @xmath294 \\ & = \bar{b}_j^2 -2\alpha\bar{b}_j(\bar{b}_j+k_j)+\alpha^2(\bar{b}_j+k_j)^2 + p_j^2\beta-\alpha(\alpha-1)(\bar{b}_j+k_j)^2 \\ & = ( 1-\alpha)\bar{b}_j^2 + \alpha k_j^2 + p_j^2\beta = \bar{b}_j^2 + p_j^2\beta(1-\beta),\end{aligned}\ ] ] which implies @xmath295 thus we may write @xmath296 for the larger ( resp . unique ) solution of the following equation when @xmath132 ( resp . @xmath34 ) : @xmath297 from ( [ eq:3.22 ] ) , we also see that @xmath298 . let @xmath299 be the solution to @xmath300 where @xmath301 by , we have @xmath302 = \frac{\log x}{t}+\frac{1}{t}\int_0^tr(t)dt + \frac{1}{2\alpha}\sum_{j=1}^n\frac{1}{t}\int_0^t f_j(t;t)dt.\ ] ] however , propositions [ prop:3.6 ] and [ prop:3.7 ] below imply @xmath303 with @xmath304,\ ] ] so that @xmath305 _ step _ 4 . here we show that in fact holds . first , @xmath306 whence @xmath307 ( which we can directly check when @xmath34 ) . next , @xmath308\\ & = \bar{\lambda}_j(-\beta + \alpha p_j\bar{r}_j ) = -(1-\alpha)\beta \bar{\lambda}_j[(1-\beta)+p_j\bar{r}_j ] , \end{aligned}\ ] ] so that @xmath309 = ( 1-\alpha)\bar{v}_j.\ ] ] therefore , @xmath310\\ & = ( 1-\alpha ) \left [ p_j^2\bar{v}^2_j+2\bar{\rho}_j\bar{v}_j - p_j^2\bar{r}_j -\beta(1-\beta)\bar{\lambda}_j^2 \right]=(1-\alpha)\bar{g}_j . \end{split}\ ] ] from , and this , we obtain @xmath192 or ( [ eq:3.16 ] ) , as desired . in the proof above , we needed the following results . [ lem:3.5]let @xmath129 . 1 . if @xmath34 , then has a unique solution @xmath284 . 2 . if @xmath132 and @xmath311 , then has a unique nonnegative solution @xmath284 . 3 . if @xmath132 and @xmath104 , then has a unique solution @xmath284 such that @xmath312 for @xmath109 $ ] , where @xmath130 is the solution to in the sense of lemma [ lem:2.1 ] ( iii ) . ( i ) when @xmath34 , is linear , whence it has a unique solution . ( ii ) for @xmath132 and @xmath313 , we put @xmath314 . since @xmath200 and @xmath134 , the larger solution @xmath202 to @xmath315 satisfies @xmath316 if and only if @xmath317 . however , this is equivalent to @xmath318 . thus , if @xmath132 and @xmath311 , then @xmath319 or @xmath320 , so that the riccati equation has a unique nonnegative solution . ( iii ) suppose @xmath132 and @xmath104 . for the solution @xmath130 to in the sense of lemma [ lem:2.1 ] ( iii ) , we consider @xmath321 let @xmath292 be as above . then , becomes @xmath322p_j(t)\\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad + \alpha[l_j(t)r_j(t)-p_j\bar{r}_j]^2=0,\quad 0\le t\le t , \end{split}\ ] ] with @xmath323 . since @xmath324 and @xmath325 ^ 2>0 $ ] , this riccati equation has a unique nonnegative solution . thus the assertion follows . [ prop:3.6]let @xmath185 , @xmath181 , and @xmath129 . let @xmath326 be the unique solution to in the sense of lemma [ lem:3.5 ] , and let @xmath296 be the larger ( resp . unique ) solution to when @xmath132 ( resp . . then 1 . @xmath326 is bounded in @xmath207 . 2 . @xmath327 . 3 . for @xmath210 such that @xmath328 , @xmath329 we assume @xmath104 and @xmath132 . since @xmath312 in @xmath207 and @xmath208 is bounded from below by proposition [ prop:3.1 ] , so is @xmath326 . let @xmath330 be the solution to the linear equation @xmath331n_j(t)+q_j+l_j^2(t)\bar{u}_j^2=0 , \quad 0\le t\le t,\quad n_j(t)=0.\ ] ] by , @xmath332 as @xmath216 , so that @xmath333 is bounded from above in @xmath207 . since @xmath334 and @xmath335 + 2[d_j(t)-l_j^2(t)\bar{u}_j][n_j(t)-u_j(t ) ] = -l_j^2(t)\left[u_j(t)-\bar u_j\right]^2\le 0,\ ] ] we have , as in the proof of proposition [ prop:3.1 ] , @xmath336 in @xmath207 . thus @xmath326 is also bounded from above in @xmath207 . combining , @xmath326 is bounded in @xmath207 . the rest of the proof is similar to that of proposition [ prop:3.1 ] , whence we omit it . [ prop:3.7 ] let @xmath185 , @xmath181 , and @xmath129 . let @xmath337 and @xmath299 be the solutions to and , respectively . then 1 . @xmath337 is bounded in @xmath207 . 2 . @xmath338 . 3 . for @xmath210 such that @xmath328 , @xmath339 the proof of proposition [ prop:3.7 ] is similar to that of proposition [ prop:3.2 ] ; so we omit it . [ rem:3.8 ] we note that the proof of lemma [ lem:3.5 ] ( iii ) is still valid under if there were a solution @xmath130 to . this implies that , to prove an analogue of theorem [ thm:3.4 ] with @xmath104 , which is relevant to problem p3 , for , one may show the existence of such @xmath340 when @xmath104 . we did not succeed in such an extension to lemma [ lem:2.1 ] ( iii ) ( see remark [ eq:2.6 ] ) . in this section , we study the large deviations probability control problem p3 for the market model @xmath2 . throughout this section , we assume ( [ eq:2.1 ] ) , ( [ eq:3.1 ] ) , ( [ eq:3.2 ] ) and @xmath341 for @xmath59 and @xmath342 , let @xmath343 be the growth rate defined by @xmath344 we have @xmath345 . following pham @xcite , we consider the optimal logarithmic moment generating function @xmath346,\quad 0<\alpha<1.\ ] ] since @xmath347 for @xmath348 , it follows from theorem [ thm:3.4 ] that @xmath349 where @xmath261 s and @xmath262 s are as in and , respectively . [ prop:4.1 ] we have @xmath350 and @xmath351 , where @xmath352 for @xmath104 , @xmath353 is equal to @xmath354 } + \frac{(p_j+q_j)^2\bar{\lambda}_j^2q_j^2\alpha } { \left[(1-\alpha)(p_j+q_j)^2+\alpha p_j(p_j+2q_j)\right]^2}.\ ] ] from this , @xmath355 . this also shows that @xmath356 if @xmath34 and @xmath357 . on the other hand , for @xmath104 , @xmath358^{1/2}\\ & \qquad\qquad\qquad\qquad + \frac{q_j^2(1-\alpha)^{1/2}}{2\left[(1-\alpha)(p_j+q_j)^2 + \alpha p_j(p_j+2q_j)\right]^{1/2}}. \end{split}\ ] ] this gives @xmath359 $ ] . this also yields @xmath360 if @xmath132 . thus the proposition follows . [ rem:4.2 ] from the proof of proposition [ prop:4.1 ] , we see that @xmath361 if @xmath132 for all @xmath33 , otherwise @xmath362 for @xmath348 , we denote by @xmath363 the optimal strategy @xmath364 in ( [ eq:3.14 ] ) . recall @xmath365 from ( p3 ) . from theorem [ thm:3.4 ] , proposition [ prop:4.1 ] , and pham ( * ? ? ? * theorem 3.1 ) , we immediately obtain the following solution to problem p3 : [ thm:4.3 ] we have @xmath366 , \quad c\in\mathbf{r}.\ ] ] moreover , if @xmath367 is such that @xmath368 , then , for @xmath369 , the sequence of strategies @xmath370 is nearly optimal in the sense that @xmath371 [ rem:4.4 ] theorem 3.1 in pham @xcite is stated for a model different from @xmath2 but the arguments there are so general that we can prove theorem [ thm:4.3 ] in the same way . we turn to the problem of deriving an optimal strategy , rather than a nearly optimal sequence , for the problem ( p3 ) when @xmath78 . we define @xmath372 by @xmath373 , \quad t\ge 0,\ ] ] where recall @xmath115 from . from ( [ eq:2.4 ] ) , @xmath374'db(t ) . \end{split}\ ] ] [ prop:4.5 ] the rate @xmath375 converges to @xmath79 , as @xmath376 , in probability . in this proof , we denote by @xmath377 positive constants , which may not be necessarily equal . for @xmath33 , we write @xmath378+[\bar{\lambda}_j - p_jk(t)]+n(t),\ ] ] where @xmath379 and @xmath380 with @xmath381 the process @xmath106 , the dynamics of which are given by @xmath382 is a positively recurrent one - dimensional diffusion process with speed measure @xmath383 . by the ergodic theorem ( cf . rogers and williams ( * ? ? ? * v.53 ) ) , we have @xmath384 ^ 2dt = \int_{-\infty}^{\infty}(\bar{\lambda}_j - p_jy)^2\nu(dy ) = \bar{\lambda}_j^2+\frac{p_j^2}{2(p_j+q_j)}\quad \mbox{a.s.,}\ ] ] where @xmath385 is the gaussian measure with mean @xmath386 and variance @xmath387 $ ] . since @xmath388 , we have @xmath389\le c\int_0^te^{-2q_j(t+s)}ds \le ce^{-2q_jt } , \quad t\ge 0.\ ] ] also , @xmath390\le c$ ] for @xmath19 . therefore , @xmath391 dt \le \frac{c}{t}\int_0^te[n^2(t)]^{1/2}dt \to 0,\quad t\to \infty.\ ] ] similarly , @xmath392 ^ 2dt = \lim_{t\to\infty}\frac{1}{t}\int_0^te\left[(\lambda_j(t)-\bar{\lambda}_j ) ( \bar{\lambda}_j - p_jk(t))\right]dt\\ & \quad=\lim_{t\to\infty}\frac{1}{t}\int_0^te\left[n^2(t)\right]dt = \lim_{t\to\infty}\frac{1}{t}\int_0^te\left[(\lambda_j(t)-\bar{\lambda}_j)n(t ) \right]dt = 0 . \end{split}\ ] ] combining , @xmath393 ^ 2dt\to \bar{\lambda}_j^2+\frac{p_j^2}{2(p_j+q_j ) } , \quad t\to\infty,\quad \mbox{\rm in probability.}\ ] ] finally , for @xmath33 and @xmath19 , @xmath394 \le 2\lambda^2_j(t)+2e\left[\xi_j^2(t)\right ] \le c\left[1+\int_0^tl^2_j(t , s)ds\right]\le c,\ ] ] so that @xmath395db_j(t)\to 0 $ ] , as @xmath376 , in @xmath396 , whence in probability . thus the proposition follows . [ thm:4.6 ] for @xmath78 , @xmath397 is optimal for problem p3 with limit @xmath398 proposition [ prop:4.5 ] implies @xmath399 for @xmath78 , so that @xmath400 thus @xmath397 is optimal if @xmath78 . [ rem:4.7 ] from theorem 10.1 in karatzas and shreve ( * ? ? ? * chapter 3 ) , we see that @xmath397 is the log - optimal or growth optimal strategy in the sense that @xmath401 we also find that @xmath402 a.s . for @xmath19 . in this appendix , we prove a generalization of the cameron martin formula that we need in the proofs of proposition [ prop:2.2 ] and theorem [ thm:3.4 ] . we refer to myers @xcite for earlier work . let @xmath68 and let @xmath403 be the set of @xmath17 real matrices . we say that @xmath404\to\mathbf{r}^{n\times n}$ ] is symmetric if @xmath405 is a symmetric matrix for all @xmath109 $ ] . let @xmath26 be the underlying complete probability space equipped with filtration @xmath406 satisfying the usual conditions . we assume that the @xmath0-valued process @xmath115 satisfies the @xmath1-dimensional stochastic differential equation @xmath407dt + c(t)db(t),\quad 0\le t\le t,\ ] ] where @xmath408 is an @xmath0-valued standard @xmath43-brownian motion and all the coefficients @xmath409\to\mathbf{r}^n$ ] and @xmath410\to\mathbf{r}^{n\times n}$ ] are deterministic , bounded measurable functions . [ thm : a1 ] let @xmath411\to\mathbf{r}^{n\times n}$ ] and @xmath412\to\mathbf{r}^{n}$ ] be deterministic , bounded measurable functions . we assume that @xmath413 is symmetric . we also assume that there exists a bounded symmetric function @xmath414\to\mathbf{r}^{n\times n}$ ] satisfying the backward matrix riccati equation @xmath415 let @xmath416\to\mathbf{r}^n$ ] be the solution to the linear equation @xmath417'v(t)-h(t)-r(t)a(t)=0 , \quad 0\le t\le t,\\ & v(t)=0 . \end{split } \label{eq : a.2}\ ] ] then , for @xmath109 $ ] , @xmath418 \\ & = \exp\left[v'(t)\xi(t)-\frac{1}{2}\xi'(t)r(t)\xi(t)\right]\\ & \qquad\times\exp\left[\frac{1}{2}\int_t^t\{v'(s)c(s)c'(s)v(s)+2a'(s)v(s ) -\mathrm{tr}(c(s)c'(s)r(s))\}ds\right ] . \end{aligned}\ ] ] we put @xmath419 $ ] for @xmath122 $ ] . then , by the it formula , @xmath420 \ -\left[k'(t)db(t ) - \frac{1}{2}\vert k(t)\vert^2dt\right ] \\ & \qquad = \left[-\frac{1}{2}\xi'(t)\{\dot{r}(t)+r(t)b(t)+b'(t)r(t ) -r(t)c(t)c'(t)r(t)\}\xi(t ) \right.\\ & \qquad\qquad\quad\ \ + \{\dot{v}(t)+(b(t)-c(t)c'(t)r(t))'v(t)-r(t)a(t)\ } ' \xi(t ) \\ & \qquad\qquad\qquad\quad\ \ + \left.\frac{1}{2}\{v'(t)c(t)c'(t)v(t)+2a'(t)v(t ) -\mathrm{tr}(c(t)c'(t)r(t))\}\right]dt\\ & \qquad = \left[\frac{1}{2}\xi'(t)q(t)\xi(t)+h'(t)\xi(t)\right.\\ & \qquad\qquad\qquad\qquad \left.+ \frac{1}{2}\{v'(t)c(t)c'(t)v(t)+2a'(t)v(t ) -\mathrm{tr}(c(t)c'(t)r(t))\ } \right]dt.\end{aligned}\ ] ] therefore , @xmath421 is equal to @xmath422 since @xmath106 is a continuous gaussian process , the process @xmath423 is a martingale ( cf . example 3(a ) in ( * ? ? ? * section 6.2 ) ) . thus @xmath424=1.\ ] ] combining , we obtain the theorem . here we summarize the results on the asymptotics for a solution to riccati or linear equation that we need in section [ sec:3 ] . for @xmath68 , we consider the one - dimensional backward riccati equation @xmath425 where @xmath426 by ( [ eq : b5 ] ) , we may write @xmath427 for the larger solution to the quadratic equation @xmath428 recall @xmath207 from ( [ eq:2.17 ] ) . [ thm : b1 ] we further assume @xmath429 then , for @xmath68 , ( [ eq : b1 ] ) has a unique nonnegative solution @xmath430 , and it satisfies the following : 1 . @xmath431 is bounded in @xmath207 . 2 . @xmath432 . 3 . for @xmath210 such that @xmath328 , @xmath433 when the condition ( [ eq : b6 ] ) is lacking , we have the following : [ thm : b2 ] we assume that , for every @xmath68 , the equation ( [ eq : b1 ] ) has a solution @xmath430 that is bounded in @xmath207 . then ( i)(iii ) in theorem [ thm : b1 ] hold . the proof of theorem [ thm : b2 ] is almost the same as that of theorem [ thm : b1 ] in @xcite , whence we omit it . we turn to the one - dimensional backward linear differential equation @xmath434 where @xmath435\to \mathbf{r})$ for $ t\in ( 0,\infty)$ and $ i=1,2 $ , } \label{eq : b8}\\ \lim_{t - t\to\infty,\ t\to\infty}b_i(t;t)=\bar{b}_i\ \ \mbox{for}\ \ i=1,2 , \label{eq : b9}\\ \bar{b}_1>0 . \label{eq : b10}\end{gathered}\ ] ] [ thm : b3 ] for @xmath68 , write @xmath436 for the solution to ( [ eq : b7 ] ) . let @xmath437 be the solution of the linear equation @xmath438 . then 1 . @xmath439 is bounded in @xmath207 . 2 . @xmath440 . 3 . for @xmath210 such that @xmath328 , @xmath441 in this appendix , we use the special case of our model @xmath2 in which @xmath14 s are constants , i.e. , @xmath442 we explain how we can statistically estimate the parameters @xmath58 , @xmath443 and @xmath444 from stock price data . this problem , for the univariate case @xmath36 , is discussed in @xcite . here we are interested in the multivariate case @xmath445 . as for the expected rates of return @xmath446 , there is as usual a structural difficulty in the statistical estimation of them ( cf . luenberger ( * ? ? ? * chapter 8) ) , whence we do not discuss it here . from ( [ eq:1.5 ] ) , we see that @xmath447/t = f(t;p_j , q_j),\quad t>0,\quad j=1,\dots , n , \label{eq : c1}\ ] ] where @xmath448 ( cf . @xcite , examples 4.3 and 4.5 ) . notice that @xmath449 . from ( [ eq:1.6 ] ) or ( [ eq:1.7 ] ) and the it formula , the solution @xmath6 to ( [ eq:1.1 ] ) is given by @xmath450 , \quad t\ge 0 , \label{eq : c2}\ ] ] for @xmath61 . since @xmath44 has stationary increments , we may define @xmath451 where cov@xmath452 denotes the covariance with respect to the physical probability measure @xmath39 . by ( [ eq : c1 ] ) and ( [ eq : c2 ] ) , we see that @xmath453 suppose that we are given data consisting of closing prices of @xmath1 assets observed at a time interval of @xmath454 consecutive trading days . for @xmath455 and @xmath61 , we denote by @xmath456 the price of the @xmath4th asset on the @xmath457th day . notice that here the time unit is the day . pick @xmath458 , and , for @xmath459 , define @xmath460 by @xmath461 for @xmath462 , we consider the estimator @xmath463 of @xmath464 , where @xmath465 . the number 252 , which is the average number of trading days in one year , converts the return into that per annum , while the number 100 gives the return in percentage . we estimate the values of the parameters @xmath58 , @xmath443 and @xmath444 by nonlinear least squares . more precisely , we search for the values of them such that the following least squares error is minimized : @xmath466 ^ 2.\ ] ] we show numerical results obtained from the following daily stock prices from september 18 , 1995 , through september 16 , 2005 : @xmath467 here we use closing prices adjusted for dividends and splits , which are available at yahoo ! finance @xcite , rather than actually observed closing prices . in this example , we have @xmath468 the estimated values of the parameters are as follows : @xmath469 = \left [ \begin{matrix } 28.7 & -14.1 & 9.1\cr 20.4 & 22.3 & 13.4\cr -1.8 & -4.6 & 24.9 \end{matrix } \right ] , \quad \left [ \begin{matrix } p_1 & q_1 \cr p_2 & q_2 \cr p_3 & q_3 \end{matrix } \right ] = \left [ \begin{matrix } 0.086 & 0.305 \cr 0.261 & 0.044 \cr 0.076 & 0.098 \end{matrix } \right].\ ] ] using the signed square root @xmath470:=\mathrm{sign}(x)\sqrt{\vert x\vert}$ ] , we write @xmath471,\quad d_{ij}(t):=\mathrm{ssr}[v_{ij}(t)],\quad t>0,\quad i , j=1,\dots , n.\ ] ] in figures [ fig : c1][fig : c3 ] , the dotted lines are the graphs of @xmath472 s , while the corresponding solid lines represent those of @xmath473 s that are obtained by using the nonlinear least squares above . we see that the fitted functions @xmath473 simultaneously approximate the corresponding sample values @xmath472 well for this data set . we have repeated this procedure for various data sets and obtained reasonably good fits in most cases . * acknowledgements . * we express our gratitude to jun sekine who suggested the use of a cameron martin type formula in the proof of theorem [ thm:2.3 ] . the work of the second author was partially supported by a research fellowship of the japan society for the promotion of science for young scientists .

we study the phase transitions of interacting bosons at zero temperature between superfluid ( sf ) and supersolid ( ss ) states . the latter are characterized by simultaneous off - diagonal long - range order and broken translational symmetry . the critical phenomena is described by a long - wavelength effective action , derived on symmetry grounds and verified by explicit calculation . we consider two types of supersolid ordering : checkerboard ( x ) and collinear ( c ) , which are the simplest cases arising in two dimensions on a square lattice . we find that the sf css transition is in the three - dimensional xy universality class . the sf xss transition exhibits non - trivial new critical behavior , and appears , within a expansion to be driven generically first order by fluctuations . however , within a one loop calculation directly in a strong coupling fixed point with striking `` non - bose liquid '' behavior is found . at special isolated multi - critical points of particle - hole symmetry , the system falls into the 3d ising universality class . = 10000

cond-mat9609138

because of the dominant role played by transport measurements , conducting to insulating transitions merit special attention in condensed matter systems . in electronic systems , examples include the anderson@xcite and mott@xcitetransitions@xcite . in bosonic systems , interest has focused on the disorder - dominated regime : the so - called superfluid insulator transition@xcite . surprisingly , the somewhat simpler problem of superfluid - insulator transitions in _ clean _ bosonic systems is less well studied . we will see that in such systems , the path from the superfluid to the insulating state generically occurs via an intermediate `` supersolid '' phase . recent work has endeavored to remedy this deficiency through numerical and mean - field studies of disorder free interacting boson models@xcite . such models are experimentally relevant both for helium on a periodic substrate , and for two dimensional josephson junction arrays ( see below ) . in the case of helium with a strong substrate potential , a tight - binding description is appropriate . this leads to the bose - hubbard hamiltonian @xmath2 where @xmath3 and @xmath4 are bosonic creation and annihilation operators obeying @xmath5 = \delta_{ij}$ ] , and @xmath6 . the parameter @xmath7 is a hopping matrix element , @xmath8 is the chemical potential , and @xmath9 , @xmath10 , and @xmath11 are on site , nearest neighbor , and next nearest neighbor interactions , respectively . for a josephson junction array , the relevant parameters are the charge @xmath12 residing on each superconducting island and its phase @xmath13 . including capacitive and josephson couplings leads to the quantum phase hamiltonian @xmath14 where the chemical potential is tuned by the applied gate voltage@xcite . the conjugate variables satisfy the commutation relation @xmath15 = i\delta_{ij}$ ] , and the strength of the josephson phase coupling is characterized by the hopping matrix element @xmath16 . finite range interactions allow for non - trivial charge ordering with fillings of less than one particle per site . if the interactions are very strong , these are simply commensurate crystals . for weak interactions , the ground state is , by contrast , a featureless superfluid ( phase ordered ) state . in the intermediate regime , as discussed below , one may obtain coexisting crystallinity ( charge order ) and superfluidity ( phase or off diagonal long range order ( odlro ) ) . such a state is termed a supersolid . the phase diagrams resulting from the above models for interacting bosons have been determined within mean - field theory @xcite and quantum monte carlo simulations@xcite . a representative ( mean field ) phase diagram is shown in fig . [ phase_diagram ] , for the case of the quantum phase model with @xmath17 and @xmath18 . the mean - field theory gives four different phases : ( 1 ) superfluid ( sf ) , ( 2 ) mott insulator ( mi ) , with homogeneous density , ( 3 ) checkerboard solid ( x ) , or mott - insulating checkerboard charge - density wave , and ( 4 ) compressible checkerboard supersolid ( xss or ss ) . for @xmath19 , `` collinear '' solid ( c ) and supersolid ( css ) phases replace the x and xss states near half the supersolid ( ss ) to crystal ( x ) transition is discussed in refs . it corresponds to the binding of vacancy or interstitial loops , and is expected to be mean - field like ( if not first order ) in two dimensions . here , we are interested in the ss - superfluid ( sf ) transition , which is more analogous to a melting transition . to study it , we will employ the techniques of landau theory and the renormalization group ( rg ) . our results are summarized below . the css sf transition is in the @xmath20dimensional xy universality class , with dynamical scaling exponent @xmath21 . moreover , the critical fluctuations lead to a reduction of the second sound velocity near the transition point . this reduction is , however , finite , so a conventional description of the superfluid properties applies within the critical regime . the xss sf transition is more interesting . at a particular value of the chemical potential in the middle of the xss lobe , the system has asymptotic particle - hole symmetry . at this multi - critical point , the transition is @xmath20dimensional ising like ( @xmath21 ) , and again the superfluid properties are conventional . away from this chemical potential , however , the superfluid and critical properties become _ strongly coupled_. within a rg analysis in @xmath0 one finds a complicated crossover regime leading at low energies to `` runaway flows '' . the absence of a stable fixed point to first order in @xmath22 is often interpreted as signaling a fluctuation driven first order transition . this conclusion must , however , be reached with some caution . cases exist in which one finds runaway rg flows in an @xmath23 expansion , but the physical system ( for @xmath23 of order one ) has a continuous critical point . the most well known of these , the normal superconducting ( ns ) transition in zero field@xcite , is actually quite similar to this problem in that the order parameter is coupled to `` spectating '' gapless ( in the ns case , gauge ) degrees of freedom . to address this question , we apply a fixed dimension rg approach directly in @xmath1 . the issue is inherently a strong - coupling one , outside of the range of applicability of a strict @xmath23-expansion . the fixed - dimension rg , while not controlled in the sense of an @xmath23-expansion , does allow at least an approximate treatment of the features associated with a putative `` strong coupling '' fixed point . in this situation , the superfluid properties are unconventional . this scenario is the bosonic analog of quantum critical modifications of electronic properties recently discussed by millis et al.@xcite . it is thereby natural to denote such anomalous superfluid behavior as `` non bose liquid . '' the non bose liquid phase is characterized , for example , by anomalous dispersion @xmath24 for the strongly - coupled phase and cdw fluctuation modes , with @xmath25 within the fixed - dimension rg approximation . the remainder of the paper is organized as follows . in section ii we give a derivation of a field theory for the order parameter fields on the basis of general symmetry arguments . it is shown that the field theory can be transformed into a `` phonon '' representation , whose form is closely related to field theories describing the influence of elastic degrees of freedom on the critical behavior of magnetic systems . section iii serves to connect the correlation functions of the field theoretic model with physical observables . a short discussion of the mean field phase diagram is given in section iv . the superfluid collinear supersolid and superfluid checkerboard supersolid transition are analyzed in sections v and vi using standard methods of wilson s renormalization group theory . more sophisticated field theoretic techniques are used in section vii to analyze the complete crossover and to calculate effective critical exponents . in the final section we give a short summary of our results and discuss some open problems . the appendices contain additional details and explicit calculations to supplement those in the main text . a hubbard stratonovich derivation of the long - wavelength effective critical action is given in appendix [ decoupling_app ] . appendix [ loop_appendix ] calculates the momentum - dependent @xmath26 terms via an exponential cut - off scheme , needed to justify the calculation in section [ sf_x_transition ] . appendix [ app_pert ] comprises the feynman diagrams and the corresponding analytical expressions used in the field theoretic rg . to study the critical phenomena associated with the transition from a superfluid into a supersolid it is useful to rewrite the original bose - hubbard hamiltonian for lattice bosons in terms of a field theory for the order parameter fields . a formal procedure for deriving such a field theory is given in appendix [ decoupling_app ] . here , we instead use general symmetry arguments to determine the most general possible action near the critical point , and hence describe the universal properties of the phase transition . this procedure is the quantum analog of landau s theory of thermal melting . within this theory , landau originally argued that such transitions must be first order . as pointed out in ref.@xcite , the presence of the periodic substrate invalidates the original argument and continuous melting is in principle allowed . here we study in detail the low order commensurate ss states occurring near 1/2-filling , where such substrate effects are strongest . there are two sorts of states which arise naturally in a model with up to next nearest neighbor interactions . if nearest neighbor interactions dominate , the ground state is a checkerboard ss , while next nearest neighbor interactions favor striped ( or collinear ) sss @xcite . the order parameters distinguishing these states are simply fourier components of the density @xmath27 : @xmath28 here @xmath29 and the @xmath30 are checkerboard and collinear order parameters , respectively . the form of the action is determined by the transformation properties of these fields . the important symmetries are translations by one lattice constant , reflections in a codimension one subspace , and @xmath31 rotations within an arbitrary plane . the transformation properties of @xmath29 and @xmath32 under these operations are summarized in table [ symmetrytable ] . the most general local action consistent with these symmetries is @xmath33 where the coordinates and fields are rescaled such that scale of the spatial gradients is fixed to be @xmath34 . the control parameters @xmath7 and @xmath35 measures the distances from the critical points , and @xmath36 and @xmath37 are `` velocities '' characterizing the crystalline order ( in an incommensurate crystal phase , @xmath36 would be proportional to the bare phonon velocity ) . _ phase fluctuations . _ eq.([critical_action ] ) assumes a _ local _ form of the action . additional non local interactions arise due to interactions with long wavelength fluctuations of the superfluid phase , @xmath38 . as usual , @xmath38 is governed by the action @xmath39 where @xmath40 is the atomic mass , and @xmath41 and @xmath42 are the `` bare '' superfluid density and velocity , respectively . we have taken @xmath43 to be a _ mass _ density , and work in units such that @xmath44 . the most relevant ( near the critical point ) coupling to the spatial order parameters allowed by the time - reversal and @xmath45 symmetries is @xmath46 note that the factors of @xmath47 in eq.([ptcoupling ] ) are dictated by the requirement ( charge conjugation ) that @xmath48 when @xmath49 . _ `` phonon '' representation_. the partition function has an alternative , purely real , representation , which is related to a _ classical _ statistical mechanical system in @xmath20 dimensions . to take advantage of insights gained in this area , we now develop an alternative representation of the phase fluctuations . for this section , we focus for simplicity on the checkerboard order , and choose @xmath50 . performing a hubbard stratonovich transformation then results in the replacements @xmath51 the auxiliary @xmath52-vector field @xmath53 has the physical interpretation of the euclidean super - current . indeed , the original phase field may now be integrated out in favor of the continuity constraint @xmath54 eq.([continuity ] ) is solved by introducing the `` phonon''like field @xmath55 , according to @xmath56 the final form of the action , expressed in terms of @xmath29 and @xmath55 , is then @xmath57 where the coefficients @xmath58 , @xmath59 , and @xmath60 . closely related models have been studied in the context of the influence of elastic degrees of freedom on the behavior of magnetic systems near a critical point @xcite . depending on the number @xmath61 of spin components , anisotropy of the elastic constants and the symmetry of the coupling between the elastic deformations and the order parameter fields one finds different stability scenarios for the various fixed points of the renormalization group recursion relations . in the case of an isotropic coupling between the order parameter field and the phonon field the value of the specific heat exponent @xmath62 is crucial for the character of the critical behavior . for @xmath63 all fixed points are found to be unstable against anisotropic perturbations of the elastic degrees of freedom . for @xmath64 the elastic degrees of freedom , isotropic as well as anisotropic , are irrelevant and the critical behavior is that of the rigid model . the effect of anisotropic couplings between the order parameter and the phonon fields has been considered in ref . it is concluded that a first order transition is to be expected for the ising , anisotropic xy and heisenberg model independent of the external conditions . in view of this earlier work it seems likely that also the above model will exhibit a topology of the flow diagram which is similar to the magnetic models and which is commonly interpreted as a signature of a fluctuation driven first order phase transition . as we will see from the calculations in sections [ sf_coll_transition][dim_reg ] this is actually the case . however , as will also be discussed in the following sections , this conclusion has to be reached with some care . near the transition point , the development of the ordered phase is reflected by the scaling behavior of the structure function . we define an euclidean density density correlation function @xmath65 where the subscript @xmath66 denotes the cumulant expectation value ( connected greens function ) . substitution of eq.([opdefs ] ) exposes the connection to the order parameters in position space : @xmath67 where non singular smooth contributions have been dropped . scaling theory gives the form of these connected correlations in the critical regime , @xmath68 where @xmath69 and @xmath70 are the correlation lengths for the ordering transitions of the @xmath29 and @xmath32 fields , respectively , @xmath71 ( @xmath72 ) and @xmath73 ( @xmath74 ) are the dynamical scaling and anomalous dimension exponents , and @xmath75 ( @xmath76 ) are universal scaling functions ( the @xmath77 signs indicate whether the system is above or below the critical point ) . additionally , there are correlation length exponents @xmath78 and @xmath79 which relate the correlation length to the deviation from the critical point , defined by @xmath80 the structure function is obtained by fourier transformation of @xmath81 . the static ( equal time ) correlator is thus @xmath82 \nonumber \\ & & \;\ ; + \sum_{ij } \tilde{\xi}^{2-\tilde{z}-\tilde{\eta}}\tilde{g}_{\pm , ij}[(\bbox{q } - \pi\hat{\bbox{e}}_i)\tilde{\xi } ] , \label{static_sf}\end{aligned}\ ] ] where @xmath83 , @xmath84 is a unit vector in the @xmath85 direction , and @xmath86 and @xmath87 are scaling functions . the full dynamical structure factor also has a scaling form , @xmath88 \nonumber \\ & & \;\ ; + \sum_{ij } \tilde{\xi}^{2-\tilde{\eta}}\tilde{h}_{\pm , ij}[(\bbox{q } - \pi\hat{\bbox{e}}_i)\tilde{\xi},\omega\tilde{\xi}^{\tilde{z } } ] , \label{euclidean_sf}\end{aligned}\ ] ] where again @xmath89 and @xmath90 are scaling functions ( simple relations exist between @xmath91,@xmath92 and @xmath89 , and may be easily derived by fourier transformation of eqs.([scalingeqns])([euclidean_sf ] ) ) . somewhat simpler results are obtained in the isotropic , ballistic limit , when @xmath21 and @xmath93 . focusing on the behavior near @xmath94 , the dynamic structure function then takes the form @xmath95.\ ] ] a similar form holds near @xmath96 . at small arguments , we expect @xmath92 to become approximately lorentzian . after analytic continuation ( @xmath97 ) , this implies the existence of a massive mode with @xmath98 for small @xmath99 . at the critical point , regularity of @xmath100 requires @xmath101 , and thus @xmath102 . for @xmath103 , this implies the behavior @xmath104 , \label{retarded_function}\ ] ] and thus a continuum of modes with linear dispersion . in addition to these quantities , the critical fluctuations of the checkerboard order parameter modify the behavior of the superfluid . to study these modifications , it is useful to define the current current correlation function @xmath105 where @xmath106 is the super - current . by introducing an infinitesimal generating field into eqs.([critical_action])([ptcoupling ] ) and differentiating the final effective action twice , one easily obtains the expression @xmath107 \label{d_formula}\ ] ] for @xmath108 in the phonon representation . the superfluid density @xmath109 and second - sound velocity @xmath110 may be extracted from the long wavelength form @xmath111 for @xmath112 . eq.([critical_action ] ) exhibits four phases in mean - field theory . in the superfluid ( sf ) phase there is no spatial order , so @xmath113 and @xmath114 . the checkerboard ( x ) state has @xmath115 but @xmath114 . a collinear ( c ) or striped phase occurs with @xmath113 but @xmath116 . lastly , there is a mixed ( m ) phase in which no symmetries persist , so @xmath117 and @xmath118 and does not point along a symmetry axis of the lattice . we have been unable to determine the detailed geometry of the mean field phase diagram . however , we expect the general topology shown in fig.[mf_phase_diagram_fig ] . for large @xmath7 and @xmath119 , there is a critical line separating the sf and c phases , while for large @xmath35 and @xmath120 , a different phase boundary separates the sf and x states . for large @xmath7 , the checkerboard order parameter @xmath29 is strongly suppressed and plays no role in the sf - c transition . we therefore consider here eq.([critical_action ] ) with @xmath113 and thereby neglect @xmath121 ( indeed , integrating out the massive @xmath29 field leads only to a small renormalization of @xmath122 and @xmath123 ) . for concreteness , we now also focus directly on the case @xmath1 , in which @xmath32 is a two - component field . in the absence of coupling to the superfluid phase mode ( i.e. @xmath124 ) , the action is identical to the classical free energy of an xy ferromagnet with cubic ( square ) anisotropy @xmath123 in @xmath125 dimensions . for positive @xmath123 ( as assumed here ) , the collinear state is favored for @xmath126 in mean - field theory . to study the critical properties for @xmath127 , however , we must include the effects of fluctuations . such fluctuations have been studied ( with @xmath124 ) by many authors @xcite . there is a general agreement between epsilon expansion and direct rg calculations as well as experimental observations @xcite that cubic anisotropy is in fact an _ irrelevant _ perturbation on the xy fixed point in three dimensions , with rg eigenvalue @xmath128 . therefore , the sf - c transition with @xmath124 should be xy - like , i.e. @xmath129 , @xmath130 , and @xmath131 . note , however , that @xmath123 is in fact a _ dangerous _ irrelevant operator , because it is necessary to select the direction of @xmath32 in the ordered phase@xcite . this asymptotic restoration of the @xmath45 symmetry should lead to the emergence of a `` pseudo - goldstone '' mode near the critical point in the ordered phase . to see this , consider following the rg flows ( for @xmath132 ) until the renormalized reduced temperature is of order one . at this point , the rescaled @xmath32 will have amplitude of order one ( enforced by the negative @xmath133 and positive @xmath134 terms ) , but almost unconstrained angle @xmath135 . the renormalized action for @xmath136 is then of the form @xmath137 \nonumber \\ & & -v ' \tilde{\xi}^{\lambda_4 } \left[\cos^4\theta + \sin^4\theta\right ] \bigg\}. \label{theta_action}\end{aligned}\ ] ] expanding the cosine and sine terms in @xmath136 gives the dispersion @xmath138 , with the anomalously small gap @xmath139 . note that the longitudinal gap @xmath140 is much larger than @xmath141 . the system appears isotropic in the sense that @xmath142 as @xmath143 . a detailed study of the effect of cubic anisotropy on the critical behavior below @xmath144 can be found in ref . @xcite . until now we have neglected the interaction with the phase @xmath38 . remarkably , the xy fixed point is stable against these fluctuations . as described in detail in ref . , the relevance of @xmath145 at the `` classical '' critical point is determined by the specific heat exponent @xmath146 . in particular , the eigenvalue @xmath147 is slightly negative at the three dimensional xy fixed point , and the long wavelength physics at the critical point is just that of a decoupled @xmath127 xy critical point and a free massless boson ( phase ) field . another way to understand this result is to integrate out the density fluctuations @xmath32 and determine their effect upon the superfluid modes . this is easily accomplished perturbatively in @xmath148 , since the coupling is precisely to the energy operator @xmath133 . at order @xmath148 , one produces only a boundary ( @xmath149 ) contribution . the first non - trivial term occurs at @xmath150 , where one finds a contribution to the effective action @xmath151 \nonumber \\ & \sim & { \sigma'}^2 \tilde{\xi}^{\tilde{\alpha}/\tilde{\nu } } \int_{\bbox{x},\tau } |\partial_\tau\theta|^2 , \label{velocity_correction}\end{aligned}\ ] ] where the subscripts @xmath152 are introduced as a shorthand notation for the integration variables @xmath153 and @xmath154 , @xmath155 is a ( short - range ) scaling function describing the energy energy correlations , and the last line is obtained via a gradient expansion of @xmath156 . the final correction term may be interpreted as a renormalization of the second sound velocity , @xmath157 where @xmath66 is an order one constant . note that for @xmath158 , as is the case here , this is a small reduction to the velocity near the critical point . for @xmath159 , however , the correction diverges , and the @xmath123 appears to be driven to zero ! this will be the case for the sf - x transition discussed in the next section and requires a more careful analysis . here we simply point out that the second sound velocity undergoes a finite _ suppression _ by critical fluctuations ( note that , once @xmath160 is negative , non - singular contributions actually dominate over the singular one exhibited explicitly in eq.([velocity_renormalization ] ) , and lead to a finite correction to @xmath161 ) . in this section we discuss the situation when @xmath162 , but @xmath7 passes through zero . in this case it is the @xmath32 modes which are weakly fluctuating , and the resulting theory is simply that of the @xmath29 field coupled to the superfluid phase . here the critical behavior for @xmath163 is simply that of a classical three - dimensional ising model . however , because the specific heat exponent in this case is positive ( @xmath164 ) , these phase fluctuations are _ relevant _ , and the problem needs to be reconsidered . to do so , we employ the technique of @xmath22expansion , developing rg equations near the upper critical dimension @xmath165 . in this section , we proceed with the simplest such method , known as a momentum - shell rg , which is sufficient to extract most aspects of the critical behavior . in section [ dim_reg ] , we use more powerful field theoretic methods within dimensional regularization to compute full crossover quantities . here the theory is regularized by including a sharp cut - off in momenta : all the functional integrals in the partition function @xmath166 are only over fields with momenta @xmath167 , the cut - off wavevector . such a cut - off serves as a crude approximation to the physical lattice cut - off of the bosonic system , but the precise form is irrelevant for the perturbative rg to the order required here , and facilitates the calculation of certain quantities . the momentum shell rg is performed by successive elimination of shells of momenta near the cut - off @xmath168 . after each shell is integrated out of the partition function , the coordinates and momenta are rescaled to keep the value of the cut - off fixed . additional rescalings of the fields allows most of the quadratic part of the action to remain fixed as well . the remaining coefficients _ flow _ under the rg . while an infinite number of operators are generated , only the relevant ( and marginal ) ones need be kept within the @xmath22 expansion , in this case @xmath169 , @xmath170 , and @xmath42 . momentum shell rg suffers , unfortunately , from one flaw : it is unable to deal properly with situations in which loop integrals must be handled at non zero external momenta ( and frequencies ) . for most problems , such momentum and frequency dependence can indeed be neglected to @xmath171 in the rg . in a cubic theory such as this one , however , the second diagram in fig . [ gamma_02 ] can potentially produce corrections at @xmath171 , because the coupling @xmath170 is only @xmath172 . we investigate this diagram in appendix [ loop_appendix ] , and find that these dangerous contributions are actually not present . the momentum shell rg is therefore well defined to leading order in @xmath22 , and we proceed here with the remainder of the calculation . we define the mode elimination step of the rg by integrating out fields with momenta in the range @xmath173 , and use the critical dynamics method of integrating out _ all _ frequencies , which are not cut off . by choosing @xmath174 , where @xmath175 is infinitesimal , we arrive at differential rg flow equations for the coupling constants . these equations are easily computed diagrammatically . because we evaluate all loops at zero external momenta ( c.f . the previous paragraph ) , the constraint that _ both _ loop momenta be within the shell is trivially satisfied by constraining just the single integration variable , due to momentum conservation . consider first the diagram in fig . [ gamma_20 ] . because of the explicit imaginary time derivative in the coupling @xmath170 , this diagram generates a renormalization of the velocity @xmath42 . because of the explicit factor of @xmath47 in the vertex , it decreases @xmath42 as in the perturbative treatment of the previous section . because this is the _ only _ renormalization of @xmath42 at one loop , the search for possible critical fixed points must proceed _ in the limit @xmath176_. this simplifies the computation of certain diagrams . in particular , any internal @xmath38 lines may be treated in this limit as simply constant @xmath177 factors . all remaining loop integrals are then proportional to the standard form @xmath178 where @xmath179 is a geometric factor . to complete the calculation , the fields and coordinates are rescaled to keep the cut - off fixed , according to @xmath180 where in this case , because of the absence of momentum - dependent loop corrections , @xmath181 ( to this order ) . the flow equations resulting from the rescaling and mode elimination are @xmath182(t - t_0 ) \ , , \label{tflow}\end{aligned}\ ] ] where we have defined @xmath183 , @xmath184 , and the dimensionless couplings @xmath185 , @xmath186 , and @xmath187 . the @xmath144 shift @xmath188 . the rg calculation is controlled for @xmath189 , where the coupling constants are of order @xmath22 , and higher loop corrections to the flow equations are higher order in @xmath22 . note that @xmath190 is strictly positive , so @xmath191 indeed flows monotonically to zero . the flow diagram resulting from eqs . ( [ kflow])([uflow ] ) is shown in fig . [ mom_shell_flow_fig ] . the three - dimensional classical ising fixed point ( @xmath192,@xmath193 ) is unstable once phase fluctuations are introduced . however , a new weak - coupling fixed point is not found within a one - loop @xmath22expansion . instead , the couplings flow into the classically unstable regime where @xmath194 . this scenario is generally interpreted as a fluctuation - induced first order phase transition @xcite . as was mentioned in the introduction , the existence of a strong coupling fixed point can not be excluded within the above line of argument . in fact , upon using a field theoretic rg approach carried out in fixed dimension , we will show in the next section that a strong coupling fixed point can be identified for large enough values of the coupling constant @xmath191 . the resulting `` non - bose liquid '' behavior will be discussed in the next section . for small values of @xmath191 `` runaway trajectories '' persist . in that case the instabilities in the trajectories may indicate a fluctuation - driven first order transition may actually describe the physics of the supersolid transition . one can understand this by matching the rg flows onto a renormalized classical theory . close enough to the critical point ( the flows may be integrated down to the scale of the correlation length ) , the renormalized theory has a negative value of @xmath169 . field theory is justified for the renormalized action , since the effective @xmath7 is now order one . with a negative @xmath169 , a standard mean field analysis then predicts a first order transition before the critical point is actually reached . note that , if the initial coupling @xmath170 ( and hence @xmath190 ) is very small , this will require an extremely close approach to the critical point , and hence result in extremely small discontinuities in thermodynamic parameters ( such as @xmath195 ) . experiments or numerical simulations which do not probe sufficiently close to @xmath196 may not resolve the true first - order transition , and instead observe `` effective '' exponents ( or more complex cross - over behavior ) . at a true fixed point , the correlation length exponent would be determined from eq . ( [ tflow ] ) as @xmath197 . a rough estimate of the measured value of @xmath78 may be obtained from an examination of fig . [ mom_shell_flow_fig ] . using values of @xmath198 and @xmath190 along a flow from the ( unstable ) ising fixed along the central trajectory in the figure suggests a gradual reduction of @xmath199 from approximately @xmath200 ( @xmath201 is the @xmath171 result for the ising value the accepted value is @xmath202 ) towards @xmath203 . in this section we use methods adapted from field theory to investigate the critical behavior of the superfluid to checkerboard supersolid transition . this allows us to go beyond the results obtained in section [ sf_x_transition ] and study the full crossover behavior near criticality . furthermore , we can study the flow of the parameters and coupling constants for arbitrary values of the second sound velocity . this will turn out to be an important technical advantage , since it allows us to identify a strong coupling fixed point above a critical value of @xmath204 ( see below ) . a perturbation theory for the effective free energy functional of the supersolid phase may be set up following the common procedure ( see e.g. ref . the elements of perturbation theory are the free propagators for the one - component density order parameter @xmath29 @xmath205 and the superfluid phase @xmath206 and the vertices , which may be read off from the anharmonic parts of eqs . ( [ critical_action])-([ptcoupling ] ) . in fig . [ elements_perturbation_theory ] we depict these elements for constructing the feynman graphs of the field - theoretical representation of the critical behavior of a supersolid . following the standard notation in field theory bare values are indicated by a subscript `` 0 '' . for @xmath207 the perturbation theory is infrared - divergent leading to non - trivial critical exponents . these anomalous dimensions are derived by studying the ultraviolet singularities of the field theory , which appear at the upper critical dimension @xmath208 , when the momentum cutoff @xmath168 is pushed to infinity ( see e.g. ref.@xcite ) . finite values are then assigned to these uv - divergent integrals through the application of a regularization prescription . we shall choose the dimensional regularization scheme as introduced by thooft and veltman @xcite ; here the ( @xmath209 ) singularities appear as poles @xmath210 . the ultraviolet divergences may then be collected in renormalization constants and absorbed into the definition of multiplicatively renormalized quantities . thus we define the renormalized fields @xmath211 and the renormalized parameters @xmath212 in eq . ( [ z_t ] ) we have taken into account the fact that the fluctuations will also shift the transition temperature . furthermore we have rendered the renormalized quantities dimensionless by introducing the explicit arbitrary length scale @xmath213 , and have finally included the geometric factor @xmath214 adapting a _ fixed dimension rg approach _ , one finds from the one loop expressions for the singular parts of the vertex functions @xmath215 , @xmath216 , @xmath217 , and @xmath218 in appendix [ app_pert ] the following results for the renormalization constants : @xmath219 for the fixed - dimension rg , the quantity @xmath220 depends on both @xmath191 and @xmath221 ( see eq . ( [ aa ] ) in appendix [ app_pert ] ) . as a consequence of the momentum dependence of the three point vertex , there are no singular contributions to the @xmath222term in the harmonic part of the superfluid phase field to any order in perturbation theory . this implies the exact identity @xmath223 the renormalization group equation serves to connect the asymptotic theory , where the infrared singularities manifest themselves , with a region in parameter space where the couplings @xmath169 and @xmath170 are finite ( but not necessarily small ) , and an ordinary `` naive '' perturbation expansion becomes applicable . it explicitly takes advantage of the scale invariance of the system near the critical point . more precisely , we observe that the bare two - point vertex function is of course independent of the arbitrary renormalization scale @xmath8 : @xmath224 introducing wilson s flow functions @xmath225 where @xmath226 denotes a logarithmic derivative with fixed bare couplings and parameters , eq . ( [ 4.1 ] ) leads to renormalization group equations for the renormalized vertex functions @xmath227 . for instance , the two point vertex function of the ising fields @xmath29 satisfies the partial differential equation @xmath228 \gamma_{02}(\mu , t , c , v , u,\sigma,{\bf k},\omega ) = 0 \ , . \label{4.8}\end{aligned}\ ] ] the rg equation ( [ 4.8 ] ) is now readily solved with the method of characteristics . the characteristics @xmath229 of eq . ( [ 4.8 ] ) define the running parameters and coupling constants into which these transform when @xmath230 . they are given by the solutions to first - order differential equations ( @xmath231 ) @xmath232 with the initial conditions @xmath233 , @xmath234 , @xmath235 , @xmath236 , and @xmath237 , namely @xmath238 \ , . \label{4.10}\ ] ] defining the dimensionless vertex function @xmath239 according to @xmath240 the solution of eq . ( [ 4.8 ] ) is @xmath241 \nonumber \\ & & \times { \hat \gamma_{02 } } \left ( t(s ) , u(s ) , \sigma(s ) , \frac{v(s)}{c(s ) } , \frac{{\bf k}}{\mu s } , \frac{\omega^2}{c(s)^2 \mu^2 s^2 } \right ) \ , . \label{4.12}\end{aligned}\ ] ] to one - loop order wilson s flow functions as derived from eqs . ( [ 4.2])-([4.7 ] ) read @xmath242 and due to the exact relation eq . ( [ z_exact ] ) one finds @xmath243 the zeros of the @xmath244functions , @xmath245 determine the values of the coupling constants where the theory becomes scale invariant . in order to study the flow diagram , it is convenient to introduce new effective coupling constants @xmath246 , @xmath247 , and @xmath248 , whose flow equations can be readily obtained from eqs . ( [ 4.2])([4.7 ] ) . they are @xmath249 , \label{rg1 } \\ \frac{1}{u}\frac{d u}{d \ell } & = & \varepsilon - \frac{3}{8}u - 24 \frac{k^2}{u(1+v)^3 } \nonumber\\ & & -k \left [ \frac{6}{(1+v)^2 } + \frac{v(1 - 3{\cal a}/2)}{(1+v)^3 } \right ] , \label{rg2 } \\ \frac{1}{v}\frac{d v}{d \ell } & = & - k \left [ \frac{1}{4 } - \frac{v(1+{\cal a}/2)}{(1+v)^3 } \right ] , \label{rg3}\end{aligned}\ ] ] where , for the purpose of comparison with eqs . ( [ kflow]-[vflow ] ) we have defined the running _ logarithmic _ scale @xmath250 , so that @xmath251 . note that @xmath220 depends on @xmath191 and @xmath221 ( see eq . ( [ aa ] ) in appendix [ app_pert ] ) . depending on the sign of the expression in the square brackets of eq . ( [ rg3 ] ) , @xmath252 , the flow of @xmath191 either tends to @xmath253 or to @xmath254 . one finds that there is a critical value @xmath255 , where @xmath256 changes sign . one of the most essential features of @xmath255 is that it can not be obtained within a strict @xmath22expansion . as can be inferred from fig . [ v_crit ] there is a critical value @xmath257 above which there is no solution of the equation @xmath258 . for @xmath259 there are two real roots , where one diverges upon approaching @xmath1 . since we are primarily interested in the critical behavior in @xmath1 dimensions the most relevant root is given by the lower curve in fig . [ v_crit ] . in what follows we will restrict our discussion to the case @xmath1 . then the flow equations have the following fixed points @xmath260 the crucial quantity for the behavior of the coupling constants is the ratio , @xmath187 , of the velocity of the second sound @xmath42 and phonon velocity @xmath36 ( first sound ) . for @xmath261 it turns out that the flow of @xmath191 tends to zero , in which case _ both _ the gaussian i and the ising i fixed points are unstable to small perturbations . this corresponds to the runaway trajectories discussed in the preceding section . however , the behavior of the renormalization group trajectories become quite different for @xmath262 . then the flow of @xmath191 tends to infinity , but it still depends on the sign of the coefficients in the flow equations for @xmath198 and @xmath190 , whether the renormalization group trajectories show runaway flow or converge into a finite fixed point . in the limit @xmath263 the flow equations reduce to the following simple form @xmath264 the flow diagram of which is shown in fig . [ flow_non_bose_liquid ] . the manifold , which separates the region in parameter space , where the flow of the renormalization group trajectories ends up in the non - bose liquid fixed point ( stable trajectory ) , from the parameter space where the flow shows runaway behavior ( unstable trajectory ) , is called the separatrix . a visualization of the complex flow behavior is given in fig . [ figure5 ] , where we have chosen the renormalization group trajectories to start at the plane @xmath265 ( bottom of the box ) . this part of the parameter space shows stable as well as unstable trajectories , where all stable trajectories are attracted to a critical surface , which finally contracts to the non - bose liquid fixed point . a numerical analysis of the flow equations also shows that approximately below @xmath266 there are no more stable trajectories . in figs . [ figure6 ] and [ figure7 ] two dimensional projections of the flow diagram are shown with starting values @xmath265 and @xmath267 , respectively . now we turn to the investigation of eq . ( [ 4.12 ] ) near a fixed point @xmath268 ; introducing the fixed point values of the wilson flow functions @xmath269 , also called the anomalous dimension of the parameter @xmath270 , one finds the following scaling law @xmath271 for @xmath216 one obtains a similar scaling law , where @xmath272 and @xmath273 are replaced by @xmath274 and @xmath275 , respectively . using the matching condition @xmath276 the following dynamic scaling law , @xmath277 is obtained , where the _ five _ independent exponents @xmath278 all other exponents are found through the scaling relations @xmath279 and @xmath280 . the crossover exponent @xmath281 describes the flow of the ratio of the two sound velocities , @xmath282 . as discussed in the previous section there is a separatrix in parameter space which separates stable from unstable trajectories . in the unstable region all trajectories show runaway behavior , but depending on the initial value of @xmath191 being smaller or larger than @xmath283 the flow of @xmath191 either tends to zero or to infinity . let us consider two extreme situations : for @xmath284 we find @xmath285 and @xmath286 with the corresponding fixed points : @xmath287 the fixed point values for the flow functions at the ising fixed point are @xmath288 , @xmath289 , @xmath290 , and @xmath291 . hence we find the following critical exponents at the ising fixed point ( to one loop order ) : @xmath292 for @xmath293 we find in @xmath1 dimensions @xmath294 and @xmath295 with the corresponding fixed points : @xmath296 the fixed point values of the flow functions at the non - bose liquid fixed point read @xmath297 , @xmath298 , @xmath299 , @xmath300 and @xmath301 , where interestingly all fluctuation contributions to the renormalization of @xmath7 cancel . hence we find the following critical exponents at the non - bose liquid fixed point ( to one loop order ) : @xmath302 now , we consider crossover effects on the most interesting physical quantities , namely the correlation functions of the density order parameter and the superfluid phase . for the discussion we will use an approximation , frequently called renormalized mean field theory , where one uses the fully renormalized flow functions , but the zero loop result for the scaling function . thus one gets @xmath303 \nonumber \\ & & \quad \times \left [ \frac{k^2}{\mu^2 s^2 } + \frac{\omega^2}{\mu^2 s^2 v^2(s ) } \right ] \ , , \\ \gamma_{\phi \phi } = \gamma_{02 } ( t , { \bf q } ) & = & \mu^2 s^2 \exp \left [ \int_1^s \zeta_{\phi } ( s^\prime ) d s^\prime / s^\prime \right ] \nonumber \\ & & \quad \times \left [ t ( s ) + \frac{k^2}{\mu^2 s^2 } + \frac{\omega^2}{\mu^2 s^2 c^2(s ) } \right ] \ , .\end{aligned}\ ] ] the most convenient way to analyze the crossover behavior of the correlation functions is in terms of effective critical exponents . we consider first the case @xmath120 and choose to define the effective dynamic critical exponents @xmath304 which are shown in fig . [ figure8 ] as a function of the flow parameter @xmath305 , where for the initial values we have chosen @xmath306 , @xmath307 and a series of ratios of sound velocities @xmath308 . upon using the matching condition @xmath309 the flow parameter @xmath305 can be related to the wave number . a corresponding effective fisher exponent for the ising field can also be defined by @xmath310 . the crossover behavior is shown in fig . [ figure9 ] for the same set of initial values as for the dynamic critical exponents . at zero external frequencies and momenta one can define an effective critical exponent for the dependence of the ising field correlation function on the control parameter @xmath311 using again the same set of initial values for the coupling constants one obtains the crossover behavior depicted in fig . [ figure10 ] . we have studied the ss - sf critical phenomena using renormalization group techniques . physical results are obtained by extrapolating from @xmath0 to @xmath1 , corresponding to bosons on a periodic substrate . coupling to the gapless phase modes of the superfluid opens the possibility of unusual quantum - critical behavior . this is _ not _ realized for the css - sf transition , in which the coupling of critical and phase modes is _ irrelevant _ , and the universality class is simply three - dimensional xy - like . by contrast , the mode - coupling terms are _ relevant _ for the xss - sf transition , and generically modify the critical behavior . the @xmath23 expansion predicts a fluctuation - driven first order transition in this case . a fixed dimension rg , however , gives a different scenario . for small values of the ratio @xmath187 runaway trajectories persist indicating again a first order transition though we can not rule out the possibility of a strong - coupling continuous critical point . for large values of @xmath191 we have identified a strong coupling fixed point . if the latter occurs , the phase and critical density modes must be strongly coupled at the transition , resulting in anomalous superfluid behavior . at the non - bose liquid fixed point the exponent for the correlation length is found to be @xmath312 to one loop order , since all fluctuation corrections cancel . it would be interesting to check whether there are two - loop contributions to this classical value . the value for the fisher exponent @xmath313 is negative , similar as in the case of a normal to superconducting transition @xcite . in either case , there are special multi - critical points in the phase diagram ( at particular values of the chemical potential ) for which the mode - coupling terms explicitly vanish by particle - hole symmetry , and simple 3d ising behavior ( and ordinary superfluidity ) is restored . an attempt to extract critical properties at the xss - sf transition in the quantum phase model has been made in ref . . in this model the multi - critical point may be exactly located at @xmath314 using particle - hole symmetry . there , they find a correlation exponent @xmath315 . this is somewhat less than the expected 3d ising behavior @xmath316 . they also quote a value @xmath21 here , in agreement with the rg . we believe the disagreement with the 3d ising result at the particle - hole symmetric point ( where the theory appears to be on extremely firm ground ) is probably due to finite - size corrections in the numerical data . away from this point ( they study @xmath317 ) , the correlation length exponent is perhaps reduced , @xmath318 , although the larger error bars make this interpretation difficult . a reduction of @xmath78 from @xmath319 is indeed predicted by the crossover analysis for a scenario , where a non - bose liquid fixed point exists . in the latter case the dynamic critical exponent is found to be @xmath25 to one - loop order . interestingly , scaling analysis of their monte - carlo data does not appear to differentiate between a dynamical scaling exponent @xmath21 or @xmath320 . however , they put a stronger constraint on the combination @xmath321 , which is consistent with @xmath322 . this gives some support to the rg prediction , though further systematic analysis of the simulation data is certainly desirable . more recently , scalettar et al.@xcite have investigated the tendency to phase separation in the bose - hubbard model as a more sensitive indicator for a possible first order transition . their preliminary results appear to confirm the existence of a first order transition in the xss - sf case , but not for the css - sf transition . this is in agreement with our rg results , which predict that the css - sf transition remains continuous ( 3d xy - like ) even away from the multi - critical point . an interesting open problem is to address the critical behavior in the xss - sf case directly in two dimensions more closely , to ascertain whether a fluctuation - driven first order transition or a continuous transition obtains in that case for small values of @xmath191 . this might either be achieved by a two loop renormalization group calculation accompanied by some resummation technique or by a self consistent calculation , which appears to capture much of the physics in the mathematically similar problem of the normal - superconducting transition in zero field@xcite . whether an analogous dual formulation of the problem can yield some insight is presently unclear . the long wavelength action eq . ( [ critical_action ] ) can be derived using the method of auxiliary fields . because the supersolid phase exhibits two broken symmetries , it is necessary to introduce two such fields , which will become order parameters for odlro and crystallinity . for concreteness , and because the most likely experimental realization of the ss is in josephson junction arrays , we work here with the quantum phase model ; the bose hubbard hamiltonian may be treated by similar methods . for simplicity , we also focus here on the case with only nearest neighbor interactions , in which case @xmath32 ordering does not arise . standard techniques can be used to translate the hamiltonian eq . ( [ phase_hamiltonian ] ) into the euclidean action @xmath323 where @xmath324 . the action enters the functional integral formulation of quantum mechanics via the generating functional ( partition function ) @xmath325[d\phi ] e^{-s_{\rm e}}.\ ] ] the josephson interaction may be decoupled using the ( hubbard stratonovich ) identity @xmath326 \nonumber \\ & & = \int [ d\bar{\psi}][d\psi ] \exp\bigg [ -\int\!d\tau \ { \psi_i e^{-i\phi_i } + \bar{\psi}_i e^{i\phi_i } \nonumber \\ & & \;\;\ ; + ( t^{-1})_{ij}\bar{\psi}_i\psi_j\ } \bigg],\end{aligned}\ ] ] where the index sums on the right hand side are implied ( we will adopt this convention in the remainder of this appendix ) , and the matrix @xmath327 . as is usually the case with such decouplings , the functional integral over the @xmath328 s is best defined by a product measure in momentum space . to obtain convergence , the contour for @xmath329 is then taken along the imaginary axis . before decoupling the charge interaction , we first separate out the zero momentum component , according to @xmath330 where @xmath331 . in momentum space , the remaining coupling is strictly negative , @xmath332 we may therefore perform a second hubbard - stratonovich to replace @xmath333 we choose to define the auxiliary field with a factor of @xmath334 to bring out the checkerboard correlations . note that because @xmath335 , the functional integral over @xmath29 is well behaved ( the theory is properly defined by omitting the zero mode , since @xmath336 ) . the last sum can be rewritten @xmath337 where @xmath338 $ ] for small @xmath99 . combining results , we obtain the full effective action @xmath339 where we have expanded @xmath340 around @xmath341 , and introduced the single - particle action @xmath342 with @xmath343 . it is useful to first perform the shift @xmath344 , which removes the @xmath38 dependence of the cosine at the cost of a ( time - dependent ) shift of the chemical potential , @xmath345 . integration over the @xmath13 and @xmath12 variables can then formally be performed independently . it is exactly equivalent to the solution of decoupled quantum - rotator models , each of which has the single - particle hamiltonian @xmath346 eq . ( [ h1def ] ) should be supplemented by the commutation relation @xmath347 = i$ ] . we will assume slow variations of the phase , so that @xmath348 is approximately constant . the mean field theory of refs . is recovered in the saddle point approximation , @xmath349 . this gives the conditions @xmath350 because we wish to obtain a landau theory , allowing for the possibility of fluctuations , we will explicitly construct the effective action @xmath351 as a function of @xmath29 and the goldstone mode @xmath38 . it is sufficient to work perturbatively in @xmath29 . this is accomplished via the cumulant expansion @xmath352 where the ( cumulant ) expectation value is evaluated using the single site theory with @xmath113 . for constant @xmath29 , the sum in eq . ( [ cumulant ] ) is just the imaginary time integral of the ground state energy , @xmath353 . unfortunately , even these expectation values can not be evaluated exactly , due to the form of eq . ( [ h1def ] ) . as discussed in ref . , the one - site hamiltonian may be rewritten in the phase representation as that of a particle in a cosine potential . the associated schrdinger equation is therefore a mathieu equation , whose solutions are not available in closed form . we therefore concentrate on the region @xmath354 , with @xmath355 , in which the checkerboard phase emerges most strongly , and analytic progress is possible . the one site problem may be recast in dimensionless form , using @xmath356 where @xmath357 , @xmath358 , and @xmath359 . the operators @xmath270 and @xmath360 obey the usual creation and annihilation operator commutation relations with respect to the number operator , @xmath361 = a$ ] , but _ commute _ among themselves @xmath362=0 $ ] . this allows the existence of states with arbitrarily ( positive or negative ) integer number , and one should note that @xmath363 . for small @xmath364 , and @xmath365 , there are two nearly degenerate low energy states : @xmath366 , with @xmath367 and energy @xmath368 , and @xmath369 , with @xmath370 and energy @xmath371 ( since we are eventually going to expand in @xmath372 and @xmath29 , @xmath373 is essentially equal to @xmath364 . all other states ( with @xmath374 ) have the much larger energies @xmath375 of order one . for @xmath376 , we may therefore work within the restricted basis of @xmath377 , in which the hamiltonian is represented @xmath378 where @xmath379 is the vector of pauli matrices in the standard representation . the energy eigenvalues for @xmath380 in this approximation are easily obtained , @xmath381 where the gap @xmath382 . the two corresponding ( normalized ) eigenstates are @xmath383 the value of @xmath244 is fixed by the mean field condition , eq . ( [ mf1 ] ) , equivalent to minimizing the effective hamiltonian density @xmath384,\ ] ] where @xmath385 , and we have taken @xmath386 . corrections to @xmath373 will be included via the landau expansion . the minimum occurs when @xmath387 , for @xmath388 , and @xmath365 for @xmath389 . for @xmath390 , @xmath391 vanishes , because an arbitrarily weak hopping matrix element breaks the degeneracy between empty and singly occupied sites . we are now ready to derive the landau expansion of the effective action . the hubbard stratonovich measure eq . ( [ dec_action ] ) gives @xmath392 where we neglected amplitude fluctuations of @xmath328 . a second contribution is obtained simply by expanding the one - site ground - state energy in @xmath29 and @xmath372 , @xmath393 up to constants . the only remaining non - trivial term is @xmath394 . it can not be calculated from the ground state energy @xmath395 , because it arises from the time dependence in @xmath396 , for which stationary quantum mechanics is not valid . it can , however , be calculated from the more general cumulant expansion , eq . ( [ cumulant ] ) , which also provides a check on the @xmath26 coefficient in eq . ( [ raw_eff_action ] ) . taking @xmath397 in eq . ( [ cumulant ] ) gives @xmath398 where we used time translation invariance to express the result in terms of @xmath399 . the imaginary time correlation function is easily expressed quantum mechanically as @xmath400 after inserting the complete set ( in this approximation ) @xmath401 , and using eqs . [ two_energies][two_states ] . fourier transforming gives @xmath402 $ ] . for @xmath403 , eq . ( [ cumulant_result ] ) then agrees with the @xmath26 term calculated earlier , and provides the `` kinetic '' correction @xmath404 the full effective action is thus @xmath405 . it is brought into the standard form of eqs . [ critical_action][ptcoupling ] by the rescaling @xmath406 after rescaling , the landau parameters are @xmath407 from these equations , we can determine the bare velocity ratio @xmath187 , as @xmath408 at the multi - critical point @xmath409 . these results are valid for @xmath410 , so that @xmath411 ( since @xmath412 ) . in this limit the @xmath411 approximation used in section [ sf_x_transition ] is valid even in the early stages of renormalization ( i.e. uniformly at all scales ) . in this appendix we consider the integral @xmath413 where @xmath414 , and @xmath415 . we have taken @xmath416 to obtain the leading term of @xmath417 , chosen an exponential cut - off for momenta , and set @xmath418 . we will also make use of the limit @xmath419 , to which the theory is driven ( within the @xmath22-expansion ) by the monotonic negative velocity renormalization . it is crucial here that _ both _ internal lines are cut - off : physically , we define the theory unambiguously by confining _ all _ momenta of the theory to be less than @xmath168 . in the momentum shell rg , we would perform an analogous integral iteratively in order to reduce @xmath168 . here we will instead perform the entire integral with fixed cut - off . it may then be differentiated with respect to @xmath168 to yield the appropriate contribution from a particular scale . any renormalizations must thus appear as logarithmic divergences , i.e. @xmath420 factors in the integral . in particular , the coefficients of @xmath421 and @xmath422 determine the renormalization of the @xmath423 and @xmath394 terms in the action , respectively . focusing first on the gradient term , consider @xmath424 . the frequency integral may be simply performed by contour integration to yield @xmath425 where we have inserted a delta function to put the momenta of the two propagators on equal footing . using the fourier representation @xmath426 $ ] and going to angular coordinates @xmath427 , where @xmath38 is the azimuthal angle , we have @xmath428 where @xmath429 . performing the @xmath430integrals gives @xmath431 the @xmath419 limits of the @xmath432integrals are well defined . taking it gives @xmath433 the @xmath434 term is @xmath435 note that this is a _ finite _ result , with no @xmath436 dependence . thus there is no renormalization ( in the scaling sense ) of the @xmath437 term . we now consider possible renormalization of the @xmath438 term . it is sufficient in this case to compute @xmath439 for small @xmath440 . a divergent ( scaling ) renormalization should appear in @xmath441 again as a @xmath436 , so @xmath440 must be kept non - zero until the end of the calculation . differentiating eq . ( [ idef ] ) gives @xmath442 \bigg[\frac{3}{{(\nu^2+p_2 ^ 2)^2 } } \nonumber \\ & & - \frac{{4p_2 ^ 2}}{{(\nu^2 + p_2 ^ 2)^3 } } \bigg ] e^{-p_1 a - p_2 a } ( 2\pi)^3\delta(\bbox{p}_1+\bbox{p}_2+\bbox{k } ) . \label{ip1}\end{aligned}\ ] ] it is convenient to take the @xmath419 limit in the integrand at this stage , which allows one to drop the second term in the first set of square brackets in eq . ( [ ip1 ] ) . performing the frequency integrals then gives @xmath443 \nonumber \\ & \times & e^{-p_1 a - p_2 a } ( 2\pi)^3\delta(\bbox{p}_1+\bbox{p}_2+\bbox{k } ) , \label{ip2}\end{aligned}\ ] ] which is identically zero ! the @xmath438 term is therefore unrenormalized . this appendix comprises the feynman diagrams to one loop order for the effective free energy functional of the supersolid to superfluid transition . we start with a list of the contributions to one loop order to the fully wavevector and frequency dependent two point vertex functions . the feynman diagrams for the vertex functions @xmath227 , with @xmath444 fields of the superfluid phase and @xmath61 ising fields , are depicted in figs . ( [ gamma_02])-([gamma_04 ] ) . the corresponding analytical expressions for @xmath445 is given by @xmath446 and @xmath447 } \nonumber \\ & & \quad\times \frac{\mu^2}{[\mu^2 + v^2 p^2 ] } \label{a2}\end{aligned}\ ] ] performing the frequency integrals using the residue theorem we find @xmath448 where we have subtracted the @xmath144shift @xmath449 and @xmath450 where @xmath451 and @xmath452 . the wavevector integral of eq . ( [ a4 ] ) can be evaluated for several limits listed below : \(b ) @xmath453 , @xmath458 , @xmath459 : @xmath460 with @xmath461 ( c ) @xmath462 , @xmath454 , @xmath463 : @xmath464 the analytical expression for @xmath465 at @xmath453 reads @xmath466 ^ 2 } \ , , \label{a8}\ ] ] and after performing the frequency integrals using the residue theorem we find @xmath467 the vertex function @xmath468 to one loop order has two contributions @xmath469 ^ 2 } \ , , \label{a10}\ ] ] and @xmath470 ^ 2 [ ( \mu / v)^2 + p^2 ] } \ , . \label{a11}\ ] ] after performing the frequency integrals we find @xmath471 @xmath472 the vertex function @xmath218 at zero external frequencies and momenta has three contributions . @xmath473 ^ 2 } \ , , \label{a14}\ ] ] @xmath474 ^ 2 [ ( \mu / v)^2 + p^2 ] } \ , , \label{a15}\ ] ] @xmath475 ^ 2 [ ( \mu / v)^2 + p^2]^2 } \ , . \label{a16}\ ] ] after performing the frequency integrals we find @xmath476 c. bruder , r. fazio , and g. schn , phys . b * 47 * , 342 ( 1993 ) ; c. bruder and g. schn , in _ kt transition and superconducting arrays _ , edited by d. kim _ et al . _ , ( min eum sa co. , seoul , 1993 ) , p. 175 . d. j. amit , _ field theory , the renormalization group , and critical phenomena _ , revised 2nd ed . ( world scientific , singapore , 1993 ) ; j. zinn justin , _ quantum field theory and critical phenomena _ ( clarendon press , oxford , 1993 ) .

mapping complex problems to simpler effective models is a key tool in theoretical physics . one important example in the realm of strongly correlated fermionic systems is the mapping of the hubbard model to a- model which is appropriate for the treatment of doped mott insulators . charge fluctuations across the charge gap are eliminated . so far the derivation of the- model is only known at half - filling or in its immediate vicinity . here we present the necessary conceptual advancement to treat finite doping . the results for the ensuing coupling constants are presented . technically , the extended derivation relies on self - similar continuous unitary transformations ( scut ) and normal - ordering relative to a doped reference ensemble . the range of applicability of the derivation of- model is determined as function of the doping and the ratio bandwidth over interaction .

1008.0522

the hubbard model @xcite is one of the most common models for the description of strongly correlated electron systems on lattices . because it contains the motion of the electrons as well as the interaction between two electrons at the same site it is capable to describe charge degrees of freedom as well as magnetic degrees of freedom . due to the rich physical behavior of the hubbard model an analytic solution is not possible except in one dimension @xcite . one common route to simplify the model for large repulsion @xmath4 is to derive an effective model which does no longer contain charge fluctuations across the charge gap . processes which change the number of doubly occupied sites ( double occupancies , dos ) are eliminated . for large enough repulsion @xmath4 and at half - filling the electrons are fixed on their lattice sites . in this mott - insulating phase the model can be mapped onto a heisenberg model describing only the energetically low - lying spin degrees of freedom . in the immediate vicinity of half - filling the motion and the interaction of doped holes is described by the extension of the heisenberg model to the @xmath0-@xmath1 model @xcite . the metallic behavior for small values of the repulsion is beyond the applicability of this mapping @xcite . in the present work the mapping of the hubbard model to the @xmath0-@xmath1 model is extended to finite macroscopic doping concentration @xmath2 . the influence of the doping @xmath2 on the resulting parameters of the @xmath0-@xmath1 model is studied for sufficiently large repulsion . our approach provides a systematic and controlled derivation of the effective coupling constants as function of the doping concentration . thereby , an important gap between the applicability of the derivation of the @xmath0-@xmath1 model and its actual applications is closed . first , we consider the half - filled case . the elimination of the charge fluctuations across the charge gap is performed by a self - similar continuous unitary transformation with various types of generators . besides the magnetic exchange couplings , the resulting @xmath0-@xmath1 model contains the motion and the interaction of holes and doubly occupied sites . since the mapping starts from a reference ensemble comprising the two spin states with equal weight and without any correlations between the spins on neighboring sites the spin state in the effective model remains unspecified . the mapping relies on the elimination of processes changing the number of holes and doubly occupied sites . note that an empty site represents a double occupancy of two holes . in the half - filled case the density of states exhibits two distinct bands for large @xmath4 ( see fig . [ fig_dos_half ] ) . the bands display equal weight @xmath5 and they are well separated for large @xmath4 @xcite . thus the states without holes or doubly occupied sites are energetically well separated from the ones with one or more holes or doubly occupied sites . if @xmath4 is decreased the bands approach each other . as soon as they touch the insulating phase is no longer the appropriate phase and metallic behavior occurs resulting in the breakdown of the mapping to the @xmath0-@xmath1 model . @xcite . ] the density of states exhibits two distinct , equally weighted bands , the lower hubbard band ( lhb ) and the upper hubbard band ( uhb ) @xcite . ] . the weight of the hole state is given by @xmath2 . the two half - filled states both carry the weight @xmath6 . note that the weight of the lower band is larger than the one of the upper hubbard band @xcite . ] the generic density of states obtained in the doped case is depicted in fig . [ fig_dos_dop ] . the effect of the doping on the density of states consists in shifting the fermi energy into the lower hubbard band for hole doping and redistributing the weight of the bands . electron doping is completely analogous in shifting the fermi energy into the upper hubbard band . since we focus here on particle - hole symmetric , bipartite lattices we will consider hole doping without loss of generality . it is obvious from the comparison of fig . [ fig_dos_half ] with fig . [ fig_dos_dop ] that the energetic separation of the hubbard bands is more subtle in the doped case than in the half - filled case . the bands are shifted depending on @xmath2 _ and _ spectral weight is transferred as well . the shift of spectral weight is a smoking gun evidence for strongly correlated fermionic systems . simple counting arguments in the limit @xmath7 tell us the distribution of weight . adding an @xmath8 electron to a site succeeds with probability @xmath9 . the first term results from the fraction of empty sites ; the corresponding weight is found at low energy because no doubly occupied site has to be created . the second term results from the fraction of sites occupied by @xmath10 electrons ; the corresponding weight is found at about @xmath11 because a doubly occupied site is created . together with the sum rule the weights shown in fig . [ fig_dos_dop ] result . we continue to use the number of doubly occupied sites ( dos ) as criterion to distinguish different sectors of the hilbert space . in order to have a quantitative measure for the energy separation of the sectors with differing number of dos the apparent charge gap @xmath12 is introduced which measures the energy separation of subspaces . it does not measure the energy gap between two pure states which is the reason why we call this separation of energy scales `` apparent '' . while the apparent charge gap is not an energy gap in a rigorous sense it is experimentally significant : it quantifies the energy needed in an mott insulator to create a charge excitation irrespective of the spin state of the system . for instance , the system can be at a temperature which implies a disordered paramagnetic spin state while it preserves the insulating properties . in the half - filled case the apparent charge gap measures the minimal energy of a doubly occupied site moving in an arbitrary spin background . the apparent charge gap is reduced for increasing values of the band width @xmath3 . if the gap vanishes the system is no longer stable against charge fluctuations and the mapping fails @xcite . obviously , the physical properties of strongly interacting fermionic systems depend considerably on the doping level @xcite . this fact leads automatically to the question how the validity of the mapping from the hubbard model to a generalized @xmath0-@xmath1 model is influenced by doping . thus the apparent charge gap has to be determined in dependence on the doping @xmath2 . keeping track of the apparent charge gap we determine the parameter range in which the mapping is still justified . this range of applicability may not be misinterpreted as phase diagram although it has some similarities @xcite . for instance , for large repulsion @xmath4 a doped @xmath0-@xmath1 model is still perfectly well defined while it displays metallic behavior . of course , it is expected that the applicability of a @xmath0-@xmath1 model decreases upon increasing doping @xcite . in view of the above , it is one of our central objectives to derive a diagram showing the range of applicability in dependence on the doping which has , to our knowledge , not been done before . our findings provide access to the limitations of the use of @xmath0-@xmath1 models in the context of planar cuprates to the extent that they can be described by a single - band hubbard model . the approach used is based on two conceptual ingredients . the first is a systematically controlled change of basis by means of continuous unitary transformations @xcite . the second is the choice of a doped reference ensemble without spin or charge order . within the range of applicability the effective @xmath0-@xmath1 model is derived . the doping dependence of the effective coupling constants is studied . the results are given in dependence on the ratio @xmath13 and on the dopant concentration @xmath2 . the method implemented here uses a self - similar truncation scheme to reduce the amount of proliferating terms in the running hamiltonian . the truncation is performed according to the range of the processes , that means rather local processes are kept while ones of longer range are neglected . hence the local processes acquire a non - perturbative dependence on the bare , initial coupling constants of the system . furthermore , recently introduced modified generators of the cut are implemented to cope with the vast amount of terms arising during the transformation @xcite . their results are very close to the previously used particle - conserving generator @xcite , but they significantly facilitate the calculation in terms of required memory and cpu time . after this introduction , the model and the method ( see sects . [ chap : model ] , [ chap : method ] ) are introduced . in sect . [ chap : gap ] results for the apparent charge gap are presented and sect . [ chap : results_coup ] provides exemplary results for the doping dependence of the coupling constants . [ chap : summ ] concludes the article . we consider the fermionic hubbard model @xcite . it describes electrons with spin @xmath14 on a lattice site @xmath15 by their creation operator @xmath16 and their annihilation operator @xmath17 . the hamiltonian consists of two terms describing the single - fermion kinetics ( @xmath18 ) and their interaction ( @xmath19 ) @xmath20 the kinetic part consists of the hopping of an electron with spin @xmath14 from site @xmath15 to site @xmath21 and vice versa . for this process to take place @xmath15 and @xmath21 have to be nearest neighbors as indicated by the bracket under the sum . the corresponding matrix element is denoted by @xmath0 . the band width @xmath3 of the model is given by @xmath22 with the coordination number @xmath23 ( number of nearest neighbors ) . in this work the lattice studied is the two dimensional square lattice with coordination number @xmath24 so that @xmath25 . the second part of the hamiltonian determines the interaction of the electrons . this term constitutes a pure on - site interaction . in @xmath19 the operator @xmath26 represents the number operator for the electrons this indicates that putting two electrons on the same site costs the additional energy @xmath4 . in the hubbard model there are four possible states per site . the site may be singly occupied by one electron with spin up or spin down @xmath27,@xmath28 , doubly occupied by two electrons with opposite spin @xmath29 or completely empty @xmath30 . the last two configurations correspond to charge fluctuations and are referred to as double occupancies ( do ) in this context . the interplay of motion and interaction of electrons in the hubbard model provides a description of the metal - insulator transition @xcite . another important field of application of the single - band hubbard model is the physics of high-@xmath31 cuprates @xcite . for a large hubbard repulsion @xmath4 in the half - filled case the density of states exhibits two separate bands , see fig . [ fig_dos_half ] , the so - called lower ( lhb ) and the upper hubbard band ( uhb ) . for infinite @xmath4 each site of the lattice is occupied by one electron which is energetically fixed to its site . if @xmath4 is finite the electron can move and virtually hop to adjacent site . thereby , dos are created but the the physics remains rather local , that means , the charge correlation length stays small . based on the locality of the important processes one may map the hubbard model onto an effective @xmath0-@xmath1 model . the generalized @xmath0-@xmath1 model conserves the number of dos . it comprises a part which describes the magnetic degrees of freedom , which is a generalized heisenberg model , and a part which describes the motion and interaction of dos reflecting the charge degrees of freedom . in order to obtain a model conserving the number of dos , processes which create or annihilate dos have to be eliminated . one systematic way to achieve this objective is the application of continuous unitary transformations to the hamiltonian . for smaller values of @xmath4 the local picture used here is no longer appropriate and the derivation of the @xmath0-@xmath1 model is not justified . the effective @xmath0-@xmath1 model is derived from the hubbard model by eliminating processes which change the number of dos . this elimination is performed using continuous unitary transformations ( cut ) @xcite . the elimination is based on a systematic change of the basis @xmath32 with a unitary operator @xmath33 and a continuous auxiliary variable @xmath34 referred to as the flow parameter . the transformation is determined by the flow equation @xmath35 \label{flow}\end{aligned}\ ] ] where @xmath36 denotes an antihermitian infinitesimal generator . at @xmath37 the transformation starts with the initial hamiltonian @xmath38 . the unitary transformation can be stopped at any arbitrary value of the flow parameter @xmath34 . usually , the effective hamiltonian is reached for @xmath39 . due to the continuity of the transformation it is readjusted to the flowing hamiltonian for every value of @xmath34 . the transformation stops automatically when the commutator @xmath40 $ ] vanishes which is generically the case for @xmath41 , i.e. , for convergence for @xmath41 . the structure of the effective hamiltonian is determined by the choice of the generator @xmath36 . we first choose the generator which leads to an effective model conserving the number of dos . to this end , we introduce the operator @xmath42\end{aligned}\ ] ] counting the number of dos . by the use of @xmath43 the repulsive part of the hamiltonian can be written as @xmath44 with @xmath45 denoting the number of sites . the kinetic part is split into three parts according to their effect on the number of dos @xmath46 where @xmath47 creates @xmath15 dos . the terms are given by @xmath48 \label{to}\\ \hat{t}_{+2 } = t_{+2}\sum_{<i , j > , \sigma } \left[\hat{n}_{i,\sigma } \hat{c}_{i,\bar\sigma}^\dagger \hat{c}_{j,\bar\sigma}^{\phantom\dagger } \left(1-\hat{n}_{j , \sigma}\right)+\right . \nonumber\\ \left.\hat{n}_{j,\sigma}\hat{c}_{j,\bar\sigma}^\dagger \hat{c}_{i,\bar\sigma}^{\phantom\dagger } \left(1-\hat{n}_{i,\sigma}\right)\right ] \\ \hat{t}_{-2 } = t_{-2}\sum_{<i , j > , \sigma } \left[\left(1-\hat{n}_{i,\sigma}\right ) \hat{c}_{i,\bar\sigma}^\dagger \hat{c}_{j,\bar\sigma}^{\phantom\dagger}\hat{n}_{j , \sigma}+\right . \nonumber\\ \left.\left(1-\hat{n}_{j,\sigma}\right)\hat{c}_{j,\bar\sigma}^\dagger \hat{c}_{i,\bar\sigma}^{\phantom\dagger}\hat{n}_{i , \sigma}\right]\,.\end{aligned}\ ] ] with @xmath49 . the terms contained in @xmath50 have no effect on the number of dos while the terms in @xmath51(@xmath52 ) increase ( decrease ) the number of dos by two . these terms are the ones that we intend to eliminate by the transformation . in the initial hamiltonian the prefactors @xmath53,@xmath54 and @xmath55 are equal , but they evolve differently under the cut . the first generator we use , the so - called _ quasiparticle conserving _ generator @xmath56 , can be expressed by the commutator @xmath57\ , . \label{pc}\end{aligned}\ ] ] this generator corresponds to the generator defined in refs . except for a global factor of two , which just implies a multiplicative renormalization of the flow parameter . the terms comprised by this generator are sketched in fig . [ fig_mku ] . the terms of the hamiltonian are classified according to their number of quasiparticle creation and annihilation operators . the @xmath58 block consists of terms with @xmath21 creation and @xmath59 annihilation operators . note that such a term requires at least @xmath59 excitations to be present in order to become active . but it is also active if more than @xmath59 excitations are present in the system . in this respect , the scheme in fig . [ fig_mku ] may not be mistaken to be a matrix . for a comprehensive presentation we refer the reader to ref . . generator highlighted in red ( dark grey ) . these terms will be eliminated . the terms in the uncoloured squares are zero initially and stay zero in the flow induced by @xmath60 . ] the quasiparticle conserving generator @xmath60 comprises all terms of the off - diagonal blocks . due to the structure of the generator the block - band structure of the hamiltonian is preserved during the flow @xcite . during the whole flow there will only be terms created which change the number of dos by @xmath61 or @xmath62 . with the definition ( [ pc ] ) of the generator the flow equation ( [ flow ] ) can be calculated . comparing the contributions on both sides of eq . [ flow ] a set of differential equations for the prefactors of the monomials in the creation and annihilation operators is obtained . these differential equations are first order in @xmath34 and they are bilinear in the prefactors entering on the right hand side . the equations do not form a closed set because in infinite systems new terms continue to arise on each application of the commutator in . for @xmath37 these terms carry the prefactor zero because they are not part of the initial hamiltonian . if we kept all these new terms in the remaining calculations we would obtain exact results for the effective model . but the number of arising terms is rising exponentially so that we have to limit them in number . for this purpose a truncation scheme is introduced which specifies the relevance of all term . less important terms are neglected , leading to a closed set of differential equations which can be solved numerically . in many previous applications a small parameter is used to classify the arising terms so that a perturbative treatment results . in contrast , we are adopting here a truncation scheme which classifies the terms according to their structure . such a cut scheme is usually called self - similar . it resembles more conventional renormalizations . effects of infinite order are present in the prefactors of the kept terms . the truncation scheme used in this work keeps or neglects terms according to their locality , i.e. , according to the range of the represented physical process . this approach is well justified if the model under study is governed by a small correlation length . this is exactly the case for a hubbard model at large @xmath4 where the propagation of charge degrees of freedom is suppressed by the high energetic cost of creating a do . before we discuss how we measure the degree of locality we have to find a unique representation for the operators to be sure to treat similar terms in the same way . to this end , the monomials are expressed as normal - ordered products of local operators . the normal - ordering we are using is not the standard one known for the fermionic or bosonic algebra because the creation or annihilation of a do can not be represented by interaction free fermions or bosons . instead , we use a reference ensemble . non - trivial operators are only those which create or annihilate fluctuations away from the reference ensemble . for a given doping concentration @xmath2 the reference ensemble is defined by the statistical operator @xmath63 + \delta | 0\rangle_i \langle 0|_i \right\ } \label{ro}\end{aligned}\ ] ] where the product extends over all lattice sites @xmath15 . in the half - filled case ( @xmath64 ) the reference ensemble is paramagnetic and the magnetic degrees of freedom are totally disordered . each site is equally probably occupied by an @xmath8 or by a @xmath10 electron ; no direction is singled out , no correlation between neighboring sites exists . charge fluctuations from this reference ensemble are the empty @xmath30 and the doubly occupied site @xmath29 . magnetic fluctuations are induced by the application by spin operators , see below . considering doping we focus on hole doping only because the model at hand is particle - hole symmetric so that electron doping leads exactly to the same results . hence we include the empty state @xmath30 in the reference ensemble besides the half - filled states with a probability given by the doping level @xmath2 . the remaining weight is again equally distributed over the two spin states . note that this extension to the doped case does not introduce any bias . there is no correlation between sites nor on each site . hence the reference ensemble is the mixture with the maximum entropy at given level of doping . based on the reference ensemble we define a term as normal - ordered if the expectation value of each of its factors of local operators vanishes with respect to this ensemble . thus a normal - ordered local operator fulfills @xmath65 based on this condition we define a basis of local normal - ordered operators ( see tab . [ table : list ] ) . of course , the identity does not fulfull the condition . but the identity is the trivial action of an operator and obviously does not create or annihilate any fluctuation away from the reference ensemble . without the identity the list of operators would not be complete . any monomial occuring during the flow is expressed in the operator basis in tab . [ table : list ] . .basis of normal - ordered local operators [ cols="^,^",options="header " , ] in this section we present analytic solutions of the flow equations which are possible for the two simplemost truncation schemes . due to the simplicity of the truncation schemes no difference between the different choices of the generator are found . for concreteness , we consider the particle - conserving @xmath56 here . the calculation of the minimal model starts by studying all processes on between adjacent sites . this nearest neighbor ( nn ) calculation is equivalent to a maximal extension of @xmath66 . to arrive at the minimal model all terms not present in the initial hamiltonian except the nn heisenberg exchange are neglected . the initial generator takes the form @xmath67 = 2\hat{t}_{+2}-2\hat{t}_{-2}\end{aligned}\ ] ] with the flow parameter @xmath34 . inserting this definition in the flow equation we calculate @xmath68.\end{aligned}\ ] ] from the commutator new terms arise @xcite . in the minimal model all terms except the nn heisenberg exchange @xmath69 are omitted . the coupling constant @xmath70 starts at @xmath71 because it is not part of the initial hamiltonian @xmath72 . for @xmath73 it evolves according to the flow equation . in this simple case the flow equation can be solved analytically for a general lattice with coordination number @xmath23 @xcite . for @xmath74 the effective model is reached . the effective coupling constants take the form @xmath75 to obtain the equations for the square lattice @xmath24 must be inserted . the variables @xmath55 and @xmath76 represent the initial , unrenormalized values of the hopping and the hubbard repulsion . for simplicity we will omit the subscript 0 and label the unrenormalized values by @xmath0 and @xmath4 henceforth . since the terms @xmath51 and @xmath52 are hermitian conjugates and we assume their coefficients to be real @xmath77 holds . from ( [ t2 ] ) we see that the terms contained in @xmath51 and @xmath52 are eliminated as it should be because they change the number of dos . the effective model is eventually given by @xmath78 in the nearest neighbor model all terms arising from a nearest neighbor calculation are included in the effective model as well as in the generators . this is the full calculation with extension 1 . it can still be solved analytically @xcite . in this truncation scheme the heisenberg exchange coupling is given by @xmath79 note the differences to the result of the minimal model . of course , this differences arises only beyond leading order . besides the heisenberg exchange the calculation contains the term @xmath80 describing the interaction of two dos @xmath81 the operator @xmath82 counts the amount of electrons compared to the mean value of the filling . for the half - filled case the mean value is @xmath83.the third term created during the flow is @xmath84 describing pair hopping processes of dos . one of the processes contained in this term is the hopping of two electrons from site @xmath21 to an empty site @xmath15 . as the empty state as well as the doubly occupied state represents a do this process does not change the number of dos . in the effective model @xmath85 and @xmath86 take the values @xmath87 in the case of doping another contribution to the flow equation also arises . it reads @xmath88 and determines the chemical potential @xmath89 . in the effective model this constant takes the value @xmath90 with the coordination number @xmath23 . in leading order in @xmath91 this yields a chemical potential which depends linearly on the doping constant @xmath2 and on the coordination number of the lattice @xmath92 in this section the influence of the choice of the generator is studied . first , we consider the rod defined in sect . [ chap : method ] for the gs - generator , gs,1p - generator ( [ gs1p ] ) , and the pc - generator ( [ pc ] ) . the results for the rod obtained in a double plaquette calculation at half - filling are shown in fig . [ fig_rod_gen ] as functions of the continuous flow parameter @xmath34 . . ] for the gs - generator , the rod converges for all values of the ratio @xmath13 . in contrast , the rod for the gs,1p - generator and even more pronounced the rod for the pc - generator show non - monotonic behavior for larger values of @xmath13 . non - monotonic behavior of the rod suggests that the intended transformation does not succeed . there is no strict statement that a successful cut has to have a monotonic rod . it is well possible that the rod displays local maxima which indicate that some energy eigenstates are re - ordered , see for instance ref . . if the cut is performed without approximation any unitary transformation is as good as any other . but since we have to truncate many terms the upturn of the rod indicates a potential loss of accuracy . if the total norm of the off - diagonal terms is large there is still a significant transformation to be done . in the course of this transformation the truncation of terms may introduce significant errors . in return , a quickly decreasing rod indicates that all coefficients to be eliminated decay fast and significant truncation errors are less likely . but we like to stress that the behavior of the rod is only an indicator for possible truncation errors which eventually may imply that the intended mapping breaks down . the faster convergence of the gs - generator is straightforward to understand because the gs - generator comprises only terms which create dos from the reference ensemble or which annihilate them , see eq . and fig . [ fig : gs ] . as long as there is a finite charge gap @xmath12 these processes are exponentially suppressed : @xmath93 . for the gs,1p - generator the processes starting from one do creating two additional dos can be more difficult to suppress if they decrease the total energy . this is possible if the dos disperse and a do at high energies decays into three dos at lower energy . the pc - generator aims in addition at eliminating processes starting from two and more dos so that there are even more processes which may decrease the total energy while the number of dos increases . hence we are not surprised to see that the pc - generator induces a flow of the rod which displays even more pronounced non - monotonic behavior . from these observations the conclusion to always favor the gs - generator suggests itself . but it is in fact a trade - off . the gs - generator is quicker to implement and more robust in its convergence , but it achieves less because it decouples only the subspace without any dos . for deriving only an extended heisenberg model this is completely sufficient and hence for this aim the gs - generator is the generator of choice . but if one is additionally interested in an explicit description of the dynamics of dos the other generators are advantageous as we will illustrate next . to see how the coupling constants are influenced by the choice of generator a few exemplary results are shown . one of the most important magnetic coupling constants besides the nearest neighbor coupling @xmath94 is the heisenberg interaction between next - nearest neighbors @xmath95 , i.e. , diagonal over a plaquette . the behavior of this exchange coupling as function of @xmath13 for the three generators is depicted in fig . [ fig_j2_gen ] . for different generators as function of @xmath13 determined in an upto4 calculation . the results for the three generators almost coincide even for large values of @xmath13 . ] the curves for all three generators almost coincide . this underlines that it is completely sufficient to use the gs or the gs,1p - generator for the determination of this coupling . all processes contributing to this coupling are included in the cut induced by the gs - generator . this can be understood from the fact , that the magnetic coupling @xmath95 describes an interaction within the subspace of half - filled states , i.e. , the subspace without any do . this subspace is decoupled from the remainder of the hilbert space by all three generators . if no truncation errors occurred , all three generators would indeed yield precisely the same result , cf.ref . . - and the @xmath96-generator agree very well whereas the results of @xmath97-generator show larger deviations . ] in contrast to a pure spin - spin coupling the term @xmath98 acts on two dos . it describes the hopping of an electron from a singly occupied site to an empty site under the condition that site @xmath99 is occupied by a do . it is a process which is not active on the subspace with zero or only one do . we do not expect that the results for the different generators agree . indeed , the results for this coupling constant ( fig . [ fig_v_n_gen ] ) show rather large deviations of the results obtained by the gs - generator from the results obtained by the other two generators . this illustrates that the gs - generator induces a different unitary transformation than the other two generators . note that the deviations do not necessarily imply that the gs - result is less accurate because it results from the representation of the hamiltonian in a different basis . in view of the above arguments , it is surprising that the results of the gs,1p - generator and the pc - generator are so close to each other . from their definitions we expect that the pc- and the gs,1p - results agree very well for processes involving a single do , but not necessarily for processes involving two dos . in conclusion , the question which generator is optimum can not be answered generally . it depends on the particular objective of the intended investigation . in the previous sections we have started to discuss the issue for which conditions the intended mapping from the hubbard model to a generalized @xmath0-@xmath1 model is possible and justified . qualitatively it is obvious that @xmath4 must be large enough . but quantitative indicators are needed . here we aim at giving a quantitative estimate for the parameter range in which the mapping is justified . the basis of the transformation from the hubbard model to the @xmath0-@xmath1 model is the elimination of charge fluctuations . these charge fluctuations correspond to changes in the number of dos . the corresponding processes are the ones that we consider to be off - diagonal . to be able to eliminate such processes the subspaces with differing numbers of dos have to be separated in energy . in sect . [ chap : method ] we introduced the residual off - diagonality ( rod ) as a measure to decide if for a given @xmath34 the off - diagonal terms are small enough to be neglected . . the rod measures to which extent charge fluctuations are eliminated as function of the flow parameter @xmath34 . ] figure [ fig_rod_doppel ] shows the behavior of the rod of the pc - generator for the double plaquette calculation at various values of @xmath13 . for small values of the flow parameter @xmath100 the rod decreases exponentially . if the ratio @xmath13 is increased to @xmath101 the rod evolves non - monotonically . for @xmath102 the rod falls below @xmath103 to rise again for larger values of @xmath34 . for even higher values of @xmath13 this increase sets in for smaller and smaller @xmath34 . we argued in the preceding section that a non - monotonic behavior of the rod as observed in fig . [ fig_rod_doppel ] is a first clue for a possible breakdown of the mapping . although the non - monotonicity indicates problems of the mapping already for @xmath104 no sign of a possible breakdown can be seen in the coupling constants , see for instance fig . [ fig_j2_gen ] . this is due to the fact , that the dominant spin coupling constants are already converged to their value in the effective model for small values of @xmath34 . thus processes appearing only at large @xmath34 have no influence on these values . to make progress in determining the range of validity of the mapping we have to study the separation of energy scales between the states without dos and the states with dos . to this end we investigate the energies of an added electron or an added hole to the system , that means the density - of - states ( dos ) which comprises the lhb and the uhb for large @xmath4 , see figs . [ fig_dos_half ] and [ fig_dos_dop ] . if @xmath3 is increased the bands approach each other and eventually touch so that there is no energy separation anymore . states with differing number of dos have the same energy so that charge fluctuations can not be eliminated . in a paramagnetic description by dynamic mean - field theory ( dmft ) the mott - insulating phase becomes unstable at this very point @xcite . turning the argument around we use the gap @xmath12 between the lower and the upper hubbard band as quantitative indicator of the energy separation of subspaces with different number of dos . if @xmath12 is finite there is good physical reason to regard the mapping of the hubbard model onto the @xmath0-@xmath1 model as justified . if @xmath12 vanishes the mapping has to break down . there is one additional aspect to which we have to draw the reader s attention . in infinite dimensions , where dmft is exact , one may suppress long - range magnetic order and consider the paramagnetic phase which does not show an spin - spin correlation between different sites so that the insulating paramagnetic phase behaves like the reference ensemble for @xmath105 . in particular , the charge gap does not depend on the spin state . but in finite dimensions , even without long - range magnetic order one has to expect that the charge gap will generically depend on the spin state . hence there is no well - defined charge gap without specifying the state of the spin degrees of freedom . thus we have to introduce the concept of the _ apparent _ charge gap @xmath12 @xcite which is designed to describe the energy separation of subspaces with different number of dos if an electron is added to the disordered reference ensemble . the apparent charge gap is not rigorously defined and it can not be measured precisely in experiment because it does not capture bandtails of the hubbard bands which carry little spectral weight . but it is an estimate for the energy separation between states without dos and states with dos . hence it provides an appropriate estimate of the range of validity of the mapping from the hubbard model to the @xmath0-@xmath1 model . we calculate @xmath12 for the effective @xmath0-@xmath1 model derived before . in the half - filled case the gap is calculated by estimating the lowest possible energy of an added do , for details see below . calculations for the half - filled case indicate a closure of the gap for @xmath106 @xcite . in the doped case the calculation of @xmath12 is divided into two steps as can be understood from the dos sketched in fig . [ fig_dos_dop ] . to calculate the apparent gap we first determine the lowest possible energy @xmath107 of a do in the upper hubbard band . in a second step we calculate the maximum energy for the destruction of a do @xmath108 , that means for adding an electron to an empty site . hence in the doped case we use @xmath109 while in the undoped case we have @xmath110 note that the seemingly discontinuous definition of @xmath12 as function of @xmath107 stems from the discontinuous evolution of the fermi level which jumps upon hole doping from the middle between the hubbard bands to the edge of the lower hubbard band . the apparent charge gap is calculated for the effective @xmath0-@xmath1 model . a full diagonalization of the hamiltonian is not feasible and it would not provide what we need , namely the charge gap above the disordered reference ensemble . hence we apply a lanczos approach in terms of operators . the lanczos approach is appropriate because we only aim at extremum eigenvalues . since we have to deal with operators acting on the reference ensemble @xcite , which is a mixed state , the liouville formulation of this method has to be used @xcite . the evolution of an operator @xmath111 is given by the liouville superoperator @xmath112\,.\end{aligned}\ ] ] the effect of this superoperator applied to the creation operator of a do consists of moving the do and changing the spin background . with this operator a basis of operators @xmath113 describing the do with momentum @xmath99 and the effect on its surrounding spins is built recursively . in the first part of the calculation the minimal energy of a do with momentum @xmath99 is calculated . the calculation starts with the vector @xmath114 where @xmath45 denotes the number of lattice sites and the vector @xmath115 determines the actual position of the do . the action of this operator is to put an @xmath8 electron on a site which is already occupied by a @xmath10 electron so that a doubly occupied site is created . from this starting vector the basis is built recursively by @xmath116 according to the rules of the lanczos tridiagonalization . the scalar product of the liouville formulation @xcite is defined as @xmath117 with the statistical operator @xmath118 of the reference ensemble ( [ ro ] ) . the prefactors @xmath119 are given through the projection onto @xmath120 @xmath121 and the @xmath122 are given by @xmath123 in this operator basis the liouville superoperator takes tridiagonal form with the coefficients @xmath119 on the diagonal the coefficients @xmath122 on the secondary diagonals . if infinitely many iterations were performed the dispersion of a do relative to the disordered spin background would be given as the lowest energy in the subspace spanned by the calculated operator basis . in real calculations only a few iterations are feasible due to the humongous number of terms in the effective hamiltonian . starting with the vector @xmath124 in ( [ vo ] ) consisting of one single operator , the commutation leads to increasingly complicated terms whose appropriate superposition describes @xmath125 . the effort grows exponentially with the number of iterations . thus we restrict ourselves to a finite basis @xmath126 . the lowest energy calculated in the subspace spanned by @xmath127 yields an upper bound @xmath128 to the real dispersion of the do . note that in the following we denote by @xmath128 this upper bound in order to keep the notation simple . in the half - filled case the apparent charge gap results from eq . . for finite doping the particle - hole symmetry is lost , the fermi energy jumps to the lhb , and the value @xmath108 has to be determined . to this end , we start from the modified vector @xmath129 this operator destroys a do by placing a single electron on an empty site . the value for @xmath108 we obtain from the calculation in a finite subspace @xmath130 is a lower bound to the true maximum energy . finally the apparent charge gap is given by in dependence on the doping level @xmath2 . as argued before the mapping to the @xmath0-@xmath1 model is justified as along as @xmath131 . extremum values of @xmath132 occur at the high symmetry points of the brillouin zone . thus we avoid costly calculation of the whole dispersion and focus on the momenta @xmath133 and @xmath134 where the lattice constant is set to unity . the calculations rely on the the nearest neighbor effective model . previous calculations in the half - filled case @xcite showed that there is no significant change in the results obtained for different truncation schemes because the main uncertainty results from the limited number of iterations in the lanczos tridiagonalization . the truncation of the effective model used plays only a minor role . since only a few iterations were feasible we additionally perform an extrapolation . the results for the gaps are extrapolated in @xmath135 with @xmath136 denoting the number of iterations . by extrapolating to @xmath137 we obtain an estimate for @xmath12 ; for more details we refer to ref . . figure [ fig_gap_durchn ] displays the extrapolation for the half - filled case . for the doped case , @xmath138 and @xmath128 are extrapolated separately in @xmath135 . of the apparent charge gap at half - filling for @xmath139 obtained by @xmath136 iterations . ] the apparent charge gap is computed for the effective @xmath0-@xmath1 model derived by a cut with nn truncation using the pc - generator or the gs,1p - generator . the gs - generator is not used in this context because the resulting effective model mixes a single do with the subspaces of two and more dos . the gap is calculated for various doping levels as function of @xmath3 . thus the value @xmath13 up to which the mapping is justified is estimated from @xmath140 . the minimum @xmath128 of the dispersion of a do is found for a vanishing momentum . the maximum energy @xmath138 for the destruction of a hole is found for a momentum @xmath134 . for various @xmath2 . ] the results for the extrapolated @xmath12 are displayed in fig . [ fig:1durchn ] for various values of the doping @xmath2 . for vanishing bandwidth @xmath141 the apparent gap is given by the hubbard repulsion @xmath4 . thus the curves of @xmath142 start at unity . then the gap decreases almost linearly until @xmath143 is reached . negative values of the gap indicate the breakdown of the mapping . the linear decrease of the gap has also been observed for the half - filled case @xcite . the decrease leads to a closure of the charge gap for @xmath144 . for the bethe lattice with @xmath145 a closure of the gap was found for @xmath146 @xcite which agrees well with our estimate in view of the different lattices and techniques . other numerical evaluations of dmft for the bethe lattice yield a closure of the insulating gap at @xmath147 @xcite or at @xmath148 @xcite . up to which a mapping of the hubbard model to the effective @xmath0-@xmath1 model is justified . it is derived from the zeros @xmath149 as function of doping . ] our results indicate that the apparent charge gap @xmath12 for the square lattice closes at @xmath150 for @xmath105 . upon doping @xmath12 vanishes even faster upon increasing bandwidth so that the range of applicability of the mapping ` hubbard model @xmath151 @xmath0-@xmath1 model ' is reduced . from the values of @xmath13 where @xmath12 beomes zero we estimate this range of applicability . the result is shown in fig . [ fig : appli ] which represents one of the central results of this work . our approach provides the first systematic derivation of this diagram of applicability as function of doping . the range of applicability decreases from @xmath150 for @xmath105 to @xmath152 for @xmath153 . then the range of applicability increases again slightly to @xmath154 for @xmath155 . the plateau in @xmath156 and the moderate increase are rather unexpected , cf . ref . we do not have an obvious explanation for it . in contrast , the decrease of the range of applicability for @xmath157 meets the qualitative expectation since a doped system has a higher mobility of charges so that the energy separation of sectors of differing number of dos becomes smeared out . the relative constant limiting value for @xmath13 below which the use of a generalized @xmath0-@xmath1 model is justified provides interesting information on the applicability of @xmath0-@xmath1 models for doped systems . the use of @xmath0-@xmath1 models is very widespread in theoretical studies for the high-@xmath158 superconductors based on cuprates . commonly used parameters are @xmath159 and @xmath160 @xcite . our results indicate that the use of @xmath0-@xmath1 models is indeed justified . but caution is required in the doping range @xmath161 where @xmath159 is at about the limit of applicability . thus our results shed light on the important question of the applicability of a commonly used model . it is remarkable that the issue of how to justify this model for significant levels of dopings has attracted so little attention so far . in the preceding section we comprehensively discussed the applicability of the derivation of a generalized @xmath0-@xmath1 model . the result of this discussion is summarized in the estimated range of applicability shown in fig . [ fig : appli ] . in the present section , we provide the coupling constants which ensue from the cut of the hubbard model to the @xmath0-@xmath1 model . results are given in the range @xmath162 because the mapping definitely breaks down beyond . all results shown are derived from upto4 calculations using the pc - generator . additionally we performed random double plaquette calculations with the gs,1p - generator to check if there are changes in the coupling constants when higher truncation schemes are applied . no significant differences are found . thus the upto4 truncation appears to be sufficient to determine the coupling constants . the results for the half - filled case ( @xmath64 ) shown in the following figures agree perfectly with the results obtained by reischl et al . @xcite . although the mapping generates a large number of terms , only few of them are really relevant in the final effective model . most others only have very small prefactors . first , we consider the chemical potential as defined in ( [ eq : mu ] ) . in leading quadratic order of @xmath91 it is proportional to @xmath2 . therefore @xmath89 will be shown in units of @xmath163 as defined in ( [ j_1 ] ) . the ratio @xmath164 shows almost no dependence on @xmath2 , see fig . [ fig : mu_n ] . as function of @xmath13 the chemical potential stays rather constant and even for @xmath144 the deviations are small . the dependence on @xmath13 is greater for smaller values of @xmath2 concentration . for various doping concentrations @xmath2 . ] the dominant terms of the effective model are the heisenberg - type spin interactions . @xmath165 the largest contribution of this type is the heisenberg exchange @xmath94 between nearest neighbors . all results are shown relative to the leading perturbative result @xmath166 . the left panel of fig . [ fig : j_1_dop_n ] shows the dependence of @xmath94 on the ratio @xmath13 . starting from @xmath167 for @xmath168 the coupling constant takes slightly smaller values for larger @xmath13 . additionally , the doping dependence of @xmath94 for various values of @xmath13 is shown relative to its value for the undoped system in the right panel of fig . [ fig : j_1_dop_n ] . @xmath94 increases with @xmath2 . the doping has a greater influence for larger values of @xmath13 , but the effect remains rather small . even for @xmath169 the doping causes a change in @xmath94 of only about @xmath170 . on @xmath13 for various values of @xmath2 ( left panel ) . the dependence on @xmath2 for two ratios @xmath13 is depicted in the right panel . the undoped value for @xmath171 is found to be @xmath172 . for @xmath169 @xmath173 takes the value @xmath174 . ] on @xmath13 for various values of @xmath2 ( left panel ) . the dependence on @xmath2 for two ratios @xmath13 is depicted in the right panel . the undoped value for @xmath171 is found to be @xmath172 . for @xmath169 @xmath173 takes the value @xmath174 . ] the heisenberg exchange between next - nearest ( diagonal ) neighbors @xmath95 as well as the exchange @xmath175 between neighbors at a linear distance of two lattice spacings are much smaller than @xmath94 . both terms appear in fourth order of @xmath91 . even for @xmath144 @xmath95 is smaller than @xmath176 , see fig . [ fig : j_2_dop2 ] . surprisingly , @xmath95 shows a slightly more significant ( relative ) dependence on the dopant concentration than @xmath94 , see fig . [ fig : j_2_dop2 ] . yet , in view of the small absolute values of @xmath95 , this doping dependence can be neglected . for various doping concentrations as function of @xmath13 . ] as can be seen in the right panel of fig . [ fig : j_3_n ] the coupling @xmath175 shows a counter - intuitive behavior . first it decreases upon doping but increases again beyond @xmath177 . on @xmath13 for various values of @xmath2 ( left panel ) . the dependence on @xmath2 for two ratios @xmath13 is depicted in the right panel . the undoped values are @xmath178 for @xmath171 and @xmath179 for @xmath169 . ] on @xmath13 for various values of @xmath2 ( left panel ) . the dependence on @xmath2 for two ratios @xmath13 is depicted in the right panel . the undoped values are @xmath178 for @xmath171 and @xmath179 for @xmath169 . ] besides the 2-spin terms in the two dimensional square lattice also allows for 4-spin interactions . the leading contribution is given by @xmath180 \label{def : j_ring } \end{aligned}\ ] ] which we sloppily call ring exchange although the complete ring exchange comprises also nearest - neighbor and diagonal 2-spin couplings @xcite . we do so since these 2-spin terms are accounted for by @xmath94 , @xmath95 , and @xmath175 in . the term describes the interaction of the four spins on a plaquette , see fig . [ rin ] , and it occurs first in order @xmath181 @xcite . its importance is discussed at length in the literature , see for instance . refs . and references therein . the magnetic excitations in planar cuprates may not be understood without considering ring exchange @xcite . . ] compared to other quartic exchange couplings such as @xmath95 or @xmath175 the ring exchange is much more important , see its values in fig . [ fig : j_ring_d ] . the ring exchange takes values of up to @xmath182 of @xmath94 . hence this term must not be neglected in an effective model . the ring exchange shown in fig . [ fig : j_ring_d ] displays nearly no dependence on the doping @xmath2 . even for doping as large as @xmath183 the change in the coefficient is less than @xmath184 percent . thus while ring exchange is an important process its doping dependence can safely be omitted . for various values of @xmath2 ( left panel ) and its doping dependence of @xmath185 ( right panel ) relative to the undoped values @xmath186 for @xmath171 and @xmath187 for @xmath169 . ] for various values of @xmath2 ( left panel ) and its doping dependence of @xmath185 ( right panel ) relative to the undoped values @xmath186 for @xmath171 and @xmath187 for @xmath169 . ] the second 4-spin term is the cross exchange @xmath188 in this term the spins are located on the same sites as for the ring exchange , but the inner products are taken of the diagonal spins . the corresponding coupling constant is shown in fig . [ kreuz ] . it takes smaller values than @xmath185 and hardly shows any doping dependence . as defined in eq . [ def : j_cross ] for various doping concentrations @xmath2 . ] the effective generalized @xmath0-@xmath1 model also contains interactions between dos . first , we consider the hubbard repulsion @xmath4 which determines the energy costs for the creation of a single do . so strictly speaking it does not represent a true interaction . since the deviations of the doped values of @xmath4 from the ones in the half - filled case are small we directly show the doped values relative to the half - filled ones in fig . [ fig : u_alle ] . this coupling constant shows nearly no dependence on the doping @xmath2 . hence the influence of doping on @xmath4 may be neglected . for various values of @xmath13 . the doped values are given relative to the undoped values of @xmath189 for @xmath171 and @xmath190 for @xmath169 . ] the following interaction terms are active only in the presence of at least two dos . thus these terms have to be seen as genuine 2-do interactions . among them density - density interactions of various distances appear . the density - density interaction between nearest neighbors reads @xmath191 where @xmath192 denotes the operator counting the number of electrons on a site compared to the average filling , cf . tab . [ table : list ] . at half - filling this term only contributes if site @xmath15 and site @xmath21 are either empty or truly doubly occupied . figure [ fig : v_alle ] only shows an increase in the coupling constant of about @xmath193 under the influence of doping for @xmath169 . . results are given relative to the values at zero doping @xmath194 for @xmath171 and @xmath195 for @xmath169 . ] a second type of interaction is correlated hopping . the most important term of this kind is the hopping of two electrons to a nearest neighbor site which is initially empty , see eq . [ eq : v_p ] . since the empty state also corresponds to a do , the effect of the term is to exchange the two dos . the results for various doping levels are depicted in fig . [ fig : v_p_t ] . as defined in eq . [ eq : v_p ] for various @xmath2 as a function of @xmath13 . ] besides the nearest neighbor pair interaction @xmath86 there are also pair interaction terms between three spins . one of these terms is the interaction of three spins on a plaquette which reads @xmath196 . \label{def : v_p_s}\end{aligned}\ ] ] the sites @xmath15 and @xmath21 are supposed to be diagonal neighbors with a common adjacent site @xmath99 . one possible process consists of the hopping of an electron from a singly occupied site @xmath21 to an empty site @xmath99 . simultaneously , an electron from the doubly occupied site @xmath15 hops to site @xmath99 forming a do on this site . the corresponding effective coupling constant @xmath197 is depicted in fig . [ fig : v_p_alle ] as function of doping . since this correlated hopping imposes an additional constraint on the state of site @xmath99 it is half as large as the nearest neighbor term @xmath86 . even for large values of @xmath13 the coupling constant is increased only by @xmath198 for large doping . on @xmath13 for various @xmath2 . in the undoped case @xmath199 takes the value @xmath200 for @xmath171 and @xmath201 for @xmath169 . ] the last class of terms considered are correlated hopping terms such as @xmath202 one of the processes described by @xmath203 is the hopping of an electron from a singly occupied site @xmath21 to an empty site @xmath15 under the condition that site @xmath99 is occupied by a do , see fig . sites @xmath15 and @xmath21 are diagonal neighbors on a plaquette and @xmath99 joint adjacent neighbor . ] due to the constraint that site @xmath99 has to be occupied by a do and site @xmath15 has to be empty in the beginning , these processes rely on the presence of two dos which justifies to view them as true interactions . the number of dos is not changed by this process . processes such as @xmath203 appear in second order of @xmath91 . the corresponding coupling constant is shown in the left panel of fig . [ fig : v_n_str_n ] as function of @xmath13 . in the right panel of fig . [ fig : v_n_str_n ] , the value for the coupling constant @xmath204 in the doped case is shown relative to its value in the half - filled case . over diagonal neighbors for various @xmath2 ( left panel ) and as function of @xmath2 for the values @xmath171 and @xmath169 ( right panel ) . the undoped values are @xmath205 ( @xmath171 ) and @xmath206 ( @xmath169 ) . ] over diagonal neighbors for various @xmath2 ( left panel ) and as function of @xmath2 for the values @xmath171 and @xmath169 ( right panel ) . the undoped values are @xmath205 ( @xmath171 ) and @xmath206 ( @xmath169 ) . ] for large values of @xmath13 , the coupling @xmath204 shows a noticeable dependence on @xmath2 . note that besides the correlated hopping defined in and illustrated in fig . [ sk ] , there is correlated hopping between three sites located on three sites in a row . the corresponding coupling constant @xmath207 shows the same behavior as @xmath204 so that we do not show it here for brevity . for too large values of @xmath13 , i.e. , @xmath208 , the curves for the coupling constants are not smooth anymore ( not shown here ) @xcite . this observation is explained by the breakdown of the mapping as it is indicated by the behavior of the apparent charge gap ( see sect . [ chap : gap ] ) . the first term to be considered is @xmath209 which was introduced before in ( [ to ] ) . the initial @xmath209 represents hopping processes by one lattice spacing without a change in the number of dos . the corresponding coupling constant @xmath55 is shown in the left panel of fig . [ fig : t_0_n ] relative to its unrenormalized value . for @xmath171 and @xmath210 for @xmath169 is depicted in the right panel . ] for @xmath171 and @xmath210 for @xmath169 is depicted in the right panel . ] the deviation of the coupling @xmath55 from its bare value is proportional to @xmath211 . it is increased upon increased doping . to examine the doping dependence the renormalized value of @xmath55 in the doped case is compared to the one in the half - filled case in the right panel of fig . [ fig : t_0_n ] . under the influence of doping the hopping parameter is increased linearly . but even for @xmath212 the parameter is changed only by a few percent . hopping also occurs between diagonal sites on a plaquette , for instance in @xmath213\ , . \label{def : t_s}\end{aligned}\ ] ] here the double bracket under the sum indicates next - nearest neighbors ( nnn ) on the square lattice . the same process also appears between third nearest neighbors with a distance of two lattice sites . the corresponding coupling constant is denoted by @xmath214 . since @xmath214 and @xmath215 show very similar behavior we only show the results for @xmath215 in fig . [ fig : t_str_n ] . between diagonal sites as function of @xmath13 for various values of @xmath2 ( left panel ) and its doping dependence for @xmath216 and @xmath212 ( right panel ) . in the undoped case @xmath215 takes the values @xmath217 ( @xmath171 ) and @xmath218 for @xmath169 . ] between diagonal sites as function of @xmath13 for various values of @xmath2 ( left panel ) and its doping dependence for @xmath216 and @xmath212 ( right panel ) . in the undoped case @xmath215 takes the values @xmath217 ( @xmath171 ) and @xmath218 for @xmath169 . ] the hopping @xmath215 decreases linearly for increasing ratio @xmath13 with slopes depending on the doping , see left panel of fig . [ fig : t_str_n ] . relative to its values at half - filling the decrease as function of doping hardly depends on @xmath13 , see right panel of fig . [ fig : t_str_n ] . it is remarkable that the constant is decreased to almost 0 for @xmath219 . this is actually the _ only _ significant dependence on doping that we found . but one has to keep in mind that the absolute values of @xmath215 are small . note that the sign of @xmath55 is positive whereas @xmath215 and @xmath214 are negative . an interesting coupling generated in second order of @xmath91 is the spin dependent hopping described by @xmath220\vec{s}_k\right ] \label{def : t_spin_s}\hspace*{1cm}\end{aligned}\ ] ] where the sum runs over two diagonal neighbors @xmath15 and @xmath21 which have a common nearest neighbor @xmath99 . the size of the corresponding hopping parameter @xmath221 , see fig . [ fig : t_spin_diag_n ] , is comparable to the spin independent parameter @xmath215 . this shows that the induced nnn spin dependent hopping processes are as important as the spin independent ones . this was first observed by reischl et al . @xcite at half - filling . for various values of @xmath2 ( left panel ) and its dependence on @xmath2 for two values of @xmath13 , namely @xmath216 ( @xmath222 ) and @xmath212 ( @xmath223 ) ( right panel ) . ] for various values of @xmath2 ( left panel ) and its dependence on @xmath2 for two values of @xmath13 , namely @xmath216 ( @xmath222 ) and @xmath212 ( @xmath223 ) ( right panel ) . ] upon doping the spin dependent hopping term is increased . but even for larger values of @xmath13 the value of @xmath221 is increased only by a few percent . the analogous process also appears between third nearest neighbors . the corresponding coupling constant behaves similar to @xmath221 so that we do not display it here . both processes concern three sites . thus it does not matter significantly whether the sites are aligned linearly or in a right angle on a plaquette . compared to the spin independent hopping @xmath215 the spin dependent hoppings @xmath221 and @xmath224 are not negligible . but all of them are fairly small compared to the bare nn hoping @xmath55 . thus one can either stick to a pure @xmath0-@xmath1 model or include more extended hopping terms . but if one opts for including hopping over two lattice spacings one should incorporate spin independent hopping @xmath215 as well as the spin dependent hoppings @xmath221 and @xmath224 . the doping dependence of the spin dependent hopping elements may be neglected . in contrast , the hopping element @xmath215 shows a rather strong dependence on the doping concentration @xmath2 . we presented a systematically controlled mapping of a fermionic hubbard model to a generalized @xmath0-@xmath1 model . the conceptual foundation of this mapping was analyzed carefully . in particular , we developed a scheme for this mapping which covers also the interesting case of substantial doping . remarkably , this issue had so far attracted only little attention . in the derivation of the generalized @xmath0-@xmath1 model we eliminate the charge fluctuations by self - similar continuous unitary transformations . processes that change the number of double occupancies are rotated away . thereby , we obtain the effective coupling constants as function of the doping @xmath2 and of the ratio @xmath13 where @xmath3 is the bandwidth and @xmath4 the local interaction . note that the generalized @xmath0-@xmath1 model comprises the magnetic degrees of freedom as well as the kinetics and the interactions of double occupancies . we extended the concept of the apparent charge gap @xmath12 @xcite from half - filling to the doped system . this gap is not the true physical gap but it measures the energy separation of subspaces with differing number of double occupancies _ irrespective _ of the spin state . we argue that as long as @xmath12 is finite the mapping to a @xmath0-@xmath1 model is justified . a vanishing @xmath12 indicates the breakdown of this mapping . by estimating the parameter where @xmath225 holds we derived a diagram of applicability of the @xmath0-@xmath1 model shown in fig . [ fig : appli ] . as expected the applicability is reduced upon doping @xmath2 . but it levels at intermediate values of doping so that the commonly assumed parameters for the description of high-@xmath158 cuprates lie within the range of applicability . to our knowledge , no such result was derived before . furthermore we find that the coupling constants of the effective model show hardly any doping dependence . the only coupling which exhibits a significant dependence on @xmath2 is the hopping parameter @xmath215 describing hopping between diagonal neighbors . relative to its value at half - filling @xmath215 exhibits a strong doping dependence . but the absolute value of this hopping element remains small . thus within a wide range of doping the @xmath0-@xmath1 model with constant coupling constants is appropriate . besides the usually considered terms , the 4-spin ring exchange on each plaquette should be included . technically we used recently developed types of infinitesimal generators for the continuous unitary transformation @xcite . they only decouple certain subspaces of the hilbert space which simplifies and accelerates the calculations . so far , the pc - generator was used which leads to a particle number conserving effective model ; the particles are the double occupancies @xcite . we extended the gs - generator introduced previously for the ground state of matrices @xcite and of many - body systems @xcite to mixed reference ensembles . the gs - generator is particularly suited to obtain the purely magnetic heisenberg model since it efficiently decouples the subspace of the reference ensemble from the remainder of the hilbert space . if , however , the dynamics of the double occupancies matters as well , the gs,1p - generator turned out to be a good compromise between efficiency and sufficient decoupling . this generator decouples the reference ensemble and the states with one double occupancy from the rest of the hilbert space . we found that the couplings derived from a faster gs,1p - calculation agree very well with the results from a slower pc - calculation . we expect that these generator or modifications of them will continue to play an important role in the systematic derivation of effective models . the present analysis for the square lattice can certainly be extended to other types of lattices such as the triangular lattice which has already been analysed by perturbative cuts @xcite , the honeycomb lattice @xcite , or more sophisticated lattices such as the kagom lattice and so on . in this way , the effects of subleading magnetic exchange processes such as ring exchange can be analysed quantitatively . another route to extend the present calculation is to also transform the observables , for instance the standard fermionic creation operator . at half - filling , one will then be able to compute the spectral weight in the upper hubbard band , that means in the subspace with one double occupancy . but there should be also weight in the subspaces with three and more double occupancies . to our knowledge , no estimate whatsoever exists for the weight in such trans - hubbard bands . more generally , the systematic derivation of effective models in other contexts can be tackled by adapting the ideas of the present work . for instance , the reliable downfolding of interacting fermionic models with many bands to models with a minimum number of bands and hubbard - type of interactions is a long standing field of research @xcite which continues to attract much attention , see for instance refs . . we think that continuous unitary transformations provide an promising approch to make systematic and controlled progress in this field .

we map the kinematics of the inner ( 200pc ) narrow - line region ( nlr ) of the seyfert2 galaxy ngc1068 using the instrument nifs and adaptative optics at the gemini north telescope . channel maps and position - velocity diagrams are presented at a spatial resolution ofpc and spectral resolution in the emission lines [ feii] m , h m and br . the [ feii ] emission line provides a better coverage of the nlr outflow than the previously used [ oiii] emission line , extending beyond the area of the bi - polar cone observed in br and [ oiii ] . this is mainly due to the contribution of the redshifted channels to the ne of the nucleus , supporting its origin in a partial ionized zone with additional contribution from shocks of the outflowing gas with the galactic disc . we modeled the kinematics and geometry of the [ feii ] emitting gas finding good agreement with the data for outflow models with conical and lemniscate ( or hourglass ) geometry . we calculate a mass outflow rate ofmyr but a power for the outflow of only l . the molecular ( h ) gas kinematics is completely distinct from that of [ feii ] and br , showing radial expansion in an off - centeredpc radius ring in the galaxy plane . the expansion velocity decelerates fromkms in the inner border of the ring to approximately zero at the outer border where our previous studies found a 10 myr stellar population . [ firstpage ] galaxies : individual : ngc1068 galaxies : active galaxies : kinematics and dynamics galaxies : nuclei galaxies : ism ism : jets and outflows

1408.4750

the kinematics of the narrow line region ( nlr ) of ngc1068 has been the subject of many studies over the years , the most recent and detailed being those of @xcite and @xcite , based on long - slit optical spectroscopy of the inner few hundred parsecs obtained with the space telescope imaging spectrograph aboard the hubble space telescope ( _ hst - stis _ ) . @xcite have shown that the nlr gas is outflowing in a hollow bi - cone , along which the gas seems to be accelerated up to 140pc from the nucleus and then decelerated . integral field optical spectroscopy of the inner 1.5kpc was obtained by @xcite using the instrument sauron at lower spatial resolution but covering a larger region of the galaxy than in @xcite . @xcite observed the outflow in [ oiii ] @xmath55007 and h@xmath14 , finding also streaming motions towards the nucleus and emission knots of h@xmath14 attributed to regions of star formation in the galaxy plane . @xcite presented optical integral field observations of the inner 400pc using the gemini multi - object spectrograph integral field unit ( gmos - ifu ) at a better spatial resolution ( @xmath1340pc ) which revealed , besides high - velocity gas emission from outflowing gas , also some emission from the gas in the galaxy disc . in the near - infrared ( hereafter near - ir ) , @xcite and @xcite have mapped the gas kinematics via integral field observations using the instrument sinfoni at the very large telescope ( vlt ) . while @xcite focus on the h@xmath11 kinematics in the inner few tens of pc which revealed inflows , @xcite focus on the br@xmath4 and coronal lines kinematics , which are dominated by outflows . in @xcite ( hereafter paper i ) we have used the gemini near - infrared integral field spectrograph ( nifs ) on the gemini north telescope operating with the adaptive optics module altair to map the flux distributions and excitation of the gas within the inner @xmath15pc , while in @xcite we have studied the stellar population . in this paper , we present detailed measurements of the gas kinematics observed in the near - ir emission lines [ feii ] @xmath2 m , h@xmath3 m and br@xmath4 using channel maps and position velocity diagrams . we concluded this is the best way to separate the multiple kinematic components which are present in the emitting gas in this galaxy . the [ feii ] gas kinematics shows a broader " outflow than previously observed in [ oiii ] and in the near - ir coronal lines , being similar to that observed in planetary nebulae and consistent with an origin in an accretion disc wind @xcite or a torus wind @xcite . we test three models with different geometries in order to reproduce the outflow . throughout this paper we adopt a distance of 14.4mpc to ngc1068 which correspond to a scale of 72pcarcsec@xmath8 at the galaxy . this paper is organized as follows . in sec . 2 we present the observations , in sec . 3 we discuss previous studies relevant to the present work , in sec . 4 we present the gas kinematics ( channel maps and _ p - v _ diagrams ) , in sec . 5 we give our interpretation to the kinematics of the [ feii ] , br@xmath4 and h@xmath11 , in sec . 6 we perform the modeling and analysis , in sec . 7 we calculate the mass outflow rate and corresponding power and in sec . 8 we present our conclusions . the spectroscopic data was obtained at the gemini north telescope , with nifs ( see @xcite ) , operating with the altair adaptive optics module . the data were obtained on the nights of november 26 and december 03 and 09 , 2008 , and in the present paper we have used spectra obtained in the h and k spectral bands at a resolving power of 5290 . nifs has a square field of view of 3 @xmath16 3 and was set to a position angle of 300 degrees for the observations , resulting in a spatial sampling of 0.044 arcsec perpendicular to the jet ( vertical direction in our images ) and 0.103 arcsec along it ( horizontal direction in our images ) . the angular resolution of the data , measured as the fwhm of the spatial profile of a star is @xmath17 at the h and k bands , corresponding to @xmath18pc at the galaxy . dithering was applied in order to cover a field - of - view of @xmath0 5 @xmath16 5 . additional details about the data can be found in paper i . on the basis of _ hst - stis _ spectra , @xcite proposed a kinematic model in which the nlr has the geometry of a hollow bicone oriented at pa of @xmath19 as illustrated in figure [ fig - das - model ] . this figure was adapted from their figure 10 and rotated to the same orientation as our data . we have identified in this figure the line of nodes ( ln ) , minor axis ( ma ) and cone axis ( ca ) as derived by @xcite . the walls of the cone corresponding to the areas intersecting perpendicularly the plane of the sky were also labeled in this figure as n , e , s and w walls . the galactic plane is inclined by @xmath20 such that the far side of the galaxy is the ne @xcite . the ne side of the cone is in front of the galactic disc , while the sw side is behind and apparently obscured by the galactic disc . the bicone axis is tilted by @xmath21 relative to the plane of the sky such that the ne side is oriented ( slightly ) toward us , while the sw side is oriented away from us . the cone is hollow but has thick " walls , corresponding to inner and outer half - opening angles of @xmath22 and @xmath20 . the nlr gas flows away from the nucleus along the cone , reaching a maximum velocity of 2000 km s@xmath8 at 140pc from the nucleus . we present the integrated flux maps from paper i in the lines [ feii]@xmath2 m , br@xmath4 and h@xmath3 m , in figures [ fig - flux - map - fe ] , [ fig - flux - map - bra ] and [ fig - flux - map - h2 ] , respectively . in fig . [ fig - flux - map - fe ] we identify the n , e , s and w walls of the bicone . in fig . [ fig - flux - map - bra ] the features indicated by arrows correspond to structures a , d , f and g previously identified in the [ o iii ] map by @xcite . in fig . [ fig - flux - map - h2 ] the structures identified by @xcite as _ northern tongue _ ( nt ) and _ southern tongue _ ( st ) are also indicated by arrows . the main feature seen in the h@xmath11 flux map is a ring which has its centre offset from the position of the nucleus by @xmath23 to the sw @xcite . we adopt as the position of the nucleus the location of the maximum of the k - band continuum emission . @xmath2 m emission line . note the hourglass ( rather than a conical ) shape of this emission . the north , south , east and west walls are indicated by the labels . spatial units are in arcsec and the position of the nucleus @xmath24 is indicated by a plus sign . the orientation is indicated by the arrows . contours are used to best delineate the structures . ] details are the same as in fig . [ fig - flux - map - fe ] , otherwise indicated bellow . the letters a , d , f and g label structures corresponding to those previously identified in the [ oiii ] flux map from @xcite . ] @xmath25 m line . details are the same as in fig . [ fig - flux - map - fe ] , otherwise indicated bellow . nt and st identify the northern and southern tongues " from @xcite . ] in a previous study , @xcite has used the same nifs data we used in the present paper to map the gas kinematics of the coronal line region , which is consistent with an origin of the emission in the inner part of the nlr . here we present and discuss the kinematics of the low ionization lines of [ feii ] @xmath2 m , h@xmath3 m and br@xmath4 , using velocity channel maps and position - velocity ( hereafter _ p - v _ ) diagrams . these two types of maps are two - dimensional cuts through the 3d integral field data : the channel maps are cuts across the wavelength ( velocity ) axis and the _ p - v _ diagrams are cuts across one of the position axes . the channel maps are presented in figures [ fig - channel - map - fe ] to [ fig - channel - map - h2 ] and show emission line flux distributions in velocity channels along the line profile from blueshifts to redshifts after the subtraction of the systemic velocity of the galaxy , adopted as @xmath26 km s@xmath8 @xcite . each velocity channel integrates 3 pixels in the velocity axis corresponding to 87.6 km s@xmath8 for [ feii ] and 88.7 km s@xmath8 for br@xmath4 and 2 pixels in h@xmath11 corresponding to 60.3 km s@xmath8 . the central velocity @xmath27 is written at the bottom of each panel . the central velocities were chosen so that the velocity range covers all the line profile from blueshifts to redshifts . in all panels we draw an auxiliary dashed gray line indicating the orientation of the galactic line of nodes adopted as 80 , as obtained from the observation of the stellar kinematics of the inner regions by @xcite . the n and e arrows show the orientation of our images in the sky and the label `` far side '' indicates the far side of the galactic disc . the _ p - v _ diagrams are presented in figures [ fig - pv - fe ] and [ fig - pv - h2 ] for [ feii ] and h@xmath11 , respectively . the bottom right panel depicts the integrated flux map of the line showing the slits of the _ p - v _ diagrams running horizontally in the figure . in the _ p - v _ panels the velocity runs along the horizontal axis ( pa @xmath28 ) and the position along the vertical axis with ne at the bottom ( negative positions ) and sw at the top ( positive positions ) . in the top left of each _ p - v _ panel the central position of the slits is written in arcsecs . in the next subsections we describe the channel maps and _ p - v _ diagrams for the ionized and molecular gas as well as our interpretation . the [ feii ] channel maps are shown in fig . [ fig - channel - map - fe ] , where we see emission in velocity channels from @xmath29 km s@xmath8 to @xmath30 km s@xmath8 . in fact we found [ feii ] emission in velocity channels ranging from about @xmath31 km s@xmath8 to @xmath32 km s@xmath8 . the channels not shown in the figure were omitted because they present a structure that does not change much from what is seen in the first and last channels . the emission seems to extend beyond the instrument field of view in the majority of the panels . the central and redshifted panels delineate a bowl " shape which can also be described as similar to an hourglass extending along the horizontal direction in our images , from ne to sw with axis along pa @xmath33 . this hourglass seems to correspond to the walls of the bipolar outflow modelled by @xcite with a hollow cone shape . the walls of this structure are oriented approximately along n and e to one side of the nucleus and along s and w to the other side . the highest blueshift structures detached from the nucleus are observed in the ne outflow up to a velocity channel centered at @xmath34 km s@xmath8 ( not presented in the figure ) . the n wall of the hourglass is observed covering a range of blueshifts and redshifts , while the e wall is observed mostly in redshift . the n and e walls of the hourglass do not keep the same pa and structure over the velocity channels due to the patchy nature of the emitting gas and also because of the hourglass geometry and its relative orientation to the line of sight . the opening angle of the walls is widest ( @xmath12@xmath35 ) at the low redshift channels . the s wall is observed mostly in blueshift , while the w wall is the faintest , observed in the lowest velocity channels . the most luminous part of the outflow is the fan - like " structure observed to the ne in the redshift channels , corresponding to the back wall of the ne outflow . radio merlin 18 cm , vla 6 cm and merlin 6 cm contours are superimposed , respectively , to the panels with velocities @xmath36 km s@xmath8 , @xmath37 km s@xmath8 and @xmath38 km s@xmath8 . the merlin 18 cm emission fills the region between the n and e walls of the hourglass and also extends towards sw partially filling that side of the hourglass . the merlin 6 cm is very compact and extends towards ne at pa @xmath33 along the axis of the hourglass . the vla 6 cm radio image is more extended than the merlin 6 cm . its emission partially fills the ne side of the hourglass and has a detached patch of emission to the sw . in the velocity channel map at @xmath39 km s@xmath8 , the contours of the _ hst _ [ oiii ] integrated flux image are shown . the emission covers most of the field and extends toward ne along pa @xmath33 . there is good correspondence between [ oiii ] and [ feii ] flux distributions within 1 arcsec from the nucleus between n and ne at the highest flux levels . at lower flux levels the correspondence is poorer and in the [ oiii ] contours we do not see the `` bowl''shape seen in the [ feii ] flux distributions of the redshift channels . [ fig - pv - fe ] presents the [ feii ] _ p - v _ diagrams in panels corresponding to adjacent slits 013 wide . the lower right panel of the figure presents the integrated [ feii ] emission line flux map . despite our efforts to clean all artifacts from the data , regularly spaced vertical lines resulting from the continuum subtraction appear in velocities @xmath40 km s@xmath8 , @xmath41 km s@xmath8 , @xmath42 km s@xmath8 and @xmath43 km s@xmath8 mainly in the central panels and should not be confused with real structures by the reader . the _ p - v _ flux distribution is dominated by emission from the ne side of the outflow in most panels ( lower half of the panels ) although weak emission from the galactic disc is also seen running vertically at @xmath44 km s@xmath8 . the most prominent feature is an inverted `` v''-shaped structure which appears in slits ranging from 108 to 000 therefore coming mainly from the n side of the hourglass structure . this structure is indicated in the panel corresponding to the slit @xmath45 by solid straight lines and the label `` inv . we observe that this feature is asymmetric with the blueshift side reaching almost twice as much velocity as the redshift side for equivalent distance from the nucleus . we have also identified in fig . [ fig - pv - fe ] the emission coming from the w wall observed in the slits 094 to 013 ( label `` w wall '' ) , and that coming from the s wall observed in the slit positions 000 to -081 ( label `` s wall '' ) . in the slits corresponding to negative positions up to @xmath46 we see the kinematic signatures of many individual high velocity dispersion clouds ( labeled hvdc ) identifiable as horizontally elongated emitting areas . the velocity dispersion of these structures range from @xmath47 km s@xmath8 to @xmath48 km s@xmath8 . each cloud appears in two or three adjacent panels indicating they have angular sizes ranging from 02 to 04 . these clouds originate in the front wall of the ne side of the hourglass . the br@xmath4 channel maps are shown in fig . [ fig - channel - map - bra ] , where we see emission in velocity channels ranging from @xmath49 km s@xmath8 to @xmath50 km s@xmath8 . we , however , found br@xmath4 emission in velocity channels ranging from about @xmath51 km s@xmath8 to redshifts beyond @xmath52 km s@xmath8 . the omitted channels beyond redshift @xmath50 km s@xmath8 show only the central component getting fainter with redshift and the channel @xmath51 km s@xmath8 shows a faint similar structure to that seen in the channel @xmath49 km s@xmath8 . the emission structure is more compact than that seen in the [ feii ] channels and is composed mainly by emission to the n - ne seen in both blueshifted and redshifted channels and to the sw in channels from @xmath53 km s@xmath8 to @xmath50 km s@xmath8 . the contours of the _ hst _ [ oiii ] line emission are overplotted on four channels corresponding to the velocities @xmath54 km s@xmath8 , @xmath55 km s@xmath8 , @xmath56 km s@xmath8 and @xmath57 km s@xmath8 . these are the same contours as those overplotted in the [ feii ] channel maps . the correlation between the [ oiii ] contours and br@xmath4 flux distribution is better than that between [ oiii ] and [ feii ] . almost every br@xmath4 feature ( covering the whole velocity range ) corresponds to an [ oiii ] feature . in particular we identify ( from * * ) in fig . [ fig - channel - map - bra ] the components a , c , d , f and g which are observed in blueshift , component e observed in blueshift and redshift and component b , the nuclear component , observed in all velocity channels . we have also identified some new structures in the sw side of the bipolar outflow which we call b1 , r1 and r2 . b1 is observed in the velocity channels from @xmath58 km s@xmath8 to @xmath59 km s@xmath8 as an area of faint br@xmath4 emission indicated in panel of @xmath60 km s@xmath8 . this area has a se - nw extension of about @xmath61 and is observed at @xmath62 arcsec sw of the nucleus , although it seems to extend to sw beyond the field of view . we attribute this emission to the front wall of the sw outflow . r1 and r2 are two redshift components ( shown in panels of @xmath63 km s@xmath8 and @xmath50 km s@xmath8 respectively ) . r1 is observed in the velocity range @xmath64 km s@xmath8 to @xmath65 km s@xmath8 at @xmath66 sw of the nucleus at pa @xmath67 and r2 in the velocity range @xmath68 km s@xmath8 to @xmath69 km s@xmath8 distant from the nucleus by @xmath66 at pa @xmath70 . we atribute the origin of r1 and r2 to the back wall of the sw outflow . direct comparison of the [ oiii ] contours to the br@xmath4 flux map at the positions of r1 and r2 reveals a correspondence of [ oiii ] emission to the br@xmath4 components . these redshift components are also observed in [ feii ] in channels from @xmath36 km s@xmath8 to @xmath71 km s@xmath8 for r1 and in channels from @xmath38 km s@xmath8 to @xmath71 km s@xmath8 for r2 . r1 and r2 are , respectively , in vertical coordinates @xmath72 and @xmath73 arcsec approximately and , therefore , are separated by @xmath74 arcsec . the ionized gas kinematics has been previously modeled by @xcite as a bi - conical outflow with axis at pa @xmath28 in which the front walls produce the blueshifted emission and back walls , the redshifted emission . their model was fitted to [ oiii ] kinematic data measured using hst stis long - slit spectroscopy . [ fig - channel - map - bra ] shows that the flux distribution in [ oiii ] ( contours ) is very similar to that of br@xmath4 and so we assume that their emission originate in the same gas and consequently the kinematics of the [ oiii ] and br@xmath4 should also be similar . the blueshift channels correspond to the front wall of the outflow and the redshift channels correspond to the back wall . from this figure it is not clear that the geometry of the outflow is conical as derived by @xcite . in our data the brightest emission comes , from a narrow region extending from the nucleus to the n - ne , being stronger in the blueshift channels . this emission comes from the front wall of the outflow which is at higher galactic latitude than the emission from the back wall , that is observed in redshift . additionally , the emission coming from the sw side of the outflow is observed in redshift corresponding to emission from the back wall of the outflow which is at higher galactic latitude than that of the front wall . @xcite derived the conical geometry for the outflow from the kinematics that they could obtain from deep stis long - slit spectroscopy of the whole region , not from the flux maps . the approximately conical geometry is better seen in the [ feii ] channel maps ( fig . [ fig - channel - map - fe ] ) , and in particular , in the redshift channels . while the br@xmath4 emission distribution is concentrated and mostly directed towards north , the [ feii ] emission distribution encompasses it and extends further spanning a wide hourglass shaped area more clearly seen to ne of the nucleus between pa @xmath75 and @xmath76 in the redshift channels . the difference between [ feii ] and br@xmath4 flux distribution and kinematics can be understood as follows . as discussed by @xcite , most of the [ fe ii ] emission comes from a region where h is only partially ionized . in the completely ionized region , as the ionization potential of fe+ is just 16.2ev , fe+ is further ionized and there is not much [ feii ] emission . extensive partially ionized zones around the nlr of seyfert galaxies are created by x - rays that have small cross sections and penetrate deeper in the gas than less energetic photons , heating the gas . the [ feii]@xmath2 m emission is excited in these partially ionized zones by collisions with electrons , as well as by shocks that may be present due to the interaction of agn jets and winds with the ambient gas . the extent of the partially ionized region determines where we may find [ feii ] emission . the larger extent of the [ feii ] emitting region compared to that of the br@xmath4 can thus be understood as caused by the existence of a partially ionized zone extending beyond the fully ionized hydrogen zone where br@xmath4 is produced . the enhancement of the [ feii ] emission to the ne in the redshift channels when compared with the br@xmath4 emission can also be attributed to the interaction of the back wall of the outflow with the far side of the galaxy disc . the shocks produced in this interaction could provide the energy for the excitation of [ feii ] , besides some contribution from collisions with electrons in the gas heated by x - rays from the agn . support for the presence of shocks is provided by line ratio maps presented in paper i. towards the borders of the line flux distributions , in particular to the ne border , there is an increase of the [ feii]/pa@xmath14 line ratio to values up to @xmath124 , consistent with shock excitation . we conclude that the [ feii ] emission can be observed beyond the region where [ oiii ] and h recombination lines can be observed , providing a better coverage of the outflow . the h@xmath11 emission is detectable mainly in the velocity range @xmath77 to @xmath78 km s@xmath8 as can be seen in fig . [ fig - channel - map - h2 ] . the distribution of h@xmath11 flux emission along the velocity channels is very different from that of the [ feii ] and br@xmath4 , indicating that they are tracing distinct gas components and processes . the strongest emission is detected in the central channels ( around zero velocity ) and exhibits a ring like morphology ( see paper i and @xcite ) . the emission of the ring in redshift appears in the velocity channels from @xmath79 km s@xmath8 to @xmath80 km s@xmath8 and it comes from the n half of the ring while in the blueshift channels from @xmath41 km s@xmath8 to @xmath81 km s@xmath8 the emission comes from the s half of the ring . in the highest blueshift velocity channel we observe nuclear emission as well as a clump of faint emission distant from the nucleus by @xmath82 at pa @xmath83 . the clump is the tip of an elongated emission structure seen in the channels with velocities @xmath84 km s@xmath8 to @xmath79 km s@xmath8 , which include the _ northern tongue _ ( nt as in @xcite ) most clearly seen from channel @xmath85 km s@xmath8 to channel @xmath86 km s@xmath8 . the _ southern tongue _ ( st as in @xcite ) is identified in channel @xmath86 km s@xmath8 and observable in channels from @xmath87 km s@xmath8 to @xmath79 km s@xmath8 . contours of the _ hst _ [ oiii ] line emission are overplotted on the velocity channel @xmath88 km s@xmath8 . a comparison of the [ oiii ] contours to the h@xmath11 maps shows that the h@xmath11 emission knots extending to the n of the nucleus in the blueshift channels also correspond to emission knots in [ oiii ] although the most extended emission is not correlated . from channel @xmath88 km s@xmath8 to @xmath89 km s@xmath8 we observe the brightest structure , indicated in the channel @xmath90 km s@xmath8 as bs ( for bright spot ) . this structure appears in our data blended to the ring at ne . in the highest redshift velocity channel we find faint nuclear emission as well as a clump of emission distant from the nucleus by @xmath91 at pa @xmath92 this redshift clump is identified as hvc ( for high velocity cloud ) in the velocity channel @xmath93 km s@xmath8 and is observed in channels from @xmath94 km s@xmath8 to @xmath78 km s@xmath8 . this clump is almost symmetrical to the clump seen in the highest blueshift channels suggesting they could be redshift and blueshift counterparts of the same gas ejection event . the h@xmath11 _ p - v _ diagrams are shown in fig . [ fig - pv - h2 ] . the integrated h@xmath11 emission line flux map is shown in the lower right panel . the most conspicuous structures in the integrated flux map are the nucleus , the ring the ne side being its brightest area , the northern and southern tongues ( nt and st ) , a bright area to the e of the ring which we call bright spot ( bs ) and an isolated cloud inside the ring , which we call high velocity cloud ( hvc ) . it is interesting to note that neither the nucleus nor the hcv are at the centre of the ring . we describe bellow the structures seen in each _ p - v _ panel , from the bottom slits ( se ) to the top slits ( nw ) . in the panel @xmath95 , there is a high velocity dispersion ( hereafter @xmath96 ) knot at position @xmath97 with velocities ranging from @xmath98 km s@xmath8 to @xmath99 km s@xmath8 . the remaining emission in this slit comes from the galactic plane and show velocities around zero , with @xmath100 km s@xmath8 . the subsequent slits @xmath101 and @xmath102 show similar structures . from @xmath103 to @xmath104 the bright knot of emission at @xmath105 corresponds to the bs , showing velocities from zero to @xmath15 km s@xmath8 . at positive positions , the slits @xmath103 and @xmath106 are tangent to the se side of the ring and its emission has negative velocities reaching up to @xmath107 km s@xmath8 . the emission in this area has a bow figure : around position @xmath108 , the velocities are more negative than around position @xmath109 and @xmath110 meaning that as one moves from the inner to the outer border of the ring the velocities change from blueshifts to zero . in the subsequent slits from @xmath111 to @xmath112 , the sw side of the ring shows velocities ranging from @xmath113 km s@xmath8 at the inner border to zero at the outer border . in the slits @xmath114 and @xmath115 we observe the hvc with a velocity of @xmath116 km s@xmath8 and @xmath117 km s@xmath8 . in the slits from @xmath115 to @xmath118 the ne side of the ring shows a similar effect to that described above : a velocity gradient across the ring , with the inner part of the ring at position @xmath119 showing redshifts of up to @xmath120 km s@xmath8 and the outer part of the ring showing zero velocities . in the slits @xmath121 to @xmath122 the nt appears at position @xmath123 with blueshifts of up to @xmath124 km s@xmath8 , and seems to be a structure detached from the ring and from the nucleus . finally , the slits @xmath118 to @xmath125 intercept tangentially the nw side of the ring , showing redshifts of @xmath126 km s@xmath8 in the inner border of the ring and velocities close to zero in the outer border of the ring . in summary , a general trend is that the ne side of the ring shows mostly redshifts , while the sw side ( which is less luminous ) shows blueshifts indicating expansion of the ring . but we , in addition , observe a velocity gradient from velocities in the range @xmath127 @xmath128 km s@xmath8 observed in the inner border of the ring to approximately zero velocities in the outer border of the ring . the h@xmath11 kinematics is completely distinct from that of the ionized gas and implies an origin for the h@xmath11 in an expanding ring in the galaxy plane while the ionized gas is observed in outflow and extends to high latitudes . this decoupling between molecular and ionized gas is a result commonly found in previous studies of the nlr of active galaxies by our group agnifs @xcite . although the morphology of the h@xmath11 ring suggests it is a circular ring in the plane of the galaxy ( paper i ) , its kinematics does not suggest rotation . in the redshift channels from @xmath79 km s@xmath8 to @xmath80 km s@xmath8 the emission comes mostly from the n half of the ring which is in the far side of the galaxy while in the blueshifted channels from @xmath41 km s@xmath8 to @xmath81 km s@xmath8 the emission comes from the s half of the ring which is in the near side of the galaxy . assuming that the ring is in the galactic plane , this kinematics is that of a ring expanding radially at an average velocity of @xmath13100kms@xmath8 . this conclusion can also be reached via the analysis of the _ p - v _ diagrams ( sec.[pv_h2 ] ) , in which the ne side of the ring shows mostly redshifts , while the sw side shows blueshifts . for an orientation of the kinematic line of nodes of @xmath129 , these results indicate expansion of ring in the plane . the ring has been also observed in co millimeter emission lines by @xcite who also concluded that it is expanding . but in the _ p - v _ diagrams we have observed something else : a velocity gradient across the ring , in which velocities of up to @xmath128 km s@xmath8 are observed in the inner border of the ring ( which are positive in the far side and negative in the near side ) and velocities close to zero in the outer border of the ring . this result indicates that the expansion in the ring has negative acceleration . this deceleration , showing that the outer border of the ring is not expanding anymore allows to reconcile our previous finding in @xcite , that the h@xmath11 ring is also the site of a young ( age @xmath13 10myr ) stellar population , with the ring kinematics . if the ring were expanding at constant velocity of @xmath12100 km s@xmath8 , as previously thought , the 100pc ring would have moved away by much more than 100pc in 10@xmath130yr , and there should be no association anymore between the h@xmath11 ring and star formation originated in this same gas . hydrodynamic jet simulations by @xcite support the formation of rings of young stellar population in the galaxy disc as a result of shock waves created during the early phases of the jet activity when the jet is still contained inside the galactic disc . they predict that these rings should have a radius of @xmath12 100 pc . the ring is formed by the expansion of gas pushed by the shock waves produced by the jet which eventually stop and pile up leading to the formation of the young stars . the h@xmath11 ring we detected in correspondence to the ring of young stellar population @xcite has the right size and kinematics predicted by the model of @xcite . only that the ring is not centered at the nucleus , as predicted by the model . one possibility to explain this is the presence of inhomogeneities in the gas density and distribution around the nucleus that could lead to the off - centered appearance after the expansion . @xcite modeled the h@xmath11 kinematics in the inner tens of pc s and proposed that there are two streams of molecular gas towards the nucleus ( indicated by nt and st in fig . [ fig - channel - map - h2 ] ) . in their model the st is in front of the central source obscuring it and flowing into it . in their measurements the nt has an almost constant velocity of @xmath131 km s@xmath8 and the st shows an increase in velocity from @xmath132 km s@xmath8 at @xmath133 ( distance from the nucleus ) to @xmath134 km s@xmath8 at @xmath135 . in our data the st has redshift velocities spanning the range @xmath136 to @xmath128 km s@xmath8 as can be seen in fig . [ fig - pv - h2 ] in slits @xmath137 and @xmath138 arcsec . the nt has blueshift velocities in the range @xmath136 to @xmath139 km s@xmath8 as seen in the slits @xmath140 , @xmath141 and @xmath142 arcsec . these velocities seem incompatible with those of @xcite . the analysis of our data led us to propose an alternative scenario for the dynamics of the h@xmath11 nt and st structures . the nt gas emission is projected in the plane of the sky co - spatially to the brightest br@xmath4 emitting structure which is seen in the channels @xmath59 km s@xmath8 to @xmath143 km s@xmath8 and is interpreted as gas in outflow at high galactic latitude . in order to survive , the molecular gas in the nt must be shielded from the ionizing radiation . we propose that the ionized gas pushes molecular gas which is thus also observed in outflow . the same scenario is also proposed for the st with the difference that this component surrounds the ionized gas in redshift which consequently produces h@xmath11 velocities in the redshift domain . additionally we note that in our data ( as well as in those from @xcite ) the nt is brighter than the st suggesting that the nt is in front of the galactic disc and the st is behind the galactic disc . in above scenario the h@xmath11 emission is not restricted to the galaxy plane . we show in figure [ fig h2+feii ] a comparison between the h@xmath11 ( green ) and [ feii ] ( red ) kinematics via their channel maps . in the channels @xmath144 km s@xmath8 and @xmath145 km s@xmath8 there is a h@xmath11 structure to the n which does not seem to be in the galaxy plane . this structure seems to connect to the nt at the channels @xmath146 km s@xmath8 and @xmath147 km s@xmath8 supporting that the nt originates in gas that is not in the plane . in fact in the channel @xmath147 km s@xmath8 the yellow color indicate that the nt is superimposed to the n wall of the [ feii ] outflow . other superposition regions ( yellow color ) between h@xmath11 and [ feii ] appear in the bs area in channels @xmath148 km s@xmath8 and @xmath149 km s@xmath8 which suggest that the enhanced h@xmath11 emission there is related to the outflow observed in [ feii ] . another high velocity structure which does not seem to be in the plane is the hvc seen from channels @xmath150 km s@xmath8 and @xmath151 km s@xmath8 . this structure seems to be associated to the r1 which we interpret as part of the back wall of the sw side of the outflow ( see [ subsec - ionized - gas ] ) . the [ feii ] integrated flux map presents an emitting structure with the shape similar to an hourglass with axis along the direction ne to sw . the ne side is brighter and more extended than the sw side . @xcite , using _ hst / stis _ long slit spectroscopic [ oiii ] data proposed a kinematic model in which a hollow conical outflow first accelerates from zero to 2000 km s@xmath8 then decelerates to zero . the radial distance from the nucleus where the velocity reaches its maximum value is 140 pc and where the velocity returns to zero is 400 pc . the inner and outer cone aperture angles are , respectively , 20@xmath152 and 40@xmath152 . the pa and the inclination of the cone axis relative to the plane of the sky are 30@xmath152 and 5@xmath152 , respectively . we have implemented this model and compared its predictions to our [ feii ] kinematics . the deviations from the model predictions to the [ feii ] kinematics led us to try distinct models but having it as a starting point . in figure [ fig - flux - mod-24 ] we present the [ feii ] integrated flux map compared to the flux map obtained for a hollow conical outflow model similar to that of @xcite . the model shows rough correspondence to the data but the opening angle of the cone is smaller than the opening angle we see in the data . we also note that the flux map shows a change of the opening angle with distance from the nucleus . integrated flux map ( hot colours ) and hollow conical outflow model ( grey levels ) . , scaledwidth=50.0% ] the conical model shown here is a modified version of the @xcite model in which the maximum velocity is lower , namely 1000 km s@xmath8 instead of 2000 km s@xmath8 as it provided a better fit to our data . in our data we do not see a decrease in velocity within our field - of - view therefore we further modified the model to keep the velocity at the maximum value up to the borders of the field . besides these changes in velocity domain , the other parameters of their cone model were kept the same . summarizing , the model consists of an accelerated hollow conical outflow with internal and external half - opening angles @xmath153 and @xmath154 respectively , inclination and position angle of the bicone axis , respectively , @xmath155 ( northeast is closer ) and pa@xmath156 , the radius where the maximum velocity occurs @xmath157 pc and the maximum velocity @xmath158 km s@xmath8 . the comparison between the [ feii ] and the model is shown in fig . 1 of the appendix ( available online as a supplementary material ) . comparison of the model with the data shows good correspondence in low velocity redshift channels to the ne . the highest blueshift channels seem to show poor agreement but this may be due to the fact that the actual outflow is patchy , in particular in the regions corresponding to the front wall of the ne side of the outflow . to the sw , most of the outflow is either also very patchy or hidden behind the galactic disc . the _ p - v _ diagrams of the hollow conical outflow model are compared to the [ feii ] data in fig . 2 of the appendix ( available online as a supplementary material ) . the model shows good agreement with the data mainly in the redshift side of the ne emission ( bottom right of the diagrams ) . p - v _ structures to the sw side show poor agreement but , as pointed out above , this is due to the patchy nature of the outflow and because the sw side is partially hidden by the galactic disc . in order to address the morphological differences between the data and the conical model and to try to meet the broader shape of the outflow at its base we next implemented a hollow spherical outflow model with constant velocity . the model consists of two thick spherical cavities with centres along the outflow axis separated by a distance @xmath159 ( where @xmath160 is the radius of the sphere to the ne and @xmath161 is the radius to the sw ) so that the spheres touch each other at the nucleus . the outflow is truncated at the distance of one radius from the nucleus to both sides so that only the hemispheres close to the nucleus compose the model . the thickness of the spherical walls was set to @xmath162 . the flux map of the data and model is shown in figure [ fig - flux - mod-15 ] . the spherical model has radii @xmath163 pc and @xmath164 pc , maximum velocity @xmath165 km s@xmath8 , inclination and position angle of the symmetry axis @xmath155 ( northeast is closer ) and pa@xmath156 , respectively . integrated flux map ( hot colours ) and hollow spherical outflow model ( grey levels ) . , scaledwidth=50.0% ] the spherical model did not fit well the data for any combination of the parameter values . this can be seen in fig . 3 of the appendix ( available online as a supplementary material ) . p - v _ diagrams of the hollow spherical outflow model are presented in fig . 4 ( only for a few slits ) of the appendix ( available online as a supplementary material ) . we conclude that this model is not a good fit to the data . with the spherical model not being able to reproduce the data we used a shape which is `` intermediate '' between a conical and a spherical shape , with conical apex ( at nucleus ) and changing opening angle as a function of distance to the nucleus . we used the shape of a lemniscate ( plot in figure [ fig - lemniscate ] ) , defined by the expression : @xmath166 the parameter @xmath167 is the horizontal length of the lemniscate . solving eq . [ eq - lemniscate ] for @xmath168 in the domain @xmath169 we obtain @xmath170 this equation draws the right side of the lemniscate ( solid black line in fig . [ fig - lemniscate ] ) and it is straightforward to verify that its opening angle at the origin is always @xmath76 . in order to construct a physical model with this shape we must be able to define a wall with finite width . a first approach to accomplish this is to define internal and external limits using lemniscates with different sizes and populating the model in between . the problem in using lemniscates with different sizes is that they always have @xmath76 opening angles at the origin and , therefore , close to the nucleus the wall would have near zero width ( see fig . [ fig - lemniscate ] ) which , in turn , gives a region with no data . we , then , modified eq . [ eq - lemniscate - y ] to allow to set the desired opening angle multiplying the whole equation by @xmath171 : @xmath172 where @xmath173 is the desired opening angle and @xmath174 is the new function for the lemniscate . the new generated geometry is that of a modified lemniscate which has improved properties relative to the conical and spherical models : an adjustable opening angle at the apex which decreases as a function of distance to the nucleus . in the velocity domain , this model consists of an acceleration region close to the nucleus , as in the conical model , where the velocity increases from zero at the nucleus to the maximum value @xmath175 at @xmath176 followed by a region of constant velocity where the velocity is kept the same up to the limits of the field . the adopted parameters and their values are : the size of the internal and external lemniscates which are , respectively @xmath177 pc and @xmath178 pc , the internal and external half - opening angles @xmath179 and @xmath180 respectively , the inclination of the lemniscate axis @xmath155 ( northeast is closer ) , the position angle of the lemniscate axis pa@xmath156 , the radius where the maximum velocity occurs @xmath157 pc and the maximum velocity @xmath181 km s@xmath8 . reduced by @xmath182 per cent . the dashed gray line ( seen only at left ) is a modified lemniscate with opening angle equal to @xmath183 and scale parameter @xmath167 reduced by @xmath182 per cent . ] the integrated flux map of the lemniscate model is shown in figure [ fig - flux - mod-27 ] overlaid to the [ feii ] data . analysing only the morphological and geometrical aspects , the lemniscate model shows some improvement over the conical model . integrated flux map ( hot colours ) and hollow lemniscate outflow model ( grey levels ) . ] in figure [ fig - channel - mod-27 ] we present the channel maps for the data ( hot colours ) and the hollow lemniscate model ( grey levels ) superimposed . for most of the channels , there is a better agreement of the model with the data than in the previous two models . the exceptions are the highest redshift channels ( @xmath184 km s@xmath8 ) , but , as we already discussed , we have to consider that the outflow is patchy and that most of the sw outflow is hidden behind the galaxy plane . taking this fact into account , the match between the data and model is very good . the _ p - v _ diagrams of the hollow lemniscate outflow model are compared to the data in figure [ fig - pv - mod-27 ] . there is good agreement to the data , mainly in the redshift side of the ne emission ( bottom right of the diagrams ) again , the patchy nature of the outflow and the fact that the sw side of the outflow is behind the galactic disc must be taken into account to explain the discrepancies . given the distinct morphology of the [ feii ] as compared to that of the [ oiii ] emitting gas and to that of the conical model , it is remarkable that the conical model previously proposed by @xcite based on long - slit spectroscopy gives such a good reproduction of our [ feii ] data , although we have tuned a bit the parameters . the conical model , however gives a poorer reproduction of the outflowing gas morphology than the lemniscate model as seen in the integrated flux map ( fig . [ fig - flux - mod-24 ] ) and in the channel maps ( figure 1 of the appendix , available online as a supplementary material ) . the spherical model fits poorly the data and results in no improvement over the conical model . the lemniscate model gives a better fit to the flux distribution , but the improvement in the kinematics relative to the conical model is modest : in the _ p - v _ diagrams , the lemniscate model reproduces better the data for the slits that are farthest away from the centre . in the slits closest to the nucleus , there is a knee " in the positive velocities at @xmath1312 from the nucleus that is also well reproduced by the lemniscate model . it may be argued that this model shows a narrower range of velocities than the data , particularly noticeable in the central slits of the _ p - v _ diagrams , but this can be understood as due to the fact that we did not include a velocity dispersion across the walls of the lemniscate . in the conical model , such velocity dispersion arises naturally because of the radial symmetry of the model which leads to a varying orientation of the velocity vectors across the walls of the cone , while , in the same region of the lemniscate model the velocity vectors are parallel . we would like to point out that , even though the geometric distribution of the gas seems to be better reproduced by the lemniscate model , there is no significant difference between the kinematics of the conical and lemniscate models , both providing similar reproduction of the data . the conical model and the lemniscate model give equally good ( or bad ! ) reproduction of the kinematic data . this is due to the fact that our modeling did not include all the effects present in the data , only a simple geometry and kinematics for the outflow . no velocity component due to gas in the disc was considered in the models , even though they seem to be present in the data . we have also not considered the patchy nature of the outflow , which is obvious in the channel maps , neither the enhanced emission in the redshift channel maps to the ne , as observed in the data . as we have already pointed out , these latter features are due to the interaction between an accretion disc outflow and the circumnuclear ism of the galaxy . due to the orientation of the outflow relative to the galaxy plane , the back wall of the ne part of the outflow is tangent to the galaxy plane , encountering much more gas on its way out than the front wall which leaves the plane at a larger angle . the absence of the effects discussed above in the modelling precludes a quantitative discrimination of the best model to fit the data . as we have stated above we can only say that the lemniscate model gives a somewhat better reproduction of the line flux distributions . an additional discriminator between the models could be obtained if the outflow could be probed at larger distances from the nucleus than we have observed , as the models predict distinct morphologies and behaviors in that regions . the outflow we are seeing on scales of tens to hundreds of parsecs originates in an accretion disc outflow which occurs on scales 23 orders of magnitudes smaller . we are thus observing the interaction of an agn outflow with the circumnuclear gas at the galaxy . thus it is remarkable that the observed geometry is so similar to that proposed in accretion disc wind models , such as those of @xcite and @xcite . the fact that the outflow is hollow supports also an origin in a finite range of radii in the accretion disc as proposed by @xcite . the only difference between the agn outflow models and the data is the fact that the opening angle of the observed outflow decreases with distance from the nucleus , what seems not to be predicted by the models and may be due to the interaction of the agn outflow and the circumnuclear gas of the galaxy . it is interesting to note that this feature is observed also in some planetary nebulae ( e.g. ngc6302 and ngc6537 ) . in order to quantify the mass outflow rate , consider a cross section of the lemniscate perpendicular to the model axis at a distance of 1 arcsec from the nucleus . this cross section has the geometry of a ring with area @xmath185 . the mass outflow rate is @xmath186 where @xmath187 is the proton mass , @xmath188 the electron density , @xmath189 is the velocity perpendicular to the cross section and @xmath190 is the filling factor . for the velocity @xmath189 we adopt the velocity given by the model at the cross section , @xmath191 km s@xmath8 , corrected for its orientation relative to the direction perpendicular to the cross section at this location , @xmath192 . the result is @xmath193 km s@xmath8 . the filling factor @xmath190 can be obtained from @xmath194 where @xmath195 is the h@xmath14 luminosity emitted by a volume @xmath196 and @xmath197 erg @xmath198 s@xmath8 @xcite assuming @xmath199 @xmath198 and @xmath200 k ( case b ) . although the temperature and density of the gas are conservative guesses , the value of @xmath201 for a wide set of possible temperatures and densities ranges from 0.66 to @xmath202 erg @xmath198 s@xmath8 . @xmath195 is determined from the flux @xmath203 using @xmath204 where @xmath205 is the distance to the galaxy and @xmath206 and @xmath207 are emission coefficients for br@xmath4 and h@xmath14 , respectively . the flux @xmath203 was obtained integrating the br@xmath4 flux map over a region with the geometry of a right triangle to the n - ne of the nucleus such that the nucleus is at one of the vertices with angle @xmath208 . this is the angle between the outflow axis ( one of the triangle sides ) and the wall of the cone model ( the triangle hypotenuse ) . the other triangle side is placed orthogonal to the axis of the outflow at 1 arcsec from the nucleus . this is the region where the br@xmath4 is brightest . the integrated flux over this triangular region is @xmath209 erg @xmath210 s@xmath8 . the triangular region over which the flux was integrated corresponds to the projection , in the plane of the sky , of one half of the volume of a cone with the same cross section area of the lemniscate model . now dividing equation [ filling - factor ] by [ eq_lum_hbeta_2 ] , solving for @xmath211 and using that the volume @xmath196 in equation [ filling - factor ] must be the difference between the volumes of the inner and outer half cones @xmath212 we obtain @xmath213 : @xmath214 although we are calculating the outflow for a lemniscate geometry and using a conical volume in equations [ filling - factor ] and [ eq_lum_hbeta_2 ] ( the volume is implicit in the integrated flux ) , this is no mistake as the volume in both equations was the same and its dependence cancel out giving place to a dependence in the area @xmath185 , which is the same as that of the lemniscate model . it is also assumed that the filling factor @xmath190 is the same for the entire cross section . inserting [ eq_fa ] in [ mass_outflow ] we obtain @xmath215 which does not depend explicitly on geometric details of the outflow . now using equation [ eq_mass_outflow_simp ] we obtain a mass outflow rate of @xmath216 . this value is in agreement with those typical of other active galaxies in @xcite , which range from 0.1 to @xmath217 and also with that obtained by @xcite , of @xmath218 , for mrk1157 as well as with the revised value for mrk1066 @xcite . as pointed out also in the above studies , such mass outflow rates are much larger than the agn accretion rate , indicating that , although the origin of the outflow is the agn , the observed outflow is actually gas from the circumnuclear region being pushed by the agn outflow ( `` mass - loaded outflow '' ) . the @xmath219 value depends mainly on the uncertainties of @xmath220 , @xmath205 , @xmath203 and @xmath206 and on the assumptions of the modeling . from these parameters the less constrained are the @xmath206 and @xmath205 . as mentioned above @xmath206 can span a range from 0.66 to @xmath221 erg @xmath198 s@xmath8 which corresponds to about @xmath222 uncertainty . the distance to ngc1068 @xmath205 is also very uncertain as in the literature we find values ranging from 10 to 30 mpc that correspond to an uncertainty of about @xmath223 . taking into account these uncertainties , the mass outflow rate is in the range 1 to @xmath224 we can use the above mass outflow rate to estimate the mechanical power of the outflow using the same method as in @xcite : @xmath225 this is a lower limit as this formula does not account for the power associated to the velocity dispersion of the gas , however , this additional term can be neglected as the velocity dispersion is much smaller than the velocity of the outflow . using the above @xmath220 , we obtain @xmath226 erg s@xmath8 , which is in good agreement with those obtained for seyfert galaxies and compact radio sources @xcite and with that found in @xcite , of @xmath227ergs@xmath8 for mrk1157 . the uncertainties for the mechanical power are , in our case , the same as those for @xmath219 and allow for @xmath228 values in the range 1.5 to @xmath229 erg s@xmath8 . adopting a bolometric luminosity of @xmath230 ergs s@xmath8 from @xcite ( the authors argue that the @xmath231 is probably within a factor of a few of the published value ) for the nuclear source , the mechanical power @xmath228 is within a factor of a few of @xmath232 . the typical power adopted in agn feedback models to significantly affect the galaxy evolution ( e.g. * ? ? ? * ) is @xmath233 and , consequently , the outflow mechanical power in ngc1068 is smaller than that required to provide significant feedback effect in the galaxy . this conclusion depends also on the adopted @xmath231 , as a 7 times smaller value would lead to the opposite conclusion . @xcite estimated @xmath234 and @xcite calculated @xmath235 . these values place the limits for our calculated mechanical power in the range @xmath236 to @xmath237 . nonetheless , the values of @xmath231 for ngc1068 must be taken as conservative lower limits , as the @xmath238 of the nuclear source is known to be high and this can lead to underestimated values of @xmath231 . these factors lead us to argue that the mechanical power in ngc1068 is smaller than that required to provide significant feedback effect to the galaxy evolution . we have mapped the inner ( 200pc ) nlr kinematics of the seyfert 2 galaxy ngc1068 on the basis of near - ir integral - field spectroscopy in the emission lines [ feii]@xmath2 m , h@xmath3 m and br@xmath4 . we have compared the ionized and molecular gas kinematics and have used three different models to reproduce channel maps and _ p - v _ diagrams of the ionized gas . our main conclusions are : * the br@xmath4 kinematics comprises both blueshifts and redshifts of up to @xmath52kms@xmath8 and its flux distribution in the channel maps is very similar to the flux distribution in [ oiii ] , confirming our conclusion of paper i that the two originate in the same bi - polar outflow . * the [ feii ] kinematics cover the same bi - polar outflow and range of velocities as observed for br@xmath4 , but the flux distributions in the channel maps show a broader ( bowl - shaped hourglass ) bi - polar morphology than the bi - conical shape previously observed in the [ oiii ] flux distribution and also observed in br@xmath4 . this is mainly due to the larger contribution of [ feii ] emission in the redshifted channels . we attribute this result to the origin of the [ feii ] emission in a partially ionized region extending beyond the fully ionized region probed by the br@xmath4 and [ oiii ] emission , with additional contribution from shocks between the nuclear outflow and gas in the galaxy disk . * we conclude that the [ feii ] kinematics provide a better coverage of the nlr outflow thus being a better tracer of the outflow than both [ oiii ] and h@xmath239 emission lines , as it extends to regions not probed by these lines . * the bi - polar bowl shape geometry ( including the partially ionized gas region ) resembles that of many planetary nebulae and that predicted by disc and torus wind models , supporting that such winds are the origin of the observed outflow . * the h@xmath11 kinematics is completely distinct from that of the ionized gas , showing an off - centered ring - like morphology at much lower velocities that suggest expansion in the galaxy plane , at an average velocity of @xmath13100kms@xmath8 as supported by previous co mm observations . * a new result regarding h@xmath11 is our finding of deceleration between the inner and outer border of the ring from @xmath128kms@xmath8 down to zero , which indicates that the expansion is being halted . this result has been predicted in recent hydrodynamic simulations and can be attributed to the interaction of a nuclear jet with the gas in the galaxy disk . this gas is pushed radially away from the nucleus and subsequently halted by the circumnuclear ism . this deceleration reconciles our previous finding of a 10myr stellar population in the ring with its kinematics . * our measurements also suggest that the two h@xmath11 linear structures nt and st ( to the north and south of the nucleus ) which apparently connect the ring to the galaxy nucleus are in outflow ( contrary to the suggested inflows in previous studies ) . * we have built models testing three distinct geometries for the outflow observed in [ feii ] : a conical , a spherical and a lemniscate ( hourglass ) . although the lemniscate model gives a better reproduction of the integrated flux distribution , its kinematics does not provide a significantly better reproduction of the velocity field than the conical model of @xcite ; in both models , the outflow is accelerated to a maximum velocity of 1000kms@xmath8 at 80pc from the nucleus , decelerating to zero in the conical model and keeping its maximum velocity constant in the lemniscate model for distances larger than 80pc from the nucleus . * the calculated mass outflow rate along the nlr is @xmath6m@xmath7yr@xmath8 . similarly to what we and others have found for other seyfert galaxies , this outflow rate is much larger than the mass accretion rate to the agn , indicating that the observed outflow is in fact gas from the inner region of the galaxy pushed by an agn outflow . * the calculated power of the outflow is @xmath240 erg s@xmath8 , which is only @xmath241 , thus smaller than the @xmath242 required to affect significantly the galaxy evolution . we thank the referee for valuable suggestions that helped to improve the paper . based on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , ministrio da cincia , tecnologia e inovao ( brazil ) and ministerio de ciencia , tecnologa e innovacin productiva ( argentina ) .

the low resolution x - ray spectra around car covering tr 16 and part of tr 14 have been extracted from a chandra ccd acis image . various analysis techniques have been applied to the spectra based on their count rates . the spectra with the greatest number of counts ( hd 93162 = wr 25 , hd 93129ab , and hd 93250 ) have been fit with a wind model , which uses several components with different temperatures and depths in the wind . weaker spectra have been fit with raymond - smith models . the weakest spectra are simply inter - compared with strong spectra . in general , fits produce reasonable parameters based on knowledge of the extinction from optical studies and on the range of temperatures for high and low mass stars . direct comparisons of spectra confirm the consistency of the fitting results and also hardness ratios for cases of unusually large extinction in the clusters . the spectra of the low mass stars are harder than the more massive stars . stars in the sequence evolving from the main sequence ( hd 93250 ) through the system containing the o supergiant ( hd 93129ab ) and then through the wolf - rayet stage ( hd93162 ) , presumably ending in the extreme example of car , share the property of being unusually luminous and hard in x - rays . for these x - ray luminous stars , their high mass and evolutionary status ( from the very last stages of the main sequence and beyond ) is the common feature . their binary status is mixed , and magnetic status is still uncertain . address for correspondence : nancy r. evans smithsonian astrophysical observatory ms 4 , 60 garden st . , cambridge , ma 02138

astro-ph0405587

@xmath0 car and the associated clusters tr 16 and tr 14 , contain some of the youngest and most massive stars in the galaxy . these stars presumably trace a sequence in descending order of mass from the enigmatic @xmath0 car , the wolf - rayet star hd 93162 , the o3if ( presumed wolf - rayet precursor ) hd 93129ab , hd 93250 ( o3 v ) , and on through other main sequence stars . an early chandra image was centered on @xmath0 car , and contained tr 16 and part of tr 14 . @xmath0 car itself from this image has been discussed by seward et al . the luminosities and hardness ratios of other stars in the clusters have been discussed by evans et al . ( 2003 , hereafter paper i ) . source variability will be discussed in a future ( in preparation . ) the chandra acis image provides low resolution spectra of sufficiently luminous sources . this paper is a discussion of the spectral properties of these stars . the low resolution spectra supplement grating spectra which can only be obtained for the very brightest x - ray sources . hetg spectra have been obtained for @xmath0 car itself ( e.g. corcoran et al , 2001 ) . a major aim of x - ray studies of this region is to understand the factors that dominate the evolution of the most massive stars in the galaxy . the large number of o stars in this part of the sky justifies the use of less than optimal chandra data ( see section 2 ) . since the surprising discovery of x - rays in o stars with the einstein satellite ( see paper i for background ) , it has been known that at least in some cases , x - ray emission originates in shocks in the strong winds produced by very massive stars . as shown by macfarlane et al . ( 1991 ) , a stellar wind distribution of shocks is expected to produce a unique x - ray emission line signature of very broad , highly asymmetric , blue- shifted x - ray emission line profiles . although chandra hetgs observations do show the expected broadness , the emission lines are found to be essentially symmetric and unshifted ( waldron and cassinelli 2001 ; miller et al . the one exception is the early o - star @xmath1 pup ( cassinelli et al . 2001 ) which shows blue - shifted lines , but the predicted large line asymmetry is minimal . the picture may be further complicated by the possibility that in some cases a binary system may contain two massive stars giving rise to `` colliding wind '' systems . recent investigations ( eg schulz et al . 2003 ) add a further complication in that some hot stars appear to produce x - rays through magnetic processes ( eg . @xmath2 ori c , a , and e ) . tr 14 and tr 16 therefore provide one of the best locations to investigate x - ray production in a broad range of massive stars . general introduction to the clusters tr 14 and tr 16 is provided in paper i. since that paper was submitted , two additional studies on the area have become available which are particularly relevant to our discussion . raassen et al ( 2003 ) have analyzed an xmm - newton high - resolution spectrum of wr 25 . albacete et al ( 2003 ) have discussed the luminosities and low - resolution spectra of other sources in the region , also using xmm - newton data . we will include these papers in our discussion later in this paper . x - rays are particularly good tracers of activity high in the atmospheres of stars . in high mass young stars , the mass loss from the atmosphere is fundamental to the future evolution of these objects . it is also linked to the dynamical and chemical history of the interstellar medium . in this paper ( paper ii ) , we focus on the information that acis - ccd low - resolution spectra provide . we are investigating three areas in particular . first , what insights can the spectra provide to the factors effecting evolution of the most massive sequence of objects ( @xmath0 car , hd 93129ab , wr 25 , and hd 93250 ) ? second , how are the x - ray spectra and intrinsic photospheric stellar parameters ( such as temperature ) or extrinsic parameters ( such as reddening ) related for the o main sequence stars from o3 through o8 ? finally , what are the characteristics of the x - ray spectra for the cool ( less massive ) , presumably pre - main sequence stars ? cool stars " refers to the photospheric temperature of low mass stars , even though their x - ray temperatures are often higher than those of high mass stars . specifically , our analysis consists of fitting the spectra to multi - component models ( where the spectra have enough counts ) , fitting simpler ciao / sherpa models to weaker spectra , intercomparing the spectra themselves in appropriate subgroups , and comparing derived temperatures and hardness ratios . succeeding sections of this paper discuss the following topics . 1 . the sources are identified . strong sources are modeled with a detailed wind model . 3 . weaker sources are fit with models available in the ciao sherpa package . 4 . spectra of sources are intercompared . information which can be obtained from acis low resolution spectra is discussed . results are summarized . @xmath0 car was imaged using the acis - i detector onboard chandra ( garmire et al . two observations were obtained ( obsid s 50 and 1249 ) within the space of 1.2 days . there data were merged by reprojecting all the events into a single reference frame . spectra were extracted for sources with a reasonable number of counts , more than 100 if possible . background subtraction was handled by extracting counts in an annulus surrounding each source to account for local variations in the diffuse x - ray emission . any contamination from point sources intruding on the background annulus was eliminated before extracting the background . in a few cases , we went below our 100-count limit . for example , weaker sources were used for cool stars to obtain representative examples of the class . the stars with extracted spectra are indicated on our chandra image shown in fig 1 . to avoid confusion , sources m and n have not been marked . they are the two sources immediately south of source h. the sources are listed in table 1 ( ordered by decreasing optical brightness ) , which lists the alphabetical identification in fig 1 . successive columns are the alphabetical identification , the star name , the source number from paper 1 , ra ( 2000 ) , dec(2000 ) , the counts , two hardness ratios defined below ( hr@xmath3 , the soft hardness ratio and its error , and hr@xmath4 , the hard hardness ratio and its error ) , the v magnitude , the spectral type and the e(b - v ) . these quantities are taken from paper i , with the final 3 columns generally adopted from massey and johnson ( 1993 , referred to as mj93 below ) . hardness ratios are formed from counts in bands 0.5 to 0.9 kev , 0.9 to 1.5 kev , and 1.5 to 2.04 kev . hd 93129ab is a particularly complicated source ( unresolved in our image ) . the a and b components have spectral types o2if * and o3.5 v((f+ ) ) respectively ( walborn , et al . in addition , component a was recently resolved by nelan and collaborators ( paper i ; benaglia and koribalski 2004 ) , resulting in an additional component 0.9 mag fainter than a. the acis detector suffered radiation damage immediately after the opening of the outer doors that protected the instrument during launch . the radiation damage increased the charge transfer inefficiency ( cti ) of the acis ccds with the effect of creating a gain shift to lower energies as well as a reduction of the spectral resolution . these effects are a function of the distance from the original location of the charge ( event ) to the ccd readout node . the effect is non - linear ( townsley , et al . 2000 ) . the observation of @xmath5 car occurred near the middle of the maximal change in cti . furthermore , the operating temperature of acis during this time was -100@xmath6 c , a temperature for which minimal calibration data exist . as a result , spectral fits may be uncertain which we explore in some detail . we judge the possible cti - induced differences between the two obsid s to be smaller than the uncertainties in the calibration . we reach this judgment solely because the observations were contiguous in time . we approximated an instrument response matrix by interpolating between a matrix at -90@xmath6 c and one at -110@xmath6 c , and introduced a correction for cti as described below , however , the data we used in this analysis do not include these corrections because they were not obtained at that temperature . ] . we can quantify the impact of cti using three bright stars in the @xmath0 car field ( sources d , e , and f ) . we extracted the spectra of these sources and found one consistent feature across the spectra : an apparent absorption line near 2.1 kev . the line is an absorption feature in the iridium coating on the mirror , so it is intrinsic to the instrument and is impressed upon all the spectra . the limitation in this approach is the presence of only one bright source ( d ) ; sources e and f are weaker . we modeled the absorption as a gaussian on a sloping background . the fits yielded the line center and line width and are shown in fig . the top sub - plot shows the three fits overlaid ; it is clear that a gain shift progression occurs from source f to source e , with the minimum of the absorption moving to lower energies . the gaussian width becomes larger in the same direction . 3 summarizes the fit parameters , now as a function of the number of pixels from the mean position of the source to the readout node measured along a readout column . the local gain shift and decreased spectral resolution resulting from cti are apparent . local effects of cti may be corrected by the use of individual response matrices . all spectral models included a global gain shift with a linear slope of 0.95 and an offset of 0.0 . these values were taken from the analysis of seward et al . on @xmath5 car itself also using the 2.1 kev iridium feature . @xmath5 car is a much brighter source with better count statistics , so the global shift determination , obtained by requiring the iridium feature to lie at its laboratory energy , is accurate . these values were used during the spectral fits to the sources described here and fixed at the values listed above . we estimate that the combination of global plus local gain shifts reduces the error on the gain to about 10% or less . response matrices and effective area files were then generated for each source in table 1 individually . for the generally low count rates of these ob stars , this approach to calibration provides spectral parameters we estimate are accurate to better than 25% . the largest discrepancies will occur at the lower energies ; the high column toward the ob stars near @xmath0 car largely eliminate the soft events , however . to confirm this approach , we have compared the fits to our spectra with fits to the same sources in an xmm observation ( albacete colombo , et al . , 2003 ) in table 6 . as discussed in section 6.1 , we find the fits differ in some cases because xmm constrains the extinction better because of its softer response . when the fits produce the same column density , the luminosities are also similar . unfortunately , we are unable to use zero order data from several hetg spectra to confirm our analysis . only one of the strong sources falls on the bi acis - s ccd chip , and the relatively few counts would not be a good test of our correction . chandra hetgs observations have finally verified that the x - ray emission from o stars arise from various regions distributed throughout their stellar winds ( e.g. , waldron and cassinelli 2001 ) which implies that all x - ray spectral model fitting efforts must incorporate the effects of stellar wind absorption . the first attempts to explore wind absorption effects were carried out during the einstein era ( e.g. , cassinelli and olson 1979 ; waldron 1984 ) . in these studies it was shown that although the wind absorption cross sections are significantly smaller than the cold ism absorption cross sections below 1 kev , since the wind column densities can be significantly larger than the observed ism column densities , the overall observed x - ray spectral distribution is predominately controlled by the absorption properties of these stellar winds . even when the ism column density is comparable to the wind column density , the effects of wind absorption still play an important role in the interpretation of observed x - ray spectra ( e.g. , waldron et al . 1998 ) . the stellar effective temperature ( strength of the uv radiation field ) is the dominant factor that determines the energy distribution of the wind absorption cross sections . these cross sections are also affected by the density distribution of the wind but these effects are most noticeable only at energies below @xmath7 0.1 kev . in the acis - i energy band ( @xmath8 0.1 kev ) the effects of density are minimal . for our stars with the greatest number of counts ( hd93129ab , hd93250 , and hd93162 ) , effective temperature range between @xmath7 50000 to 54000 k ( howarth and prinja 1989 ; lamers and leitherer 1993 ) , so their wind cross sections are very similar . figure 4 shows a comparison of the wind cross sections with the ism cross sections for these stars . since the wind is subjected to a strong uv radiation field , the dominant h and he absorption cross sections seen in ism cross sections are greatly reduced revealing a highly structured collection of wind cross sections below 1 kev from lower abundant elements ( e.g. , cno ) . for example , the energy shift of the o k - shell edge illustrates that the wind is in a higher state of ionization which leads to the commonly used phrase `` warm - wind '' absorber model . a slight difference exists between in our fig 4 and waldron et al . ( fig 2 ) because fig 4 is based on an effective temperature of @xmath7 52,000 whereas , the waldron et al plot is based on an effective temperature of 40,000 . following the wind absorption x - ray fitting procedure outlined by waldron et al . ( 1998 ) , we obtain wind absorption model fits to our three acis - i spectra using the mekal emissivity model ( of mewe , kaastra , and liedahl ) . the ism column densities are fixed parameters in our model fits . the adopted values are given in table 2 and were determined from their e(b - v ) values given in table 1 as described in shull and van steenberg ( 1985 ) . ( the fits in sections 3 and 4 are made for spectra merged between two 10 ksec images . ) we start by assuming a single component model and add components where necessary to obtain statistically better fits . all single component fits were clearly not able to explain the spectral distributions . one star ( hd93250 = source e ) only required two components , whereas , the other two ( hd 93129ab = source f and hd 93162 = source d ) required three components . the fitting parameters are given in table 2 . these fits include the cti correction as determined in sec . 2 . to handle this calibration issue we developed a special algorithm to apply artificial broadening and energy shifts to all lines . the algorithm extracts the line energy and strength from the input emissivity . each line is then characterized by a gaussian line profile with the appropriate line broadening ( hwhm in table 2 ) and energy shift . this modified emissivity is then folded through the acis - i rmf and arf to obtain our best fit models . we point out that this initial approximate approach to address the calibration problems assumes that all lines are subjected to the same broadening and shifts . in addition , the continuum and cross section energies are not shifted , which does not appear to be an important effect . although the broadening and shifts do tend to improve the @xmath9 , from a statistical point of view the difference is essentially minimal . however , a visual inspection of the resultant best - fit spectra clearly show that broadening and shifts indicate better fits to spectra . the following provides additional information about the parameters in table 2 . for the variables t@xmath10 , n@xmath11 , and em , the quoted best - fit values and errors are based on taking the @xmath9 90% confidence range for a given variable and averaging to obtain the best - fit value . the error is then associated with the difference . all variable upper or lower limits represent the upper or lower limits on the 90% confidence range . the errors on the observed flux , f@xmath12 , and intrinsic flux , f , for each component are determined by comparing the model predicted counts with the observed counts in specified energy bands . for each star these energy bands are dependent on the location at which each component is expected to provide maximum flux . the relative errors in observed and predicted counts are then compared to the corresponding observed count statistic error , and the adopted error is the larger of the two . the results of our fits , shown in figs . 57 , indicate that all three targets show evidence for a hard x - ray component with a temperature @xmath13 10 mk which appears to be deeply embedded within the stellar wind ( with a column density n@xmath11 @xmath8 10@xmath14 @xmath15 ) . the softer components , based on the derived n@xmath11 , all appear to be distributed farther out in the wind . although this distribution appears consistent with a shock model , the derived temperature distribution ( increasing inward ) may be a problem for a shock model scenario as discussed by waldron and cassinelli ( 2001 ) . the remaining sources fall into two groups : a group for which spectral fits are potentially meaningful ( sources b = -59@xmath162600 , c = hd 93205 , and g = hde 303308 ) and a group for which spectral fits yield little more than a temperature _ or _ a column density ( sources i , j , k , m = -59@xmath162635 , and n = hd 93343 ) . we fit the spectra of the first group with three types of raymond - smith models : all parameters free to vary , freezing the plasma energies of the o stars at 0.4 kev , and freezing the column densities at @xmath17 @xmath15 . the optical extinction typical of the hot stars corresponds to this column density and represents a minimum extinction where there is no additional circumstellar wind absorption . the adopted temperature is confirmed by the fit in table 4 to source c , the strongest of the normal o stars . the fits were carried out using the fitting codes sherpa and xspec . figure 8 shows the results while table 3 lists the model parameters . in table 3 we include the formal errors from the fits , although intercomparison of the results shows that the actual uncertainties are larger than the formal errors . we have suppressed the error bars in the upper portion of fig . 8 for legibility . the lower plot shows the residuals of the fit to the spectrum of source c ; the residuals for sources b and g are similar in vertical range , but contain fewer data points spanning the energy scale . sources b and g ( o6 v and o3 v respectively ) have nearly identical spectral shapes ( and parameters ) and differ from source c ( o3 v ) in that c shows evidence for emission lines ( table 4 ) . the lines in c were all fit using zero - width gaussians and are significant at @xmath899.7% for 2 parameters of interest ( line center , line flux ) . two of the features fit with gaussians ( 0.579 and 0.892 kev ) match strong features seen in grating spectra of hot stars ( e.g. @xmath1 pup , cassinelli , et al . , 2001 ) ; one of the gaussians ( 0.686 kev ) did not have a match . thus , the low resolution spectra pick out some emission lines accurately , but not all . the temperature and absorption for source c in table 3 are not exactly the same at that in table 4 . however , when either of the parameters is frozen at a value similar to those in table 4 , a similar combination is produced . this is a good illustration of the non - uniqueness of the fitting procedure ( even where the @xmath9 can be used to select between fits ) . in table 5 ( below ) two temperature fits were made . for source c , the two temperatures bracket the single temperature in tables 3 and 4 . for the weakest sources , the spectral fits essentially permit only an adjustment of the model normalization and perhaps one additional parameter . within the errors , it is largely irrelevant whether one fixes the model temperature and fits the column or vice versa . as part of our exploration of the information that can be obtained from acis resolution spectra , we have also fit raymond - smith models to the strongest spectra , those fit with the wind - models ( table 5 ) . experience has shown that the absorption needs to constrained in the fits . because the absorption is reasonably constant throughout the clusters , we have fixed the column density at @xmath70.3@xmath1810@xmath14 @xmath15 as in table 3 . the simplest model combined with this extinction that provides reasonable fits to the spectra is a two temperature model . one temperature fits result in poor reduced chi - squared values , as seen in table 3 . ( the errors in table 3 are poisson errors from the solution plus 25% because of the calibration uncertainty added in quadrature . ) table 5 provides fits for the two temperature models for sources c , d , e , and f. the reduced @xmath9 is markedly smaller in the two temperature fit than for a single temperature . these temperatures will be discussed in conjunction with the physical parameters of the stars . when we compare the temperatures in table 3 and those in table 5 , several features emerge . table 3 is partly an exploration of the robustness of the fitting results , to see which values are returned most consistently when different parameters are held constant . the fit which fixes the absorption in table 3 is the most similar to the fit in table 5 . for these fits , the single temperature in table 3 always falls between the two temperatures in table 5 . furthermore the mean temperature in table 5 and the absorption constrained fit in table 3 order in the same way , c , e , f , and d from the coolest to the hottest . for the two stars most likely to have significantly more absorption than the interstellar absorption , source d = wr 25 and source f = hd 93129ab , the reduced @xmath9 in table 3 shows a marked increase when the absorption is held to the ism absorption . an exploration of the spectral information independent of response matrix or calibration issues is justified because of the large number of o stars in the region . tr 14 and tr 16 contain many massive stars ( including several o3 stars , the hottest spectral types ) . in addition , ground - based observations have provided extinction measures for these stars . in this section , we provide direct comparisons between spectra with a variety of optical parameters . fig . 9 shows the comparison of the spectra of 3 o3 v stars in the image . source e = hd 93250 is the most x - ray luminous normal star in the chandra image and we use it as the reference spectrum in the comparisons . in fig 9 , in order to compare spectral slopes , sources c and g have been scaled so that they approximately match source e between 0.8 and 1 kev . the figure shows that the spectral slope from 0.8 to 2 kev is the same for sources c and g , however , source e is clearly harder . sources c and g have relatively more counts between 0.5 and 0.8 kev than source e. thus the brightest source in x - rays also has the hardest spectrum . from the optical results , all 3 stars have similar e(b v ) ( table 1 ) . thus the fact that source e has the hardest x - ray spectrum should not be caused by differential absorption , but apparently reflects a different source temperature . it is interesting to note that source e with the hardest spectrum is not known to be a binary ( paper i ) , but the softer source c is a binary , and source g may be a binary . 10 shows the comparison of the standard o3 x - ray spectrum ( source e ; solid line ) with the o3 star ( source l ; squares ) which has a high optical absorption [ e(b v ) = 0.94 mag , table 1 ] . source l lies very close to the line of sight of a dust lane , and also very close to the wolf - rayet star hd 93162 . ( source l has been scaled to match source e at high energies to see whether the difference in spectra can be accounted for by extra absorption . ) 10 shows that while the spectral slopes of sources e and l match for energies higher than 1 kev , the softer flux from source l is much smaller than that of source e , which is consistent with the higher optical extinction in source l. ( massey and johnson were particularly careful in the spectroscopic observations of the absorbed star because of their surprise at the early spectral type for a comparatively faint star . ) source l is a luminosity class i star , where source e is class v , but the major differences between the spectra seem to be due to different extinction . note that in table 3 , the column density appropriate to the e(b - v ) is between the standard cluster extinction and that found in a free fit . by inference , the temperature would be between the temperatures for the two solutions , which is comparable to the temperature found for source e and the standard cluster extinction . this is consistent with the working hypothesis that the sources are a similar temperature , but with a different extinction . 11 shows the comparison of source f ( hd 93129ab ) with the reference spectrum . source f is the brightest star in the tr 14 - 16 region list of massey and johnson . it is also the most unusual of the low - resolution x - ray spectra . 11 shows prominent emission lines at the locations of ionized mg , si , and perhaps s. this is in marked contrast to the smooth spectrum of source e at this resolution , which is the second brightest star in the massey and johnson list . the helium - like mg xi and si xiii lines are the strongest lines in the high - resolution spectra of the o4f star @xmath1 pup in this energy region ( cassinelli , et al . , 2001 ; kahn , et al . , 2001 ) . as discussed in sec . 2 , hd 93129ab is a complex source ( though unresolved in our image ) , consisting of components aa , ab , and b. since many of the other massive stars we detected are also multiple systems ( table 8 , paper i ) , the multiple components are not the cause of the unique spectrum in fig . 11 , though binarity may have some affect on the spectral properties . the unique property of hd 93129aa is that it is the most luminous star in the two clusters after @xmath0 car . the implication is that the spectral emission is related to the extreme luminosity , presumably coupled with a high mass loss rate and possibly circumstellar material . 12 shows the comparison of the later o stars . the spectra have not been scaled , so there is a range of count rates . however , the spectra are similar in that they all have soft counts . counts decrease with energy for energies greater than about 0.8 kev . 10 in paper i shows that there in no indication that l@xmath10/ l@xmath19 is a function of photospheric spectral type . this is consistent with table 1 in this paper , which is ordered by optical magnitude . the group of later o stars shows that neither the x - ray counts ( column 6 ) nor the optical spectral types ( column 12 ) is tightly correlated with the optical magnitude . binarity for this group of x - ray sources ( table 8 , paper i ) is mixed . source b has no velocity evidence for binary motion , and the evidence for sources p and n needs to be confirmed . the other three ( o , m , and h ) do show binary motion . in other words , the level of x - ray strength is not uniquely dependent on optical luminosity , spectral type , or binarity . the similarity of the x - ray spectra ( fig . 12 ) and l@xmath10/ l@xmath19 show that within the range of optical luminosity , spectral type , and multiplicity , the x - ray production is similar . for the four sources that were fitted with temperatures ( b , h , m , and n , table 3 ) , the constrained fit ( requiring that the absorption be appropriate for the cluster ) found the temperatures for h , m. and n similar within the errors . source b has a somewhat cooler temperature from the fit . however , the similarity of the spectra in fig . 12 underscores the fact that care must be taken in interpreting the fits when each bin has relatively few counts . 13a shows the comparison between the reference o3 spectrum ( source e ) and an o6 spectrum ( source b ) , scaled to match near 1 kev . the slopes for e @xmath13 0.8 kev are similar , however , the o6 star has proportionately more counts than the o3 star at lower energies . the optical absorption [ e(b v ) ] for the two stars is similar . in fig . 13b , a second o3 star ( source c ) is compared with the o6 star . in this case , the spectra are virtually identical . thus , stars with optical spectral type o3 can have different x - ray spectra ( fig . 9 ) . on the other hand , stars with optical spectral types o3 and o6 can have identical x - ray spectra ( fig 13b ) . this is in accord with the hardness ratios in paper i. the three sources with particularly soft hr@xmath3 ( soft hardness ratios ) are sources g ( hde 303308 ) , c ( hd 93205 ) , and b ( -59@xmath162600 ) . hr@xmath3 for source e , on the other hand , is harder , which agrees with fig 13a . one of the aims of this paper is to explore how well the hardness ratios reflect the details of the spectra . for these three sources , we can accurately differentiate between harder spectra ( source e ) and softer spectra ( b and c ) using hardness ratios . we note , however , that sources e and c do not show orbital motion in their velocities ( table 8 , paper i ) . source c is a binary ( o3 v + o8 v ) . does this invalidate the interpretation of the hardness ratios ? in cases where there is a colliding wind , the spectrum of the binary will be different from a non - interacting case , where the composite spectrum is simply the sum of the spectra of the two stars . because l@xmath10/ l@xmath19 is approximately constant for o stars , in the case of a non - interacting composite spectrum , the x - ray spectrum will be dominated by the brightest star in the optical . this situation would explain the similarity between the spectra of source b and source c in fig . 14 shows the comparison between source a , an o8.5 star and source h , an o8 star , also shown in fig 12 . neither spectrum has been scaled and fig . 14 shows that at energies higher than 1 kev , source a has about a factor of two more counts than source h. on the other hand , if source h were scaled to have the same flux as source a for energies greater than 1 kev , source a would have fewer soft counts than source h. this is in keeping with the extra absorption in source a ( table 1 ) . source h is a multiple system , with components o7 v , o8 v , and o9 v ( which is itself a binary ; albacete colombo , et al . source a , on the other hand , has no information about radial velocity variation . 12 shows that source h has a spectrum with a slope very similar to other late type stars , so the interpretation in the previous section that for a non - interacting binary system , the spectrum will be dominated by the most luminous star appears to hold . in contrast , the spectrum of source a is both harder and more absorbed than source h ( and the other late o stars in fig . this would be consistent with additional circumstellar material , possibly a colliding wind system or an interacting binary . 15 shows the comparison between the spectra of the wolf - rayet star wr25 = hd93162 ( source d ) and the o3 standard star ( source e ) . fig 15a shows both spectra , unscaled ( source d : with error bars ; source e : solid line ) . the spectra are strikingly similar and uncomplicated . the dip near 2 kev is an instrumental effect . in fig 15b , the wolf - rayet spectrum ( solid line ) has been scaled to emphasize the similarity of the spectra at high energies . however at low energies , the wolf - rayet star has relatively fewer counts than source e ( dashed line ) , appropriate for the its higher absorption . in general , wr stars have higher mass loss rates than o stars , which may account for the lower count rate at low energies in the wolf - rayet star . a major motivation for intercomparing the spectra of various sources in this series of plots is to investigate empirical characteristics of the spectra . experience with spectral fits to these comparatively low resolution spectra shows that the fits are not unique . the fit in table 2 for source d with three components actually has a somewhat worse reduced @xmath9 than the simpler two component fit in table 5 . we have no particular expectation that the wolf - rayet star ( source d ) will have a spectrum similar to the o3 v star ( source e ) , which has a comparatively normal optical spectrum . ( we have no explanation , however , for the higher x - ray temperature of source e than other o3 v stars in fig . 9 , particularly since there is no velocity evidence to support the possibility that source e might be a colliding wind binary . ) the spectral comparison in fig . 15 supports the interpretation that the wolf - rayet star has a higher temperature x - ray component and larger absorption than source e. the fits to source d are discussed further in sec . 6.3 , and compared with the high resolution xmm results of the same source ( raasen , et al . 2003 ) . despite very different photospheric spectral types of source l ( o3 i ) and source a ( o8.5 v ) , the x - ray spectra are identical , as shown in fig . source a is also shown in fig . 14 , which illustrates that it is an absorbed spectrum , confirming a large extinction in the optical . however , the spectra in fig . 16 are not hard simply because their soft counts have been absorbed ; the count rate at 2 kev and higher show that they are intrinsically hard . in table 3 , most of the solutions listed have constraints either on the temperature of the source or the absorption because of fits are not definitive when too many parameters are included in a fit of weak spectra . for source a , when both temperature and absorption are fitted , both are larger than the standard constraints for the cluster . although the temperature in the fit with constrained extinction ( 10.2 k ) is not very hard , it is harder than constrained fits for other o stars in table 1 , except for sources d and f ( and l ) which are more evolved . fig 17 shows the spectra of three stars ( sources i , j. and k ) which are fainter and cooler than the o star x - ray sources . in paper i , we overplotted all the o and b stars in the clusters on the x - ray sources . sources not coinciding with an o or b source are listed as cooler stars in tables 2 and 4 in paper i. sources i , j , and k are clearly among the very strongest of these sources . they have v magnitudes of 16.82 , 14.34 , and 13.16 mag ( table 1 ) respectively , appropriate for sources cooler than the o stars and fall within the pre - main sequence band . for comparison , a representative hot star spectrum ( source b , o6 v ) is shown in fig . all three cool stars have fairly flat spectra . in contrast , the hot star spectrum has proportionately more soft counts . this is in accord with previous findings that cool stars produce x - rays through magnetic processes at higher temperatures than the wind shock processes in hot stars . the goal of this investigation is to explore what physical x - ray parameters can be determined from acis low - resolution spectra . we have described a particularly useful chandra pointing for this purpose , because the region around @xmath0 car contains a large number of extremely massive stars as well as many low mass stars . we have investigated the spectra with detailed wind models , with sherpa spectral fits , and with comparisons between spectra . the first step is to discuss whether the results from these approaches are consistent . as discussed above , the spectra of the weak sources do not constrain both the extinction and the source temperature . for the stars in this field , the extinction is surprisingly uniform , with e(b - v ) approximately 0.5 mag ( mj93 ) . the hot stars typically have individually determined reddenings ( e.g. mj93 ) , which corresponds to a hydrogen column density n@xmath20 = 3 x 10@xmath21 @xmath15 ( seward , 2000 ) . table 3 lists a fit for each source for which the ism n@xmath20 has been fixed to this value ( restricted fitting " ) . we will examine the temperatures using this fixed n@xmath20 value to see if they are reasonable . this value of absorption would be the minimum value , assuming there is no absorption in the wind ( the warm absorber " ) . from the temperature fits , we can see that this minimum absorption produces a high temperature . if the absorption is allowed to increase in the fit , the fitted temperature decreases . we stress that any fit is not unique ; by selecting the restricted fits for discussion , we have imposed a reasonable condition . the cool stars ( sources i , j , and k ) have harder spectra than the hot stars ( fig . 17 ) . in table 3 , two of the three cool stars , sources i and k , have temperatures of 1.4 and 5.2 kev respectively from this fitting , i.e. well above the standard range for hot stars . the other two sources for which the standard e(b - v ) is a poor choice are a and l , which we know from the optical results to have e(b - v ) about twice that value . again this restricted fitting results in temperatures kt of 1.1 and 9.7 kev respectively , larger than the standard range for hot stars . for the remaining sources , a combination of n@xmath20 ( consistent with optical reddening ) and kt can be found for a raymond - smith plasma that matches the spectrum . a recent paper by albacete colombo et al ( 2003 ) discusses xmm - newton x - ray observations of many of the same targets as we do . because our fitting procedures are somewhat different than theirs , the results from temperature fitting ( their table 4 ) will not be identical to ours . our table 3 presents fits generally restricting either the temperature or the absorption . similarly , our table 5 presents fits for a few sources with two temperatures but a fixed extinction . since the foreground extinction is reasonably constant , so we have forced an extinction which is at least a lower limit , rather than allowing what seem to be unphysically large values that sometimes result from allowing the extinction to be a free parameter . table 6 shows the comparison between the single component fit in table 3 and dominant component in the xmm fit . clearly , we match the shape of the fitted spectral function ; the differing bandpasses ( xmm goes 0.2 kev softer ) has a significant impact on the precision and value of the fitted column density , particularly for high values of the column density ( sources b , c , d , e ) . when the two sets of fits yield approximately the same column , we obtain approximately the same luminosity ( sources a , g ) . remaining luminosity differences are easily attributed to the wider xmm bandpass plus the poorly - calibrated phase during which our data were obtained . in general , our analysis of more complex multicomponent fits agrees with that albacete colombo et al ( 2003 ) in identifying sources with unusually hot components , which we will discuss in turn . for source d = wr25 , both analyses find one component a little above 3 kev and one close to 0.8 kev , although the absorption for the cooler component is larger than our canonical absorption . for source c = hd 93205 the temperatures in both fits are close to 1.0 kev and 0.2 kev , with approximately the same absorption . for source e = hd 93250 , the temperatures for both components are hotter than our results ( 5 rather than 3 kev and 0.8 rather than 0.3 kev ) , with similar absorptions . for source a = cpd -59 2629 , our fig . 14 clearly shows that there is a hot component , and we know extra absorption is present from the optical results . a hot component is found by albacete colombo et al . , as one of the components of the fit . since we only did a one temperature fit , the derived temperature is lower than for the hottest component of their two temperature fit . for source g = hde 303308 , albacete colombo et al . found temperatures and extinctions which differ markedly for their two components . for our three fits , we found a single temperature and extinction solution which was fairly close to standard values for the cluster for each . for source b = cpd -59 2600 , both our constrained fits and the single temperature fit of albacete colombo et al . found temperatures and extinctions close to those typical for the cluster . direct comparisons of c and g ( fig . 9 ) and c and b ( fig . 13 ) show that all three spectra are very similar , which agrees with similar values from our temperature fits . in paper i we derived hardness ratios from 3 energy bands , 0.5 to 0.9 kev , 0.9 to 1.5 kev , and 1.5 to 2.04 kev , referred to as soft ( s ) , medium ( m ) , and hard ( h ) bands . included are the soft and hard hardness ratios : @xmath22 @xmath23 we discuss hr@xmath3 but include hr@xmath24 in table 1 for convenience . fig . 18 in paper i shows that hr@xmath3 and hr@xmath4 are correlated . hardness ratios are affected by reddening , however , fig . 18 in paper i shows that the effect for a reasonable range of reddening is small compared to the observed range in the ratios . for the following discussion , for any source best fit with a two temperature model , we adopted the mean of the fitted values . in fig . 18 the relation between @xmath25 and the temperature from the restricted fitting " is shown . the general trend in fig . 18 is the same whether the temperature used for the brightest sources ( d , e , and f ) is the mean of the two temperature fit , or the temperature from the single temperature fit . we stress that the temperature / absorption fits ( especially for the weak sources ) depends on which parameters are fixed . the stars in fig . 18 show two very different behaviors . the weaker stars have a range of hardness ratios , and a small to moderate range in temperatures . the three strongest sources , however , have much larger temperatures and harder hardness ratios . fig . 19 shows the temperatures as a function of photospheric spectral type . as in fig . 18 , the weaker sources have a relatively small temperature range , which is not correlated with photospheric spectral type . on the other hand , the 4 stars with the hottest spectral types ( o3 stars ) divide between relatively cool and much hotter x - ray temperatures . in other words , the photospheric temperature alone is not the determinant of the x - ray temperature . the known binary properties of the sources from radial velocity studies ( table 8 , paper i ) provide a similarly mixed result . of the two hot o3 sources , one is a binary , one is not . for the cooler o3 stars , one is binary , one needs confirmation " . among the later spectral types , one does not have binary velocity variations , two are binaries , and one needs confirmation " . fig 20 shows the x - ray temperature as a function of the v magnitude . for the cluster stars , this is related to the luminosity of the stars , and probably also to the mass loss rate . the fainter stars ( x s ) show no relation between the x - ray temperature and the mass of the stars . the x - ray temperatures are markedly larger for the massive stars ( squares ) . the correlation would be even more prominent except that the hottest star in x - rays is the wolf - rayet star , which may well be somewhat faint in v because it has already lost mass . the components we found ( table 2 ) for wr 25 = hd 93162 ( source d ) can be compared with those found from fitting a high resolution xmm - newton spectrum ( raassen , et al . 2003 ) . the combination of temperatures and wind absorption values they find is somewhat different from those in table 2 , in that the xmm spectrum is fit with two components , while the chandra spectrum is fit with three . this makes detailed comparison difficult . raassen et al . find lines seen in the xmm - newton spectrum are from the same species as those in hd 93129ab ( source f , fig . the two features which are present between 1.0 and 2.0 kev in the xmm spectrum can be identified in the chandra spectrum of hd 93162 . the contrast in these features , which is much smaller than those in hd 93129a , is commensurate with that in the xmm spectrum . the fits in table 2 require much larger line widths for hd 93162 ( source d ) and hd 93250 ( source e ) than for hd 93129ab ( source f ) . fig 3 shows that some of the width is accounted for by the distance of the sources from the acis readout . however , raassen et al . list 2480 km sec@xmath26 as the terminal velocity for wr 25 . this would contribute to the broadening found in table 2 , but is only about 10% of the required amount . the high velocities in wr 25 contribute to the reduction in the contrast in the lines . raassen et al . find a prominent fe k shell feature near 6.5 kev . while the xmm spectrum shows that this line is clearly present , it is not obvious in our spectrum ( fig . 15 ) , presumably because it has relatively few counts . because shock velocities in single stars are not usually strong enough to produce this feature , they propose that it arises in systems that are colliding wind binaries . the table they assemble of fe k@xmath27 detections in binary and single stars , however , shows mixed results on this question . the number of massive stars in tr 14 and tr 16 makes this field a good place to attempt to identify factors contributing to x - ray flux , particularly unusually strong x - ray flux . the well - known relation x - ray luminosity and bolometric luminosity l@xmath10 = 6 x 10@xmath28 l@xmath19 ( see the recent discussion in albacete colombo , et al 2003 ) implies a relation with luminosity , and probably with mass loss and terminal velocity . the four unusually strong , hard x - ray sources ( @xmath0 car itself , wr 25 , hd 93129ab and hd 93250 ) are all at the very end of their main sequence lifetimes or beyond . as discussed in sec . 2 above , hd 93129ab is made up of at least three components which are not resolved in our x - ray image , with spectral types for the a and b components of o2if * and o3.5 v((f+ ) ) respectively . the v magnitudes of aa , ab , and b are approximately 7.6 , 8.5 , and 8.9 mag respectively . based on the standard relation between x - ray luminosity and bolometric luminosity , it would be expected that the brightest component aa would also dominate in the x - ray region . in addition , fig . 20 shows that stars of the magnitudes ab and b would be expected to have a lower x - ray temperature than a star as bright as aa . while we can not measure the individual x - ray brightnesses for the three components , it is likely that the x - ray spectrum is heavily dominated by the brightest star in the system , hd 93129 aa . the x - ray properties of @xmath0 car itself are set out by albacete colombo et al . it has an unusually large x - ray luminosity and x - ray to bolometric luminosity ratio . they fit the x - ray spectrum with four raymond - smith components with temperatures up to 4.4 kev . it also has a strong fe k @xmath27 line . stellar radius increases in very late - stage main sequence stars , yielding increased mass loss . this characteristic should affect @xmath0 car itself , wr 25 , hd 93129ab and hd 93250 . clearly there are several possible variants to this scenario , in particular a binary system ( paper i ) , or a confined magnetic field . as discussed carefully by raassen et al . , information on the binarity of wr 25 is inconclusive . for @xmath0 car and hd 93129ab , binary effects may certainly be important . hd 93250 has no velocity evidence of a binary companion . fig . 10 in albacete colombo shows that hard spectra are not correlated with binary companions . schulz et al . ( 2003 ) suggest that @xmath2 ori c , a , and e all are magnetic stars because they are very close to the zero age main sequence . this adds another possibility to the list of complexities . one candidate source for confined magnetic fields is source a = tr16 - 22 . it has an unusually high l@xmath10/l@xmath19 , as well as a surprisingly hard spectrum , particularly for an o8.5 star . if this is the case , it would be an example of a peculiar property determining the x - ray production . it will be very interesting to see what the radial velocity studies undertaken by albacete colombo et al . ( 2003 ) reveal . albacete colombo et al . do find that it is not variable in their 43 ksec xmm observation . in other words they do not see a flare which would be a clue to x - rays produced by a pre - main sequence companion . if it turns out that magnetic fields play a role for a significant fraction of o stars , it may even be that a close binary companion can suppress a magnetic field , thus actually reducing the x - ray flux . qualitatively , the comparison of the x - ray spectra of wr 25 , hd 93129ab , and hd 93250 shows dramatic differences . is the prominence of emission features related simply to evolutionary changes ? certainly decreasing in mass loss rate in this sequence causes the x - ray spectrum to sample different levels of the atmosphere . the e(b - v ) of wr 25 is larger than for hd 93129ab and hd 93250 , as would be expected from the high mass loss rates of wr stars . however , the x - ray spectrum of the wolf - rayet star @xmath29 vel provides a cautionary note about a simple interpretation ( skinner , et al . 2001 ; willis et al . 1995 ; stevens et al . 1996 ) . willis , et al . ( 1995 ) and stevens , et al . ( 1996 ) obtained rosat and asca observations of @xmath29 vel respectively . with the rosat observations , they mapped a dramatic increase in x - ray flux close to the periastron in the 78@xmath30 orbital period . the asca observations also show a dramatic change in two spectra before and after periastron . after periastron , the x - ray flux was both larger and harder , and also has emission lines very similar to those seen in hd 93129ab . this was confirmed by a chandra hetg spectrum at the same phase ( skinner et al . the proposed explanation is that the large x - ray flux appears when the system is viewed along the shock cone of the colliding wind ( around the o star , which has a weaker wind than the wr star ) . at other phases , the extended wr star dominates the spectrum . thus the `` wind - shock '' phase in @xmath29 vel produces a spectrum much more like the o3 if hd93129ab than wr 25 . even the spectrum in @xmath29 vel at the phase dominated by the wolf - rayet star is softer than wr 25 . in other words , sources may be made up of several components , and it is only the relative importance which changes as the stars evolve . furthermore , as noted above , emission lines are seen in the xmm spectrum of wr 25 ( raasen , et al 2003 ) . the addition of a hard component to a spectrum like hd 93129ab could swamp the emission features . in the previous section we have discussed properties such as binarity and magnetic fields which may mean that the evolutionary development may differ from evolution of a single object . ignoring these complexities , we suggest here a simple scenario which relates the changes in the x - ray spectra to evolutionary changes in these very massive stars . first it is clear that hd 93205 (= source c ) should become a colliding wind binary in the future since it is in a short period ( 6@xmath30 ) orbit with an o8 v star . however , so far its spectrum resembles cooler o stars , and the x - ray bright o3 v star hd 93250 ( = source e ; e.g. figs 9 and 19 ) . hd 93250 is more x - ray luminous , as well as more bolometrically luminous . the mass loss rates of the sequence of o3 stars hd 93205 , hd 93250 , and hd 93129ab (= source f ) are log m = -5.8 , -4.6 , and -4.9 m@xmath31 yr@xmath26 respectively ( howarth and prinja , 1989 ) from iue high resolution spectra . thus it is likely that the increase in x - ray luminosity from hd 93205 to hd 93250 is related to the increased mass loss in the more luminous , and presumably more massive star . the o3 if supergiant system hd 93129ab is not only more x - ray luminous , but it also has very broad emission lines . circumstellar material accumulated from the mass loss of the more evolved primary star in the system could account for the emission , even though the mass loss rate found by howarth and prinja is actually smaller than that for hd 93250 . for hd 93162 = wr 25 = source d , the mass loss rate log m -4.4 m@xmath31 yr@xmath26 ( raasen , et al . it is an unusually strong x - ray source compared with other wr stars . the strong hard x - rays , the large mass loss rate , and the large e(b - v ) are all presumably related . the @xmath29 vel system demonstrates that a wr system can produce broad emission lines , although they are related to the wind - shock cone . it is not clear whether wr 25 is a binary system , as discussed by raasen et al . , although they find a fe k@xmath27 feature which they suggest signifies a binary system . as suggested above , in our scenario emission lines related to circumstellar material such as those in hd 93129ab may be dominated by the unusually strong high temperature component in wr 25 . although this scenario for the four strongest x - ray sources is simplified , it does provide a working model which links the observed features in the x - ray spectra with the underlying evolutionary changes . among the points found from the spectral comparisons in this study are : @xmath32 among the hottest ( o3 ) stars with similar photospheric temperatures ( fig . 9 ) , two have identical spectra , but one is significantly harder . the harder spectrum is produced by the more bolometrically luminous star . a harder spectrum in a more luminous star may result partly from extra absorption from a larger mass loss in the wind . however , it also has a higher x - ray flux , which means the difference can not result only from the removal of soft counts . @xmath32 extra absorption is clearly seen in the spectrum of source l = tr 16 - 244 ( fig . . this may be from foreground absorption from the nearby dust lane , circumstellar absorption , or absorption related to the proximity to the wolf - rayet star wr 25 ( hd 93162 ) . @xmath32 the comparison ( fig 14 ) of the spectra of the heavily reddened o 8.5 star source a with source h ( normal extinction ) for the region shows that the reddened star also has a harder spectrum with an unusually large number of counts at 2 kev for an o star . @xmath32 the wolf - rayet star source d ( fig 15 ) has a spectrum very similar to the luminous o3 star source e ( hd 93250 ) in that it has significant number of hard counts , and no particularly prominent lines at this resolution . the high extinction of the wolf - rayet star is apparent from the comparison of the counts below 1 kev . @xmath32 what happens to the x - ray properties as the very massive stars evolve ? qualitatively , the sequence hd 93250 to hd 93129ab to hd 93162 = wr 25 is a sequence of very massive stars from the main sequence through the first stages of the wolf - rayet phase toward @xmath0 car . the x - ray spectra for these stars are all hard , but with dramatic changes , from relatively smooth through prominent emission lines ( hd 93129ab ) back to comparatively smooth for wr 25 . presumably this results from increasing mass loss driven by evolution . different regions would then dominate the spectra because of different optical depths . however , there may be complicating factors governing the x - ray spectra such as binary interactions ( colliding winds ) or magnetic fields . @xmath32 how are the stellar photospheric parameters and absorption related the the x - ray spectra ? sources a and l which have high absorption in the optical have extra absorption of soft x - ray counts , as would be expected . the x - ray spectra of the hot stars can be divided into two groups . for normal main sequence stars from o3 to o8 , the x - ray temperatures have no dependence on spectral type or luminosity . for more luminous stars discussed in the immediately preceding bullet , have higher x - ray temperatures . this distinction is picked up by the hardness ratios also . financial assistance was provided from the chandra x - ray center nasa contract nas8 - 39073 for nre , ems , fds , mik , jn , and sjw . wlw was supported in part by chandra grants go2 - 3027a and go2 - 3028 . we thank the anonymous referee for very thorough comments which considerably improved the presentation and discussion of the paper , and a useful conversation with fred bruhweiler . f & hd93129ab & 103 & 10 43 57.51 & -59 32 53.1 & 2132.4 & 0.45 & 0.03 & -0.49 & 0.03 & 6.90 & o3 i & 0.55 + e & hd93250 & 22 & 10 44 45.06 & -59 33 55.2 & 1572.8 & 0.13 & 0.03 & -0.58 & 0.04 & 7.41 & o3 v & 0.49 + c & hd93205 & 25 & 10 44 33.80 & -59 44 15.6 & 532.6 & -0.24 & 0.05 & -0.72 & 0.09 & 7.76 & o3 v & 0.40 + d & hd93162 & 28 & 10 44 10.44 & -59 43 11.2 & 4989.9 & 0.61 & 0.02 & -0.42 & 0.02 & 8.11 & wn7 & 0.66 + g & hde303308 & 10 & 10 45 05.95 & -59 40 06.3 & 232.1 & -0.31 & 0.08 & -0.63 & 0.14 & 8.19 & o3 v & 0.46 + b & -59@xmath332600 & 88 & 10 44 41.87 & -59 46 56.5 & 179.9 & -0.18 & 0.09 & -0.82 & 0.15 & 8.65 & o6 v & 0.51 + o & -59@xmath332603 & 83 & 10 44 47.36 & -59 43 53.4 & 52.2 & -0.06 & 0.16 & -0.72 & 0.23 & 8.82 & o7 v & 0.46 + m & -59@xmath332635 & 44 & 10 45 12.78 & -59 44 46.4 & 61.3 & -0.06 & 0.16 & -0.72 & 0.23 & 9.27 & o8.5 v & 0.54 + p & -59@xmath332641 & 2 & 10 45 16.58 & -59 43 37.3 & 84.7 & -0.12 & 0.14 & -0.65 & 0.20 & 9.28 & o6 v & 0.61 + h & -59@xmath332636 & 5 & 10 45 12.93 & -59 44 19.4 & 114.0 & 0.05 & 0.12 & -0.77 & 0.17 & 9.31 & o8 v & 0.60 + n & hd93343 & 45 & 10 45 12.28 & -59 45 00.6 & 64.4 & -0.03 & 0.16 & -0.64 & 0.22 & 9.60 & o7 v & 0.56 + l & tr16 - 244 & 102 & 10 44 13.26 & -59 43 10.4 & 142.7 & 0.66 & 0.13 & -0.37 & 0.11 & 10.78 & o3 i & 0.94 + a & tr16 - 22 & 55 & 10 45 08.26 & -59 46 07.3 & 226.4 & 0.51 & 0.10 & -0.43 & 0.09 & 10.93 & o8.5 v & 0.78 + i & & 71 & 10 44 57.96 & -59 47 9.5 & 51.3 & 0.82 & 0.27 & -0.06 & 0.17 & 16.82 & & + j & & 70 & 10 45 0.99 & -59 45 15.6 & 50.4 & -0.16 & 0.18 & -0.72 & 0.27 & 14.34 & & + k & & 85 & 10 44 46.54 & -59 34 12.4 & 90.8 & 0.44 & 0.16 & -0.38 & 0.15 & 13.16 & & + & n@xmath34 /10@xmath21 ( @xmath15 ) & 2.9 & 2.6 & 3.4 + 1st & t@xmath10 ( mk ) & 5.6@xmath35 3.2 & 6.4 @xmath35 0.9 & 7.8 @xmath35 4.1 + & kt ( kev ) & 0.48 & 0.55 & 0.67 + & em /10@xmath36 ( @xmath37 ) & 2.1 @xmath35 1.9 & 3.2 @xmath38 1.0 & 4.4 @xmath35 4.0 + & n@xmath11 /10@xmath21 ( @xmath15 ) & @xmath39 10 & 5.6 @xmath35 1.5 & @xmath39 12 + & f@xmath12/10@xmath40 ( erg @xmath15 s@xmath26 ) & 0.87 @xmath35 0.10 & 7.1 @xmath35 0.3 & 3.8 @xmath35 0.6 + & f /10@xmath40 ( erg @xmath15 s@xmath26 ) & 3.5 @xmath35 0.4 & 60.4 @xmath35 2.4 & 16.6 @xmath35 2.6 + 2nd & t@xmath10 ( mk ) & 8.2 @xmath35 1.8 & 27.2 @xmath35 16.0 & 14.1 @xmath35 3.5 + & kt ( kev ) & 0.71 & 2.33 & 1.21 + & em /10@xmath36 ( @xmath37 ) & 3.7 @xmath35 1.2 & 5.6 @xmath35 4.3 & 9.8 @xmath35 3.1 + & n@xmath11 /10@xmath21 ( @xmath15 ) & 10.6 @xmath35 3.3 & 75.0 @xmath35 47.8 & 7.6 @xmath35 3.2 + & f@xmath12/10@xmath40 ( erg @xmath15 s@xmath26 ) & 4.2 @xmath35 0.5 & 2.4 @xmath35 0.4 & 24.3 @xmath35 3.8 + & f /10@xmath40 ( erg @xmath15 s@xmath26 ) & 58 . @xmath35 7 . & 38 . @xmath35 6 . & 120 . @xmath35 19 . + 3rd & t@xmath10 ( mk ) & @xmath8 10 & & @xmath8 7 + & kt ( kev ) & @xmath8 0.9 & & @xmath8 0.6 + & em /10@xmath36 ( @xmath37 ) & 0.9 @xmath35 0.6 & & 13.9 @xmath35 13.1 + & n@xmath11 /10@xmath21 ( @xmath15 ) & @xmath8 20 & & @xmath8 25 + & f@xmath12/10@xmath40 ( erg @xmath15 s@xmath26 ) & 0.8 @xmath35 0.1 & & 6.1 @xmath35 1.0 + & f /10@xmath40 ( erg @xmath15 s@xmath26 ) & 2.6 @xmath35 0.4 & & 250 . @xmath35 39 . + & & & & + & @xmath9 / dof & 97.3 / 81 & 144.4 / 116 & 849.3 / 224 + & energy shift ( ev ) & 29 . & -140 . & -56 . + & hwhm ( ev ) & 7.0 & 147 . & 126 . + & total f@xmath12/10@xmath40 ( erg @xmath15 s@xmath26 ) & 5.9 @xmath35 0.6 & 9.4 @xmath35 0.9 & 34.3 @xmath35 5.4 + & total f /10@xmath40 ( erg @xmath15 s@xmath26 ) & 65 . @xmath35 7 . & 99 . @xmath35 10 . & 387 . @xmath35 61 . + source & kt & kt@xmath41 & t@xmath41 & kt@xmath42 & t@xmath42 & @xmath43/dof & dof + & ( mean ) & ( kev ) & ( mk ) & ( kev ) & ( mk ) & & + c & 0.62 & 1.06 & 12.3 & 0.19 & 2.2 & 2.39 & 46 + d & 2.08 & 3.29 & 38.4 & 0.86 & 10.0 & 3.23 & 217 + e & 1.59 & 2.85 & 33.0 & 0.33 & 3.9 & 2.36 & 109 + f & 1.72 & 2.59 & 30.1 & 0.84 & 9.8 & 2.50 & 103 + + & & & & + a & xmm & [email protected] & [email protected] & 30@xmath351 + & chan & [email protected] & [email protected] & 16 + & & & & + b & xmm & [email protected] & [email protected] & [email protected] + & chan & 0.3f & [email protected] & 1.8 + & & & & + c & xmm & [email protected] & [email protected] & 35@xmath352 + & chan & [email protected] & [email protected] & 6 + & & & & + d & xmm & [email protected] & [email protected] & 256 + & chan & [email protected] & [email protected] & 19 + & & & & + e & xmm & [email protected] & [email protected] & 34@xmath352 + & chan & [email protected] & [email protected] & 6 + & & & & + g & xmm & [email protected] & [email protected] & 25@xmath352 + & chan & [email protected] & [email protected] & 26 + & & & & + +

we propose a major index statistic on 01-fillings of moon polyominoes which , when specialized to certain shapes , reduces to the major index for permutations and set partitions . we consider the set of all 01-fillings of a moon polyomino with given column sum whose empty rows are , and prove that this major index has the same distribution as the number of north - east chains , which are the natural extension of inversions ( resp . crossings ) for permutations ( resp . set partitions ) . hence our result generalizes the classical equidistribution results for the permutation statistics inv and maj . two proofs are presented . the first is an algebraic one using generating functions , and the second is a bijection on 01-fillings of moon polyominoes in the spirit of foata s second fundamental transformation on words and permutations .

0902.3637

given a multiset @xmath4 of @xmath5 positive integers , a word on @xmath4 is a sequence @xmath6 that reorders the elements in @xmath4 . when @xmath7:=\{1,\dots , n\}$ ] the word is a permutation . a pair @xmath8 is called an _ inversion _ of @xmath9 if @xmath10 and @xmath11 . one well - known statistic on words and permutations is @xmath12 , defined as the number of inversions of @xmath9 . the _ descent set _ and _ descent statistic _ of a word @xmath9 are defined as @xmath13 in @xcite p. macmahon defined the _ major index _ statistic for a word @xmath9 as @xmath14 and showed the remarkable result that its distribution over all words on @xmath4 is equal to that of the inversion number over the same set . precisely , for the set @xmath15 of all words on @xmath4 , @xmath16 the proof of given by macmahon relied on combinatorial analysis , while an elegant bijection @xmath17 such that @xmath18 was constructed later by foata @xcite . a _ set partition _ of @xmath19 $ ] is a family of nonempty sets @xmath20 which are pairwise disjoint and whose union is @xmath19 $ ] . if all the blocks @xmath21 have at most two elements , the set partition is called a _ matching_. if the size of every block is exactly two , the matching is called a _ perfect matching_. a set partition @xmath22 can be represented as a simple graph with vertex set @xmath19 $ ] drawn on a horizontal line , where two vertices are joined by an arc if and only if they represent consecutive elements in the same block of @xmath22 under the numerical order . two statistics , the numbers of crossings and nestings , arise naturally from such graphical representations , and have been the central topic in many recent research articles , e.g.,@xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , to list a few . a _ crossing _ . _ nesting _ ) of a set partition is a pair of arcs @xmath23 , @xmath24 such that @xmath25 ( resp . @xmath26 ) . a crossing can be viewed as a generalization of an inversion on permutations . hence it is natural to ask what would be the major index for matchings ( set partitions ) that extends macmahon s equidistribution results . a definition of such a statistic , denoted by pmaj , was proposed by chen et al . in @xcite . for completeness , we present the definition for perfect matchings here . the general case for set partitions is very similar . label the arcs of a perfect matching @xmath27 of @xmath28 $ ] by @xmath29 in decreasing order with respect to their left endpoints from left to right . suppose @xmath30 are the right endpoints of arcs from right to left . then to each right endpoint @xmath31 of an arc associate a word @xmath32 on @xmath19 $ ] where the words @xmath32 are defined backwards recursively : let @xmath33 , where @xmath34 is the label of the arc that ends with @xmath35 . in general , after defining @xmath32 , assume that the left endpoints of the arcs labeled @xmath36 lie between @xmath31 and @xmath37 . then @xmath38 is obtained from @xmath32 by deleting entries @xmath36 and adding the entry @xmath39 at the very beginning , where @xmath39 is the label of the arc that ends with @xmath37 . then @xmath40 chen et al . @xcite computed the generating function for @xmath41 and showed that it is equally distributed with the number of crossings over all matchings with fixed left and right endpoints . a similar result holds for set partitions . the aim of this paper is to extend the major index to certain 01-fillings of moon polyominoes which include words and set partitions . in section [ s : definitions ] we introduce the necessary notations , and describe the definition of the major index for 01-fillings of moon polyominoes . we explain in section [ examples ] how the classical definition of major index on words and set partitions can be obtained by considering moon polyominoes of special shapes . in section [ s : proof ] we show that the maj statistic is equally distributed as ne , the number of north - east chains , by computing the corresponding generating functions . in the fillings of special shapes , ne corresponds to the number of inversions for words and crossings for set partitions . therefore , our main result , theorem [ main ] , leads to a generalization of and the analogous result for set partitions . in section [ s : foata ] , we present a bijective proof for the equidistribution of maj and ne , which consists of three maps . the first is from the fillings of left - aligned stack polyominoes to itself that sends @xmath42 to @xmath43 , constructed in the spirit of foata s second fundamental transformation . the other two are maps that transform a moon polyomino to a left - aligned stack polyomino with the same set of columns , while preserving the statistics maj and ne , respectively . composing these three maps yields the desired bijection . a _ polyomino _ is a finite subset of @xmath44 , where we represent every element of @xmath44 by a square cell . the polyomino is _ convex _ if its intersection with any column or row is connected . it is _ intersection - free _ if every two columns are comparable , i.e. , the row - coordinates of one column form a subset of those of the other column . equivalently , it is intersection - free if every two rows are comparable . moon polyomino _ is a convex intersection - free polyomino . if the rows ( resp . columns ) of the moon polyomino are left - aligned ( resp . top - aligned ) , we will call it a _ left - aligned stack polyomino _ ( resp . _ top - aligned stack polyomino _ ) . ferrers diagram _ is a left - aligned and top - aligned stack polyomino . see figure [ moonexample ] for an illustration . we are concerned with 01-fillings of moon polyominoes with restricted row sums . that is , given a moon polyomino @xmath1 , we assign a @xmath45 or a @xmath46 to each cell of @xmath1 so that there is at most one @xmath46 in each row . throughout this paper we will simply use the term _ filling _ to denote such 01-filling . we say a cell is empty if it is assigned a @xmath45 , and it is a @xmath46-cell otherwise . given a moon polyomino @xmath1 with @xmath47 columns , we label them @xmath48 from left to right . let @xmath49 be an @xmath47-tuple of nonnegative integers and @xmath3 be a subset of rows of @xmath1 . we denote by @xmath50 the set of fillings @xmath27 of @xmath1 such that the empty rows of @xmath27 are exactly those in @xmath3 and the column @xmath51 has exactly @xmath52 many 1 s , @xmath53 . a _ north - east ( ne ) chain _ in a filling @xmath27 of @xmath1 is a set of two @xmath46-cells such that one of them is strictly above and to the right of the other and the smallest rectangle containing them is contained in @xmath1 . the number of ne chains of @xmath27 will be denoted @xmath54 . see figure [ moon ] . of a moon polyomino @xmath1 for @xmath55 and @xmath56 with @xmath57.,width=94 ] the set @xmath50 was first studied by a. kasraoui @xcite , who showed that the numbers of north - east and south - east chains are equally distributed . the generating functions for these statistics are also presented in ( * ? ? ? * theorem 4.4 ) . to define the major index for fillings of moon polyominoes , we first state it for rectangular shapes , which is essentially the classical definition of major index for words , ( c.f . section [ examplerec ] ) . let @xmath58 be a filling of a rectangle whose @xmath5 _ nonempty _ rows @xmath30 are numbered from top to bottom . we define the _ descent statistic _ for @xmath58 as @xmath59 that is , descents of a rectangular filling are ne chains in consecutive nonempty rows . for each nonempty row @xmath31 of @xmath58 , let @xmath60 denote the rectangle that contains the row @xmath31 and all rows in @xmath58 above the row @xmath31 . [ defrec ] the major index for the rectangular filling @xmath58 is defined to be @xmath61 clearly , the empty rows of @xmath58 do not play any role in @xmath62 . that is , if we delete them , the major index of the resulting filling remains the same . [ defmajal ] let @xmath27 be a filling of a moon polyomino @xmath1 . let @xmath63 be the list of all the maximal rectangles contained in @xmath1 ordered increasingly by height , and denote by @xmath64 the filling @xmath27 restricted on the rectangle @xmath65 . then the major index of @xmath27 is defined to be @xmath66 where @xmath67 and @xmath68 are defined by . in particular , @xmath69 when @xmath27 is a rectangular shape . it is also clear that @xmath70 is always nonnegative since @xmath71 , @xmath72 . consider the filling from figure [ moon ] . it has four maximal rectangles and @xmath73 . for a moon polyomino using definition 2.2.,width=415 ] the major index can be equivalently defined in a slightly more complicated way . however , this way is useful in proofs , especially when one uses induction on the number of columns of the filling , ( c.f . theorem 4.1 ) . we state this equivalent definition next . first we need some notation . if the columns of @xmath1 are @xmath74 from left to right , it is clear that the sequence of their lengths is unimodal and there exists a unique @xmath75 such that latexmath:[$|c_1| \leq \cdots \leq |c_{k-1}| < |c_k| \geq length of the column @xmath51 . the left part of @xmath1 , denoted by @xmath77 , is the set of columns @xmath51 with @xmath78 and the right part of @xmath1 , denoted by @xmath79 , is the set of columns @xmath51 with @xmath80 . note that the columns of maximal length in @xmath1 belong to @xmath79 . we order the columns of @xmath1 , @xmath74 , by a total order @xmath81 as follows : @xmath82 if and only if * @xmath83 or * @xmath84 , @xmath85 and @xmath86 , or * @xmath84 , @xmath87 and @xmath88 is to the left of @xmath89 , or * @xmath84 , @xmath90 and @xmath88 is to the right of @xmath89 . similar ordering of rows was used in @xcite . for every column @xmath85 define the rectangle @xmath91 to be the largest rectangle that contains @xmath88 as the leftmost column . for @xmath92 , the rectangle @xmath91 is taken to be the largest rectangle that contains @xmath88 as the rightmost column and does not contain any columns from @xmath77 of same length as @xmath51 . let @xmath93 be the ordering of the columns of @xmath1 and let @xmath94 be the restriction of @xmath27 on the rectangle @xmath95 , @xmath96 . then @xmath97 where @xmath98 and @xmath99 are defined by . [ majmoonex ] consider the filling @xmath27 from figure [ moon ] . the order @xmath81 on the columns of @xmath1 is @xmath100 . so , @xmath101 , @xmath102 , @xmath103 , @xmath104 , @xmath105 , and @xmath106 , as illustrated in figure [ moonmaj1 ] . by definition 2.2 , @xmath107 . for a moon polyomino using definition 2.2.,width=415 ] definitions [ defmajal ] and 2.2 are equivalent . some of the rectangular fillings @xmath94 in definition 2.2 may contain @xmath108 , in which case @xmath109 and formula can be simplified . more precisely , there are uniquely determined indices @xmath110 such that @xmath111 . that is , among the rectangle @xmath112 s the maximal ones are @xmath113 . following the notation in definition 2.2 , the filling @xmath114 is the filling @xmath115 of the @xmath75-th maximal rectangle contained in @xmath1 . then , after cancellation of some terms , the right - hand side of becomes the right - hand side of . it is well - known that permutations and set - partitions can be represented as fillings of ferrers diagrams ( e.g. @xcite , @xcite ) . here we will describe such presentations to show how to get the classical major index for words and set partitions from definition 2.2 . for any rectangle , label the rows from top to bottom , and columns from left to right . fillings of rectangles are in bijection with words . more precisely , let @xmath116 be a word with letters in the set @xmath117 $ ] . the word @xmath9 can be represented as a filling @xmath27 of an @xmath118 rectangle @xmath1 in which the cell in row @xmath119 and column @xmath120 is assigned the integer @xmath46 if @xmath121 , and is empty otherwise . conversely , each filling of the rectangle @xmath1 corresponds to a word @xmath9 . it is readily to check that @xmath122 and @xmath123 . this follows from the fact that @xmath124 . as explained in @xcite , general fillings of ferrers diagrams correspond to multigraphs . here we briefly describe the correspondence when restricted to 01-fillings with row sum at most 1 . the ferrers diagram is bounded by a vertical line from the left , a horizontal line from above , and a path consisting of east and north steps . following this path starting from the bottom left end , we label the steps of the path by @xmath125 . this gives a labeling of the columns and rows of the diagram . draw @xmath5 vertices on a horizontal line and draw an edge connecting vertices @xmath126 and @xmath127 if and only if there is a 1 in the cell of the column labeled @xmath126 and the row labeled @xmath127 ( see figure [ ferrers1 ] ) . the resulting graph has no loops or multiple edges and every vertex is a right endpoint of at most one edge . the ne chains correspond to crossings of edges in the graph . if the column sums are at most one , the graph is a matching and the major index of the filling is equal to the major index of the matching as defined in @xcite . as explained in @xcite , the case of set partitions can be reduced to that of matchings . hence fillings of ferrers diagrams include set partitions as well . restricting to the triangular ferrers diagram , the 01-fillings considered in this paper also include _ linked partitions _ , a combinatorial structure that was studied in @xcite . in the case when @xmath1 is a top - aligned stack polyomino , @xmath70 has a simpler form in terms of the des statistic , just as in words and permutations . explicitly , if the nonempty rows of @xmath27 are @xmath30 , denote by @xmath128 the filling @xmath27 restricted to the largest rectangle contained in @xmath1 whose bottom row is @xmath31 . [ stackprop ] let @xmath27 be a filling of a top - aligned stack polyomino and @xmath128 be as defined above . then @xmath129 let @xmath130 be all the nonempty rows of @xmath27 that are of same length as the last row @xmath131 . denote by @xmath132 and by @xmath58 the filling of the largest rectangle of @xmath1 containing @xmath131 . we proceed by induction on the number @xmath5 of rows of @xmath27 . the claim is trivial when @xmath27 has only one row or is a rectangle . otherwise , @xmath133 which completes the proof . consider the filling @xmath27 of a top - aligned stack polyomino in figure [ stackmaj ] with @xmath55 and @xmath134 . @xmath27 has five nonempty rows and , by proposition [ stackprop ] , @xmath135 . using proposition [ stackprop].,width=529 ] proposition 3.1 will be used in the proof of lemma 5.5 . in this section we state and prove the main theorem which gives the generating function for maj over @xmath50 , the set of all fillings of the moon polyomino @xmath1 with column sums given by the integer sequence @xmath2 and empty rows given by @xmath3 . let @xmath136 be the column lengths of @xmath27 and let @xmath137 the number of rows in @xmath3 that intersect the column @xmath51 . suppose that , under the ordering defined in section [ s : definitions ] , @xmath138 . then for @xmath139 , define @xmath140 the numbers @xmath141 have the following meaning : if one fills in the columns of @xmath27 from smallest to largest according to the order @xmath81 , then @xmath142 is the number of available cells in the @xmath127-th column to be filled . [ main ] for a moon polyomino @xmath1 with @xmath47 columns , @xmath143{0pt}{}{h_i}{s_i}_q\ ] ] where @xmath141 is defined by . we postpone the proof of theorem [ main ] until the end of this section . suppose @xmath1 is the first moon polyomino in figure [ moonexample ] , @xmath144 , and @xmath145 . the @xmath81 order on the columns of @xmath1 is : @xmath146 . the fifth row intersects all columns except @xmath147 , so , @xmath148 and @xmath149 . therefore , @xmath150 by theorem [ main ] , @xmath151{0pt}{}{h_i}{s_i}_q= \genfrac[]{0pt}{}{2}{1}_q \genfrac[]{0pt}{}{1}{0}_q\genfrac[]{0pt}{}{2}{2}_q\genfrac[]{0pt}{}{3}{1}_q \genfrac[]{0pt}{}{1}{1}_q = ( 1+q)(1+q+q^2)=1 + 2q+2q^2+q^3 $ ] . the fillings in @xmath50 are listed in figure [ theoremexample ] . .,width=340 ] let @xmath152 be a permutation of @xmath117 $ ] and let @xmath1 be a moon polyomino with columns @xmath74 . suppose the shape @xmath153 with columns @xmath154 is also a moon polyomino and @xmath155 with @xmath156 . then @xmath157 that is , the generating function @xmath158 does not depend on the order of the columns of @xmath1 . the formula seems to depend on the order @xmath81 of the columns of the moon polyomino . however , note that @xmath82 if @xmath159 ; and columns of same length are consecutive in the order @xmath81 . hence it suffices to compare the terms in the right - hand side of that come from columns of fixed length . if @xmath160 are all the columns of @xmath1 of fixed length , then @xmath161 thus the contribution of these columns to the right - hand side of is @xmath162{0pt}{}{h_{i_{j+r}}}{s_{i_{j+r}}}_q = \genfrac[]{0pt}{}{h_{i_{j+1}}}{s_{i_{j+1 } } , \dots , s_{i_{j+k } } , h_{i_{j+1}}- ( s_{i_{j+1}}+ \cdots + s_{i_{j+k } } ) } _ q.\ ] ] the last q - multinomial coefficient is invariant under permutations of @xmath163 , and the claim follows . the right - hand side of formula also appeared in the generating function of ne chains , as shown in @xcite : [ kastheorem ] @xmath164{0pt}{}{h_i}{s_i}_q.\ ] ] combining theorems [ main ] and [ kastheorem ] , we have [ equal ] @xmath165 that is , the maj statistic has the same distribution as the ne statistic over the set @xmath166 . since @xmath54 is the natural analogue of inv for words and permutations , theorem [ equal ] generalizes macmahon s equidistribution result . we shall prove a lemma about the major index for words and then use it to prove theorem [ main ] . [ perm ] let @xmath167 be a word such that @xmath168 for all @xmath126 . consider the set @xmath169 of all the words @xmath170 that can be obtained by inserting @xmath47 many @xmath5 s between the letters of @xmath9 . then the difference @xmath171 ranges over the multiset @xmath172 . it follows that @xmath173{0pt}{}{k+m}{m}_q.\ ] ] the same statement holds if @xmath167 is a word such that @xmath174 for all @xmath126 and @xmath169 is the set of all the words @xmath170 obtained by inserting @xmath47 many 1 s between the letters of @xmath9 . we give an elementary proof for the case of inserting @xmath5 s . the case of inserting 1 s is dealt with similarly . for a word @xmath175 we define the _ descent sequence _ @xmath176 by letting @xmath177 , where @xmath32 is obtained by inserting one @xmath5 in the @xmath126-th gap of @xmath9 , i.e. , @xmath178 . note that @xmath179 hence , if @xmath180 then @xmath181 is a shuffle of the sequences @xmath182 and @xmath183 which begins with @xmath184 and ends with @xmath45 . moreover , every such sequence is @xmath181 for some word @xmath9 . it is not hard to see that the multiset @xmath185 depends only on @xmath181 and not on the letters of @xmath9 . first we consider the case when @xmath186 , i.e. , @xmath187 $ ] . inserting @xmath39 @xmath5 s between @xmath188 and @xmath189 increases the major index by @xmath190 . hence , if one inserts @xmath191 many @xmath5 s between @xmath188 and @xmath189 for @xmath192 , the major index is increased by @xmath193 letting @xmath194 range over all @xmath195-tuples with @xmath196 , the increased amount ranges over the multiset @xmath197 . to deal with the other possible descent sequences , we note that any descent sequence @xmath181 can be obtained by applying a series of adjacent transpositions to @xmath198 , where we only transpose adjacent elements @xmath199 which correspond to ascent and descent positions , respectively . suppose @xmath200 is obtained by transposing the @xmath126-th and @xmath201-st element in @xmath202 , where @xmath137 and @xmath203 correspond to an ascent and a descent in @xmath204 , respectively . consequently , @xmath205 and @xmath206 , while @xmath207 if and only if @xmath208 for @xmath209 . it suffices to show that the multisets @xmath185 and @xmath210 are equal . suppose @xmath211 is obtained by inserting @xmath212 many @xmath5 s in the @xmath127-th gap of @xmath9 , @xmath213 . let @xmath214 be the word obtained by inserting @xmath215 @xmath5 s in the @xmath127-th gap of @xmath204 where @xmath215 are defined as follows . + if @xmath216 then @xmath217 if @xmath218 then @xmath219 this defines a map between sequences @xmath220 with @xmath216 ( resp . @xmath218 ) , and @xmath221 with @xmath222 ( resp . @xmath223 ) , which is a bijection . the @xmath5 s inserted in the @xmath127-th gap of @xmath9 for @xmath224 contribute to @xmath171 the same amount as do the @xmath5 s inserted in the same gaps in @xmath204 to @xmath225 . the major index contributed to @xmath226 by the segment @xmath227 is the sum of @xmath228 ( which equals the contribution of @xmath229 to @xmath230 ) and @xmath231 similarly , the major index contributed to @xmath232 by the segment @xmath233 is the sum of @xmath126 ( which equals the contribution of @xmath234 to @xmath235 ) and @xmath236 now , one readily checks that @xmath237 . this completes the proof . [ gessel ] lemma [ perm ] can be proved in many ways , for example , in ( * ? ? ? * theorem 1.2 ) foata and han constructed a transformation on the shuffle class sh@xmath238 , where @xmath204 is a nonempty word whose letters are positive integers . alternatively , one can use the theory of @xmath239-partitions @xcite , which leads to a stronger result , namely , equation holds if @xmath167 with @xmath240 , @xmath241 and @xmath169 is the set of all words @xmath170 obtained by inserting @xmath47 copies of @xmath242 between the letters of @xmath9 . to show this , it is enough to consider the case when @xmath9 is a permutation of length @xmath75 on @xmath75 distinct numbers between 1 and @xmath243 such that the integers @xmath244 are missing from it and @xmath170 varies over the set @xmath169 of all permutations obtained by inserting @xmath245 in @xmath9 from left to right . we outline the proof in this general case assuming that the reader is familiar with the theory of @xmath239-partitions . we adopt the notation and definitions from section 4.5 in @xcite , which is also a good reference for basic results on @xmath239-partitions . let @xmath246 be the antichain with @xmath75 elements labeled with the letters of @xmath9 . let @xmath247 be a disjoint union of @xmath246 and a chain of @xmath47 elements labeled @xmath248 . we denote by @xmath249 the generating function of all @xmath9-compatible functions @xmath250 \rightarrow \mathbb{n}$ ] ( see @xcite for the definitions ) . lemma 4.5.2 in @xcite asserts that @xmath251 all @xmath247-partitions compatible with some @xmath211 are obtained by combining a @xmath246-partition compatible with @xmath9 and a integer partition with at most @xmath47 parts . therefore , @xmath252 applying lemma 4.5.2 in @xcite to all @xmath211 , we have @xmath253 combining equations , , and gives . let @xmath254 be the smallest column of @xmath1 in the order @xmath81 and @xmath255 . note that @xmath254 is the leftmost or the rightmost column of @xmath1 . let @xmath27 be a filling of @xmath1 with @xmath256 nonempty cells in @xmath254 and let @xmath257 be its restriction on @xmath258 . from definition 2.2 , we derive that @xmath259 where @xmath260 is the filling of the rectangle @xmath261 determined by the column @xmath254 . @xmath260 is obtained by adding a column with @xmath256 many @xmath46-cells to the rectangular filling @xmath262 . using the bijection between fillings of rectangles and words described in section [ examplerec ] , one sees that if @xmath254 is the leftmost ( resp . rightmost ) column in @xmath27 , this corresponds to inserting @xmath256 maximal ( resp . minimal ) elements in the word determined by the filling @xmath262 . when the column @xmath254 varies over all possible @xmath263 fillings , by lemma [ lemma ] , the value @xmath264 varies over the multiset @xmath265 with generating function @xmath266{0pt}{}{h}{s}_q$ ] , where @xmath267 is the value @xmath268 ( resp . @xmath269 ) if @xmath254 is the leftmost ( resp . the rightmost ) column of @xmath27 , as defined in . therefore , @xmath270{0pt}{}{h}{s}_q \sum_{m ' \in { \textbf{f}}({\mathcal{m } } ' , { \textbf{s}}';a ) } q^{{\mathrm{maj}}(m')}\ ] ] where @xmath271 is obtained from @xmath2 by removing the first ( resp . last ) component if @xmath254 is the leftmost ( resp . rightmost ) column of @xmath27 . equation now follows by induction on the number of columns of @xmath1 . the objective of this section is to give a bijective proof to theorem 4.5 . for the set @xmath15 of all words of a multiset @xmath4 , the equidistribution of maj and inv was first proved by macmahon by combinatorial analysis . this raised the question of constructing a canonical bijection @xmath272 such that @xmath273 . foata answered the question by constructing an elegant map @xmath274 @xcite , which is referred to as the _ second fundamental transformation _ we begin this section by reviewing foata s map @xmath272 . let @xmath275 be a word on @xmath276 and let @xmath34 be an integer . if @xmath277 , the @xmath34-factorization of @xmath9 is @xmath278 , where each @xmath191 is a letter less than or equal to @xmath34 , and each @xmath279 is a word ( possibly empty ) , all of whose letters are greater than @xmath34 . similarly , if @xmath280 , the @xmath34-factorization of @xmath9 is @xmath278 , where each @xmath191 is a letter greater than @xmath34 , and each @xmath279 is a word ( possibly empty ) , all of whose letters are less than or equal to @xmath34 . in each case one defines @xmath281 with the above notation , let @xmath282 and let @xmath283 . the second fundamental transformation @xmath274 is defined recursively by @xmath284 if @xmath9 has length @xmath46 , and @xmath285 if @xmath9 has length @xmath286 . the map @xmath274 has the property that it preserves the last letter of the word , and @xmath287 . foata s map @xmath274 is constructed recursively with certain `` local operations '' to eliminate the difference caused by adding the last letter in the words . inspired by this idea , we construct a foata - type bijection @xmath288 with the property @xmath289 . the map @xmath290 can be defined directly for left - aligned stack polyominoes . but we describe it first for ferrers diagrams in section 5.1 , because this case contains all the essential steps and is easy to understand . this map is a revision of the foata - type bijection presented in the preprint of @xcite for set partitions , which correspond to fillings with row and column sums at most 1 . then in section 5.2 we extend the construction to left - aligned stack polyominoes and prove that @xmath289 . in section 5.3 , we construct two bijections , @xmath291 and @xmath292 , that transform a filling of a moon polyomino to a filling of a left - aligned stack polyomino and preserve the statistics maj and ne , respectively . composing these maps with @xmath290 defined on left - aligned stack polyominoes yields a bijection on @xmath0 that sends maj to ne . the empty rows in the filling @xmath293 do not play any role in the definitions of @xmath70 or @xmath54 . therefore , in what follows we assume that @xmath294 and describe @xmath290 for fillings without empty rows . for @xmath295 , one can first delete the empty rows of the filling , apply the map @xmath290 , and reinsert the empty rows back . let @xmath296 be a ferrers diagram and @xmath297 a filling of @xmath296 . the bijection @xmath290 is defined inductively on the number of rows of @xmath297 , which preserves the first row of @xmath297 . if @xmath297 has only one row , then @xmath298 . otherwise , we denote by @xmath299 the filling obtained by deleting the top row @xmath300 of @xmath297 . let @xmath301 and let @xmath302 be the filling obtained by performing the algorithm @xmath303 described below . then @xmath304 is obtained from @xmath302 by adding the top row @xmath300 . [ [ section ] ] * * ~~ if @xmath305 is the @xmath46-cell of @xmath300 , then denote by @xmath306 the set of all rows of @xmath307 that intersect the column of @xmath305 . the @xmath46-cells in @xmath306 that are strictly to the left of @xmath305 are called left and the @xmath46-cells in @xmath306 that are weakly to the right of @xmath305 are called right the cell @xmath305 is neither left nor right . let @xmath308 ( critical left cell ) be the topmost left @xmath46-cell , and @xmath309 ( critical right cell ) be the leftmost right @xmath46-cell that is above @xmath308 . if there is more than one such cell the @xmath309 is defined to be the lowest one . denote by @xmath310 the set of all rows weakly below the row of @xmath309 that intersect the column of @xmath309 and @xmath311 is the set of all rows in @xmath306 that do not intersect the column of @xmath309 . if @xmath308 does not exist then both @xmath310 and @xmath311 are empty . if @xmath308 exists but @xmath309 does not , then @xmath310 is empty and @xmath311 contains all the rows in @xmath306 weakly below @xmath308 . [ expressions ] let @xmath312 and @xmath313 be two @xmath46-cells with coordinates @xmath23 and @xmath24 , respectively . we _ swap the cells @xmath312 and @xmath313 _ by deleting the 1 s from these two cells and write 1 s in the cells with coordinates @xmath314 and @xmath315 . for a cell @xmath305 , denote by @xmath316 the column of @xmath305 , and @xmath317 the length of @xmath316 . 1 . 2 . 3 . see figure [ alg1 ] for illustration of the steps . * @xmath319note : * this step is well - defined since it can not occur before @xmath320 reaches @xmath308 ( that would contradict the definition of @xmath309 ) and , after that , @xmath321 is always below @xmath308 . therefore , the @xmath46-cell @xmath322 always exists . the fact that the square @xmath323 belongs to @xmath324 follows from the definition of @xmath310 and lemma 5.2 ( c ) . on @xmath310.,width=566 ] when the algorithm @xmath318 stops , we continue processing the rows of @xmath311 by algorithm @xmath325 . 1 ( _ * borrowing * _ ) 3 . 4 . see figure [ alg2 ] for illustration of the steps in @xmath325 . on @xmath311.,width=566 ] let us note the following easy but useful properties of the algorithm @xmath303 . 1 . the pointers @xmath321 and @xmath320 process the rows of @xmath310 and @xmath311 from top to bottom . however , while @xmath320 always moves from one row to the next one below it , @xmath321 sometimes stays on the same row ( cf . step ( b2.2 ) ) and sometimes `` jumps '' several rows below ( cf . step ( b1 ) ) . 2 . pointer @xmath321 always points to a right @xmath46-cell in @xmath318 and to a left @xmath46-cell in @xmath325 . further we show some nontrivial properties which will be used to show that the map @xmath290 has the desired property @xmath326 . [ prop ] suppose in @xmath318 @xmath321 is pointing to @xmath312 and @xmath320 is pointing to @xmath313 . then * in the rows between @xmath312 and @xmath313 there are no 1 s weakly to the left of @xmath312 . * if @xmath322 is the lowest left @xmath46-cell above @xmath312 , then in the rows between @xmath322 and @xmath312 there are no 1 s weakly to the left of @xmath312 . * the @xmath46-cell @xmath312 is in a column with same length as the column of the critical right cell @xmath309 . moreover , when @xmath318 stops @xmath321 points to the last row in @xmath310 . the first two parts are proved by induction on the number of steps performed in the algorithm . * the only step which leaves the gap between @xmath321 and @xmath320 nonempty is ( b2.2 ) . but after this step , the number of 1 s in the rows between @xmath321 and @xmath320 weakly to the left of @xmath312 does not change . * the case when the last step is ( b2.2 ) follows from the induction hypothesis . if the last step is ( b1 ) or ( b2.3 ) the claim follows from part ( a ) . the case when the last step was ( b2.1 ) and the cells @xmath312 and @xmath313 formed an ne chain the claim also follows from part ( a ) and the induction hypothesis . if in ( b2.1 ) , the cells @xmath312 and @xmath313 do not form an ne chain , in addition to the induction hypothesis , one uses the fact that there are no 1 s in the rows between these two cells which are weakly to the left of @xmath313 . * the first part follows immediately from the definition of the steps . if the second part is not true , then the last step must be ( b2.2 ) . and the column of @xmath312 is longer than the column of @xmath313 . since this is the last step of @xmath318 , it follows that the column of @xmath312 intersects the top row of @xmath311 . however , the column of @xmath309 does not intersect the top row of @xmath311 by definition of @xmath311 . this contradicts the fact that @xmath312 and @xmath309 are in columns of same length . [ alg ] after algorithm @xmath318 the number of ne chains decreases by the total number of left @xmath46-cells in @xmath310 , whereas after algorithm @xmath325 the number of ne chains increases by the total number of right @xmath46-cells in @xmath311 . to prove the first part note that pointer @xmath320 points to each left @xmath46-cell in @xmath310 exactly once in the algorithm @xmath318 when that happens , step ( b1 ) is performed during which , by lemma [ prop](a ) , the number of ne chains decreases by one . in the other steps , the number of ne chains remains unchanged . this is trivial for steps ( b2.1 ) and ( b2.2 ) , whereas for step ( b2.3 ) it follows from lemma [ prop ] ( a ) and ( b ) . for the second part , note that pointer @xmath320 points to each right @xmath46-cell in @xmath311 exactly once in the algorithm @xmath325 with possible exception being the top @xmath46-cell in @xmath311 to which it never points . in those steps the number of ne chains is increased by one , while otherwise it remains the same . when the top @xmath46-cell is right , then borrowing occurs . it follows from parts ( b ) and ( c ) of lemma [ prop ] that the number of ne chains is also increased by one . applied inductively on @xmath327 . the fillings @xmath328 are restrictions of @xmath297 on the last @xmath126 rows.,width=566 ] [ ferrer ] the map @xmath329 is a bijection . the proof is quite technical and is given in the appendix . here we only present an example of the bijection @xmath290 in figure [ bijex ] and explain how it includes foata s second fundamental transformation . if the moon polyomino is rectangular , then the algorithm @xmath303 has a simpler description because some of the cases can never arise . namely , if the cell in the second row forms an ne chain with the top cell ( a descent ) then @xmath330 , and only the steps ( c1 ) and ( c2 ) are performed . otherwise , @xmath331 and , since all columns are of equal height , only the steps ( b1 ) and ( b2.1 ) are performed . in both these cases , via the correspondence between words and rectangular fillings ( c.f . section [ examplerec ] ) the algorithms @xmath325 and @xmath318 for rectangles are equivalent to foata s transformation for words @xcite . next we extend the bijection @xmath290 to fillings of left - aligned stack polyominoes which sends maj to ne . let @xmath1 be a left - aligned stack polyomino . as in the case of ferrers diagrams , we can assume that @xmath332 and only consider fillings without empty rows . suppose @xmath27 is a filling of @xmath1 with top row @xmath300 . the map , which is again denoted by @xmath290 , is defined inductively on the number of rows of @xmath27 . if @xmath27 has only one row , then @xmath333 . otherwise , let @xmath296 be the maximal ferrers diagram in @xmath1 that contains the top row and @xmath334 be @xmath296 without the top row . let @xmath335 and @xmath336 . to obtain @xmath337 we perform the following steps . ( see figure [ stack1 ] ) . 1 . delete the top row of @xmath27 and get @xmath338 . 2 . apply @xmath290 to @xmath339 and get @xmath340 . 3 . apply the algorithm @xmath303 to the filling @xmath341 and leave the cells in @xmath342 outside of @xmath307 unchanged . denote the resulting filling by @xmath343 . @xmath337 is obtained by adding row @xmath300 back to @xmath343 . for left - aligned stack polyominoes.,width=566 ] it is a bijection from @xmath0 to itself since every step is invertible . that the step 3 is invertible follows from the proof of theorem [ ferrer ] . [ mtf ] using the notation introduced above , @xmath344 where @xmath345 is defined with respect to the top row @xmath300 of @xmath297 as in section [ ferrers ] . let @xmath4 be the filling @xmath27 restricted to the rows which are not completely contained in @xmath296 . from the definition of @xmath346 , we have @xmath347 note that @xmath348 , hence and give the first equality in . the second one is readily checked using proposition [ majstack ] for major index of stack polyominoes . the rest of this subsection is devoted to proving that the map @xmath290 has the desired property @xmath289 . we shall need one more lemma . let @xmath305 be a @xmath46-cell in a filling @xmath27 of a left - aligned stack polyomino . we say that @xmath305 is _ maximal _ if there is no 1 in @xmath27 which is strictly above and weakly to the left of @xmath305 . clearly , a maximal cell must be the highest @xmath46-cell in its column . for a maximal cell @xmath305 in @xmath27 , we use @xmath349 to denote the index of the leftmost column ( the columns are numbered from left to right ) to the right of @xmath305 that contains a @xmath46-cell which together with @xmath305 forms an ne chain . if such a column does not exist set @xmath350 . [ maxcell ] a @xmath46-cell @xmath305 in @xmath27 is maximal if and only if the highest @xmath46-cell @xmath351 in the same column in @xmath352 exists and is maximal . moreover , @xmath353 when they are both maximal . we proceed by induction on the number of rows of @xmath27 and keep the notation as in figure 11 . the case when @xmath27 has one row is trivial . suppose that the top row @xmath300 of @xmath27 has a @xmath46-cell @xmath354 in the column @xmath355 . the cell @xmath354 is maximal in both @xmath27 and @xmath257 and @xmath356 . note that a @xmath46-cell @xmath357 in @xmath27 ( resp . @xmath257 ) is maximal if and only if @xmath305 is in column @xmath358 and is maximal in @xmath339 ( resp . @xmath343 ) . denote by @xmath359 , @xmath360 , and @xmath361 the highest @xmath46-cells in column @xmath358 in the fillings @xmath339 , @xmath340 , and @xmath343 , respectively . by the induction hypothesis , @xmath359 is maximal in @xmath339 if and only if @xmath360 is maximal in @xmath342 . @xmath343 is obtained from @xmath342 by applying the algorithm @xmath303 with respect to the row @xmath300 to @xmath307 . @xmath360 is a left cell in this algorithm , and since @xmath303 does not change the relative position of the left @xmath46-cells , @xmath361 is maximal in @xmath343 . it only remains to prove that @xmath353 . suppose @xmath362 ( they are equal by the inductive hypothesis ) . if @xmath363 then @xmath364 . in @xmath342 it corresponds to a @xmath46-cell @xmath365 in a column left of @xmath354 such that @xmath366 form an ne chain . since the algorithm @xmath303 preserves the column sums and the relative positions of left @xmath46-cells , we have @xmath367 , and hence @xmath368 . if @xmath369 then @xmath370 . the fact that @xmath371 means that there is no left @xmath46-cell in @xmath342 above and to the right of @xmath360 and , since @xmath303 does not change the relative positions of the left @xmath46-cells , there will be no left @xmath46-cell above and to the right of @xmath361 in @xmath343 . so , in this case , we conclude @xmath372 . [ thmphi ] if @xmath27 is a filling of a left - aligned stack polyomino @xmath1 , then @xmath289 . again we proceed by induction on the number of rows of @xmath27 . the case when @xmath27 has one row is trivial . we use the notation as in figure 11 , and let @xmath373 . by proposition [ alg ] , the algorithm @xmath303 on @xmath342 decreases ne by one for each left @xmath46-cell in @xmath374 and increases it by one for each right @xmath46-cell in @xmath375 . therefore , @xmath376 since @xmath303 preserves the column sums , @xmath377 hence @xmath378 one the other hand , from lemma [ mtf ] and the induction hypothesis , one gets @xmath379 so , it suffices to show that @xmath380 to show equation , note that for a filling of a ferrers diagram @xmath297 the number of rows in @xmath381 is determined by the column indices of @xmath308 ( critical left cell ) and @xmath309 ( critical right cell ) . the @xmath308 is the topmost left @xmath46-cell which , if exists , is the topmost _ maximal _ cell besides the @xmath46-cell in the first row . by lemma [ maxcell ] , @xmath308 in @xmath382 is in the same column as @xmath383 in @xmath384 , or neither of them exists . the latter case is trivial , as @xmath381 is empty . in the first case , let @xmath385 denote the length of row @xmath300 . then @xmath309 , the critical right cell of @xmath299 , is in the column @xmath386 if @xmath387 and does not exist otherwise . similarly , the critical right cell of @xmath307 , @xmath388 , is in the column @xmath389 if @xmath390 , and does not exist otherwise . by lemma [ maxcell ] , @xmath391 , and this implies . now we consider the case when @xmath1 is a general moon polyomino . we label the rows of @xmath1 by @xmath30 from top to bottom . let @xmath153 be the unique left - aligned stack polyomino whose sequence of row lengths is equal to @xmath392 from top to bottom . in other words , @xmath153 is the left - aligned polyomino obtained by rearranging the columns of @xmath1 by length in weakly decreasing order from left to right . for the definitions that follow , we use an idea of m. rubey @xcite to describe an algorithm that rearranges the columns of @xmath1 to obtain @xmath153 ( figure [ alpha ] ) . + 1 . set @xmath393 . if @xmath258 is left aligned go to ( 4 ) . if @xmath258 is not left - aligned consider the largest rectangle @xmath306 completely contained in @xmath258 that contains @xmath147 , the leftmost column of @xmath258 . update @xmath258 by letting @xmath258 be the polyomino obtained by moving the leftmost column of @xmath306 to the right end . go to ( 2 ) . + 4 . set @xmath394 . + .,width=566 ] in this section we give two bijections @xmath395 which preserve the major index and the number of ne chains , respectively . the sequence @xmath271 is obtained by rearranging the sequence @xmath2 in the same way @xmath153 is obtained by rearranging the columns of @xmath1 . let @xmath58 be a filling of a rectangle @xmath306 with column sums @xmath398 . first , we describe a transformation @xmath399 which gives a filling of @xmath306 with column sums @xmath400 and preserves the major index . recall that a descent in @xmath58 is a pair of 1 s in consecutive nonempty rows that form an ne chain . define an _ ascent _ to be a pair of 1 s in two consecutive nonempty rows that does not form an ne chain . suppose the first column of @xmath58 has @xmath75 many 1 s : @xmath401 from top to bottom . for rectangles.,width=566 ] 1 . let @xmath402 be the rectangle obtained by adding one empty column to @xmath58 from right . process the @xmath46-cells @xmath401 from bottom to top ( see figure [ tau ] ) : 2 . for @xmath403 do the following : + let @xmath404 be the lowest @xmath46-cell above @xmath405 in @xmath402 and @xmath406 be the highest @xmath46-cell below @xmath405 in @xmath402 . if @xmath404 does not exist or @xmath407 , and @xmath406 does not exist , just move @xmath405 horizontally to the last column of @xmath402 ; 2 . if ( i ) @xmath404 does not exist or @xmath407 , but @xmath406 exists , or ( ii ) both @xmath404 and @xmath406 exist , and @xmath404 is to the right of @xmath406 : + let @xmath408 be the maximal chain of consecutive ascents in @xmath402 . move @xmath409 horizontally to the column of @xmath410 , for @xmath411 , and move @xmath412 horizontally to the last column of @xmath402 . ( if @xmath412 is in the last column of @xmath402 , it remains there . ) 3 . if @xmath413 , and ( @xmath404 is weakly to the left of @xmath406 or @xmath406 does not exist ) : + let @xmath414 be the maximal chain of consecutive descents in @xmath402 . move @xmath409 horizontally to the column of @xmath410 , for @xmath411 , and move @xmath412 horizontally to the last column of @xmath402 . 3 . delete the first column of @xmath402 . note that each iteration in step 2 decreases the number of 1 s in the first column of @xmath402 by one and increases the number of 1 s in the last column by one , while the other column sums remain unchanged . now we are ready to define the map @xmath291 . suppose @xmath415 . we perform the algorithm @xmath416 on @xmath27 to transform the shape @xmath1 to @xmath153 . while we are in step 3 , instead of just moving the first column of @xmath306 to the right end , we perform the algorithm @xmath399 on the filling @xmath58 of @xmath306 . @xmath417 is defined to be the resulting filling of @xmath153 . [ lemmatau ] the transformation @xmath399 is invertible . moreover , it preserves the descents and hence @xmath418 . the map @xmath399 is invertible since @xmath419 can be obtained by taking @xmath420 , where @xmath421 is the rotation of rectangles by @xmath422 . moreover , each iteration of step 2 of @xmath399 preserves the positions of descents , hence the second part of the claim holds . [ propf ] the map @xmath423 is a bijection and @xmath397 . the first part follows from lemma [ lemmatau ] and the fact that @xmath416 is invertible . the filling @xmath417 is obtained after several iterations of step 3 in the algorithm @xmath416 combined with application of @xmath399 . for the second part of the claim , it suffices to show that maj is preserved after one such iteration . let @xmath257 be the filling of the shape @xmath258 obtained from @xmath27 after one step of @xmath291 in which the rectangle @xmath58 is transformed into @xmath424 . we shall use definition 2.2 to compare @xmath70 and @xmath425 . let @xmath426 ( resp . @xmath427 ) be the maximal rectangles of @xmath27 ( resp . @xmath257 ) , ordered by height . in particular , @xmath64 and @xmath428 are of same size . suppose @xmath429 , then @xmath430 . we partition the index @xmath126 into three intervals , according to @xmath431 , @xmath432 , and @xmath433 , and compare the contribution of @xmath64 s in each interval . 1 . for @xmath434 , the rectangles @xmath64 and @xmath435 are of smaller height than @xmath58 and contain several consecutive rows of @xmath58 . using lemma [ lemmatau ] we know that the descents in @xmath436 ( resp @xmath437 ) @xmath434 , are in the same rows as in @xmath438 ( resp . @xmath439 ) . and descents formed by one cell in @xmath58 and one cell on the right of @xmath58 are preserved since @xmath399 preserves the row sum of @xmath58 . consequently , @xmath440 2 . for @xmath432 , by lemma [ lemmatau ] , we have @xmath441 3 . for @xmath442 , the height of the rectangle @xmath64 or @xmath443 is greater than or equal to that of @xmath58 . each of these rectangles contains several consecutive columns of @xmath58 excluding the first one . the filling of each subrectangle of @xmath58 consisting of the columns @xmath444 ( @xmath445 ) is equal , up to a rearrangement of the empty rows , to the filling of the subrectangle of @xmath424 consisting of the columns @xmath446 . to see this , note that the shifting of the @xmath46-cells @xmath447 horizontally in the algorithm @xmath399 can be viewed as moving the @xmath46-cell @xmath409 vertically to the row of @xmath448 and moving @xmath449 to the last column of @xmath402 in the row of @xmath412 . thus , the descent positions in the fillings @xmath436 ( resp . @xmath450 ) , @xmath442 are the same as in @xmath438 ( resp . @xmath451 ) . finally , one checks that the descents formed by one cell in @xmath58 and one cell outside @xmath58 are also preserved . therefore , @xmath452 combining , , and gives @xmath453 . suppose @xmath415 . to obtain @xmath456 , we perform the algorithm @xmath416 to transform the shape @xmath1 to @xmath153 and change the filling when we move columns in step 3 so that the number of 1 s in each row and column is preserved . let @xmath58 be the rectangular filling in step 3 of @xmath416 that contains the column @xmath147 of the current filling . suppose @xmath147 contains @xmath75 many @xmath46-cells @xmath457 from top to bottom . shade the empty rows of @xmath58 and the cells in @xmath58 to the right of @xmath457 . let @xmath458 denote the number of empty white cells in @xmath58 above @xmath459 . if @xmath460 is the rectangle obtained by moving the column @xmath147 from first to last place , fill it to obtain a filling @xmath461 as follows . 1 . the rows that were empty remain empty . shade these rows . 2 . write @xmath75 many 1 s in the last column from bottom to top so that there are @xmath458 white empty cells below the @xmath126-th 1 , @xmath462 . 3 . shade the cells to the left of the nonempty cells in the last column . fill in the rest of the rectangle by writing 1 s in the unshaded rows of @xmath460 so that the unshaded part of @xmath58 to the right of the first column is the same as the unshaded part of @xmath461 to the left of the last column . see figure [ mapg ] for an example . .,width=566 ] [ propg ] the map @xmath463 is a bijection and @xmath455 . clearly , every step of @xmath292 is invertible . to see that ne is preserved , it suffices to show that it is preserved in each step of @xmath292 , when one column is moved from left to right . first note that @xmath464 . this follows from the fact that the 1 s in the first column of @xmath58 form @xmath465 ne chains from top to bottom , while the 1 s in the last column of @xmath461 form @xmath465 ne chains from bottom to top and the remaining parts of @xmath58 and @xmath461 are essentially equal . the numbers of ne chains in @xmath27 and @xmath257 respectively made of at least one 1 which is outside of @xmath58 and @xmath461 respectively are equal because each step of @xmath292 preserves the row and column sums . let @xmath290 denote the foata - type bijection described in section [ stack ] . for a general moon polyomino @xmath27 the map @xmath466 is a bijection with the property @xmath467 the maps @xmath291 and @xmath292 permute the columns of @xmath27 together with the corresponding number of 1 s , while @xmath290 preserves the column sums . moreover , all three of them preserve the empty rows . therefore , @xmath468 is indeed a map from @xmath50 to itself . finally , follows from proposition [ propf ] , proposition [ propg ] , and theorem [ thmphi ] . the authors would like to thank ira gessel @xcite for suggesting a connection between the enumeration of permutations by major index and the theory of @xmath239-partitions . to show that the map @xmath329 is a bijection , it suffices to describe the inverse of the algorithm @xmath303 . first , we need to determine what @xmath310 and @xmath311 were . for that we use the following properties of @xmath303 . 1 . [ r21 ] when @xmath318 stops , the pointer @xmath321 is at the lowest row of @xmath310 pointing at a right @xmath46-cell @xmath359 . the algorithm @xmath325 does not change the cell @xmath359 and when the whole algorithm @xmath303 terminates , there is no @xmath46-cell below @xmath312 that together with it forms an ne chain . [ r22 ] after performing @xmath325 , all the right @xmath46-cells in @xmath311 have at least one @xmath46-cell below them with which they form an ne chain . if @xmath311 was nonempty then , after performing @xmath303 , the lowest row of @xmath311 contains a left @xmath46-cell . [ borrowing ] if any borrowing occurred and the right cell @xmath469 and the left cell @xmath470 were swapped , then right after this step the cell @xmath314 is a right @xmath46-cell in @xmath310 which forms an ne chain with a left @xmath46-cell from @xmath311 and this is the lowest @xmath46-cell in @xmath310 with this property . therefore , if @xmath306 has no left @xmath46-cells then @xmath471 . otherwise , find the lowest right @xmath46-cell @xmath354 in @xmath306 such that there is no @xmath46-cell in @xmath306 below it that together with @xmath354 forms an ne chain . then , by properties [ r21 ] and [ r22 ] all the nonempty rows in @xmath306 below @xmath354 are in @xmath311 . if there is no @xmath46-cell @xmath354 with that property , then @xmath311=@xmath306 . suppose @xmath257 is a filling obtained by applying the algorithm @xmath303 to the filling @xmath293 . next we show that by applying @xmath472 and @xmath473 to @xmath257 one obtains the filling @xmath27 . steps ( ib@xmath4751 ) and ( ib@xmath4752 ) clearly invert the steps ( c@xmath4752 ) and ( c@xmath4751 ) , respectively . immediately after the borrowing step in @xmath325 , by lemma [ prop ] , ( b ) and ( c ) , and property [ borrowing ] , we can conclude that borrowing has been performed if and only if there is a right @xmath46-cell in @xmath474 which forms an ne chain with a left @xmath46-cell from @xmath311 and does not form an ne chain with any left cells in @xmath310 . it is clear that step ( id@xmath475 ) detects whether there was any borrowing and inverts it if it occurred . so , the first algorithm @xmath472 is indeed the inverse of @xmath325 . in @xmath473 , when @xmath320 points to a left @xmath46-cell we need to invert either ( b1 ) or ( b2.3 ) . using lemma [ prop ] ( c ) one sees that step ( ib1 ) detects exactly which of ( b1 ) and ( b2.3 ) occurred and inverts it . if @xmath320 points to a right cell , it is clear that ( ib2.1 ) and ( ib2.2 ) invert the steps ( b2.1 ) and ( b2.2 ) of @xmath318 . moreover , if both @xmath321 and @xmath320 point to right @xmath46-cells , and there is no left @xmath46-cell above them , then clearly the algorithm @xmath473 leaves them unchanged . that is , the algorithm @xmath473 preserves all rows that are above the rows in @xmath310 . therefore , we have @xmath476 this implies that @xmath477 is injective . since @xmath50 is finite , it follows that @xmath303 is bijective . chen , i. gessel , c.h . yan , a.l.b . yang , a major index for matchings and set partitions , j. combin . theory ser . a 115 ( 2008 ) 10691076 . a longer version containing a bijective proof is on arxiv : math/0701096 . i. gessel , applications of quasi - symmetric functions and noncommutative symmetric functions in permutation enumeration , talk presented at the canadian mathematical society winter meeting 2008 , and private communication .

the andreev conductance across normal metal ( n)/superconductor ( sc ) junctions with relativistic dirac spectrum is investigated theoretically in the blonder-tinkham-klapwijk formalism . it is shown that for relativistic materials , due to the klein tunneling instead of impurity potentials , the local strain in the junction is the key factor that determines the transparency of the junction . the local strain is shown to generate an effective dirac-gauge field . a remarkable suppression of the conductance are observed as the strength of the gauge field increases . the behaviors of the conductance are in well agreement with the results obtained in the case of n / sc junction . we also study the andreev reflection in a topological material near the chiral-to - helical phase transition in the presence of a local strain . the n side of the n / sc junction is modeled by the doped kane-mele ( km ) model . the sc region is a doped correlated km- ( kmtj ) model , which has been shown to feature-wave spin - singlet pairing . with increasing intrinsic spin-orbit ( so ) coupling , the doped kmtj system undergoes a topological phase transition from the chiral-wave superconductivity to the spin-chern superconducting phase with helical majorana fermions at edges . we explore the andreev conductance at the two inequivalent dirac points , respectively and predict the distinctive behaviors for the andreev conductance across the topological phase transition . relevance of our results for the adatom-doped graphene is discussed .

1701.00591

the andreev reflection ( ar ) is an electron - hole conversion process taking place between a normal metal and a superconductor with the excitation energy of the incident electrons is lower than the superconducting gap energy , which was originally proposed by a. f. andreev in 1964 [ ] . this conventional andreev reflection is also called the andreev retro - reflection since the reflected holes will retrace the path of the incident electrons . a recent theoretical observation on the andreev reflection at the interface between a sheet of pure graphene and a sheet of thin film graphene s - wave superconductor induced by proximity effect unvieled another andreev reflection process , known as the specular andreev reflection [ , ] . this process takes place when an incident electron with its excitation energy greater than the fermi energy but lower than the superconducting gap in the conduction band is converted into a hole in the valence band associated with a reverse on the velocity parallel to the n / sc interface . moreover , at the energy higher than the fermi energy the andreev conductance is dominated by the andreev specular reflection . after the discovery of the andreev specular reflection in graphene superconductor junction [ , ] , the andreev specular reflection is also predicted to happen in a 2@xmath6 semiconductor - superconductor junction with finite rashba spin - orbit ( so ) coupling [ ] . in addition , various theoretical works concerning the andreev reflection have been done on a variety of n / sc junctions with underlying honeycomb lattice structure such as graphene/@xmath7-wave superconductor [ ] , bilayer - graphene / s - wave superconductor [ ] and the topological materials [ , ] . these works all relies on the assumption of a clean and smooth n / sc interface . recently , the ar on a graphene - based superconducting junction has been experimentally realized in ref . and , the andreev conductance across a low disorder van der waals interface formed between bilayer graphene and superconducting nbse@xmath8 has been measured . they found the conductance across the n / sc junction was suppressed when the fermi level across the junction was tuned to lie within the superconducting gap , which gave a solid evidence on the transition between andreev retro - reflection ( intra - band process ) to andreev specular reflection ( inter - band process ) on a graphene honeycomb lattice . moreover , due to the limitation on the fabrication technology of transparent n / sc junctions , a finite tunneling transparency on the n / sc junction has been also measured . the above limitations strongly suggest that the previous theoretical calculations on the andreev conductance under the assumption of a transparent n / sc junction are not sufficient to account for the most recent experimental observations . in this paper , we try to simulate realistic n / sc junction with different degree of tunneling transparency . to do so , we first realize that for relatisvistic materials , the impurity potentials in the junction do not suppress the current . instead , due to the phenomena associated with the klein paradox , the impurity potentials tend to enhance the tunneling current [ ] . therefore , unlike the conventional blonder-tinkham-klapwijk ( btk ) formalism [ ] in which the transparency of the tunneling is simulated by the impurity potential across the junction , it is not feasible to control the transparency for the n / sc junctions made by relativistic materials via impurity potentials . in this work , we show that the local strain in the junction is the key factor that determines the transparency of the junction made by relativistic materials . in particular , we describe how to impose different degrees of transparency by adding a narrow homogeneous local strain parallel to the n / sc interface [ , ] for a system with underlying honeycomb lattice . it is shown that a @xmath1-gauge field on the n / sc interface arises and can be used to investitage the andreev reflection in various degrees of tunneling transparency . we choose the graphene normal metal/@xmath5-wave superconductor n / sc junction as an example and compute the andreev conductance via the btk formalism [ ] in the presence of a @xmath1-gauge potential . as the strength of the @xmath1-barrier is varied , the behaviors of andreev conductance obtained via our approach are in good agreement with the results obtained in the case of 1d normal metal/@xmath9-wave n / sc junction in ref . . next , we apply our theory to the case of a @xmath0 topological quantum spin hall insulator . due to the broken valley - degeneracy [ , ] , the andreev conductance from the incident electrons near the two dirac points exhibits entirely different behaviors . we provide experimental profiles of the andreev conductance within certain parameter regimes for future experiments . in this section , we start by briefly review the electron motions in a sheet of graphene monolayer , within the tight - binding formalism in sec . [ sec : graphene - review ] . the configurations of the lattice structure for a single - layered graphene sheet is schematically illustrated in the left region in fig . [ fig : localstrain]-(a ) . the derivations of the effective @xmath1-gauge field via applying a narrow homogeneous local strain on graphene will be given in sec . [ sec : deltapotential ] . finally , we apply the btk formalism [ ] to compute the andreev conductance across a graphene normal metal/@xmath7-wave n / sc junction in the presence of a @xmath1-gauge field . the results will be illustrated in sec . [ sec : ar - d+id ] . the electronic motion in an uniform undoped graphene monolayer is often formulated by the nearest - neighbor tight - binding model [ , ] : @xmath10 where @xmath11 annihilates ( creates ) an electron on the @xmath12 sublattice in the @xmath13-th unit cell while the electron operators in the momentum space is simply given by the fourier transform of @xmath14 : @xmath15 with @xmath16 being the position vector of the @xmath13-th unit cell and @xmath17 being denoted the total number of the unit cells . nearest - neighbor lattice vectors are @xmath18 with unit length @xmath19 as shown in fig . [ fig : localstrain]-(a ) . here , we set @xmath20 in what follows . the constant prefactor @xmath3 represents the hopping strength between two nearest - neighbor electrons for an uniform graphene monolayer . @xmath21 is a @xmath22-dependent function which characterizes the band structures . the undoped single - layered graphene based on the tight - binding hamiltonian in eq . ( [ eq : graphenetb ] ) features the well - known dirac band structure with linear spectrum on the dirac points as shown in fig . [ fig : localstrain]-(b ) : @xmath23 the linear dispersion is governed by the linearized hamiltonian around the dirac points , which is given by @xmath24 subjected to the condition @xmath25 . in the momentum space , the @xmath26 dirac hamiltonians around the dirac points take the form of @xmath27 & = \hbar v_f \left(\pi^{y * } q_x - \pi^x q_y \right ) \label{eq : kph}\end{aligned}\ ] ] and @xmath28 & = -\hbar v_f \left(\pi^y q_x - \pi^x q_y \right ) , \label{eq : kmh}\end{aligned}\ ] ] which acts on a two - dimensional spinor @xmath29 . the valley indices @xmath30 refer to the electronic states @xmath31 near @xmath32 . @xmath33 is defined as the fermi velocity for the tight - binding model of graphene . here , @xmath34 denote the pauli matrices : @xmath35 which are used to label the sublattices . we also defined a @xmath36 unit matrix , @xmath37 for later use , which is also for the sublattices . one possible way to introduce disorders in graphene is to change the bond spacing between two different sites via applying a local strain , which effectively varies the hopping strength @xmath3 as in eq . ( [ eq : graphenetb ] ) [ , ] . to account for the effect of the local strain , we may change the hopping strength as @xmath38 in the tight - binding hamiltonian in eq . ( [ eq : graphenetb ] ) . the magnitude of @xmath39 can be in general bond - dependent and spatially inhomogeneous over one bond - spacing . for simplicity , @xmath40 here is assumed to be uniform over the bond spacing , thus it will not acquire fourier components in the fourier transformation . under these assumptions , the linearized hamiltonian for the change of the hopping strength around @xmath41-valley takes the form @xmath42{\nonumber \\}[5pt ] & \ = & & -\int \ , d { \mathbold}{r } \ , \psi^{\dagger}_{+}({\mathbold}{r } ) \ , \begin{pmatrix } 0 & i\mathcal{a}_x -\mathcal{a}_y \\ - i\mathcal{a}_x -\mathcal{a}_y & 0 \end{pmatrix } \ , \psi_{+}({\mathbold}{r } ) , { \nonumber \\}[5pt ] & \ = & & -\int \ , d{\mathbold}{r}\ , \psi^{\dagger}_{+}({\mathbold}{r } ) \left ( \pi^{y * } \mathcal{a}_x - \pi^x \mathcal{a}_y \right)\ , \psi_{+}({\mathbold}{r } ) , \label{eq : straink } \end{aligned}\ ] ] where @xmath43 denotes the field operator for the @xmath44-valley while @xmath45 is a complex function . here , @xmath46 and @xmath47 are real functions . if we assume that the affect of the local strain only extends over one lattice spacing and only influences on the horizontal bonds along the @xmath48-direction , as shown in fig . [ fig : localstrain ] ( a ) . the complex function @xmath49 in eq . ( [ eq : aofr ] ) can be reduced to the form of a @xmath1-function . in unit of @xmath50 , @xmath49 simply takes the form @xmath51 consequently , only the real part of @xmath49 survives ; the imaginary part vanishes : @xmath52 the linearized hamiltonian for change of the hopping amplitude near @xmath53-valley is related to the one for the @xmath41-valley by time - reversal transformation @xmath54 , i.e. @xmath55 : @xmath56 combining eq . [ eq : kph ] and eq . ( [ eq : kmh ] ) , eq . ( [ eq : straink ] ) and eq . ( [ eq : strainkprime ] ) , the linearized hamiltonian in the presence of a homogeneous local strain in real space is given by @xmath57 0 & -\left [ \pi^y \left ( \hat{q}_x + \mathcal{a}_x \right ) - \pi^x \left(\hat{q}_y + \mathcal{a}_y\right ) \right ] \end{pmatrix } \psi({\mathbold}{r } ) , \label{eq : hrealspace}\end{aligned}\ ] ] where @xmath58 and the momentum operator @xmath59 . the fact that the reverse of sign in the terms containing the complex vector @xmath60 for different valleys in eq . ( [ eq : hrealspace ] ) implies that @xmath61 can be viewed as a gauge field [ , ] . combining eq . ( [ eq : aofr ] ) , eq . ( [ eq : vectora ] ) and eq . ( [ eq : hrealspace ] ) , it is clear that the effect of a homogeneous local strain in the distance over one horizontal bond on graphene can be simply regarded as an effective @xmath1-gauge field of a series of localized impurities along the @xmath48-direction , which couples the electrons from the sublattices @xmath62 and @xmath63 . throughout this article . the three phases for the bond - dependent @xmath7-pairing are defined as @xmath64 , as shown in the sc region in ( a ) . the thicker purple horizontal bonds which locate at the n / sc interface represent the modified hopping strength @xmath65 due to the local strain . the first brillouin zone is shown in ( b ) . ] in this section , we dedicate our efforts to investigating the andreev reflection through a graphene normal metal/@xmath5-wave spin - singlet superconductor n / sc junction with an effective @xmath1-gauge field laying on the n / sc interface via btk formalism . unlike the well - known case of superconductivity in graphene via promixity to a superconducting electrode [ ] , the @xmath5-wave spin - singlet superconducting order is induced in graphene at finite doping by on - site electron - electron coulomb repulsion [ , , ] . due to the @xmath67 point group symmetry of the underlying honeycomb lattice , the @xmath5-wave superconducting order in the @xmath22-space takes the form @xmath68 with the bond - dependent order parameter @xmath69 with @xmath70 [ , , ] . as depicted in fig . [ fig : localstrain]-(a ) , the n / sc junction being considered is composed of a sheet of graphene normal metal ( n ) occupies the region of @xmath71 connecting to a @xmath0 superconducting ( sc ) thin film which occupies @xmath72 with a sharp n / sc interface in between ( i.e. at the position of @xmath73 ) , which implies that the translational invariance along the @xmath74-direction is broken . we assume that the n / sc junction is homogeneous and infinitely extended in the @xmath48-direction , therefore the translational symmetry is preserved in @xmath48 . the sharp n / sc junction signifies that the bulk value of the superconducting pairing amplitude denoted as @xmath75 is reached at a negligibly small distance from the interface , which can be achieved via adjusting the doping or gate voltage in the sc region [ , , ] . due to the valley and spin degeneracy , the electronic motions can be described by two sets of decoupled dirac - bogoliubov - de gennes ( dbdg ) equations [ , ] for the @xmath41 and @xmath53-valley , each containing four equations . thus , it suffices to consider the set for @xmath76-valley only : @xmath77 \theta(x ) \delta_{-}({\mathbold}{q } ) ^{\dagger } & \mu - \mathcal{h}_{-}({\mathbold}{q } ) \end{pmatrix } \begin{pmatrix } u \\[4pt ] v \end{pmatrix } = \frac { \epsilon_{{\mathbold}{q } } } { t } \begin{pmatrix } u \\[4pt ] v \end{pmatrix } , \label{eq : dbdg - eq } \end{aligned}\ ] ] where @xmath78 is the @xmath79 linearized single - particle hamiltonian of graphene at the @xmath53-valley and @xmath80 denotes the chemical potential . here , @xmath81 and @xmath82 are the two component @xmath83-dependent wavefunctions for the electron ( electron - like ) and hole ( hole - like ) excitations at the excitation energy @xmath84 which is measured relative the chemical potential @xmath80 . here , we introduce an unit step electrostatic potential @xmath85 to the sc region , where @xmath86 can be tuned independently through doping or gate voltage . to justify the assumption of sharp n / sc junction as mentioned previously , the energy scales must satisfy @xmath87 such that the fermi wavelength @xmath88 in sc is much shorter than that in n where @xmath89 [ , , ] . the linearized superconducting pairing matrix @xmath90 with @xmath91 pairing symmetry at the @xmath53-valley is given by @xmath92 3 & 0 \end{pmatrix } , \label{eq : d+id - linearized } \end{aligned}\ ] ] which is related to the one for the @xmath93-valley by @xmath94 . apparently , the @xmath95-wave superconducting pairing at low energy features the @xmath9- and @xmath96-pairing symmetry [ , ] . to study the andreev reflection across a n / sc junction , we may image that an incident electron comes from @xmath97 toward the n / sc junction and scatters by the potential at the interface . while scattering with the potential , electrons may reflect back to the n region either as normal electrons or holes , or may tunnel through the barrier into the sc region as dirac - bogoliubov quasiparticles . for convenience , electrons are assumed to go through elastic scattering processes at the interface . hence , all the scattering basis for the incident , reflected and transmitted states inside the n and sc region , which can be solved via eq . ( [ eq : dbdg - eq ] ) respectively , are characterized by the same excitation energy @xmath98 . the real space eigenfucntions of eq . ( [ eq : dbdg - eq ] ) in general take the form of plane - wave solutions , i.e. @xmath99 u_{b\uparrow } \\[4pt ] v_{a\downarrow } \\[4pt ] v_{b\downarrow } \end{pmatrix}. \label{eq : wavefunc - realspace } \end{aligned}\ ] ] please note that the solutions of eq . ( [ eq : dbdg - eq ] ) in the sc region may be either an evanescent mode which decays exponentially with the increase in the distance from the interface at the energy @xmath98 of the incident electron smaller than the superconducting gap , namely @xmath100 or a propagating mode at @xmath101 . furthermore , since the hamiltonian in the normal metal is diagonal in the spin subspace , here we only consider the incident electrons to be spin - up in eq . ( [ eq : wavefunc - realspace ] ) . here , we denote @xmath102 . therefore , the total wavefunction @xmath103 in the n ( sc ) region can be expressed as a superposition of various eigenstates of eq . ( [ eq : dbdg - eq ] ) with positive excitation energy at the region @xmath104 : @xmath105 & \psi_{sc}({\mathbold}{r } ) = t_e \ , \psi^{(e)}_{sc } ( \bar{q}_x , \ , q_y ) e^{i\bar{q}_x x + iq_y y } + t_h \psi_{sc}^{(h ) } ( -\bar{q}^ { \prime}_x , \ , q_y ) e^{-i\bar{q}^{\prime}_x x + iq_y y } . \end{aligned}\ ] ] with the incident state being normalized to unity . @xmath106 and @xmath107 , which depend on the energy @xmath98 and the wavevector @xmath83 of the incident state , represent the reflection and transmission coefficient for the electron branch ( with subscript @xmath108 ) and hole branch ( with subscript @xmath109 ) . due to the assumption of elastic scattering , @xmath110 are the eigenstates of eq . ( [ eq : dbdg - eq ] ) with the same excitation energy @xmath98 . please note that the transverse component of the wavevector @xmath111 is a conserved quantity during the scattering process due to the translational symmetry in @xmath48 while the longitudinal components for the electron and hole ( electron - like and hole - like ) states @xmath112 ( @xmath113 ) can be determined via their ( quasiparticle ) dispersion relations at a given @xmath98 and @xmath114 . due to the presence of a @xmath1-gauge field as well as the restriction that the dbdg equations are of first order , the wavefunction continuity at the interface as being widely used in various literatures under the assumptions of ideal n / sc interfaces is not an appropriate boundary condition . detailed derivations for the boundary condition for our theory are provided in the following : the dbdg equations for the @xmath53-valley in the presence of an effective @xmath1-gauge field as shown in eq . ( [ eq : vectora ] ) are given by @xmath115 -i\hat{q}_x + \hat{q}_y - \frac{\delta t}{t } \delta(x ) -\frac{\mu}{t } & 0 & \frac{3\delta_0}{t } & 0 \\[3pt ] 0 & \frac{3\delta_0}{t } & 0 & \frac{\mu}{t } - i\hat{q}_x - \hat{q}_y + \frac{\delta t}{t } \delta(x ) \\[3pt ] \frac{3\delta_0}{2 t } ( i\hat{q}_x + \hat{q}_y ) & 0 & \frac{\mu}{t } + i\hat{q}_x - \hat{q}_y + \frac{\delta t}{t } \delta(x ) & 0 \end{pmatrix } \begin{pmatrix } f_{a\uparrow } \\[3pt ] f_{b\uparrow } \\[3pt ] g_{a\downarrow } \\[3pt ] g_{b\downarrow } \end{pmatrix } = \frac{\epsilon}{t } \begin{pmatrix } f_{a\uparrow } \\[3pt ] f_{b\uparrow } \\[3pt ] g_{a\downarrow } \\[3pt ] g_{b\downarrow } \end{pmatrix}. \label{eq : dbdg - matrix}\end{aligned}\ ] ] taking the momentum operator @xmath116 and integrating along the @xmath74-direction over a small distance across the interface , eq . ( [ eq : dbdg - matrix ] ) become @xmath117-\frac{3 \delta_0 } { 2 t } \ , \left [ g_{b \downarrow}^{sc}(0 ) - g_{b \downarrow}^{n}(0)\right ] = \frac{\delta t}{t } f_{b\uparrow } ( 0 ) , { \nonumber \\}[4pt ] & \left[f_{a\uparrow}^{n}(0 ) - f_{a\uparrow}^{sc}(0)\right ] = \frac{\delta t}{t } \ , f_{a\uparrow}(0).\ , { \nonumber \\}[4pt ] & \left[g_{b \downarrow}^{sc}(0)- g_{b \downarrow}^{n}(0 ) \right ] = \frac{\delta t}{t } \ , g_{b \downarrow } ( 0 ) , { \nonumber \\}[4pt ] & \left[g_{a \downarrow}^{n}(0)- g_{a \downarrow}^{sc}(0 ) \right ] + \frac{3 \delta_0}{2 t } \ , \left[f_{a \uparrow}^{n}(0)- f_{a \uparrow}^{sc}(0 ) \right ] = \frac{\delta t}{t } \ , g_{a \downarrow}(0 ) . \label{eq : bc - integrate}\end{aligned}\ ] ] in eq . ( [ eq : bc - integrate ] ) , @xmath118 while @xmath119 with @xmath120 denotes the @xmath121 sublattice and @xmath122 denotes the spin @xmath123 . as we shall see in eq . ( [ eq : bc - integrate ] ) , the boundary condition results in an ambiguity of @xmath124 at @xmath125 which results from the situation that the dbdg equations are of first order . we further impose the following conditions to resolve the wavefunction ambiguity : in the situation of @xmath126 and @xmath127 [ ] , we expect no tunneling current across the n / sc junction . the conditions are quite straightforward : in the case mentioned above , the original n / sc junction reduces to two disconnected semi - planes of graphene sheet ; thus , electron tunneling is forbidden , giving rise to no tunneling current . the issue of wavefunction ambiguity at the origin now changes to problem of the electron tunneling across a junction of two pure graphene semi - planes with a @xmath1-gauge field taking the form of eq . ( [ eq : vectora ] ) in between . based on the condition of current conservation and the requirement of no tunneling current across the junction at @xmath128 and @xmath126 , we choose the wavefunctions at the origin as @xmath129 & g_{a\sigma } ( 0 ) = g_{a\sigma}^{sc},~~g_{b \sigma } ( 0 ) = g_{b \sigma}^{n } , \label{eq : bc - ambiguity}\end{aligned}\ ] ] where @xmath130 denotes spin . please note that due to the relation of time reversal partner between the incident electrons and the reflected holes within the dbdg formalism [ , ] for ar in a single - layered graphene , the hole wavefunctions @xmath131 share the same boundary conditions with the electrons wavefunctions @xmath132 as shown in eq . ( [ eq : bc - ambiguity ] ) . leading to @xmath133 , { \nonumber \\}[4pt ] f_{b \downarrow}^{sc}(0 ) & = \eta f_{a \uparrow}^{sc}(0 ) , { \nonumber \\}[4pt ] g_{b \downarrow}^{sc}(0 ) & = \eta g_{b \downarrow}^{n}(0 ) , { \nonumber \\}[4pt ] g_{a \downarrow}^{n}(0 ) & = \eta g_{a \downarrow}^{sc}(0 ) + \frac { 3\delta_0}{2 t } \ , \left [ f_{a \uparrow}^{sc}(0 ) - f_{a \uparrow}^{n}(0)\right ] , \label{eq : bc - final}\end{aligned}\ ] ] where @xmath134 . for the detailed derivations for the boundary conditions in eq . ( [ eq : bc - ambiguity ] ) , we refer the readers to the appendix [ app : bc - graphene ] . once the reflection and transmission coefficients are obtained , following the btk formalism [ ] , the normalized differential conductance can be computed by summing over all possible incident states , leading to @xmath135,\end{aligned}\ ] ] where @xmath136 is the ballistic conductance of graphene [ , ] . . ] in the absence of the local strain @xmath137 , the andreev conductance for the graphene @xmath138-wave superconductor n / sc junction is reproduced@xcite as shown in fig . [ fig - hu - dt-1 ] and [ fig - hu - dt-2 ] and the specular - ar to retro - ar transition marked by @xmath139 can be easily identified for the case of @xmath140 . the behaviors of the normalized differential conductance @xmath141 in fig . [ fig - hu - dt-1 ] and [ fig - hu - dt-2 ] can be qualitatively explained via the aspect of the linear band structure of graphene normal metal and the electron - hole conversion processes at different fermi energy as shown in fig . [ fig : btk - scattering ] . for the case of @xmath142 as in fig . [ fig - hu - dt-2 ] , at zero bias @xmath143 , the phase spaces for the incident electrons and reflected holes are identical to each other , giving rise to the maximum andreev conductance . however , once @xmath144 is increased but still stays lower than @xmath80 , it is clear that the phase space of hole band shrinks and results in a monotonic decline in @xmath145 until the bias reaches the fermi energy @xmath146 , where there is no density of states for reflected hole , thus leading to a conductance dip as shown in fig . [ fig - hu - dt-1 ] . in the regime @xmath147 , a conduction band electron is reflected as a conduction - band holes via the andreev reflection , it is called the intraband andreev reflection ( or andreev retro - reflection ) . once the bias exceeds the fermi energy @xmath148 , the incident electrons from the conduction band are converted as valence - band holes , leading to the interband andreev reflection ( or andreev specular reflection ) . for this case , @xmath145 increases again with the increasing @xmath144 due to the increase of phase space in the hole band . at @xmath149 , the tunnelling process returns to the normal metal - normal metal tunnelling and the andreev conductance saturates . apparently , the fermi energy serves as the `` critical energy '' for the transition between the intra to interband andreev reflection . on the contrary , only the andreev retro - reflection process exists as @xmath150 , thus leading to a monotonic decline of @xmath145 as shown in fig . [ fig - hu - dt-1 ] . in the following , we discuss the normalized conductance for the situation of non - zero barrier @xmath151 . the vanishing of @xmath145 for the case of @xmath127 signifies no electron tunneling , as expected . for the situation of finite potential barrier , our results qualitatively capture the most significant features of andreev conductance in the presence of @xmath1-barrier : in fig . ( [ fig - hu - dt-1 ] ) and ( [ fig - hu - dt-2 ] ) , as the barrier strength @xmath152 is increased @xmath145 dramatically decreases down to zero due to heavily scattering of electrons by the potential barrier , in well agreement with the previous results for the case of the @xmath2 metal/@xmath9-wave n / sc junction in ref . . the cusps for the normalized conductance in fig . [ fig - hu - dt-2 ] at @xmath153 with non - zero @xmath154 can be simply understood as the competition between the graphene density of states and the effect of the @xmath1-barrier : the density of states for graphene at low energy is linearly proportional to the excitation energy , signifying that increasing the energy will enhance the conductance . on the contrary , the effect of @xmath154 tends to suppressed the conductance . therefore , the competition between the density of states and the @xmath1-barrier gives rise to the cusps in fig . [ fig - hu - dt-2 ] . superconducting junction with varying barrier height @xmath152 and fixed fermi energy @xmath155 and superconducting pairing strength @xmath156 . the superconducting gap energy is around @xmath157 . ] with varying barrier height @xmath152 and fixed fermi level @xmath158 and superconducting pairing strength @xmath159 . the superconducting gap energy is found to be @xmath160 . ] in the past decades , much effort has been devoted to searching for the novel topological states of matters . the two examples of particular interest are the topological insulators [ , ] , which have insulating bulk states while the edge or surface supports time - reversal symmetry protected conducting states , and the topological superconductors which support gapless , charged neutral majorana edge ( or surface ) states [ ] with superconducting bulk states . recently , the doped kane - mele ( km ) model , which was originally proposed in ref . and , with large onsite electron - electron repulsive interaction on a @xmath0 periodic honeycomb lattice has been theoretically shown to feature time - reversal broken @xmath161-wave superconducting state in the bulk via renormalized mean - field theory [ , ] . moreover , it was also found that the system undergoes a topological phase transition from the helical superconducting to the chiral superconducting order as the strength of the intrinsic spin - orbit coupling is decreased , and two pairs of counter - propagating helical majorana zero modes have been found theoretically at the edges of a finit - sized zigzag ribbon of the tight - binding km @xmath3-@xmath162 model in spite of the time - reversal broken @xmath161-wave superconducting order [ , ] . via the numerical simulation by density functional theory , the kmtj model may be realized via doping adatoms such as indium or thallium on a graphene sheet , which generates an effective kane - mele type intrinsic so coupling ( @xmath163 mev ) [ ] , more larger than the un - doped graphene . besides graphene - based systems , our model is also applicable to other compounds with underlying honeycomb lattice such as in@xmath164cu@xmath165vo@xmath166 [ , ] , @xmath167-cu@xmath165v@xmath165o@xmath168 [ , ] , mos@xmath165 [ ] and silicene [ ] . those materials have been proposed to exhibit chiral-@xmath6-wave superconducting state around half - filling . these exotic features discovered in the kmtj model motivates us to seek the corresponding experimental signatures . in the following , we investigate the andreev reflection across a n / sc junction with the normal side being modeled by the doped km model while the sc region is a doped correlated km @xmath3-@xmath4 model with @xmath5-wave spin - singlet superconducting order . the kane - mele model which can be viewed as a spinful haldane model [ ] is composed of the nearest - neighbor ( nn ) tight - binding hamiltonian @xmath169 as in eq . ( [ eq : graphenetb ] ) and the next - nearest neighbor ( nnn ) hopping intrinsic spin - orbit ( so ) interaction @xmath170 : @xmath171 here , @xmath172 annihilates ( creates ) an electron on either the @xmath62 or @xmath63 sublattice on the @xmath13-th unit cell . @xmath173 denotes the nnn indices , @xmath174 is the coupling strength of the intrinsic so interaction and @xmath175 represent spins . @xmath176 is an orientation dependent factor : @xmath177 for an electron makes a right turn while moves from the @xmath13-th site to its @xmath178-th nnn site , and @xmath179 for left turn . the doping is characterized by @xmath180 with @xmath80 being the value of the chemical potential . the mean - field hamiltonian on a periodic lattice in terms of the basis @xmath181 is given by the @xmath182 matrix : @xmath183 0 & h^{-}_{{\mathbold}{k } } - \mu & -\delta_{{\mathbold}{k } } & 0 \\[3pt ] 0 & -\delta_{{\mathbold}{k}}^{\dagger } & \mu - h_{-{\mathbold}{k}}^{+^ * } & 0\\[3pt ] \delta_{{\mathbold}{k}}^{\dagger } & 0 & 0 & \mu -h_{{\mathbold}{-k}}^{-^ * } \end{matrix}\right ) { \nonumber \\}= & \ , \left ( \begin{matrix } h_{{\mathbold}{k}}^{+ } - \mu & 0 & 0 & \delta_{{\mathbold}{k } } \\[3pt ] 0 & h_{{\mathbold}{k}}^{- } - \mu & -\delta_{{\mathbold}{k } } & 0 \\[3pt ] 0 & -\delta_{{\mathbold}{k}}^{\dagger } & \mu - h_{{\mathbold}{k}}^{- } & 0\\[3pt ] \delta_{{\mathbold}{k}}^{\dagger } & 0 & 0 & \mu -h_{{\mathbold}{k}}^{+ } \end{matrix}\right ) \label{eq : kmtj - h}\end{aligned}\ ] ] with @xmath184 f^{\ast}({{\mathbold}{k } } ) & \mp\gamma({{\mathbold}{k } } ) \end{pmatrix } , \\[3pt ] & \gamma({{\mathbold}{k } } ) = 2\lambda_{so } \left [ 2 \ , \text{cos}\frac{3k_{x}}{2 } \text{sin}\frac{\sqrt{3 } k_{y}}{2}- \text{sin}\sqrt{3 } k_{y}\right ] . \end{aligned}\ ] ] in the second line of eq . ( [ eq : kmtj - h ] ) , we have applied the relations of @xmath185 [ ] and @xmath186 . the electronic excitations near the @xmath44-valley in the n and sc sides are described by the linearized dbdg equations , which take the form @xmath187 \theta(x ) \bar{\delta}_{\tau}({\mathbold}{q } ) ^{\dagger } & \mu - \mathcal{h}_{\tau}({\mathbold}{q } ) \end{pmatrix } \begin{pmatrix } u_\tau \\[4pt ] v_\tau \end{pmatrix } = \frac{\epsilon_{{\mathbold}{q}}}{t } \begin{pmatrix } u_\tau \\[4pt ] v_\tau \end{pmatrix}. \label{eq : dbdg - eq - kmtj } \end{aligned}\ ] ] here , @xmath188 is an eight - component wavefunctions in the momentum domain near the @xmath44-valley with the first four components @xmath189 for particles while the last four components @xmath190 for holes . @xmath191 where the @xmath192 unit matrix @xmath193 together with the three pauli matrices @xmath194 are for the spin subspace in the hilbert space while the matrices @xmath195 are for the sublattices as already defined in eq . ( [ eq : pauil - sublattice ] ) and eq . ( [ eq : tau0-sublattice ] ) . here , we assume the magnitudes of intrinsic so coupling in the n and the sc region can be adjusted independently , thus we introduce @xmath196 \lambda^{\prime}_{so}/t , ~~\text{for}~x>0 , \end{cases}\end{aligned}\ ] ] where @xmath174 indicates the intrinsic so coupling in n while @xmath197 in sc . here , @xmath198 is a @xmath199 matrix for the linearized @xmath5-wave pairing near the @xmath44 dirac point , which takes the form @xmath200 -\delta_\tau ( { \mathbold}{q } ) & 0 \end{pmatrix}. \end{aligned}\ ] ] around @xmath53 , @xmath201 is given by the eq . ( [ eq : d+id - linearized ] ) in the previous section , which is related to the one for the @xmath41-valley by @xmath94 . for the reason that the km@xmath202 model exhibites an effective spin - singlet @xmath203 superconducting order near the two dirac points @xmath204 [ ] , the valley degeneracy no longer exist and we ought to consider the normalized andreev conductance contributed from @xmath41 and @xmath53 , respectively . here , we assume the electron scattering only occurs within one valley , thus the normalized andreev conductance can be simply evaluated by taking the average of the individual contributions from @xmath204 . the andreev conductance contributed from one valley can be similarly obtained via the btk formalism . the average of the normalized andreev conductance is expressed as @xmath205 where @xmath206 is the andreev conductance from the @xmath204 , respectively . the results are illustrated in fig . [ fig : kmtj - mixed - nostrain ] and fig . [ fig : kmtj - mixed - strain ] . in the following , we qualitatively discuss the conductance behaviors for fig . [ fig : kmtj - mixed - nostrain ] and [ fig : kmtj - mixed - strain ] . as the @xmath1-barrier is switched off , the andreev conductance @xmath207 from the @xmath53-valley , as shown in fig . [ fig : kmtj - mixed - nostrain]-(c ) , monotonically decreases with the increasing bias @xmath144 due to the shrinking of the phase space in the hole band as shown in fig . [ fig : btkscattering-2 ] . the resulting behavior of the andreev conductance @xmath207 is similar to the case in fig . [ fig - hu - dt-1 ] , hence only the andreev retro - reflection process involves . due to the inversion symmetry breaking in the kmtj model , the andreev conductance @xmath208 from the @xmath209 in fig . [ fig : kmtj - mixed - nostrain]-(b ) behaves entirely different from @xmath207 . averaging @xmath208 and @xmath207 gives rise to distinctive andreev conductance behaviors for different values of @xmath210 . for small intrinsic so coupling , we find that the average andreev conductance @xmath211 increases as the bias is increased at low bias , which @xmath211 behaves in a similar manner to the andreev specular reflection . on the contrary , for large intrinsic so coupling @xmath210 , @xmath211 monotonically decreases with increasing bias . we argue the andreev retro - reflection may dominate @xmath211 in this situation . before we conclude , the effect of electron scattering by the edge states on the andreev conductance deserves some discussions here . as mentioned in the previous section , via the bulk - edge correspondence [ , ] , the kmtj model supports chiral or helical majorana edge states depending on the topological phase in the bulk . accordingly , we anticipate that as the electron scattering by the quasiparticles in the edge states at the n / sc interface is considered , the andreev conductance may exhibit distinctive behavior rather than a relatively smooth crossover as the sc region is tuned to undergo chiral - to - helical topological phase transition . we expect this distinctive behavior on the andreev conductance at the topological critical point will serve as an experimentally signature to probe the topological phase transition . in conclusion , we have investigated the andreev reflection based on blonder - tinkham- klapwijk formalism in a graphene normal metal/@xmath7-wave superconducting junction in the case of a finite barrier laying on the n / sc interface . in order to investigate the electron scatterings on a n / sc junction with different transparencies , an effective dirac @xmath1-guage field has been introduced by applying a homogeneous local strain parallel to the interface . in the absence of local strain , i.e @xmath137 , our results successfully reproduce the normalized andreev conductance @xmath145 curves in ref . . at the other extreme parameter regime of @xmath127 , the andreev conductance vanishes because the n / sc junction is disconnected and therefore electron tunneling is forbidden . for the case of finite barrier and at the energy of the incident electron lower than the superconducting gap @xmath212 , the andreev conductance dramatically decreases down to zero as the barrier strength @xmath154 is increased since the incident electrons scatter heavily with the barrier . our results are qualitatively consistent with the results of the andreev conductance originally obtained for the case of a @xmath2 normal metal/@xmath9-wave superconducting junction . we further investigate the andreev reflection across the n / sc junction with the n region being described by the doped kane - mele model while the sc region features @xmath5-wave spin - singlet pairing induced by strongly electron correlations . the normalized andreev conductance contributed from the @xmath41 and @xmath53-valleys illustrate entirely different behaviors due to the different effective superconducting pairing symmetry near the two dirac points . our results provide spectra of the normalized andreev conductance within certain parameter regimes for future experiments . this work was supported by ministry of science and technology ( most ) , taiwan . we also acknowledge support from tcecm and academia sinica research program on nanoscience and nanotechnology , taiwan . this work is also supported by the most grant no . 104 - 2112-m-009 -004 -my3 , the moe - atu program , the ncts of taiwan , r.o.c . in this section , we repeat the derivations on the boundary conditions for the electronic transport through a graphene normal metal - normal metal ( n / n ) junction in the low - energy limit in the presence of a effective @xmath1-gauge field in between . this issue was originally addressed by castro neto @xmath213 in ref . and . the schrodinger s equations for the graphene tight - binding model in the low - energy limit near the @xmath214-valley read @xmath215 \psi_-({\mathbold}{r } ) = ( \epsilon / t ) \ , \psi_-({\mathbold}{r } ) { \nonumber \\}\hfill { \nonumber \\}\rightarrow \ , & \left [ \begin{matrix } 0 & -\partial_x + i \partial_y - \mathcal{a}_y \\ \partial_x + i \partial_y - \mathcal{a}_y & 0 \end{matrix}\right ] \psi_-({\mathbold}{r } ) = -(\epsilon / t ) \ , \psi_-({\mathbold}{r } ) . \label{eq : se - km}\end{aligned}\ ] ] expressing the two - component wavefunction as @xmath216 and substituting the effective @xmath1-potential to @xmath217 in eq . ( [ eq : se - km ] ) yields @xmath218 integrating the schrodinger s equations in eq . ( [ eq : se - km ] ) over a infinitesimal region across the origin , i.e. @xmath219 , we obtain the boundary condition @xmath220 since the schrodinger s equations with linear dirac spectrum are of first order , we can not demand the wavefunction to be continuous at the origin , therefore giving rise to the wavefunction ambiguity , @xmath221 and @xmath222 , as we shall see in eq . ( [ eq : se - bc - km ] ) . here , we choose the undetermined wavefunctions in following the way : @xmath223 and eq . ( [ eq : se - bc - km ] ) become @xmath224 later , we will show that the choice of the undetermined wavefunctions at @xmath225 in eq . ( [ eq : wavefunction - bc - km ] ) will lead to the conservation of probability current . thus , current conservation justify our choice of the undetermined wavefunctions . we may image that an incident electron far from the interface in the graphene sheet at the region of @xmath226 moves toward the interface and get scattered with the potential at the origin . by solving the schrodinger s equation , the right - moving state in @xmath83-space for the incident electron is given by @xmath227 while the left - moving state for the reflected electron can be obtained by simply reverse the sign of @xmath228 in eq . ( [ eq : km - incidentstate ] ) , which is given by @xmath229 the total wavefunctions with normalized incident state in the left and right side of the graphene sheet are @xmath230 with @xmath231 and @xmath54 being denoted as the reflection and transmission coefficients . here , the superscripts @xmath232 and @xmath233 stand for left and right , respectively . the boundary conditions are give by @xmath234 @xmath235 apparently , @xmath236 when @xmath237 as expected and the probability current is conserved , that is @xmath238 . likewise , the schrodinger equations for the @xmath239-valley are given by @xmath240 \psi_+(\textbf{r } ) = ( \epsilon / t ) \ , \psi_+(\textbf{r } ) { \nonumber \\}\hfill { \nonumber \\}\rightarrow \ , & \left [ \begin{matrix } 0 & \partial_x + i \partial_y + \mathcal{a}_y \\ - \partial_x + i \partial_y + \mathcal{a}_y & 0 \end{matrix}\right ] \psi_+(\textbf{r } ) = ( \epsilon / t)\ , \psi_+(\textbf{r } ) .\end{aligned}\ ] ] express @xmath241 , we have @xmath242 next , integrating from @xmath243 to @xmath244 yields the boundary conditions : @xmath245 apparently , we can immediately see the undetermined wavefunctions @xmath246 and @xmath247 appears on the r.h.s . in eq . ( [ eq : bc - kp ] ) . for the same reason of current conservation , we choose the undetermined wavefunction to be @xmath248 please note that due to the relation of the time - reversal partner for the states near the @xmath41 and @xmath53 wavevector , the choice of the undertermined wavefunctions at @xmath125 in eq . ( [ eq : bc - ambiguity - kp ] ) is identical to that in eq . ( [ eq : wavefunction - bc - km ] ) . to calculate the transmission and reflection coefficients , we first prepare the normalized right - moving and left - moving states for the incident and reflected electrons in the momentum space : @xmath249 for the incident electrons . ] the phase @xmath250 is defined in the way shown in fig . [ fig : incidentangle ] . the total wavefunction on the left - hand side , denoted as @xmath251 , can be expressed as a superposition of the incident and reflected wavefunctions , namely @xmath252 the total wavefunction on the right - hand side @xmath253 is given by @xmath254 in the above , we neglect the phase factor and prefactor @xmath255 since it will play no role on finding @xmath231 and @xmath54 . written @xmath256 , the b.c.s can be found to be @xmath257 we can solve for the transmission and reflection coefficients : @xmath258 we can immediately check that once @xmath259 , there is no transmitted particles since the graphene has been cut into two separate pieces , and the probability current is conserved , that is @xmath238 . in this section , we derive the boundary condition for the electron scattering across the kane - mele @xmath7 -wave superconducting n / sc junction in the presence of a @xmath1-barrier for the @xmath204-valleys , respectively we integrate the above equations over a infinitesimal distance across the interface and eq . ( [ eq : bc - kmtj-1 ] ) becomes @xmath117-\frac{3 \delta_0}{2 t } \ , \left [ g_{b \downarrow}^{sc}(0 ) - g_{b \downarrow}^{n}(0)\right ] = \frac{\delta t}{t } f_{b\uparrow } ( 0 ) , { \nonumber \\}- & \left[f_{a\uparrow}^{sc}(0 ) - f_{a\uparrow}^{n}(0)\right ] = \frac{\delta t}{t } \ , f_{a\uparrow}(0),{\nonumber \\ } & \left[f_{b \downarrow}^{sc}(0)- f_{b \downarrow}^{n}(0 ) \right]+ \frac{3 \delta_0}{2 t } \ , \left[g_{b\uparrow}^{sc}(0 ) - g_{b\uparrow}^{n}(0 ) \right ] = \frac{\delta t}{t}\ , f_{b \downarrow } ( 0),{\nonumber \\}- & \left[f_{a \downarrow}^{sc}(0)- f_{a \downarrow}^{n}(0 ) \right ] = \frac{\delta t}{t } \ , f_{a\downarrow } ( 0 ) , { \nonumber \\ } & \left[g_{b \uparrow}^{sc}(0)- g_{b \uparrow}^{n}(0 ) \right ] = \frac{\delta t}{t } g_{b\uparrow } ( 0 ) , { \nonumber \\ } & \left[g_{a \uparrow}^{sc}(0)- g_{a \uparrow}^{n}(0 ) \right ] -\frac{3 \delta_0}{2 t } \ , \left[f_{a \downarrow}^{sc}(0)- f_{a \downarrow}^{n}(0 ) \right ] = - \frac{\delta t}{t } \ , g_{a \uparrow}(0),{\nonumber \\ } & \left[g_{b \downarrow}^{sc}(0)- g_{b \downarrow}^{n}(0 ) \right ] = \frac{\delta t}{t } g_{b \downarrow } ( 0 ) , { \nonumber \\ } & \left[g_{a \downarrow}^{sc}(0)- g_{a \downarrow}^{n}(0 ) \right ] + \frac{3 \delta_0}{2 t } \ , \left[f_{a \uparrow}^{sc}(0)- f_{a \uparrow}^{n}(0 ) \right ] = \frac{\delta t}{t } \ , g_{a \downarrow}(0 ) . \label{eq : bc - integrate}\end{aligned}\ ] ] in the above equations for the boundary conditions , @xmath262 while @xmath263 with @xmath120 denotes the @xmath264 sublattice and @xmath122 denotes the spin @xmath123 . similarly , the wavefunction ambiguity at @xmath73 also exists in eq , ( [ eq : bc - integrate ] ) due to the fact that the dbdg equations are of first order . following the similar approach , we are able to determine @xmath265 via the requirements of no tunnelling current across the n / sc junction as @xmath126 and @xmath127 . based on eq . ( [ eq : bc - ambiguity ] ) , the boundary conditions at @xmath125 can be written as @xmath266 , { \nonumber \\}f_{a \uparrow}^{n}(0 ) & = \eta \,f_{a \uparrow}^{sc}(0 ) , { \nonumber \\}f_{b \downarrow}^{sc}(0 ) & = \eta \ , f_{b \downarrow}^{n}(0 ) - \frac { 3 \delta_0}{2 t } \ , \left [ g_{b \uparrow}^{sc}(0 ) - g_{b \uparrow}^{n}(0)\right ] , { \nonumber \\}f_{a \downarrow}^{n}(0 ) & = \eta \ , f_{a \downarrow}^{sc}(0 ) , { \nonumber \\}g_{b \uparrow}^{sc}(0 ) & = \eta \ , g_{b \uparrow}^{n}(0 ) , { \nonumber \\}g_{a \uparrow}^{n}(0 ) & = \eta \,g_{a \uparrow}^{sc}(0 ) - \frac{3 \delta_0}{2 t } \ , \left [ f_{a \downarrow}^{sc}(0 ) - f_{a \downarrow}^{n}(0)\right ] , { \nonumber \\}g_{b \downarrow}^{sc}(0 ) & = \eta \ , g_{b \downarrow}^{n}(0 ) , { \nonumber \\}g_{a \downarrow}^{n}(0 ) & = \eta \ , g_{a \downarrow}^{sc}(0 ) + \frac { 3\delta_0}{2 t } \ , \left [ f_{a \uparrow}^{sc}(0 ) - f_{a \uparrow}^{n}(0)\right ] , \label{eq : bc - final}\end{aligned}\ ] ] where @xmath267 . following the same procedures , the boundary condition for @xmath41-valley are given by @xmath268 , { \nonumber \\}f_{b \downarrow}^{sc}(0 ) & = \eta \ , f_{b \downarrow}^{n}(0 ) , { \nonumber \\}f_{a \downarrow}^{n}(0 ) & = \eta \ , f_{a \downarrow}^{sc}(0 ) + \frac { 3 \delta_0}{2 t } \ , \left [ g_{a \uparrow}^{sc}(0 ) - g_{a \uparrow}^{n}(0)\right ] , { \nonumber \\}g_{b \uparrow}^{sc}(0 ) & = \eta \ , g_{b \uparrow}^{n}(0 ) + \frac { 3 \delta_0}{2 t } \ , \left [ f_{b \downarrow}^{sc}(0 ) - f_{b\downarrow}^{n}(0)\right ] , { \nonumber \\}g_{a \uparrow}^{n}(0 ) & = \eta \,g_{a \uparrow}^{sc}(0 ) , { \nonumber \\}g_{b \downarrow}^{sc}(0 ) & = \eta \ , g_{b \downarrow}^{n}(0 ) - \frac{3 \delta_0}{2 t } \ , \left [ f_{b \uparrow}^{sc}(0 ) - f_{b \uparrow}^{n}(0)\right ] , { \nonumber \\}g_{a \downarrow}^{n}(0 ) & = \eta \ , g_{a \downarrow}^{sc}(0 ) . \label{eq : bc - final-2}\end{aligned}\ ] ] j. t. ye , y. j. zhang , r. akashi , m. s. bahramy , r. arita , and y. iwasa , science * 338 * , 11931196 ( 2012 ) . j. vu@xmath270i@xmath270evi@xmath270 , m. o. goerbig , and m. v. milovanovi@xmath270 , phys . b * 86 * , 214505 ( 2012 ) .

we report a measurement of the production cross section for hadrons in collisions at . using a data sample derived from an integrated luminosity of collected with the upgraded collider detector ( cdfii ) at the fermilab tevatron , we analyze hadrons , , partially reconstructed in the semileptonic decay mode . our measurement of the inclusive production cross section for hadrons with transverse momentum and rapidity is , where the uncertainties are statistical , systematic , and from branching fractions , respectively . the differential cross sections are found to be in good agreement with recent measurements of the cross section and well described by fixed - order next - to - leading logarithm predictions . fermilab - pub-09 - 069-e

0903.2403

measurements of the @xmath0-hadron production cross section in high energy @xmath1 collisions provide an excellent test of perturbative quantum chromodynamics ( qcd ) . next - to - leading order ( nlo ) calculations of the @xmath0-hadron cross section @xcite have been available for more than a decade . these nlo calculations have been recently supplemented with corrections for terms proportional to @xmath10^k$ ] and @xmath11^k$ ] with @xmath12 , where @xmath13 is the strong coupling constant , @xmath14 is the transverse momentum of the @xmath0 hadron , and @xmath15 is the mass of the bottom quark . these calculations are referred to as fixed order next - to - leading logarithm ( fonll ) @xcite . a number of @xmath0-hadron cross - section measurements has been performed at the tevatron . in the 1992 - 1996 running period , tevatron run i , measurements were performed at @xmath16 by the cdf and d0 experiments using several different techniques and final states @xcite . these measurements were consistently higher than theoretical expectations from nlo calculations . since 2001 , the tevatron has been running at @xmath17 . to date , the cdf collaboration has made two inclusive @xmath0-hadron cross - section measurements at the higher center - of - mass energy . the first was an inclusive measurement using @xmath18 decays , where @xmath4 denotes a @xmath0 hadron @xcite . this measurement was performed for @xmath7 @xcite and was the first to map out the @xmath0-hadron cross section down to zero transverse momentum . the @xmath19 cross - section result , along with an improved fonll calculation @xcite , prompted studies @xcite suggesting that prior discrepancies had been resolved with improved measurement and calculational techniques . more recently , the cdf collaboration has performed a measurement of the @xmath20 meson cross section using @xmath21 of data through the fully reconstructed decay chain @xmath22 , with @xmath23 , for @xmath20 transverse momentum , @xmath24 , greater than @xmath25 and @xmath26 . with large statistics and a high purity signal , the total uncertainty on this measurement is smaller than @xmath27 @xcite . the result is also in good agreement with fonll calculations . the difference between the original nlo calculation and the current fonll calculation arises from a number of different factors , as discussed in ref . @xcite . while improved theoretical calculations compare favorably to recent measurements , more experimental input is needed . in particular , cross - section measurements using different @xmath4 decay modes can be complementary to one another since they might be acquired using different trigger paths and be subject to different background contributions . for example , ref . @xcite showed that previous @xmath0-hadron cross - section measurements were inconsistent when comparing semileptonic and @xmath28 final states . in this paper , we present the first measurement of the @xmath4 cross section using semileptonic decays at @xmath2 . this analysis takes advantage of the distinct semileptonic @xmath0-hadron decay signature and provides a measurement that is complementary to those made in @xmath29 modes . we reconstruct signals of semileptonic @xmath0-hadron decays , @xmath30 , where charm mesons , @xmath31 and @xmath32 , are reconstructed in fully hadronic modes , using a data sample acquired with a dedicated semileptonic trigger path . throughout this paper , any reference to a specific charge state implies the antiparticle state as well . previous measurements of the @xmath0-hadron cross section at the tevatron using semileptonic decays were performed using inclusive @xmath33 @xcite and the semielectronic @xmath34 @xcite modes . there are no previously published @xmath0-hadron cross - section results from the tevatron utilizing the @xmath35 mode . we improve upon prior semileptonic results with a comparable data sample by extending the measurement to lower values of @xmath36 than have been accessible in previous analyses . the trigger path used in this analysis , which will be described in detail below , accepts muons with @xmath37 while previous measurements were limited to leptons with @xmath38 . we begin in sec . [ sec : overview ] with an overview of the measurement technique . the cdf ii detector and trigger system will be described in sec . [ sec : detector ] , followed by a discussion of trigger selection in sec . [ sec : trigger ] . section [ datasamplesection ] will describe the event selection and candidate reconstruction . sections [ sec : eff ] and [ sec : acc ] then describe the measurements of signal reconstruction efficiency and acceptance , followed by an assessment of the background contributions in sec . [ sec : bkg ] . in sec . [ sec : results ] , we present our results along with a comparison to theoretical predictions . we select events with an identified muon in conjunction with a fully reconstructed charm meson ( @xmath31 or @xmath32 , generically referred to as a @xmath39 meson ) such that the @xmath40 topology is consistent with a @xmath0-hadron decay . to measure the cross section , we count the number of signal events and correct for detector acceptance , identification inefficiency , and contributions from background sources . incorporating these factors , along with the measured @xmath1 luminosity , @xmath41 , we extract the cross section , @xmath42 , times branching ratio ( @xmath43 ) in the following way : @xmath44 where @xmath45 is the number of observed candidate @xmath46 ( @xmath47 ) events and @xmath48 is the background fraction , _ i.e. the fraction of events not originating from the decay of a @xmath0 hadron . the acceptance @xmath49 is defined as the fraction of events in which the muon and the charm decay particles pass through active regions of the cdf ii detector . the detection efficiency @xmath50 is the fraction of these events that are reconstructed . _ the product of branching ratios , @xmath43 , is shorthand for @xmath51 in the @xmath46 mode and @xmath52 for the @xmath47 mode . the generic @xmath0 hadron , @xmath4 , is an admixture of all weakly decaying @xmath0 hadrons , including @xmath20 , @xmath53 , @xmath54 , @xmath55 , and @xmath56 . to extract the @xmath4 cross section , we utilize the measurement of the inclusive @xmath35 branching ratio from the delphi and opal experiments @xcite . to apply this branching ratio to tevatron data , we assume that the fragmentation fractions at high energy ( @xmath57 ) are the same at lep and the tevatron . charm branching ratios are taken from cleo , argus , and aleph measurements @xcite . under the assumption that @xmath0 and @xmath58 quarks are produced in equal rates at the tevatron , we include both @xmath59 and @xmath60 final states and introduce the factor of 2 in the denominator of eq . ( [ crosssectionequation ] ) to report the cross section for a single @xmath0-hadron flavor . to extract the number of candidate signal events , @xmath45 , we reconstruct @xmath61 with @xmath62 , as well as @xmath63 with @xmath64 and @xmath62 , events using data that were acquired via a dedicated semimuonic @xmath4 trigger that requires a well measured muon and a track displaced from the primary vertex . the @xmath32 sample is a subset of the @xmath65 sample . we use these two samples to make separate , but correlated , measurements . the selection criteria are chosen to preferentially select @xmath4 decays and reject combinatoric backgrounds . there are two types of physics backgrounds that exhibit a @xmath40 signature that is similar to that of the expected signal . one arises from direct @xmath66 production where one charm quark decays to a muon and the other fragments into a @xmath39 meson . the second physics background arises from @xmath67 events where one @xmath0 hadron decays to a muon while the other decays to a @xmath39 meson . the contributions from these sources are assessed using a combination of data and monte carlo ( mc ) simulation . the acceptance @xmath49 is determined by using mc simulation . we generate @xmath0 quarks based on input distributions in transverse momentum @xmath14 and rapidity @xmath68 taken from theoretical calculations as well as previous cross - section measurements . fragmentation and decay are performed as part of the mc simulation . the passage of generated decay products in the cdf ii detector is then simulated . based upon the output of these simulated events , the fraction of produced events that fall within the active detector volume is calculated . we factorize the determination of the trigger and detector efficiency , @xmath50 , into a product of relative efficiencies and an absolute efficiency . the advantage of this technique is that we can measure relative efficiencies from the data ( as will be described in sect . [ sec : eff ] ) and minimize our reliance on simulation . the sensitivity of this analysis is limited by the systematic uncertainties . to limit our dependence on theoretical input and on detector simulation we estimate systematic uncertainties from data wherever possible . the cdf ii detector comprises a charged - particle tracker in the magnetic field of a solenoid surrounded by a full coverage calorimeter and by muon detectors . in this section we give a short description of those detector components that are relevant to this analysis . a more detailed description of them and of the entire detector is given in refs . @xcite . we utilize a cylindrical coordinate system with the @xmath69 axis oriented in the direction of the proton beam and the @xmath70-@xmath71 plane perpendicular to the beamline . we also define a polar angle , @xmath72 , measured from the @xmath69 axis and pseudorapidity is defined as @xmath73 $ ] . two devices inside the 1.4 t solenoidal magnetic field are used for measuring the momentum of charged particles : the silicon vertex detector ( svx ii ) and the central tracking chamber ( cot ) . the svx ii consists of double - sided microstrip sensors arranged in five cylindrical shells with radii between @xmath74 and @xmath75 . the svx ii detector is divided into three contiguous sections along the beam direction for a total @xmath69 coverage of @xmath76 . the cot is a cylindrical drift chamber containing 96 sense wire layers grouped into eight alternating superlayers of axial and stereo wires . the active volume covers @xmath77 and @xmath78 to @xmath79 in radius . the central muon detector ( cmu ) surrounds the central electromagnetic and hadronic calorimeters . the calorimeters have a depth corresponding to 5.5 interaction lengths . the cmu detector covers a pseudorapidity range @xmath80 0.6 and is segmented into two barrels of 24 modules , each module covering @xmath81 in @xmath71 . every module is further segmented into three submodules , each covering @xmath82 in @xmath71 and consisting of four layers of drift chambers . the smallest drift unit , called a stack , covers @xmath83 in @xmath71 . adjacent pairs of stacks are combined together into a tower . a track segment is identified by at least two hits out of the four layers of a stack . a second set of muon drift chambers , the cmp , covers the same @xmath84-range as the cmu and is located behind an additional steel absorber of @xmath85 interaction lengths . muons that produce a track segment in both the cmu and cmp systems are called cmup muons . a third set of muon drift chambers , the cmx , covers a pseudorapidity range @xmath86 and is separated by 6.2 interaction lengths from the nominal interaction point . luminosity is measured using gas cherenkov counters ( clc ) mounted at small angles to the beamline . the clc measures the rate of inelastic @xmath87 collisions . the inelastic @xmath87 cross section at @xmath88 gev is scaled from measurements at @xmath89 1800 gev using the calculations in ref . the integrated luminosity is determined with negligible statistical uncertainty and a 6% systematic accuracy @xcite . the cdf experiment uses a three - level trigger system . at level 1 ( l1 ) , data from every beam crossing are stored in a pipeline capable of buffering data from 14 beam crossings . the level 1 trigger either rejects events or copies them onto one of the four level 2 ( l2 ) buffers . events that pass the level 1 and level 2 selection criteria are sent to the level 3 ( l3 ) trigger , a cluster of computers running a speed - optimized version of the offline event reconstruction code . to select heavy flavor events , we rely heavily upon charged - particle tracking in the trigger . at level 1 , charged tracks with @xmath90 and @xmath91 are found by the extremely fast tracker ( xft ) @xcite , which uses information from the axial wires in the cot to perform @xmath70-@xmath71 track finding with high efficiency ( @xmath92 ) and good transverse momentum resolution ( @xmath93 ) . the track extrapolation system ( xtrp ) takes the track information from the xft and provides extrapolation information so that the xft tracks can be matched to track segments found in the muon detectors @xcite . tracks are matched to track segments in the cmu , cmp and cmx to identify muon candidates at level 1 . at level 2 , the silicon vertex trigger ( svt ) combines information from the axial layers of the svx ii with the charged - particle tracking information from the xft . owing to the high precision information from the silicon detector , the svt provides a precise measurement of the track impact parameter @xmath94 ( distance of closest approach to the beamline ) in the transverse plane with a resolution of approximately @xmath95 . this resolution arises from approximately equal contributions of @xmath96 from resolution and intrinsic spread of the position of the @xmath97 interaction . tracks with large impact parameter are utilized to identify heavy flavor decay and strongly reject light ( @xmath98 ) quark events . for the data sample utilized in this analysis , level 3 event reconstruction included full cot tracking but did not include tracking in the silicon detectors . events accepted by the level 3 trigger are stored for subsequent analysis . the flexibility of the cdf ii trigger permits the selection of samples with no leptons ( all hadronic modes ) , one lepton ( semileptonic modes ) and two leptons ( @xmath99 and @xmath100 events . ) for this analysis we utilize all three types of triggers , and we describe them in the following section . a single trigger path is defined as a specific set of level 1 , level 2 , and level 3 selection criteria in the cdf ii trigger system . the primary trigger path utilized in this analysis is referred to as the @xmath101svt path . data acquired through other trigger paths are utilized in this analysis to determine the selection efficiencies described in sect . [ sec : eff ] . * @xmath101svt path . at level 1 , this path identifies a muon candidate with @xmath102 by requiring that an xft track be matched to segments in the cmu and cmp detectors . at level 2 the trigger utilizes information from the svt to identify displaced tracks . we require a single track ( which is not the muon candidate ) with @xmath103 and an impact parameter between 120 @xmath104 and @xmath105 to be identified , where the impact parameter measured by the svt is relative to the beamline as determined over a large sampling of events . at level 3 , full event reconstruction is done , except for track reconstruction in svx ii . in order for the event to be accepted , level 3 requires confirmation of the muon identified at level 1 and confirmation of the xft track identified at level 2 . the invariant mass of the two tracks must be less than @xmath106 , the azimuthal opening angle between the two tracks , @xmath107 , must be @xmath108 90@xmath109 , and the tracks must be consistent with coming from the same @xmath1 interaction vertex . * * inclusive muon path . this trigger requires that one muon candidate with @xmath110 be identified by the xft , cmu , and cmp . this path provides a sample of @xmath111 decays where only one of the muons is required in the trigger . the second muon in the event is unbiased , and can be utilized to directly measure the efficiency of the trigger selection . * * heavy flavor path . this trigger path is for hadronic decays of @xmath4 and @xmath39 hadrons . two oppositely charged xft tracks with @xmath112 and @xmath113 are required at level 1 . those tracks are then required to have @xmath114 as measured by the svt at level 2 . the level 3 selection requires confirmation of the level 2 quantities using full reconstruction . from this path , we select large samples of @xmath115 , @xmath116 , and @xmath117 decays . since only two of the three tracks are required in the trigger , the third track is unbiased with respect to the trigger . we can therefore use this sample to measure the efficiency of the svt . * * @xmath4 semileptonic backup path . this trigger path is used to determine the level 3 selection efficiency . its requirements are the same as the @xmath101svt trigger , except that it has no requirements at level 3 , and only a fraction of events that pass the level 1 and level 2 trigger requirements are recorded . * @xmath28 dimuon path . the @xmath28 trigger paths which require one muon in the cmu system and a second muon in either the cmu or cmx system are used to measure the svx ii offline efficiency . the @xmath28 cmu - cmu trigger requires , at level 1 , at least two cmu muons be found with @xmath118 . there are no additional selection criteria applied at level 2 . at level 3 , an opposite sign muon pair is required with invariant mass between @xmath119 and @xmath120 , which is efficient for @xmath28 and @xmath121 decays into dimuons . the two muon tracks must also have @xmath122 and @xmath123 130@xmath109 , where @xmath124 is the separation of the two tracks along the beamline ( @xmath125 ) and @xmath126 is the separation of the two tracks in azimuth at @xmath125 . the @xmath28 cmu - cmx trigger is similar , except that instead of two cmu muons , one cmu and one cmx muon must be identified . the cmx muon must have @xmath127 . this analysis uses data derived from @xmath3 of @xmath97 collisions at @xmath128 that were collected between october 2002 and may 2003 . thanks to the trigger utilized in this analysis , this relatively small sample of data is sufficient to significantly improve upon previous cross - section measurements in semileptonic channels , and yields a result that is limited by the systematic uncertainties . we begin the signal reconstruction by requiring the candidate muon , pion , and kaon tracks to satisfy cot fiducial and quality selection criteria , which include requirements on the number of hits assigned to the track and a loose selection on fit quality of the track . muon selection criteria are motivated by the trigger requirements . the muon track must have @xmath129 and @xmath130 and must be matched to cmu and cmp track segments that satisfy fiducial and quality selection criteria , which include requirements on the number of hits used in the track segment and the quality of the match between the central track and the cmp track segment @xcite . the candidate muon must also be matched to a cmup muon found by the level 1 trigger . in order to reduce combinatoric background , the pion and the kaon candidates must have @xmath131 and @xmath132 , must satisfy svx ii fiducial and quality cuts , which include requirements on the number of hits assigned to the track and a loose selection on the fit quality of the track @xcite . the two tracks must both come from a common displaced vertex and be consistent with the decay of a charmed meson . in addition , at least one of the candidate tracks must be matched to an svt track that passes the displaced - track - trigger requirements , have @xmath133 , and have hits in all of the svx ii layers used by the svt . the pion and kaon candidates as well as the pion and muon candidates must have opposite charges . we combine the pion and kaon candidates to form @xmath134 candidates . we form the combined momentum vector of the muon and the @xmath31 candidate and require that this vector have @xmath135 . the number of @xmath31 signal events is determined by fitting the @xmath136 invariant mass distribution between @xmath137 and @xmath138 to a gaussian peak plus a linear background . the @xmath31 mass plots are shown for different regions of @xmath139 in fig . [ d0fitmassplot ] . while the signal yield is rather low in the highest @xmath14 region , clean signals are observed in the lower @xmath14 regions . the signal yields are reported in table [ yieldtable ] . figure [ d0fitmassplottotalnosvt ] shows the @xmath31 mass distribution for all events with @xmath140 9.0 gev/@xmath141 and @xmath135 . for the decay @xmath142 , @xmath62 , the muon and @xmath31 reconstruction is the same as described above . we additionally require the presence of the pion from the @xmath143 decay with @xmath144 mev/@xmath141 and @xmath132 that passes cot fiducial and quality selection criteria . since this pion tends to have low @xmath14 , we refer to this as the `` soft pion '' ( @xmath145 . ) we require that the muon and soft pion have opposite charges and only consider events where the @xmath136 invariant mass is between @xmath146 and @xmath147 . as can be seen in fig . [ d0fitmassplot ] , this mass window is wide enough so that no real @xmath31 decays are lost . we require @xmath148 between any two of the decay products of the @xmath4 must be less than @xmath149 . in order to minimize the background , we look at the mass difference @xmath150 between the @xmath151 and the @xmath136 . we fit the mass difference distribution to a gaussian plus the background shape @xmath152 , where @xmath153 and @xmath0 are free parameters in the fit to the data . the mass difference plots are shown in fig . [ dstarfitmassplot ] . in order to make comparisons between the two measurements , we present the data in @xmath14 bins of the @xmath46 system . the yields are included in table [ yieldtable ] . figure [ dstarfitmassplottotalnosvt ] shows the mass difference for all events with @xmath154 9.0 gev/@xmath141 and @xmath155 . the efficiencies of the cdf ii trigger and detector components for the cross - section measurement are determined whenever possible by using data collected by complementary trigger paths . the trigger paths utilized are described in section [ sec : trigger ] . we have separated the overall efficiency , @xmath50 , into nine separate measurements , denoted @xmath156-@xmath157 . the first eight terms are extracted using relative efficiencies and the ninth term is measured as an absolute efficiency . to calculate @xmath158 , we measure the efficiency for events to pass selection criterion eight _ relative to events that already have passed criterion nine . this yields @xmath159 . repeating this procedure for selection criterion seven , we calculate the relative efficiency for events to pass selection seven relative to events that have already passed selections eight and nine : @xmath160 . repeating this process for each of the selection criterion , we arrive at a final efficiency which is _ @xmath161 the absolute efficiency measurement ( @xmath157 ) is the efficiency for finding a charged - particle track in the cot . this absolute efficiency is determined using a combination of data and mc simulation based upon geant @xcite . a brief description of these efficiency measurements is provided below . details of all of the efficiency measurements and parameterizations may be found in ref . we perform these measurements in bins of kinematic variables . for example , the level 1 trigger efficiency depends upon the momentum of the muon . the efficiency corrections are then applied according to the kinematics of each event . for the @xmath101svt trigger path , the level 1 muon trigger requirement consists of an xft track with @xmath129 that is matched to track segments in the cmu and cmp subdetectors . we measure the efficiency of this trigger with respect to the offline cmup muon reconstruction efficiency . to do this , we reconstruct the @xmath23 signal in events that were collected via the inclusive muon trigger path . the trigger muon ( @xmath110 ) in these events is biased for the efficiency measurement , but the other decay muon of the @xmath162 can be used to determine the trigger efficiency . we refer to the @xmath28 decay muon that is unbiased by the trigger as the probe " muon , since it can be used to directly measure the trigger efficiency . we then compare the number of times the probe muon satisfied the trigger to the number of probe muons that were within the acceptance of the trigger . we require both muons of the @xmath162 to pass cot , cmu , and cmp fiducial and quality selection criteria . the probe muon must also have @xmath163 and @xmath164 4 gev/@xmath141 . we find a signal yield of approximately 1900 events . sideband subtraction is used to parameterize the efficiency in terms of muon @xmath14 and @xmath84 . we find the plateau efficiency ( @xmath165 ) to be @xmath166 . for the @xmath101svt signal path , the xft tracks found in level 1 are used as input for the level 2 svt trigger . the trigger path requires at least one xft track , not associated with the muon candidate , having @xmath168 ; the svt further requires that the impact parameter for this track be between @xmath169 and @xmath105 . the overall efficiency of this selection is given by the product of the xft and svt efficiency . we require that either the kaon or pion from the @xmath31 decay satisfy this selection . the xft efficiency is measured for pions and kaons with @xmath170 2.0 gev/@xmath141 and @xmath132 . given their different ionization energy losses , we treat kaons and pions separately in this measurement . in order to obtain pure samples of kaons and pions , we reconstruct @xmath171 events , @xmath172 , @xmath173 events , and @xmath174 , @xmath173 events that were collected by the heavy flavor trigger path . we have roughly 200 000 , 9500 , and 18 000 signal events respectively in the three modes . two of the tracks from each decay are required to pass the trigger requirements , and the third track is used to probe the xft efficiency . we apply both cot and svx ii fiducial and quality selection criteria on the tracks that fired the trigger , and cot fiducial and quality selection criteria on the tracks used to probe the xft efficiency . for the resonant decays , a cut is placed at @xmath175 around the @xmath71 mass . sideband subtraction is used to measure the pion and kaon yields . we parameterize the efficiencies in terms of @xmath14 , @xmath84 and @xmath71 . we additionally parameterize the xft efficiencies over the time that the data was acquired . the average efficiencies are approximately @xmath166 for pions and @xmath176 for kaons . the second component of the level 2 trigger efficiency is the svt efficiency for tracks that have already been identified by the xft . the svt efficiency does not depend on the particle species , so we measure the efficiency using a high statistics sample of @xmath178 events that were collected by the inclusive @xmath110 muon trigger path . the technique here is similar to the technique utilized to measure the level 1 efficiency . we use tracks that have @xmath170 2.0 gev/@xmath141 , @xmath132 , and @xmath179 inclusive muon trigger , both muons are used as probe tracks . we impose cot and svx ii fiducial and quality selection criteria on the probe track , and require the track to be matched to an xft track . therefore the svt efficiency is measured relative to a track found offline with the required number of svx ii hits . there are approximately 71 000 tracks that pass these selection criteria . the efficiency is parameterized in terms of @xmath14 , @xmath94 , and track isolation , which is a measure of the number of tracks found within an angular region of @xmath180 . for tracks with @xmath181 , the plateau efficiency for the svt is approximately @xmath166 . the @xmath84 dependence of the efficiency is small , and further mitigated because the tracks used in the efficiency measurement sample the detector in a manner similar to the signal . since the svx ii hit efficiency is not part of the level 2 svt efficiency calculation ( it contributes to the svx ii efficiency described below ) the inefficiency here is dominated by the track - finding and fitting algorithms utilized in the svt . the track isolation dependence of the svt efficiency requires special treatment . the occupancy in the silicon detector arises from a number of sources , including very low momentum tracks from the underlying @xmath97 interaction along with a spectrum of tracks originating from multiple @xmath97 interactions . these effects are not modeled reliably in the monte carlo simulation , and so we reweight the data based on the isolation of the signal events . in addition to the raw yields , table [ yieldtable ] also lists the yields after correction for the svt efficiency . at level 3 , our trigger path requires that a muon with @xmath183 be reconstructed . in addition , a track identified in the cot ( other than the muon ) must be matched to an svt track meeting the level 2 trigger requirements . the muon track and cot - svt track must have an invariant mass of less than @xmath184 and @xmath185 . to find the efficiency of this trigger , we look at @xmath186 events that were acquired on the @xmath4 semileptonic backup path . we parameterize the efficiency in terms of the @xmath14 of the muon and the @xmath14 of the svt track . for muons with @xmath187 , the level 3 efficiency is approximately @xmath188 . in order to determine the efficiency for attaching svx ii hits to a cot track , we look at muon tracks from @xmath23 decays collected on the @xmath28 dimuon trigger path . we use this sample in order to have a large sample of tracks with negligible fake rate . the tracks are required to pass cot quality and fiducial , as well as svx ii , selection criteria . we consider only tracks that are expected to pass through the four layers of the svx ii used in the svt trigger . the average efficiency is observed to be @xmath190 , where the inefficiency is dominated by dead silicon ladders and poor efficiency near the ends of the silicon ladders . to account for the dependence on svx ii coverage , we correct the efficiency as a function of the @xmath69 position of the track at its origin . we require hits in at least three layers of the svx ii on the @xmath31 decay track that is not required by the trigger . we find the efficiency by looking in our signal sample with and without the svx ii hit requirement for the second track . this yields an svx ii efficiency for the untriggered track of @xmath192 . the efficiency for the untriggered track is higher than that for the triggered track because of the looser requirement on the number of svx ii hits . in order to find the efficiency for linking a muon cot track to a cmu track segment , we reconstruct @xmath23 events using both the inclusive muon and semileptonic backup trigger paths . the muon tracks must pass cot fiducial and quality selection criteria . one of the decay muons from the @xmath28 must satisfy the muon trigger requirements . for events acquired on the semileptonic backup trigger path , the probe track must satisfy the svt trigger requirements . the probe tracks are extrapolated to the cmu , and are required to pass cmu fiducial selection criteria . the probe tracks must have @xmath194 and @xmath130 . the efficiency is found to be independent of @xmath14 , @xmath71 , and @xmath84 , and so we use a single efficiency value of @xmath195 . we find the efficiency for linking a cmp track segment to a cot track that has already been linked to a cmu track segment by reconstructing @xmath23 candidates in events collected by the @xmath28 dimuon trigger path . this trigger path does not require hits in the cmp , so either decay muon from the @xmath28 can be used to probe the cmp efficiency . we require the muon to pass cot and cmu fiducial and quality selection criteria . the muon is extrapolated out to the cmp and required to pass cmp fiducial selection criteria . we examine only muons with @xmath170 4.0 gev/@xmath141 and @xmath130 . the efficiency is parameterized in terms of @xmath14 and @xmath84 . the average cmp efficiency is @xmath176 . to determine the absolute cot track - finding efficiency , we use a combination of data along with a detailed mc simulation of the cdf ii tracking system . the techniques described here have been utilized in previous cdf analyses @xcite . we have verified that the mc simulation accurate models the charged - particle kinematics of our signal sample @xcite . we have additionally performed detailed studies of the cot simulation and find that the data is well - modeled by the simulation @xcite . first , we use a mc simulation only technique to compare charged particles generated by the pythia mc program to charged tracks reconstructed after full mc detector simulation and event reconstruction . this yields our measurement of absolute cot track - finding efficiency . to assess the systematic uncertainty on this measurement , we use a combination of data and mc simulation to `` embed '' simulated mc kaons , pions and muons into data events acquired via the inclusive muon trigger path . the mc simulated cot hits from the simulated track are incorporated into the detector data from a real event . to accurately emulate track reconstruction in @xmath46 events , the embedded mc simulated track is placed within a region in @xmath84-@xmath71 space that is within @xmath196 of the trigger muon . the hybrid event , which consists of data plus hits from a single simulated track , is then reconstructed in the same way as our signal sample . the simulated track is considered to be correctly found if a reconstructed track is matched closely in @xmath14 and @xmath71 to the simulated track embedded in the event . we vary the matching criteria to assess the rate at which tracks may be found but reconstructed with incorrect parameters . for tracks with @xmath197 ( which includes all muons and kaons , and the pions from the decay of the @xmath31 ) , the track - finding efficiencies derived from mc - only simulation and track embedding agree to within @xmath198 and are insensitive to the details of the matching criteria . the cot efficiency for muons is approximately @xmath199 . for pions and kaons with @xmath200 the cot efficiency is greater than @xmath166 . for pions with @xmath201 , which are used in reconstructing the @xmath142 decay , we observe a significant difference in efficiency estimates from the different techniques , and we use these variations to assess a systematic uncertainty on the low @xmath14 efficiency measurement . it is additionally necessary to account for uncertainty in the amount of material in the inner detector region , which also affects the absolute track - finding efficiency . we make a conservative estimate of the systematic uncertainty by varying the total amount of material by 25% . this rather large variation in the amount of passive material in the detector translates into a systematic uncertainty on the tracking efficiency of approximately @xmath202 . for a given signal event that is within the acceptance of the detector , the absolute efficiency to reconstruct the event is the product of the efficiency terms outlined in this section . in many cases , the efficiency depends upon the track momentum as well as its position within the detector . the details of these efficiency parameterizations can be found in ref . @xcite . for signal events that are well away from turn - on effects , the overall trigger and reconstruction efficiency is approximately @xmath203 . our detector acceptance is defined by the selection criteria and the active regions of the detector that cover those criteria . we determine the detector acceptance from mc simulation by first generating single @xmath0 hadrons according to measured momentum and rapidity spectra . these hadrons are then decayed using the evtgen package @xcite and then fed into a geant simulation of the detector . the mc simulation of the detector is utilized to map out the active regions of the detector . at this stage , we are not using the simulation to quantify detector inefficiencies , since those are measured using techniques described in the previous section . we generate @xmath0 hadrons based upon two distinct distributions in transverse momentum and rapidity . the first is based on nlo @xcite qcd theory with mrsd@xmath204 parton distribution function @xcite , a @xmath0 quark mass of @xmath205 , and renormalization and factorization scales equal to @xmath206 . peterson fragmentation is used , with @xmath207 . the other spectrum is based on the measured @xmath208 cross - section @xcite , which we refer to below as the `` cdf mc sample . '' we use the production fractions @xmath209 of 1:1:0.270 @xcite . we assume that @xmath56 decays provide a negligible contribution to our signal because of the low probability for a semleptonic @xmath0-baryon decay that would also yield a @xmath31 meson . to demonstrate the degree to which the mc simulation reproduces the data , figs . [ datatomccomppt](a , b , c ) show the @xmath14 distributions of the @xmath46 , @xmath101 , and @xmath31 for the mc simulation after all selection criteria and efficiencies have been applied , and for data after all selection criteria and corrections for the svt efficiency . the data are seen to have higher @xmath139 than the mc simulation based upon the @xmath208 cross - section result @xcite , while they are seen to have lower @xmath139 than the spectrum from the mrsd@xmath204 mc simulation , referred to below as the `` mrsd@xmath204 mc sample . '' we therefore measure the cross sections by taking the average of the cross sections from the two methods and assign one - half the difference as a systematic uncertainty . we perform additional comparisons between the data and the two mc simulated samples . distributions such as @xmath210 , @xmath211 , and @xmath212 are all seen to agree well between the data and both mc simulated samples . these comparisons can be found in ref . table [ acctimeseff ] shows the acceptance times efficiency ( excluding the svt efficiency ) as a function of @xmath36 bin calculated using the cdf and mrsd@xmath204 samples @xcite . there is significant uncertainty in the branching ratios for @xmath213 where the @xmath214 decays to @xmath215 . an incorrect modelling of this fraction can affect the @xmath139 spectrum and therefore the acceptance . to conservatively assess the systematic uncertainty on the measured cross section , we modify @xmath216 by @xmath217 . while fitting the @xmath31 peak to a gaussian allows us to determine how many @xmath60 events are in our sample and account for noncharm background , it is possible that there are real @xmath46 events that are not part of our @xmath218 signal . one source of this background is direct @xmath219 production . it is possible for direct charm to mimic our signal by having one charm quark decay to a muon , and the other charm quark fragment into a @xmath31 . the second background source is @xmath220 production , where one @xmath0 hadron decays to @xmath215 , and the other @xmath0 hadron follows the decay chain @xmath221 . since these background sources are irreducible , we estimate their contributions and correct the total number of observed signal events @xcite . in order to estimate the fraction of events in our sample from direct charm , we use the impact parameter of the @xmath31 @xcite . as the @xmath31 mesons from direct charm are created at the primary vertex , they should point back to the primary vertex within an error determined by the detector resolution . the @xmath31 from @xmath0 decays , on the other hand , are created at the secondary @xmath0 vertex , and are less likely to point back at the primary vertex . the @xmath31 impact parameter distribution for our signal sample is shown in fig . [ d0d0dat ] . although dominated by @xmath31 from @xmath0-hadron decays , this distribution contains a prompt component arising from direct @xmath219 production . we use mc simulation to determine the expected impact parameter distributions for @xmath31 mesons from @xmath0-hadron production and direct production . to determine the impact parameter distribution for @xmath31 mesons from @xmath0 decays , we use the same mc simulated sample that is used to find the acceptance above . in order to get the @xmath94 distribution of the @xmath222 from direct charm , we generate @xmath62 events originating from the primary vertex . the generated prompt @xmath65 sample has lower @xmath212 than does our signal sample , so we reweight the @xmath14 spectrum of the mc simulated @xmath31 so that it matches the @xmath212 spectrum in @xmath223 events from the pythia mc simulation @xcite . in the direct charm and @xmath224 samples , we calculate the impact parameter using generator level mc simulated quantities . in order to estimate the uncertainty on the @xmath31 impact parameter due to detector resolution , we look at the sideband subtracted @xmath31 impact parameter uncertainty distribution in our signal sample acquired on the @xmath101svt trigger . we see that the average uncertainty is approximately 34 @xmath225 , so we smear the mc simulated impact parameters using a gaussian resolution function with a standard deviation of 34 @xmath104 . we do a bin - by - bin @xmath226 fit to determine the ratio of events from direct charm to those from @xmath0 hadron decays . we find that using the @xmath94 distribution from the @xmath4 events generated with the @xmath14 spectrum measured in the cdf @xmath29 cross section measurement @xcite , we measure a charm fraction of @xmath227 , while using the @xmath94 distribution from the @xmath0 events in the mrsd@xmath204 sample gives a charm fraction of @xmath228 . so as a final result , we use the average charm fraction of @xmath229 . for the differential cross section , we must also account for the @xmath14 spectrum of @xmath46 candidates arising from direct @xmath219 production . for normalization , we use the total @xmath219 yield as calculated above . to determine the @xmath212 spectrum , we use mc simulation , where the generated events are reweighted to match the predicted direct charm spectrum @xcite . from this reweighted mc sample , we estimate the bin - by - bin contribution from direct @xmath219 background . the presence of @xmath220 events can mimic the signal in the cases where the @xmath230 and @xmath31 come from different @xmath0 hadrons . an example of this is where the @xmath0 hadron decays as @xmath231 and the other follows the decay chain @xmath232 . this configuration provides a @xmath233 candidate with the proper charge correlation , constituting an irreducible background to genuine @xmath234 decays . to investigate this background source , we look at events where one @xmath0 hadron decays to a @xmath31 meson , and the other follows the decay chain @xmath235 , providing the wrong - sign @xmath236 combination . the right - sign @xmath220 background is then estimated by counting wrong - sign @xmath236 events and correcting this by the expected ratio of the two decay chains @xmath237 and @xmath238 . we begin by looking for wrong - sign @xmath236 events , where we have a @xmath239 peak in events where the pion and muon have the same charge , @xmath240 . this is complicated by two factors . first , there is a large background in the wrong - sign sample from real @xmath233 signal events where the @xmath241 decay is incorrectly reconstructed as @xmath242 . the reflection does not yield a narrow @xmath31 peak in the pion - kaon invariant mass distribution but instead yields a broad peak which we parameterize using mc simulated @xmath31 decays . we use this parameterization and fix normalization of the reflection peak relative to the signal peak based upon the mc simulation . the mc simulated wrong - sign mass distribution is found to be insensitive to the input @xmath243 spectrum . the second complication is that the doubly cabibbo - suppressed decay of the @xmath244 is also expected to yield a small peak in the @xmath245-@xmath246 invariant mass distribution . based upon measured branching ratios @xcite , the number of doubly cabibbo - suppressed events is expected to be @xmath247 times the number of events in our signal peak . after correcting for these effects , we find a wrong - sign peak that is @xmath248 the size of the right - sign peak . to convert this number into an estimate of the number of right - sign @xmath249 events , we look at a generator level sample of @xmath249 mc simulation generated using pythia and decayed with evtgen using our best knowledge of branching ratios . in this sample , we look for events where one @xmath0 decayed to a muon , and the other one produced a @xmath31 . in order to get a sufficiently large event sample , we use somewhat looser selection criteria on the @xmath249 than are in our sample , and do not apply any of our efficiency measurements to the events . we find the ratio to be @xmath250 . to take into account the effect of the looser event selection criteria , we apply a 50% systematic , giving a final ratio of @xmath251 . combining this result with the wrong - sign contribution of @xmath248 yields a @xmath249 background fraction of @xmath252 . in addition to backgrounds from @xmath255 and @xmath249 events , a background also comes from the decays of a single @xmath0 hadron . these events can occur when the decay @xmath256 or @xmath257 is followed by one of the charm quarks decaying to a muon , and the other decaying through @xmath62 . these events also can come from a decay of @xmath258 , where the tau lepton decays to a muon , and the @xmath141 decays through @xmath62 . because these events come from real @xmath4 decays , the impact parameter of the @xmath31 will not necessarily be consistent with production at the primary vertex . the muon and the pion from the decay will have opposite charges , as is the case in @xmath259 events . we account for this contribution by retaining events with @xmath253 and @xmath254 decays in the mc simulation utilized for the acceptance measurement and correct for it in the unfolding procedure described in sect . [ sec : results ] . since the branching fractions for some of these decays are not well known , we allow the number of events from @xmath253 and @xmath254 to vary by @xmath217 , which translates in a @xmath260 systematic uncertainty on the measured cross section . in fitting the @xmath31 mass distribution , we account for combinatoric background using the sidebands of the mass distribution . this technique is not available for hadrons that are misidentified as muons , so we must estimate the contribution of fake muons independently . for @xmath219 background events described above , the background is assessed by looking at the impact parameter of the @xmath31 , independent of whether the muon is real or fake . therefore , the only source of fake muons that we must account for are fake muons reconstructed in association with real @xmath261 decays . to evaluate this , we use a sample of @xmath262 decays from the heavy flavor trigger path that is enhanced in @xmath263 decays . we measure the rate of tracks near the @xmath264 which are fiducial to the cmu and cmp detectors and have @xmath265 . because of their smaller nuclear interaction cross section , kaons are misidentified as muons at a higher rate than pions . to conservatively assess a muon misidentification rate , we assume that all of these tracks are kaons . we then apply a @xmath266 probability that a fiducial kaon is reconstructed as a muon , as measured in ref . @xcite . from this , we determine that the rate of fake muons in our signal sample is approximately @xmath267 , small enough to neglect . in order to extract the @xmath0-hadron differential cross section from the measured @xmath269 distribution , we unfold the observed @xmath36 distribution on the basis of the mc simulated signal . we determine a weight @xmath270 that is defined as @xmath271}_i \to { [ \mu^-d^0]}_j ) } { n({[\mu^- d^0]}_j ) } , \ ] ] where @xmath272}_i \to { [ \mu^-d^0]}_j)$ ] is the number of @xmath0 hadrons with @xmath7 in @xmath36 bin @xmath273 decaying to @xmath46 in @xmath269 bin @xmath274 , which pass all selection criteria , and @xmath275}_j)$ ] is the number of @xmath46 events in @xmath139 bin @xmath274 , which pass all selection criteria . we generate mc simulated @xmath4 events with @xmath276 as described in section [ sec : acc ] to properly model the detector acceptance . we also include events where the @xmath46 is produced via a @xmath253 or @xmath254 decay in the denominator in order to take that background into account . for the determination of the weights @xmath270 , it is the shape of the @xmath36 distribution , not its absolute normalization , that matters . we list the weights from the cdf and mrsd@xmath204 mc simulated samples in tables [ cdfweights ] and [ mrsd0weights ] . then , to unfold the number of @xmath0 hadrons @xmath277 in a given @xmath36 bin @xmath273 we use the formula @xmath278 where @xmath279 is the number of events in @xmath269 bin @xmath274 from data , shown in table [ yieldtable ] . we now have all of the terms from eq . ( [ crosssectionequation ] ) required to extract the cross section . the statistical uncertainty on the cross section for @xmath36 bin @xmath273 is given by @xmath280 ^ 2,\ ] ] where @xmath281 and @xmath282 are the acceptance and efficiency for the bin @xmath273 respectively . the statistical uncertainty on the total cross section is obtained by summing the uncertainties in quadrature over all bins of @xmath36 . we check the analytical calculation of the statistical uncertainty using mc simulated pseudo experiments which model the signal and background distributions along with bin migration of the signal . we verify that the mc simulation yields the same statistical uncertainty on the differential and integrated cross section as we observe using the above analytical formula . the uncertainty is calculated by generating 1000 distinct mc simulated experiments corresponding to the total sample @xmath46 yield we observe . the distribution of fitted yields in each @xmath14 bin is observed to be gaussian , and we take the standard deviation of the distribution to be the statistical uncertainty in each bin . the systematic uncertainties are determined in general by varying the efficiency or acceptance in question by @xmath283 and calculating the shift in the measured cross section . in the case of the efficiency measurements , which are parameterized based upon fits to the data ( _ e.g . , the level 1 trigger efficiency is parameterized as a function of muon @xmath14 and @xmath84 ) the variation in efficiency is taken from the error matrix determined in the fits . _ using this technique , the uncertainty derived based upon the svt efficiency fit appears to be underestimated . the variation in the measured points is not reflected by the statistical uncertainty in the fit for the svt efficiency . in this case , we assess the systematic uncertainty by replacing the fitted efficiency with the bin - by - bin points that are derived in the efficiency measurement . the difference between the result and the value obtained from the svt efficiency data points is taken as a fit modelling uncertainty and combined in quadrature with the 1@xmath42 systematic uncertainty described above to obtain the total systematic uncertainty @xcite . the systematic uncertainty on the mc simulated @xmath14 spectrum is estimated by taking the fractional difference between the cross section found using the @xmath284 and the reweighted mrsd@xmath204 nlo mc simulated samples . the degree to which final state radiation affects the @xmath31 lineshape is assessed with the photos mc simulation @xcite . systematic uncertainty contributions are shown in table [ systematicstable ] . it is worth noting that the systematic uncertainty arising from the mc simulated @xmath14 shape on the total cross section is less than the systematic uncertainty due to the mc simulation @xmath14 shape in most of the @xmath14 bins . this is because uncertainty on the @xmath14 shape primarily affects the fraction of the total cross section in a specific @xmath14 bin . when integrating over all bins , this effect is diluted . also , when finding the acceptance for a specific @xmath14 bin , one can only use the mc simulated events within that bin , while the overall measurement uses all of the mc simulated events . this leads to a lower statistical uncertainty in the mc simulated samples . after applying all corrections to the data , we obtain the total cross section times branching ratio @xmath285 for @xmath0 hadrons with @xmath286 and @xmath7 . using the current world average branching ratios @xmath287 @xcite and @xmath288 @xcite , we measure a total cross section of @xmath289 for @xmath0 hadrons with @xmath290 and @xmath7 . the differential cross section times branching ratio is shown in table [ xsectable ] , and displayed in fig . [ diffcorr](a ) . figure [ diffcorr](b ) shows the differential cross section . for the @xmath292 , @xmath293 , @xmath294 channel , the efficiencies are the same as those for the @xmath295 , @xmath296 , except that we must additionally account for the soft pion identification efficiency . the systematics for the @xmath32 measurement are the same as those for the @xmath31 measurement listed in table [ systematicstable ] unless mentioned specifically . after applying all corrections to the data , we obtain the total cross section times branching ratio @xmath292 , with @xmath293 and @xmath297 decays of @xmath298 for @xmath0 hadrons with @xmath290 and @xmath7 and using the notation @xmath299 as shorthand for the product of @xmath300 , @xmath301 , and @xmath302 . using the current world average branching ratios @xmath303 @xcite , @xmath304 @xcite , and @xmath305 @xcite , we find a total cross section of @xmath306 for @xmath0 hadrons with @xmath286 and @xmath7 . the differential cross section times branching ratio is shown in table [ xsectabledstar ] , and displayed in fig . [ difftimesbrdstar ] . we also show the differential cross section using branching ratios measured elsewhere . we consider the @xmath307 cross - section results to be a modest extension and cross check of the primary @xmath308 cross section result presented in this paper . the @xmath309 sample is of limited statistical power and , being a subset of the @xmath310 sample , there is nothing to be gained by averaging the two results . our results agree well with previous measurements of the @xmath4 cross section at cdf @xcite and with the predictions of fonll @xcite . the cdf collaboration previously measured the @xmath4 cross section using @xmath208 of @xmath311 in the same range of @xmath14 and @xmath68 @xcite . the fonll prediction for the total cross section with @xmath312 and @xmath7 is @xmath313 @xcite . figure [ comboplot1 ] shows the differential cross sections plotted together . to compare these results with the cross section measured in the exclusive decay channel @xmath22 @xcite , we must account for the different rapidity ranges . reference @xcite measured the cross section for @xmath26 , while the result presented here and the inclusive @xmath314 result from ref . @xcite measure the cross section for @xmath7 . figure [ comboplot2 ] shows the comparison between these results , where we have scaled all cross sections to @xmath26 assuming that inclusive @xmath4 production is flat in rapidity . furthermore , we correct for the differences in fragmentation fractions for @xmath20 ( @xmath315 ) and @xmath4 ( @xmath316 ) @xcite . figure [ comboplot2 ] shows good agreement between these recent @xmath0-hadron cross section measurements at @xmath2 , as well as agreement between these complementary measurements and the fonll prediction . the results presented here also compare favorably to the recent theoretical work using the general - mass variable - flavor - number scheme ( gm - vfns ) presented in ref . . the trigger and analysis strategy shown here for the @xmath46 and @xmath47 final state are quite different than those utilized in measurements using @xmath28 final states . in addition to being statistically independent , many of the systematic uncertainties of this result are distinct from those of previous measurements . this result therefore provides a unique and independent measurement of the @xmath4 cross section at the tevatron . we have measured the @xmath0-hadron production cross section in @xmath35 and @xmath317 decays using the cdf ii detector at the fermilab tevatron . the muon plus charm hadron decay signature was acquired using a dedicated trigger that takes advantage of the muon signature as well as the long @xmath4 lifetime . by selecting an event sample based upon decay length information measured in the trigger , it is possible to acquire clean @xmath4 signal samples at lower @xmath36 than have been measured before . complementary data samples are used to measure the trigger efficiencies for this analysis and monte carlo simulation is used to measure the acceptance . after accounting for backgrounds that mostly arise from heavy flavor ( @xmath220 and @xmath219 production ) the cross section is measured to be @xmath318 for @xmath0 hadrons with @xmath286 and @xmath7 . we additionally present the differential cross section @xmath9 in this kinematic region , which extends to lower @xmath36 than previous measurements in semileptonic modes . we find the results presented here to be in good agreement with other recent measurements of the @xmath4 cross section and in good agreement with fixed order next - to - leading logarithm calculations . we thank the fermilab staff and the technical staffs of the participating institutions for their vital contributions . this work was supported by the u.s . department of energy and national science foundation ; the italian istituto nazionale di fisica nucleare ; the ministry of education , culture , sports , science and technology of japan ; the natural sciences and engineering research council of canada ; the national science council of the republic of china ; the swiss national science foundation ; the a.p . sloan foundation ; the bundesministerium fr bildung und forschung , germany ; the korean science and engineering foundation and the korean research foundation ; the science and technology facilities council and the royal society , uk ; the institut national de physique nucleaire et physique des particules / cnrs ; the russian foundation for basic research ; the ministerio de ciencia e innovacin , and programa consolider - ingenio 2010 , spain ; the slovak r&d agency ; and the academy of finland . 99 p. nason , s. dawson , and r.k . b303 * , 607 ( 1988 ) . p. nason , s. dawson , and r.k . b327 * , 49 ( 1989 ) . m. cacciari , m. greco , and p. nason . j. high energy phys . , 9805 ( 1998 ) 007 . ( cdf collaboration ) , phys.rev . d * 65 , 052005 ( 2002 ) ; f. abe _ et al . _ ( cdf collaboration ) , phys . * 79 , 572 ( 1997 ) ; b. abbott _ ( d0 collaboration ) , phys . b * 487 , 264 ( 2000 ) ; b. abbott _ ( d0 collaboration ) , phys . rev . lett . * 85 , 5068 ( 2000 ) . ( cdf collaboration ) , phys . * 71 , 500 ( 1993 ) ; * * * * * f. abe _ ( cdf collaboration ) , phys . lett . * 71 * , 2396 ( 1993 ) . s. abachi _ ( d0 collaboration ) , phys . 74 * , 3548 ( 1995 ) . d. acosta _ et al . _ ( cdf collaboration ) , phys . rev . d * 71 * , 032001 ( 2005 ) . b. aubert _ ( _ ar _ collaboration ) , phys . lett . * 100 , 051802 ( 2008 ) ; s. dobbs _ ( cleo collaboration ) , phys . d * 76 , 112001 ( 2007 ) ; m. artuso _ ( cleo collaboration ) , phys . * 80 , 3193 ( 1998 ) ; r. barate _ ( aleph collaboration ) , phys . b * 403 , 367 ( 1997 ) ; h. albrecht _ ( argus collaboration ) , phys . b * 340 , 125 ( 1994 ) ; d. decamp _ et al . _ ( aleph collaboration ) , phys . b * 266 , 218 ( 1991 ) . * * * * * * j. bartelt _ ( cleo ii collaboration ) , phys . rev . * 80 , 3919 ( 1998 ) ; h. albrecht _ ( argus collaboration ) , z. phys . c * 66 , 63 ( 1995 ) ; f. butler _ ( cleo collaboration ) , phys . 69 * , 2041 ( 1992 ) . * * d. acosta _ et al . _ ( cdf collaboration ) , phys . d * 71 , 032001 ( 2005 ) . methods phys . res . , sect . a * 447 * , 1 ( 2000 ) . t. affolder _ et al . _ , nucl . methods phys . res . , sect . a * 526 * , 249 ( 2004 ) . g. ascoli _ et al . _ , methods phys . res . , sect . a * 268 * , 33 ( 1988 ) . d. acosta _ et al . methods phys . res . , sect . a * 494 * , 57 ( 2002 ) . m. m. block and r. n. cahn , rev . phys . * 57 * , 563 ( 1985 ) . * .the signal yields for the number of @xmath31 and @xmath32 events per @xmath139 bin before ( and after ) correcting for svt efficiency . since the svt efficiency depends upon the relative track isolation , a correction is performed based upon the topology of the signal events . [ yieldtable ] [ cols= " > , > , > , > , > " , ]

we present a new method to introduce phase - space fluctuations in transport theories , corresponding to a full implementation of the boltzmann - langevin equation for fermionic systems . it is based on the procedure originally developed by bauer et al . for transport codes employing the test particle method . in the new procedure , the pauli principle is carefully checked , leading to a good reproduction of the correct fluctuations in the `` continuum limit '' ( ) . accurate tests are carried out in one and two dimensional idealized systems , and finally results for a full 3d application are shown . we stress the reliability of this method , which can be easily plugged into existing tranport codes using test particles , and its general applicability to systems characterized by instabilities , like for instance multifragmentation processes . fluctuations ; stochastic collision integral ; fermionic systems ; transport theories . + pacs numbers : 24.10.cn ; 24.60.ky , 25.70.pq

0802.0377

in recent years , the dynamics of heavy ion collisions at intermediate energy has been extensively investigated within the framework of transport theories , such as the nordheim approach , in which the vlasov equation for the one - body phase space density , @xmath1 , is supplemented with a pauli - blocked boltzmann collision term @xcite . the basic ingredients that enter the resulting transport equation , often called boltzmann - uehling - uhlenbeck ( buu ) equation , are the self - consistent mean - field potential and the two - body scattering cross sections . these transport models hence describe the time evolution of the reduced one - body density in phase - space and , consequently , they are suited for the description of one - body observables , such as inclusive particle spectra in nuclear collisions , average collective flows and excitations . however , they can not provide a reliable description of fluctuation phenomena , such as multi - fragmentation processes , i.e. the break - up of excited nuclear systems into many pieces . in fact , neither fluctuations of one - body observables nor many - body correlations can be addressed with this class of mean - field models . hence suitable extensions , including fluctuations of the one - body density , have to be considered . an intense theoretical work on fluctuations in nuclear dynamics has started in the past years , also stimulated by the availability of large amounts of experimental data on fragment formation in intermediate energy heavy ion collisions and the possibility to observe a liquid - gas phase transition @xcite . in order to introduce fluctuations in transport theories , a number of different avenues have been taken , that can be essentially reconducted to two different classes of models . one is the class of molecular dynamics models @xcite while the other kind is represented by stochastic mean - field approaches @xcite . in molecular dynamics models the many - body state is represented by a simple product wave function , with or without antisymmetrization . the single particle wave functions are assumed to have a fixed gaussian shape . in this way , though nucleon wave functions are supposed to be independent ( mean - field approximation ) , the use of localized wave packets induces many - body correlations both in mean - field propagation and hard two body scattering ( collision integral ) , which is treated stochastically . hence this way to introduce many - body correlations and produce a trajectory branching is essentially based on the use of empirical gaussian wave packets . if wave functions were allowed to assume any shape , the method would become identical to standard mean - field descriptions . while the wave function localization appears appropriate to describe final fragmentation channels , where each single particle wave function should be localized within a fragment , the use of fixed shape localized wave packets in the full dynamics could affect the correct description of one - body effects , such as spinodal instabilities and zero sound propagation @xcite . on the other side , in the so - called stochastic mean - field approaches , the stochastic extension of the transport treatment for the one - particle density is obtained by introducing a stochastic term representing the fluctuating part of the collision integral @xcite , in close analogy with the langevin equation for a brownian motion . this can be derived as the next - order correction , in the equation describing the time evolution of @xmath2 , with respect to the standard average collision integral , leading to the boltzmann - langevin ( bl ) equation . thus , the system is still described solely in terms of the reduced one - body density @xmath2 , but this function experiences a stochastic time evolution in response to the random effect of the fluctuating collision term . in this way density fluctuations are introduced , that are amplified when instabilities or bifurcations occur in the dynamics . this procedure is suitable also for addressing multifragmentation phenomena , since fragments can be associated with the regions where the spacial density becomes larger , which finally can be reconstructed by sampling the one - body distribution function . a specific method for solving the boltzmann - langevin equation by direct numerical simulation was introduced in refs.@xcite . in this numerical implementation , the one - particle density @xmath3 is represented on a lattice of grid points in phase space and the collision integral is treated by considering all possible transitions between phase space cells , adding a noise term whose features are related to the average rate of transitions between two specified initial cells and two final cells . the numerical implementation of this method has only been possible in two dimensions ( 2d ) because it requires too large computer resources in 3d . hence several approximate solutions of the bl equations have been formulated , mostly based on the projection of the bl noise only on a given dynamical variable ( such as the local quadrupole tensor of the momentum distribution ) or on @xmath4 space only @xcite . more tractable fluctuating terms , such as a stochastic force added to the mean - field potential , have also been proposed and extensively applied to multifragmentation studies @xcite . however , the implementation of the full structure in phase space of the original bl term can still be considered as an important goal to reach . in fact , this would allow to treat a more general class of phenomena , where the correct description of fluctuations and correlations in @xmath5 space is essential ( such as particle production and fragment velocity correlations for instance ) . moreover , also in the multi - fragmentation mechanism , that is dominated by spacial density fluctuations , a more accurate representation of the full phase space dynamics , including fluctuations , would allow to improve the description of the fragment kinematical properties and correlations . we also stress the general interest of this effort . indeed transport phenomena occur in many physical systems , for which a more precise description of the time evolution of the one - body distribution function , including the effect of many - body correlations , would be important . a first attempt to introduce a fluctuating collision term in a 3d transport approach was made by bauer et al . this method can be implemented relatively easily into standard transport codes that adopt the scattering of pseudo - particles ( or test particles ) as a method of solution of the collision integral and it consists essentially in forcing similar two - body collisions to occur for neighboring test particles , defined according to a given distance in phase space , so that effectively two nucleons are involved in each particular collision event . the distance should reproduce the phase - space shape of the nucleon wave function . in this way the random nature of the two - body scattering , that in the standard codes applies only to test particles and is washed - out when using a huge number of them , is transferred to entire nucleons . however , in the procedure proposed in ref.@xcite the pauli blocking is checked only for the collision of the two original test particles and not for the entire swarm affected , leading to some unpleasant features in the description of fermionic systems . indeed the pauli - blocking violation introduces important inaccuracies in the fluctuations of the one - body density . in the present manuscript we present a new method to reconstruct the phase space nucleon wave function in mean - field approaches , in such a way that the pauli - blocking is checked for the entire cloud of moved particles this will improve the description of the fluctuation variance , approaching the one expected for fermionic systems . we will also pay special attention to the definition of the phase - space metric that would optimize the value of this variance . the paper is organized as follows : we will first recall the main ingredients of the bl theory , in order to connect the formalism with the numerical implementation adopted ( section 2 ) . then we will discuss in more detail the methods that have been proposed so far to solve the boltzmann - langevin equation ( section 3 ) . the new procedure that we follow to build fluctuations is presented in section 4 . several results demonstrating and analyzing in detail the method are discussed from section 5 to 8 . conclusions and perspectives are drawn in section 9 . within the semi - classical framework , the stochastic transport equation of motion for the one - body distribution function @xmath2 can be expressed in the following form , @xmath6,f\}\ = \ k[f]\ = \ \bar{k}[f]\ + \ \delta k[f]\ , \ ] ] where the left side describes the collisionless propagation of the individual particles in their common self - consistent one - body field , while the right side expresses the effect of the residual binary collisions . the collision term @xmath7 $ ] has a stochastic character . for example , the distance a particle travels in the medium before colliding is stochastic , as is the resulting scattering angle . at the boltzmann level , eq.([ble ] ) includes only the _ average _ part of the collision term , @xmath8 $ ] . usually , in nuclear systems , the quantum statistics is taken into account by adding suitable fermi blocking in the single - particle final states , leading to the uehling - uhlenbeck collision term @xcite . the resulting nuclear boltzmann equation exists in many implementations that differ with respect to both the physical input ( such as the types of constituents included , the form of their effective hamiltonian , and their differential interaction cross sections ) and the numerical methods employed ( whether of test particle or lattice type ) and various names have been employed in the literature , including buu ( for boltzmann - uehling - uhlenbeck ) , vuu ( for vlasov - uehling - uhlenbeck ) , landau - vlasov and bnv ( for boltzmann - nordheim - vlasov ) @xcite . the boltzmann - langevin treatment includes also the _ part of the collision term , @xmath9$]@xcite . in the simple physical scenario where the residual interaction can be considered as binary collisions that are well localized in space and time , the average part is given by @xcite : @xmath10 \ , \ ] ] where @xmath11 is a short - hand notation for @xmath12 and @xmath13 is the associated fermi blocking factor . @xmath14 is the degeneracy factor . the basic transition rate is simply related to the differential cross section for the corresponding two - body scattering process , @xmath15 being @xmath16 the relative velocity , and it thus has corresponding symmetry properties , @xmath17 since it arises from the same elementary two - body processes , the stochastic part of the collision term is fully determined by the basic transition rate as well , as a manifestation of the fluctuation - dissipation theorem . with the collisions assumed to be local in space and time , the correlation function for the fluctuating part of the collision term has the following form , @xmath18 where @xmath19 denotes the average with respect to the ensemble of possible trajectories resulting from the current one - body density @xmath2 . furthermore , for elastic scattering , the correlation kernel is given by @xcite : @xmath20 @xmath21 \ , \ ] ] with the short - hand notations @xmath22 and @xmath23 ( @xmath24 is the dimension of the space considered ) . the symmetry properties ( [ wsym ] ) of the transition rate ensure that the following sum rules hold , @xmath25 where @xmath26 is the kinetic energy for a specified momentum . these sum rules express the fact that each of the elementary binary collisions conserves particle number , momentum , and energy , respectively . the numerical implementation of the boltzmann - langevin equation is accomplished by correctly simulating the basic stochastic process , i.e. the stochastic transition rate among phase - space cells . one can realize that this point is very delicate in finite systems , when we work with a relatively small number of nucleons , but still need to build a smooth distribution function . there are different ways to overcome this difficulty . as mentioned in the introduction , one possibility , proposed in ref.@xcite , is to solve the bl equation on a lattice ; phase space is therefore divided into a number of cells , each one having volume @xmath27 ( @xmath28 denotes a point in phase space : @xmath29 ) . each transition involves four locations , and the collision integral arises from the sum of the transitions evaluated for all possible combinations of the cells : from this consideration one recognizes that the practicality of the method is limited by the huge computational effort required . in fact , only two - dimensional implementations exist @xcite . each transition represents a basic stochastic process , and , following the bl theory , the actual number of such transitions in a time step @xmath30 is dispersed around the mean value according to a poisson distribution , so its variance amounts to : @xmath31 where the average is given by : @xmath32 @xmath33[ni_med ] + this is a compact form of the fluctuation - dissipation theorem , since it predicts that the fluctuations are simply related to the mean number of dissipative processes . then , a noise term @xmath34 is added to the mean number @xmath35 , and thus the actual number of transitions is a random number picked from a normal distribution , the center of which is the mean value , and where the width is given by eq.([sigma_ni ] ) . in refs . @xcite this procedure is proven to yield correct results for a 2d system of fermions interacting through hard two - body scattering . starting from a non - equilibrium situation ( two touching fermi spheres ) the method is successful in reproducing the expected fluctuations , preserving also the average trajectory of the system . in particular , the fluctuations introduced build the expected statistical value for the equilibrium one - body density variance : @xmath36 evaluated considering phase - space cells of volume @xmath37 , for which the fluctuating transition rate is implemented . here @xmath38 denotes the equilibrium value of the one - body distribution function and @xmath39 . of course , eq.([variance ] ) holds also for volumes larger than @xmath40 and fluctuations are scaled accordingly . also the co - variance , i.e. the correlation between density fluctuations in two different phase - space points , @xmath28 and @xmath41 , is well reproduced @xcite . the same authors of ref.@xcite also verified that , for the same idealized system , but prepared at low spatial density , early fluctuations developed in momentum space are subsequently transmitted into density fluctuations and amplified by the nuclear mean field @xcite , leading to large instabilities and a statistical population of fragments . the method of ref.@xcite was originally developed as a way to overcome the problems arising from the solution of transport equations with the test particle method , which we now discuss briefly . within the test particle method , each nucleon is represented by a collection of @xmath42 test particles , that are propagated according to the mean - field interaction and random two - body scattering . the use of a large number of test particles allows to have a smooth distribution function and a good coverage of the phase space for the pauli blocking . on the other side , since collisions are treated stochastically for the single test particles , the dispersion around the average number of nucleon collisions ( eq.([sigma_ni ] ) ) is automatically divided by @xmath42 . hence fluctuations which are introduced by the test particle algorithm can also be seen as the correct ones expected from the bl approach , divided by the factor @xmath42 . various paths have been followed to overcome this problem . for instance , one can choose @xmath43 , i.e. work with whole nucleons @xcite , but in this case one has to solve the problem of the induced numerical errors on the smooth path of the dynamics , i.e. the mean - field and the pauli - blocking factors . alternatively , bauer et al . proposed a method to introduce a correlation between close particles in phase - space @xcite . the method follows the idea , applied in extended time - dependent - hartree - fock ( tdhf ) calculations , of evolving the single - particle density including a statistical mixing of slater determinants @xcite . the jump from one configuration to the other is possible under the important assumption that any coherence between determinants can be ignored ( decorrelation approximation ) . the choice of the single - particle basis for the determinant is however somewhat arbitrary , apart from the requirement of momentum - energy conservation . these ideas can be translated into the semi - classical approximation using the test particle method . the mixing between slater determinants is realized in the collision integral by means of the following procedure . 1 . first of all , the nucleon - nucleon ( nn ) cross section is reduced by a factor @xmath42 . 2 . then two test particles @xmath44 and @xmath45 are chosen as colliding partners , and will be moved from their positions @xmath46 and @xmath47 , to new positions @xmath48 and @xmath49 , according to the corresponding transition probability including the pauli blocking of the final states @xcite . if the two particles can collide , the scattering actually involves two `` clouds '' of neighbouring test particles , corresponding to two entire nucleons ( @xmath50 particles ) . the contiguity criterion is based on the following definition of phase - space distance : @xmath51 where @xmath52 is the fermi momentum and @xmath53 is the radius of the considered system . ( it should be noticed that the choice of this phase - space metric is rather arbitrary ) . 4 . in order to ensure energy and momentum conservation , the final states are adjusted using the average momenta of the two clouds as initial states . two important remarks have to be done : * the pauli blocking is checked only for the collision of the test particles @xmath44 and @xmath45 and not for each particle of the cloud . we will see in the following that this choice induces a strong violation of the pauli principle ; * the effect of the collision is a mere translation ( in momentum space ) of the two clouds to final positions , as sketched in fig . [ trasla ] . this method is conceptually simple , and moreover the implementation into existing buu or bnv codes is really straightforward . indeed it was immediately used to test the effects of the fluctuations on fragment mass spectra in a @xmath54 reaction at @xmath55 @xcite . its validity has been checked later by chapelle et al . @xcite , for the same idealized 2d system for which the bl lattice simulation method was implemented . accordingly , a different metric , involving only distances in momentum space , was used . as shown in @xcite , the final equilibrium value obtained for the fluctuation variance does not reproduce the expected one . the variance profile , as a function of the energy , appears to be proportional to @xmath56 rather than to @xmath57 . actually , this would be the result for a classical system , obeying to the boltzmann statistics . this can be considered as a consequence of the fact that pauli blocking is checked only for the first two colliding partners , and the system progressively evolves towards an equilibrium configuration which is more consistent with the one of a classical system . from this discussion it is evident that a proper treatment of the pauli blocking is a fundamental requirement for any numerical implementation of the boltzmann - langevin theory for fermionic systems . however , apart from these problems , the pseudo - particle correlation method represents a simple and practical way to implement fluctuations into transport codes . therefore , it is worthwhile to think about some improvements to make it more accurate . this is the aim of the present work , as we will discuss in the next section . the method devised by bauer et al . is able to agitate the phase space function and build fluctuations , although their strength does not reproduce the expected value for fermionic systems . however , a careful modification of the original procedure can considerably improve the results . the new procedure can be summarized in the following steps : 1 . the choice of the two colliding partners closely traces the standard recipe . if the two test particles are allowed to collide , two clouds of @xmath42 particles will be moved , with conditions specified below . 2 . only particles within a sphere , in coordinate space , around the center of mass of the two partners @xmath44 and @xmath45 can belong to the clouds . the distance criterion is @xmath58 , where @xmath59 is a free parameter ( see later discussion ) . then , for this considered space sphere , a grid is introduced in momentum space , around @xmath44 and @xmath45 , the size of each cell being @xmath60 . 3 . given the momentum space cells @xmath61 and @xmath62 containing the partners @xmath44 and @xmath45 , we consider the cells @xmath63 and @xmath64 , corresponding to the final positions , in a rotated frame , as indicated in fig . [ ruota ] . for a given set of initial and final cells , the number of test particles that will be actually moved to final states is equal to the minimum between the occupation of the initial cells , @xmath65 , and the availability of the final ones @xmath66 : @xmath67 where @xmath68 is the maximum number of test particles that can stay inside a cell . it should be noticed that this choice corresponds to the maximum of possible moves . 4 . surrounding momentum space cells are searched with the same prescriptions , until two entire nucleons are found . the search procedure is symmetric with respect to the center of momentum of the two partners and random in the vicinity of the original cells ( i , j ) . finally , the two clouds are moved to the new states of the rotated frame ( see fig.2 ) , so that the conservation of energy and momentum is automatically obtained . however , a further check is performed . in fact , due to the finite number of test particles , the one - body density is not perfectly homogeneous inside each cell . this causes slight violations of the conservations laws . hence the origin of the momentum space grid is eventually slightly displaced in order to have a perfect energy and momentum conservation . this method involves two parameters , namely the radius of the sphere in @xmath69 space , @xmath59 , and the size of the momentum cells , @xmath60 . the radius @xmath59 fixes the spatial extension of the nucleon , that in turn influences its spreading in momentum space . hence this parameter fixes somehow the extension of the nucleon wave packet in phase - space and could affect the transport dynamics . it can be constrained by physical arguments , and in general it should depend on the physical properties of the system , such as its dilution . the method proposed here has several advantages with respect to the original one of ref . it is still very flexible , and can be easily implemented into existing transport codes ; moreover , it is quite fast , and does not require huge amounts of resources . the pauli blocking is locally checked in each collision , so in principle this procedure satisfies the basic requirements to preserve the average evolution and to yield the expected fluctuations . besides , differently from the original method , where the nucleon is essentially spherical in phase space , here it is allowed to assume any shape in momentum space , since the only restriction is given by the cell size @xmath60 . in other words , it is possible to take into account momentum deformations of the nucleon wave packet , due to the momentum distribution of the system in particular conditions . for instance , when equilibrium is approached , the pauli principle allows collisions essentially for nucleons lying on the surface of the fermi sphere ; therefore such nucleons should not have a spherical shape in momentum space . it is also interesting to make a connection between the proposed method and the lattice simulation method to build fluctuations . the numerical procedure described in ref . @xcite , where the phase space is discretized by a fixed grid of cells , allows to construct the correct value of the fluctuation amplitude in each cell of the grid , generally taken of volume @xmath40 ( but in principle it can be even smaller ) , on the basis of the fluctuating gain and loss terms . on the other hand , our procedure is based on a different philosophy : instead of constructing fluctuations in fixed cells , when a collision occurs we move all together the phase space volume containing one nucleon . the shape of this volume is not fixed a priori and depends on the location of the four points ( @xmath70 ) that have been chosen as initial and final centroids of the considered two - body collision . this means that the `` correlation volume '' , i.e. the volume of the sphere that envelops all test particles that move all together in each collision , is not fixed , but has to be considered as a dynamical variable that generally exceeds @xmath40 and could even reach rather large values . in other words , compared to what is done in ref . @xcite , we build fluctuations in volumes containing one nucleon , but these volumes do not correspond to single cells of a fixed grid : instead , their contour changes from one collision to another , according to the prescription adopted to construct the clouds around the two first colliding test particles @xmath44 and @xmath45 . but how can the two proposed methods be connected to the original bl equation ? it should be noticed that the boltzmann - langevin equation ( [ ble ] ) is a semi - classical equation , derived in the `` continuum limit '' ( @xmath0 ) . this implies that the evolution of the system under consideration is completely determined by a smooth distribution function , i.e. it behaves as a fluid in phase space . in other words , the volume @xmath71 of the phase space cells for which the fluctuating transition rate is evaluated , is supposed to be much larger than @xmath40 . the boltzmann - langevin equation is essentially based on the idea of a fluctuating collision rate among phase space cells containing many nucleons . the formalism does not provide any additional information about the `` structure '' , i.e. the shape of the single nucleon wave packet . in both numerical implementations described above , two - body collisions among nucleons are treated as a stochastic process . only the definition of nucleon wave packet and/or of the elementary cells where fluctuations are built change from one method to the other . as discussed above , this ingredient is not contained in the bl equation and such information is beyond its derivation . hence , both procedures can be considered as correct implementations of the boltzmann - langevin theory in the `` continuum limit '' , the definition of `` nucleon wave packet '' being rather arbitrary . however , it is also very interesting to investigate the differences between the various possible methods used to build the nucleon clouds , that in turn influence the correlation volume . in particular , one can also try to force the procedure in order to reduce the wave packet smearing and to correlate particles inside smaller volumes ( close to @xmath40 ) . in the present first implementation of the method , similarly to what was done for the lattice simulations of ref.@xcite , we neglect the evolution in coordinate space , extending @xmath59 to the whole coordinate space . the cell size @xmath60 is then constrained by the following arguments : it should be small enough to allow an accurate check of the pauli blocking , but large enough to contain a sufficient number of test particles to reduce numerical uncertainties . in any case , @xmath60 has not to exceed the volume where , at most , one nucleon can be accommodated : @xmath72 , being @xmath73 the volume in the coordinate space ( @xmath74 ) . in the following we take @xmath75 , unless expressly indicated . since we neglect the coordinate space , we will refer only to volumes @xmath76 in momentum space . the expected equilibrium value of the variance @xmath77 in such volumes is : @xmath78 where @xmath79 is the number of nucleons that can be accommodated , at most , in the considered volume @xmath76 . it should be noticed that , according to our procedure to build the nucleon wave packet , the expected value of fluctuations , ( eq.[scaling ] ) , can be reproduced only in volumes @xmath76 larger than the average correlation volume , while it is underestimated in smaller cells , since fluctuations are built on a larger scale . in order to illustrate how our procedure works , we will first consider the very simple case of particles moving only along one direction . we take a large system , containing 1000 nucleons with average occupancy @xmath80 . so the total extension in momentum space of the system is @xmath81 . then the system is divided into @xmath82 smaller cells that can contain at most half nucleon , whose occupancy can be either @xmath83 or @xmath84 . hence @xmath85 . the occupancy is randomly chosen at the beginning and collisions are performed until equilibrium has been reached . the initial ( @xmath86,@xmath87 ) and final ( @xmath88,@xmath89 ) states of a collision are chosen randomly among the @xmath90 cells . we neglect energy and momentum conservation in this simple example . moreover , for simplicity , we take @xmath91 and @xmath92 . hence only half of the considered space is independent and , in this situation , the probability for a collision to happen is proportional to the product @xmath93 . according to eq . ( [ scaling ] ) , the expected fluctuation value , in a given momentum space volume @xmath76 , will be @xmath94 . the general prescription ( see sect . [ improved ] ) states that when a collision happens , a full nucleon has to collide . therefore , we move together the cell @xmath95 and the closest occupied cell that finds an empty cell in the final position . this example can be seen as a simple implementation of the general procedure when considering nucleons represented by only two test particles . one can easily realize that the `` correlation radius '' , i.e. the distance between the two cells that move together , changes from one event ( collision ) to another : sometimes the two co - moving cells are neighbours , sometimes they can be rather far . of course , for practical reasons , we have to restrict our search within a given radius . for instance , considering 10 cells on each side ( with respect to @xmath95 ) , we are able to reconstruct the nucleon and to perform the collision in more than 98@xmath96 of the cases . ( see text ) , in the idealized case of particles moving only along one direction . ] in fig . [ flu_joseph ] we present the results for the fluctuation variance as a function of the number @xmath97 of `` nucleon '' cells ( of volume @xmath98 ) that are contained inside the volume @xmath76 where the function @xmath2 is evaluated . one thousand events have been considered . the fluctuation value presented in the figure is multiplied by @xmath97 . as one can see in the figure , the correct value of the fluctuations , i.e. @xmath99 , is approximately recovered ( within @xmath100 ) for volumes @xmath101 . the slight deviation is due to the upper bound considered for the `` correlation radius '' and to the constraints imposed by the total mass conservation , that are also responsible for the decrease of fluctuations observed for large @xmath97 values . from this simple example , it already appears that the exact fluctuation value @xmath102 is reached only asymptotically . this is due to the arbitrariness of the nucleon shape and of the centroid position . hence , when testing a given procedure to solve the stochastic collision integral , the resulting fluctuations have to be compared to the expected values in large volumes , in which all different possibilities ( nucleon configurations ) may be accommodated . these considerations would apply also to the original procedure proposed by bauer et al . thus , in ref.@xcite , the comparison between the obtained fluctuation variances and the expected ones should have been done considering also volumes containing many nucleons , and not only cells of volume @xmath103 where , as seen in fig.[flu_joseph ] , fluctuations are naturally suppressed by the fact that the nucleon wave packet may take any shape . however , this does not affect the conclusions drawn in @xcite , mainly based on the wrong profile of the density variance ( and not on its absolute value ) . in the next section we will move to describe more complex geometries and different procedures to build the nucleon wave packets here we will discuss our procedure to build fluctuations in the context of a system of fermions having a given density and temperature , i.e. initialized according to a fermi - dirac distribution . this is a situation that is easily reached in the course of a dissipative nuclear reaction , after the initial collisional shock @xcite . since collisions happen mostly on the fermi surface , let us first consider only particles within a thin stripe @xmath104 around the fermi momentum , where @xmath80 . for those colliding particles the relative velocity @xmath105 is constant , since only particles with opposite momenta can collide , in order to stay on the fermi surface . so the transition rate is essentially governed by the product @xmath106 . this simplified system will allow us to study the sensitivity of the results to the ingredients of the procedure used to build , step by step , the nucleon clouds . as also discussed before , we will show that the correct fluctuation amplitude is recovered in the `` continuum limit '' , i.e. in large volumes . however , in the following we will mostly concentrate on the fluctuation amplitude in volumes of size @xmath98 , in order to probe the extension of the nucleon wave packet . the latter could influence significantly the evolution of actual nuclear collisions , where it could be important to work with more compact nucleon configurations . using this simple model we will find that building the maximum fluctuation value ( e.g. @xmath107 ) in such a volume , i.e. correlating particles inside @xmath103 , is a quite difficult task and additional efforts are required in order to reduce the nucleon cloud smearing . in order to keep the particles on the fermi surface , the original procedure described in sect . [ improved ] was slightly modified so that the modulus of the momentum of the particles does not change : only particles belonging to cells with opposite momenta can collide . then , for symmetry reasons it will result @xmath108 . the transition probability finally depends only on two values of the distribution function , @xmath109 and @xmath110 . we use spherical coordinates @xmath111 as independent variables . the collision partners of particles in the cell @xmath112 will be the ones in the cell @xmath113 . the choice of these coordinates implies the grids used in the procedure to be fixed along the @xmath114 axis . they can slide along the @xmath115 axis , where periodic conditions are taken into account . the procedure is as follows : a test particle @xmath44 is chosen at random ( together with the corresponding partner @xmath45 ) and a grid is constructed , centered around @xmath44 and @xmath45 ( see fig.2 ) . the rotation angles , also taken from a flat distribution , fix the final states and the new frame corresponding to these final positions . we consider a shell with @xmath116 cells , whose size is chosen to accommodate one nucleon at most , so that @xmath117 . we have checked that the results do not depend on the number of cells , i.e. the number of nucleons considered , by performing the same calculations on a @xmath118 system . we use @xmath119 test particles per nucleon . the growth of fluctuations is described by the variance @xmath77 of the distribution function , as discussed before . usually it is evaluated performing an ensemble average over a large number of events having the same initialization . however , in this scheme the mean value @xmath120 does not depend on the position @xmath5 and is not changing in time . considering the large number of cells employed , we can also follow the evolution of the system by calculating in a single event the following quantity : @xmath121 where @xmath122 denotes the average with respect to the cells and @xmath123 . in order to understand how the recognition of the nucleon wave packet can be influenced by the details of our procedure , we first further simplify the problem by using a single fixed grid to make the collisions , i.e. test particles are placed at ( and can only move to ) the center of the cells of the grid . in this case we have a single reference frame , so the effect of a collision ( in the space @xmath124 ) is a mere translation to other states of the same grid . apart from the use of test particles , and related differences in the method followed to solve the collision integral , in this way the procedure becomes similar to the one used in ref . we will take , as initial conditions , @xmath2 strictly equal to @xmath125 in each cell of the fixed grid , corresponding to @xmath126 nucleons . by construction , once collisions are performed , the only possible values for @xmath2 are @xmath83 , @xmath125 and @xmath84 . the maximum fluctuations in @xmath103 ( @xmath107 ) would be reached only if , in each cell , @xmath2 is either zero or one . achieving this result is not a trivial task : cells starting with @xmath127 can be partially emptied , and cells with @xmath128 partially filled . the dot - dashed curve in fig.[full_esa ] , upper panel , shows the variance of the distribution function calculated in the cells of the fixed grid , with increasing number of collisions ( i.e. as a function of time ) . it saturates approximately when almost all nucleons have suffered one collision . in the equilibrium configuration the number of cells with the three possible values of @xmath2 is roughly the same , as shown in fig . [ full_esa ] ( lower panel ) , so the fluctuation variance is @xmath129 , which is about @xmath130 of the maximum . , as a function of the number of collisions , for three different cases : fixed grid with @xmath131 ( dot - dashed ) , fixed grid with randomly distributed test particles ( full ) , moving grid ( dashed ) . lower panel : distribution of the number of cells as a function of their occupation @xmath2 at equilibrium , for the three cases indicated above . ] , as a function of the number of collisions , for three different cases : fixed grid with @xmath131 ( dot - dashed ) , fixed grid with randomly distributed test particles ( full ) , moving grid ( dashed ) . lower panel : distribution of the number of cells as a function of their occupation @xmath2 at equilibrium , for the three cases indicated above . ] the fluctuations reached in @xmath103 also depend on the variance of the initial configuration . if we initialize the system distributing the test particles randomly , the occupation in the cells has an average value @xmath132 and a width given by @xmath133 , where @xmath134 is the average number of test particles per cell ( @xmath135 ) . the distribution function can assume any value between @xmath83 and @xmath84 , as shown by the full curve in fig . [ full_esa ] , lower panel . the initial continuous distribution of values of @xmath2 causes a reduction of the fluctuations in @xmath103 ( full curve in fig . ( [ full_esa ] ) , upper panel ) with respect to the previous case . as expected , the fact that @xmath2 varies continuously between @xmath83 and @xmath84 contributes to enhance the smearing of the nucleon wave packet and hence to reduce fluctuations in @xmath103 . now we turn to a more general situation , relaxing the condition of the fixed grid . hence here we would like to test the full procedure , described in section 4 , that we will finally follow in the full 3d case . the two reference frames used to make the collision ( the one concerning the initial states and the one for the final states ) are now shifted along the @xmath115 axis , since the angle @xmath115 of the initial and final states can assume any value between @xmath83 and @xmath136 , according to the initial and final positions of the colliding test particles , @xmath44 and @xmath45 . this carries two important consequences . first of all , even being the pauli blocking checked for each collision in a particular frame , the prescription @xmath137 can be violated in another frame . however , we checked that this violation concerns only about @xmath138 of the cells . second , the correlation between the cells representing the reconstructed nucleon , which is maximum in the search frame , will be systematically broken in another ( shifted ) reference frame , increasing the nucleon delocalization and thus reducing fluctuations in @xmath103 . the dashed curve in fig . [ full_esa ] shows the fluctuations obtained in @xmath103 in this case . since this feature is due to the arbitrary position of the centroid of the constructed nucleon with respect to a fixed reference frame , we will name it _ centroid effect_. we notice a fluctuation reduction of @xmath139 with respect to the fixed grid case ( dash - dotted line in the same figure ) . up to now , the step of the grid has been chosen so that @xmath140 . however , in general , the size of the search cell has to be smaller than @xmath103 , especially when the nucleon wave packet is rather delocalized in @xmath141 space ( being more compact in coordinate space ) . in fact , small search cells are required for the pauli blocking to be accurate . hence we want to study the sensitivity of the results to the size of the search cell . let us start with the fixed grid scheme . in fig . [ smallphi_full ] we plot the fluctuation variance obtained dividing the step in the @xmath115 direction , @xmath142 , by two or four ; the volume of the search cell is then @xmath143 , respectively . this is equivalent to shift , for each collision , the origin of the grid by @xmath125 or @xmath144 @xmath142 respectively , and acts also as a sort of centroid effect , thus reducing the variance in @xmath103 ( see solid and dashed lines in fig . [ smallphi_full ] ) . , also shown in fig . [ full_esa ] . ] the value obtained with @xmath85 is lower than the corresponding case with @xmath140 , indicated as `` fixed grid '' in the figure . moreover , it further reduces for @xmath145 . nevertheless , this smearing effect is partly compensated by the fact that the procedure used to select the cells belonging to the nucleon clouds favours compact , symmetric configurations . it is interesting to evaluate the effect of the use of a small search cell on the fluctuations also in the general case , i.e. when the fixed grid condition is relaxed . intuitively , we expect a small effect on the variance , as long as @xmath146 is not too small : the main reduction of fluctuations comes already from the _ centroid effect _ ( see previous subsection ) , which is equivalent to an infinitely small step . in fact , the differences between calculations with @xmath85 and @xmath145 ( fig . [ smallphi_full ] , dotted and dash - dotted curves respectively ) are smaller than in the previous calculation with fixed grid . the reduction with respect to the case with @xmath140 ( denoted as `` centroid effect '' in the figure ) is also smaller . in conclusion , it appears from our calculations that the smearing of the nucleon wave packet is rather dependent on the procedure adopted to build nucleons and is particularly affected by the so - called _ centroid effect _ , related to the use of randomly distributed test particles for the phase - space mapping . as a consequence , fluctuations calculated in volumes @xmath103 appear reduced . as shown above , the use of a fixed grid to locate the initial and final cells of the colliding test particles would help to increase the value of the variance in @xmath103 . however , energy and momentum conservation would not be exact , due to the phase - space discretization . on the other hand , the standard test particle method allows a good mapping of the phase space and a good resolution of the average collision integral , with perfect energy and momentum conservation . however , when constructing fluctuations , the _ centroid effect _ may destroy the correlations already built and increase the nucleon cloud smearing . however , the expected statistical value of the fluctuation variance is recovered in large volumes . this last point has been checked for all the simplified situations considered here ( fermi surface ) . for instance , this analysis is shown in fig.[fluct_new ] in the case of the fixed grid with @xmath140 and test particles randomly distributed inside the system ( last case of section 5.2 ) . ) as a function of the number @xmath97 of nucleon cells @xmath103 , obtained in the fixed grid case with @xmath147 and test particles randomly distributed . ] here fluctuations are evaluated in volumes @xmath76 , by considering an ensemble of 800 events . as expected , the fluctuation variance , rescaled by @xmath97 , almost approaches the expected value as the volume @xmath76 increases . as explained above for the idealized case presented in fig.3 , the maximum value , @xmath99 , is not exactly reached because of the upper bound imposed to the correlation radius in the nucleon reconstruction procedure and of mass conservation constraints . in the fermi energy domain , two colliding nuclei may be schematically represented by touching fermi spheres in @xmath148 space , so it is interesting to study the development of fluctuations starting from this initial distribution . in our simplified 2d scheme , we can mimic the two spheres using a `` chess - board '' initialization , that is subdividing our momentum space in four regions where @xmath2 is alternatively zero or one . although the system is in a non - equilibrium situation , the variance , defined as in eq.([sigmaf ] ) , starts from its maximum value ( since @xmath149 ) . it can be shown that in the fixed grid case the equilibrium configuration consists of a random distribution of full and empty cells , and the variance keeps constant in time . however , a very different result comes out either changing the search step or removing the fixed grid constraint . in both simulations the variance rapidly decreases because the initial correlation is destroyed by the _ centroid effect_. as an illustrative example , in fig . [ f1_full ] we plot @xmath77 for the choice @xmath150 , fixed grid . comparison with the corresponding curve for the configuration @xmath151 ( dashed , see section 5.4 ) shows that the two calculations converge to the same value . ( dashed curve ) in the fixed grid case for @xmath150 . ] we get a confirmation of these findings by removing the fixed grid constraint . also in this case , at the end we find no difference with the case described before ( initialization with randomly distributed test particles , section 5.3 ) , see fig . [ f1_rand ] . moreover , in this case the convergence between the two calculations is rather quick . then we can conclude that the final outcome does not depend on the initial conditions of the system , but only on the final equilibrium situation , as it should be . but for @xmath147 and without the fixed grid constraint . ] guided by the simplified schematic cases illustrated above , let us now discuss the full case of a nuclear fermi gas of particles at a given temperature and density interacting through hard two - body scattering . as in the previous examples , since we focus on fluctuations in momentum space , only one large cubic cell is present in coordinate space , and all particles can be chosen to collide ( no restrictions in @xmath69 space ) . the size of the box is @xmath152 , and we consider @xmath153 nucleons , so that the density has the saturation value @xmath154 ; each nucleon is represented by a collection of @xmath119 test particles . besides , we do not consider any distinction between neutrons and protons , so that one nucleon occupies at least a phase - space volume @xmath155 ( @xmath156 ) . we initialize the momenta so as to reproduce a fermi - dirac profile corresponding to a temperature of @xmath157 . finally , we consider a constant cross section @xmath158 . in these calculations the volume @xmath60 corresponds to a cube of side @xmath159 ( and coincides with @xmath160 ) . we notice that , for the full fermi gas case , a grid in cartesian coordinates is easier to use , with respect to spherical coordinates , in the nucleon construction procedure . however , spherical coordinates will be adopted to analyze the resulting fluctuations . calculations are stopped when the fluctuation variance saturates . first of all , we checked the effect of our fluctuating collision integral on the average evolution of the system . in fig . [ effe ] we plot the energy profile of the distribution function at the initial time ( solid ) and at the time when the fluctuation variance saturates and we stop the calculation ( dashed line ) . slight changes are observed , probably due to the finite extension of the nucleon packet , that induces a discretization of the phase space . ( solid ) and at the time @xmath161 ( dashed line ) . ] we also compared the average collision rate with analytical expectations : we find that the number of collisions is slightly larger than the analytical one , but differences are within @xmath162 ; on the other hand , similar deviations are also found in usual buu implementations , due to the approximate mapping of phase space induced by the finite number of test particles . in the following we analyze in more detail the capability of the procedure to build the fluctuations in the considered 3d scenario . we investigated the region in @xmath5 space where collisions are most probable and the extension of the nucleon cloud . from the left panel of fig . [ polar ] it is evident that most collisions involve nucleons lying on the surface of the fermi sphere , as expected . besides , the right panel illustrates the extension of the clouds in the radial direction , calculated according to the definition : @xmath163 where @xmath164 denotes the center of momentum of the cloud and @xmath165 is the modulus of the momentum of the test particles contained in the cloud . from this distribution one can deduce that the clouds correspond to a relatively narrow stripe , @xmath166 , in the region where @xmath2 is different from either zero or one . right panel : distribution of the size of the nucleons on the same axis.,title="fig : " ] direction . right panel : distribution of the size of the nucleons on the same axis.,title="fig : " ] moreover , we have checked that the extension of the constructed wave packet is similar in the three directions @xmath167,@xmath168,@xmath169 , as expected . we notice also that the tail of the distribution ( fig.10 , right ) extends up to large @xmath170 values , @xmath171 . although the nucleon clouds extends , on average , over a volume @xmath172 , that is not much larger than @xmath103 , the _ centroid effect _ , i.e. the fact that the nucleon centroid may be located anywhere , causes a significant smearing of fluctuations , as already discussed in section 5 . in fact , we find that , for our fermi gas calculations , fluctuations in @xmath103 are considerably underestimated , as shown in fig.[flu_vp ] , where the variance @xmath77 is reported as a function of the energy @xmath173 . , in cubic cells of size @xmath103 , for a fermi gas at equilibrium . ] in the following , to test our procedure , we will evaluate fluctuations considering big volumes , that contain many nucleons . however , it should be noticed that , in the geometry considered , it is not trivial to predict which is the size and shape of the most appropriate volume to recover the expected value of fluctuations . while , from one side , it is desirable to consider volumes containing a large number of nucleons , from the other side one should keep in mind that spurious fluctuation reduction , due to the conservation of the total number of particles , may be a problem . hence we have performed a systematic analysis of fluctuations as a function of the coordinates @xmath174 , integrating over the angle @xmath115 . the corresponding volume employed for the calculation of the variance is given by : @xmath175 we consider volumes corresponding to fixed angular spreads @xmath176 and/or to a given step @xmath177 . in fig . [ spicchi_p ] we fix the angular spread , @xmath178 , and we consider four different steps , @xmath177 , corresponding to different volumes , i.e. to a different number of nucleons @xmath79 that can be contained inside . and different steps @xmath179 . ] the considered volumes can contain up to a rather large number of nucleons ( about 90 ) . the variance , multiplied by @xmath97 , is plotted as a function of the energy @xmath180 . we find that the peak of the variance is around the fermi energy , as expected . the rescaled variance gets larger when we increase the step @xmath177 . however , even in the best situation , the expected equilibrium value at the peak ( @xmath181 ) is underestimated by a factor @xmath182 . therefore , for a fixed value of @xmath183 , we tested different angular spreads . and different angular spreads ( curves with circles ) . the thick full and dashed lines are explained in the text . ] in fig.[spicchi_teta ] we present the obtained results , rescaled by @xmath79 . now we can observe that , increasing the angular spread , the variance first increases , then decreases again , due to the mass conservation constraints . the variance gets larger when considering the angular spread @xmath184 , but the expected value is still underestimated by about a factor @xmath185 , at the peak . this result can be understood on the basis of the results obtained in @xmath103 and displayed in fig.[flu_vp ] . in fact , in order to recover the expected fluctuation value , the volume employed for the calculation of the variance must be much larger than @xmath186 in all the directions @xmath187 . due to the considered geometry , while this can be achieved in the angular coordinates , this can not be easily obtained along the @xmath188 direction without losing locality for the value of @xmath2 and related fluctuations . actually , in the case of @xmath189 , the corresponding extension along the @xmath188 direction , @xmath190 , of the volume considered to evaluate the function @xmath2 is approximately equal to @xmath191 at @xmath192 . we thus expect fluctuations to be reduced , with respect to @xmath193 , by a factor @xmath194 , being @xmath195 the reduction factor in @xmath103 , as evaluated from fig.[flu_vp ] . the expected fluctuation reduction can be extracted also from the schematic calculations described in sect.5 . in fact , from the inspection of fig.[full_esa ] , it appears that , in the fixed grid scheme , fluctuations in @xmath103 are reduced by a factor @xmath196 . relaxing the fixed grid constraint in one dimension , fluctuations are further reduced by a factor 0.64 with respect to the fixed grid case . we thus expect a fluctuation reduction by a factor @xmath197 ( at the fermi surface ) . this is close to our result at @xmath198 ( see the dashed line with circles in fig.13 , for instance ) . one may try to introduce this correction in the results of fig . 13 . in the considered geometry , the reduction factor @xmath199 of the fluctuations in @xmath103 depends on the energy @xmath180 , i.e. one can write , for the variance in @xmath103 : @xmath200 where @xmath201 indicates the expected fluctuation value , that in our case should coincide with @xmath193 . by approximating the volume dependence of the rescaled variance by a linear behaviour for small volumes ( see fig.[flu_joseph ] for instance ) , the fluctuation variance presented in figs . 12 - 13 can be rewritten as : @xmath202 we notice that @xmath190 is also depending on @xmath180 . eqs.([sig1])([sig2 ] ) allow to extract the suppression factor @xmath203 , as well as the function @xmath201 . the latter is displayed in fig.13 ( thick dashed line ) . one can see that the expected fluctuation value , @xmath193 ( thick solid line ) , is well reproduced at the fermi surface , while it is underestimated , within @xmath204 , expecially for outer regions . this underestimation indicates that in our procedure , due to the finite extension of the nucleon cloud , the regions far from the fermi surface are less involved in the building of fluctuations than they should be and can be considered as an intrinsic limitation of the method . however , the overall shape of the fluctuation variance is reasonably reproduced , relative to @xmath205 : this is a nice indication that the pauli blocking is effective , and the equilibrium distribution is almost unchanged by the procedure . therefore , it really represents a remarkable improvement with respect to the original method by bauer et al . ( see ref.@xcite ) , which makes us confident about applications to nuclear reactions . we can directly visualize the effect of the fluctuating term by selecting a thin region in momentum space around the fermi momentum . for this purpose , we adopt a new set of coordinates @xmath206 ; for fixed @xmath188 and @xmath104 , the distribution function depends only on two coordinates , namely : @xmath207 with @xmath208 . this set of coordinates correspond to displacements of equal length along the three directions : @xmath209 . we choose @xmath210 ( approximately equal to the fermi momentum ) and @xmath211 , and in fig . [ tetaeta ] we plot the distribution function @xmath212 at two different times ( initial and final times ) . we recognize the sinusoidal profile on the @xmath213 axis due to the modulation given by the @xmath214 factor . at the initial time , @xmath2 is nearly uniform , and its fluctuations are simply due to the numerical noise induced by the finite number of test particles . at later times , we observe the growth of fluctuations , evidenced by the typical structure with `` peaks and holes '' . at time @xmath215 ( left ) and @xmath161 ( right).,title="fig : " ] at time @xmath215 ( left ) and @xmath161 ( right).,title="fig : " ] in actual nuclear collisions the extension of the nucleon wave packet in phase - space may affect the transport dynamics and it may be important to correlate the test particles belonging to the nucleon clouds inside smaller volumes . in the context of the simplified model ( fermi surface ) discussed in sect.5 we attempted to find a method to reduce the nucleon cloud smearing and enhance the fluctuations in @xmath103 without losing the consistency of the original procedure . first of all , it should be remembered that the collision probability is given by @xmath216 ( see sect . [ details ] ) : therefore , we need to consider only two locations in @xmath148 space . the main idea is to force the system to fully empty the cells associated with the initial state or fully fill the final state when a collision occurs . this condition is not automatically achieved for all cells contributing to a given collision . in fact , the number of particles , @xmath217 , taken from a cell @xmath95 and employed to build the nucleon obey to the constraint : @xmath218 being @xmath42 the number of test particles per nucleon . hence the number of test particles that are really taken from a cell is not always equal to @xmath219 , but instead to @xmath220 , being @xmath221 the idea is to try to avoid the situation when the number of particles moved is less than @xmath222 . in this way fluctuations get to a larger value , since we favour the transitions where cells are completely filled or emptied . in order to implement this idea , when constructing the nucleon clouds we choose , among the possible ones , the cells for which the probability @xmath223 is large . , as a function of the number of collisions , in the fixed grid scheme . solid line : no optimization ; dashed line : optimized procedure . ] the outcome of this method is plotted in fig . [ opt_full_esa ] for the fixed grid case , where the result of fig . [ full_esa ] is also shown for comparison ( solid line ) . as the dashed curve in fig . [ opt_full_esa ] clearly shows , this optimization procedure is successful in enhancing the fluctuations , which eventually reach almost their upper limit @xmath107 . actually , this maximum value is not obtained due to the fact that only adjacent cells are considered in the nucleon construction . in this way , some half - filled cells remain `` isolated '' and are not involved in a collision anymore . indeed , at the time when we stop the calculation , isolated cells amount to about @xmath138 of the total . the method works also in more general situations , for instance in the case corresponding to fluctuating initial conditions ( see sect.5.2 ) . , as a function of the number of collisions , in the fixed grid scheme with fluctuating initial conditions . solid line : no optimization ; dashed line : optimized procedure . ] the comparison between variances with and without the optimization procedure is shown in fig . [ opt_full_rand ] . as in the previous case , the optimization favours a faster growth of the fluctuations . their saturation value for @xmath224 is now enhanced , by @xmath225 with respect to the standard procedure . this is an indication that the procedure is able to increase the number of empty and full cells . this feature is also evident from the distribution of values of @xmath2 plotted , for the same two simulations of fig.16 , in fig . [ opt_dndf_full ] . in the end , this procedure helps the nucleon packets to keep a similar , more compact shape in @xmath148 space during the time evolution of the system , from one collisional event to another . as a consequence , nucleons are more localized and fluctuations in @xmath103 appear enhanced . for the same two calculations of fig . [ opt_full_rand ] . ] we have proposed a new numerical implementation of the full boltzmann - langevin equation in 3d . the stochastic character of the two - body collision integral , that in standard transport codes applies only to the single test particles , leading to a strong reduction of fluctuations , is recovered , at the nucleon level , by a careful reconstruction of the nucleon wave packet once collisions occur . this is achieved by moving , in each collisional event , two entire test particle clouds , corresponding to one nucleon each , as proposed in ref.@xcite . however , while in the original procedure the pauli blocking was checked only for the centroids of the nucleon clouds , here we propose a new method that allows to construct nucleon wave packets by carefully checking the pauli blocking for the entire swarm of test particles . the last point is essential in the construction of the correct fluctuation value in fermionic systems . the nucleon wave packet may take , in principle , any shape . thus the correlation volume , i.e. the volume of the sphere that contains all test particles that move together is generally larger than @xmath226 . moreover , the nucleon centroid is chosen randomly among the test particle distribution , leading to a partial overlap of nucleon configurations from one collision to another . all these effects lead to a reduction of fluctuations in the cells of volume @xmath226 . however , the correct fluctuation value is recovered in larger volumes , where all possible nucleon configurations may be accommodated . the procedure , and its numerical ingredients , has been tested carefully in several idealized situations , studying in particular the relation between the results obtained for the fluctuations in cells of volume @xmath226 and the ingredients of the model . considering a fermionic system at equilibrium , at a given density and temperature , we find that our procedure builds , within a good approximation , the expected profile of the fluctuation variance , @xmath227 , as a function of the nucleon energy . this can be considered as a significant improvement with respect to the original procedure @xcite , that opens interesting possibilities of applications to nuclear collisions . in fact , the method proposed can be easily implemented into existing transport codes , and could allow to treat , in a more complete way , phenomena where large fluctuations or bifurcations occur and fluctuations are important . g.f.bertsch and s.das gupta , phys . rep . * 160 * ( 1988 ) 189 a.bonasera , f.gulminelli and j.molitoris , phys * 243 * ( 1994 ) 1 r.botet et al . , * 86 * ( 2001 ) 3514 j.b.elliott et al . * 88 * ( 2002 ) 042701 m.dagostino et al . , nucl . phys . * a699 * ( 2002 ) 795 . ph.chomaz and f.gulminelli , nucl . * a647 * ( 1999 ) 153 ; f.gulminelli , ph.chomaz , al.h.raduta , ad.r.raduta , phys.rev.lett . * 91 * ( 2003 ) 202701 g.tabacaru et al . , . phys.j . * a18 * ( 2003 ) 103 a.ono , phys . rev . * c59 * ( 1999 ) 853 j.aichelin , phys * 202 * ( 1991 ) 233 h.feldmeier , nucl.phys . * a515 * ( 1990 ) 147 a.ono , h.horiuchi , t.maruyama and a.ohnishi , phys . * 68 * ( 1992 ) 2898 ; a.ono , h.horiuchi , t.maruyama and a.ohnishi , prog . theor . phys . * 87 * ( 1992 ) 1185 . m.papa , t.maruyama , a.bonasera , phys . rev . * c64 * ( 2001 ) 024612 m.colonna and ph.chomaz , phys . lett . * b436 * ( 1998 ) 1 j.rizzo , m.colonna , a.ono , phys . rev.*c76 * ( 2007 ) 024611 s.ayik , c.gregoire , phys . * b212 * ( 1998 ) 269 , and refs . therein . j.randrup and b.remaud , nucl . phys . * a514 * ( 1990 ) 339 ph.chomaz , g.f.burgio , and j.randrup , phys.lett . * b254 * ( 1991 ) 340 g.f.burgio , ph.chomaz , and j.randrup , nucl . phys . * a529 * ( 1991 ) 157 e.suraud et al . phys . * a542 * ( 1992 ) 141 ; e.suraud et al . , nucl . phys . * 580 * ( 1994 ) 323 ; f.s.zhang and e.suraud , phys * c51 * ( 1995 ) 3201 m.colonna et al . , nucl . phys . * a642 * ( 1998 ) 449 w.bauer , g.f.bertsch , and s.das gupta , phys . * 58 * ( 1987 ) 58 , proc . r. soc . london * a119 * ( 1928 ) 689 , phys.rev . * 43 * ( 1933 ) 552 w. bauer , g.f . bertsch and h. schultz , phys.rev.lett . * 69 * ( 1992 ) 1888 b. borderie , b. remaud , m.f . rivet and f. sebille , phys.lett . * b302 * ( 1993 ) 15 b. a. li and w. bauer , phys . lett . * b254 * ( 1991 ) 335 ; phys . * c44 * ( 1991 ) 450 f. chapelle , g.f . burgio , ph . chomaz , and j. randrup , nucl . phys . * a540 * ( 1992 ) 227 g.f . burgio , ph . chomaz , and j. randrup , phys . * 69 * ( 1992 ) 88 p.g . reinhard and e. suraud , ann . of phys . * 216 * ( 1992 ) 98 ; p.g . reinhard and e. suraud , nucl . a545 * ( 1992 ) 59c

we argue for finiteness of flux vacua around type iib cy singularities by computing their gauge theory duals . this leads us to propose a geometric transition where the compact 3-cycles support both rr and ns flux , while the open string side contains 5-brane bound states . by a suitable combination of s duality and symplectic transformations , both sides are shown to have the same ir physics . the finiteness then follows from a holomorphic change of couplings in the gauge side . as a nontrivial test , we compute the number of vacua on both sides for the conifold and the argyres - douglas point , and we find perfect agreement . . @firstpageps . @empty oddhead evenheadoddhead 0

hep-th0611002

the study of string vacua in flux compactifications of type iib has attracted much attention , in part as a setting where many properties of the landscape of vacua are under control ( see @xcite for a recent review ) . one of the first issues to be addressed here is whether the number of realistic string vacua is finite @xcite . recently it has been shown in @xcite that the ashok - douglas index of supersymmetric vacua is finite . this is a crucial step towards proving the finiteness of flux vacua . the aim of the present work is to understand the physics underlying the previous result . it was shown in @xcite that the index of supersymmetric vacua is finite around smooth points in moduli space ; the analysis may be restricted then to singularities of the moduli space where the curvature diverges . the finiteness proof @xcite is based on weil - petersson geometry and a detailed analysis of degenerations of hodge structures on the moduli space . however , from the string theory point of view it is not clear which is the physical mechanism responsible for this . specially , why singularities leading to very different field theories all give a finite number of vacua . our approach may be summarized as follows . we construct calabi - yau s where singularities are easily embedded and argue for finiteness of vacua around them by computing their dual gauge theories . we establish a precise correspondence between flux and gauge degrees of freedom . this shows that the gauge theories are generalized versions of the ones obtained through the dijkgraaf - vafa correspondence @xcite ; but they still have finitely many vacua . as we shall see , the underlying reason for this is the topological nature of the chiral ring of such theories . in section [ sec : generalcy ] we discuss the type iib noncompact model that can embed ade singularities and study the nonperturbative superpotential generated by fluxes . next , in section [ sec : fluxcurve ] we derive the formula for counting vacua in the previous setup ; this involves nontrivial steps because of the noncompact nature of the model . then in section [ sec : dualym ] we construct the dual gauge theory after the geometric transition , applying s - duality to the dirac - born - infeld ( dbi ) action . the field theory turns out to be a generalization of the usual @xmath0 sym encountered in geometric transitions . the argument for finiteness of vacua is presented in section [ sec : finite ] ; it is based on the holomorphic dependence of the gauge effective superpotential on nondynamical fields ( couplings ) . finally , in section [ sec : examples ] we show that the computations from the gravity and gauge side agree for the conifold and argyres - douglas singularities . section [ sec : concl ] contains our conclusions . we start by studying moduli stabilization in type iib theory in a calabi - yau threefold . since we are interested in analyzing a neighborhood of an ade singularity , it is enough to consider noncompact threefolds of the form @xmath1 the nontrivial dynamics comes from the complex curve @xmath2 : @xmath3 . ( [ eq : cy1 ] ) may be thought as a decoupling limit @xmath4 of an adequate compact variety @xcite , although this will not be necessary for our purposes . for concreteness , let us consider the case of a hyperelliptic curve where we can realize singularities of the a - type : @xmath5 @xmath6 is a polynomial of degree @xmath7 , and will play the role of the superpotential in the gauge theory : @xmath8 @xmath9 is a deformation of the singular curve @xmath10 . its effect is to split @xmath11 . if all the roots of @xmath12 are different then the singular curve has just odp ( conifold ) singularities . we will also encounter more complicated singularities , where three or more roots coincide . the fact that ( [ eq : cy2 ] ) is the same variety that appears in the dijkgraaf - vafa duality @xcite is not a coincidence ; the ( generalized ) gauge dual will play a major role in proving the finiteness of the number of vacua . furthermore , such local string models have been considered recently in the context of soft supersymmetry breaking @xcite . for our future computations , it is crucial to remark the following . in the four dimensional effective field theory ( eft ) , the moduli @xmath13 have a very different interpretation . fluctuations in @xmath14 have infinite energy and hence are non - dynamical ; each arbitrary choice of @xmath14 will give a different 4d theory so they can be interpreted as couplings . on the other hand , the @xmath15 s are dynamical and are interpreted as scalar fields in vector multiplets . their gauge theory meaning will become clear in section [ sec : dualym ] . as shown in @xcite , the periods of the noncompact threefold reduce to periods of the hyperelliptic curve : @xmath16 with @xmath17 the cycle @xmath18 surrounds the cut @xmath19 $ ] ; @xmath20 is the noncompact cycle dual to @xmath18 , running from @xmath21 to infinity . the @xmath22-periods need to be regulated ; this will be discussed shortly . therefore all the computations can be done directly on the hyperelliptic curve @xmath23 of genus @xmath24 . when @xmath25 @xmath26 this implies that @xmath27 ( and @xmath28 ) is a differential of the third kind on @xmath2 @xcite . for any value @xmath29 , there are two points on the riemann surface @xmath2 ; let @xmath30 denote the points corresponding to @xmath31 . then @xmath32 is a holomorphic differential only on the punctured surface @xmath33 . the details of the homology of @xmath2 and the effect of the punctures were considered in @xcite and we follow their conventions . a choice of homology cycles is shown in figure [ fig : homology ] ; @xmath34 runs through the @xmath35-th cut , from @xmath36 to @xmath37 . from these noncompact cycles we construct @xmath38 . besides , @xmath39 and @xmath40 circle the punctures at @xmath37 and @xmath36 respectively . the canonical symplectic basis of @xmath2 is @xmath41 , @xmath42 . in @xmath2 , @xmath43 so @xmath44 is not independent ; however , in @xmath45 , @xmath46 . this means that we can take @xmath44 to be an independent cycle and use this to fix the values of the meromorphic differentials at infinity . a symplectic basis for @xmath47 is hence @xmath48 , @xmath49 . ( 0,0 ) and @xmath45.,title="fig : " ] # 1#2#3#4#5 @font ( 11325,9543)(601,-8923 ) ( 601,-4336)(0,0)[lb ] ( 2926,-4336)(0,0)[lb ] ( 11926,-4336)(0,0)[lb ] ( 7051,464)(0,0)[lb ] ( 2626,-1411)(0,0)[lb ] ( 4651,-1411)(0,0)[lb ] ( 10576,-1411)(0,0)[lb ] ( 6676,-2011)(0,0)[lb ] ( 6751,-3200)(0,0)[lb ] in the holomorphic decomposition @xmath50 , there is a unique basis of holomorphic differentials @xcite @xmath51 such that @xmath52 where the period matrix @xmath53 is defined to be the symmetric matrix @xmath54 they can be constructed as linear combinations of the differentials @xmath55 the third kind differential @xmath56 has residues @xmath57 at @xmath58 respectively . an adequate linear combination of ( [ eq : diff1 ] ) and ( [ eq : diff2 ] ) will give the unique third kind differential @xmath59 such that @xmath60 @xmath61 every holomorphic differential on @xmath45 can be written as a linear combination of @xmath62 . such differentials are meromorphic differentials on @xmath2 with at most simple poles . a more symmetric description follows from taking @xmath44 ( instead of @xmath39 ) to be an independent cycle ; hence the basis of allowed differentials will be @xmath63 where @xmath64 is a superposition of @xmath65 and @xmath66 fixed by @xmath67 , @xmath68 . the complex moduli of @xmath69 are stabilized by turning on 3-form fluxes @xmath70 , which generate the nonperturbative superpotential @xcite @xmath71 in the noncompact model , the axio - dilaton @xmath72 is fixed , corresponding to a coupling of the 4d eft . upon integrating over the @xmath73 fibers given by @xmath74 , ( [ eq : superp1 ] ) reduces to the superpotential on the hyperelliptic curve @xmath75 the fluxes through all the compact cycles are quantized : @xmath76 @xmath77 . however , the fluxes through the noncompact cycles can vary continuously and , in fact , we will argue that they have to diverge . we denote @xmath78 these quantities will play the role of running gauge couplings . given that the @xmath22-cycles extend to infinity , and both @xmath27 and @xmath79 are differentials of the third kind , we need to regulate their @xmath22 periods . following @xcite we introduce a cut - off at large distances @xmath80 , replacing @xmath37 and @xmath36 by @xmath81 and @xmath82 . for the noncompact approximation to be consistent , ( [ eq : superp2 ] ) has to be finite in the limit @xmath83 . we write @xmath84 for the regularized version of @xmath20 , running from @xmath82 to @xmath81 through the @xmath85 $ ] cut . the @xmath81 dependence of @xmath86 is most easily obtained @xcite by doing a monodromy around infinity @xmath87.=3/2 $ ] , which follows from @xmath88=3 $ ] . ] in @xmath45 this corresponds to @xmath89 giving @xmath90 where @xmath91 are single valued contributions . comparing with ( [ eq : rtoinf ] ) , @xmath92 from ( [ eq : prepcutoff ] ) , we see that all the periods have the same @xmath93 dependence . it was shown in @xcite that the cutoff dependence of @xmath79 is exactly the one needed to cancel the logarithmic divergence from ( [ eq : prepcutoff ] ) and yield a finite cutoff independent @xmath94 : @xmath95 the @xmath96 where defined in ( [ eq : fluxes2 ] ) and @xmath97 are a set of finite energy scales . therefore ( [ eq : renormgeom ] ) may be interpreted as a _ geometric renormalization _ of certain bare coupling constants @xmath98 . this is the geometric analog of the rg running of the gauge couplings ( see sections [ sec : dualym ] and [ sec : finite ] ) . in this section we develop the necessary tools to count supersymmetric flux vacua on the hyperelliptic curve ( [ eq : cy2 ] ) . we will show that the index formula @xmath99 of @xcite is still valid in our case . this is a priori not obvious , the main issues being that the curve is noncompact so many quantities need a regulator and the punctures contribute extra moduli that have to be included . furthermore , having fluxes @xmath96 that can vary continuously would immediately lead to an infinite number of vacua . and finally , we will have to introduce a tadpole cancellation condition . counting supersymmetric flux vacua is equivalent to studying the geometry of the moduli space @xmath100 of @xmath45 . there are different ways of parametrizing it ; while from the eft it is natural to work with the @xmath101 , in the geometrical side it is more convenient to use the coefficients @xmath15 of the deformation @xmath102 . more specifically , we parametrize @xmath100 by combinations @xmath103 of the @xmath15 ( @xmath104 ) such that @xmath105 giving directly the basis of holomorphic differentials introduced in ( [ eq : zeta ] ) plus @xmath64 . this is an efficient and symmetric way of taking into account the modulus from the puncture at p and will simplify our formulas . @xmath2 becomes singular when two branch points coincide ; this leads us to define the discriminant @xmath106 where @xmath107 . we denote the zero locus by @xmath108 ; the moduli space is therefore @xmath109 @xmath108 is codimension one in @xmath100 and corresponds to conifold - like singularities : around two coinciding roots we can always perform a holomorphic change of variables to rewrite the curve as @xmath110 higher order argyres - douglas singularities @xcite occur when three or more roots coincide , and will be discussed in sections [ sec : finite ] and [ sec : examples ] . the moduli space is a special kahler manifold , with metric @xmath111 which can be derived from the kahler potential @xmath112 the covariant derivative is @xmath113 ( @xmath114 ) and the curvature tensor is @xmath115 a displacement in @xmath100 deforms the complex structure of @xmath2 , so we expect the holomorphic differentials @xmath116 to mix with the antiholomorphic ones . it is easy to show that the covariant derivative of a @xmath117 form gives a pure @xmath118 form : @xmath119 and the relation with the curvature is @xmath120 we want to count the vacua that preserve @xmath0 supersymmetry . in the limit @xmath4 , supersymmetric solutions are given by @xmath121 , where @xmath122 . as explained before , this limit corresponds to taking into account only a neighborhood of the singularity , so that supergravity effects are negligible . solutions to these equations may be viewed in two equivalent ways . if we want to stabilize at a particular point in the moduli space , @xmath123 is an on - shell condition that restricts the possible values of the fluxes to a subspace . indeed , since @xmath124 gives by construction a basis of @xmath125 , @xmath126 implies that @xmath127 on the other hand , a holomorphic differential is uniquely specified by giving its @xmath128-periods . indeed , the @xmath22-periods are then functions of the period matrix : @xmath129 the other possible point of view is that we can turn on arbitrary fluxes through all the cycles ; this will lift almost all the degeneracy of the @xmath130 supersymmetric moduli space , leaving only some number of @xmath0 supersymmetric vacua . therefore , if we specify arbitrarily both the @xmath128 and @xmath22 fluxes , ( [ eq : tonshell2 ] ) stabilizes the complex moduli of the curve : @xmath131 the ingredient that makes the number of vacua finite in compact calabi - yau manifolds is the tadpole cancellation condition @xcite . there is no such constraint in the noncompact case , since the flux can go off to infinity . however , the fluxes can not be arbitrarily large , because once their associated energy is of order @xmath132 , the noncompact approximation breaks down : our local variety will be mixed with far away cycles in the cy . therefore , in counting the total number of vacua , we have to impose by hand a tadpole condition . by analogy with the compact case @xcite , we require that @xmath133 using the on - shell formula ( [ eq : tonshell1 ] ) and recalling ( [ eq : metric ] ) , the tadpole condition becomes @xmath134 where @xmath135 . @xmath136 is the maximum value of @xmath137 , fixed by data of the compact cy that we choose to embed ( [ eq : cy2 ] ) . from ( [ eq : tadpole1a ] ) , the counting of supersymmetric vacua may be rephrased in terms of the geometry of @xmath2 : over each point @xmath138 in moduli space we have a ` solid sphere ' @xmath139 , with volume @xmath136 . each of these allowed points determines a point in flux space ; the number of such points will give the number of supersymmetric vacua . furthermore , ( [ eq : tadpole1a ] ) shows why degeneration limits may produce an infinite number of vacua : if @xmath140 develops a null direction , the tadpole condition will not bound the number of flux points . in other words , from this analysis it is not clear how configurations where one flux goes to infinity and another goes to minus infinity , in a correlated way such that @xmath141 stays finite , will be ruled out . the gauge theory analysis will shed light on this point . finally , even with the tadpole condition , the number of solutions to the equations of motion ( [ eq : eom1 ] ) with continuous fluxes @xmath96 will be infinite . fortunately , there is a simple way out of this problem . recall that the noncompact hyperelliptic curve should be considered as part of a compact cy . instead of parametrizing the fluxes with arbitrary energy scales @xmath97 , we take them to be integers . then ( [ eq : renormgeom ] ) will _ fix _ the energy scales at particular values , depending on the fluxes . this approach was also taken in @xcite to study the consequences of the klebanov - strassler solution @xcite and leads to the usual exponentially large hierarchies of energy scales , as we show later . now we have all the elements to count vacua on complex curves with punctures ; the derivation of the formula for the density of vacua continues as in @xcite : the number of supersymmetric vacua is given by @xmath142 @xmath143 with @xmath144 here , @xmath145 because of @xmath146 , we can replace @xmath147 in ( [ eq : matrixd2w ] ) . the main simplification in the noncompact case is that , since @xmath148 , @xmath149 , and then @xmath150 therefore _ the number of supersymmetric vacua coincides with the supersymmetry index _ , which is topological and , as we shall see , much easier to compute . on the contrary , in the compact case , when gravity is not decoupled , the supersymmetric index gives just a lower bound to the number of vacua . the final result is @xmath151 where @xmath152 . as expected , this coincides with @xcite when @xmath4 . the index @xmath153 is introduced for clarity reasons , to mean that this is the result from the closed string side . in this section we construct the supersymmetric gauge theory which is dual to the previous gravity configuration . the analysis will be done along the lines of the dijkgraaf - vafa ( dv ) correspondence , based on geometric transitions connecting open and closed superstrings . however our situation is more general and will require additional techniques . let us first quickly review the dv case , which corresponds to the flux subspace @xmath154 , @xmath155 and @xmath156 for all @xmath157 . the large n duality between open / closed topological strings was derived in @xcite . the role of the holomorphic matrix model and the relation to @xmath0 sym was considered in @xcite . on the other hand , in @xcite the dv relation was derived purely from the field theory side , using the chiral ring relations and the konishi anomaly . close to the semiclassical limit @xmath158 , @xmath159 , the geometry ( [ eq : cy2 ] ) corresponds to a product of @xmath7 independent deformed conifolds . they are cones over @xmath160 and , while the @xmath73s are collapsed to zero , the @xmath161s have finite size as measured by @xmath162 . in the geometric transition the @xmath7 3-spheres @xmath18 are collapsed and we blow - up the conifolds at @xmath21 by introducing @xmath7 @xmath163 s . then the @xmath164 fluxes @xmath165 will disappear and , instead , we will have @xmath165 d5 branes wrapping the corresponding @xmath163s . the dv correspondence states that the large @xmath166 limit of the closed string theory on the deformed threefold is equivalent to the open string theory on the resolved threefold , with the previous relation between rr fluxes and d5 branes . @xmath167 plays the role of a tree - level superpotential for the chiral superfield @xmath168 in the @xmath130 vector multiplet of a pure @xmath169 sym ; this potential breaks @xmath130 to @xmath0 . classically , the number of vacua is given by the number of ways of choosing @xmath165 eigenvalues of @xmath168 equal to @xmath14 , with @xmath170 . this breaks @xmath171 . @xmath172 is the bare gauge coupling of @xmath169 , while @xmath173 are relative changes in the @xmath174-angles of the @xmath175 factors @xcite . furthermore , the complex moduli measure gaugino condensation @xmath176 ( @xmath177 projects onto @xmath178 ) . we return now to the general flux configuration @xmath179 , @xmath180 . denote @xmath181 , @xmath182 and @xmath183 , i.e. , @xmath184 and @xmath185 with @xmath186 and @xmath187 relatively prime . consider first the effect of the geometric transition around the semiclassical regime . in the open string side we end with @xmath165 d5-branes and @xmath188 ns5-branes wrapping the i - th @xmath189 . the @xmath96 do not have a brane analogue since the @xmath22-cycles remain 3-cycles ; their meaning will become clear later . our aim is to find a gauge theory interpretation for these @xmath7 @xmath190 5-brane states . the basic requirement is that the infrared limit of this configuration shall be given by composite fields @xmath101 with an effective superpotential @xmath191 we omitted a @xmath192 factor as compared to ( [ eq : superp2 ] ) . we expect each @xmath190 5-brane to decay to @xmath193 copies of an @xmath194 bound state @xcite ; here @xmath195 , @xmath196 with @xmath197 and @xmath198 coprime . however , the generic point in flux space will give @xmath7 different types of bound states and it is hard to see how this may come from a unique uv gauge theory . instead , the straightforward way of getting a gauge theory is if on each @xmath163 we have _ the same _ type of bound state . combining this with the requirement that the sum of fluxes @xmath199 remains constant implies that we will have @xmath200 copies of the bound state of type @xmath201 distributed over all the different @xmath163s . the physical mechanism that may be responsible for this is already known , namely , eigenvalue tunnelling in matrix models . consider what happens when we tune the couplings @xmath202 from ( [ eq : w1 ] ) so that the @xmath7 cuts come very close together : @xmath203 , @xmath204 . in this limit , the process of eigenvalue tunnelling between different cuts becomes relevant ; this will result in rr flux transfer until we end with the same @xmath201 bound states in all the cuts . the tunnelling is explained by @xmath205 branes wrapped around an @xmath161 interpolating between two @xmath73s in the resolved geometry @xcite . this object is a domain wall from the eft point of view , with tension @xmath206 . after the tunnelling has taken place , we can tune back the couplings to their initial values . we will now start to argue that the previous gauge theory is indeed the dual to our gravity configuration . the key elements entering into the argument are @xmath207-duality ( decay to bound states ) and moving the @xmath18 cycles around , which is associated to an @xmath208 symmetry transformation . we work in the deformation side . denote the deformed threefold defined in ( [ eq : cy1 ] ) and ( [ eq : cy2 ] ) by @xmath209 ; the limit @xmath210 is a singular cy @xmath211 with ( generically ) conifold degenerations . recall that @xmath207-duality acts by @xmath212 transformations @xmath213 this does nt change the geometry of the hyperelliptic curve ( off - shell ) . on the other hand , the curve ( [ eq : cy2 ] ) has a symmetry group @xmath214 of matrices mixing the canonical cycles @xmath215 . these transformations are generated by all the possible interchanges of the roots @xmath216 . the generators are @xcite @xmath217 @xmath218 and @xmath22 is a symmetric matrix with integer coefficients . note that @xmath219 is invariant under @xmath214 because the loop around infinity does nt change under monodromies of the roots . the first step is to use @xmath207 duality to set the total ns flux @xmath220 and hence @xmath221 . the transformation doing this is @xmath222 for some integers @xmath223 solving @xmath224 . we denote with tildes the transformed quantities after @xmath207 duality . next we set @xmath225 , @xmath157 with @xmath208 transformations . this is done with the ` diagonal ' @xmath226 which mix the @xmath18 and @xmath227 cycles only : @xmath228 primes refer to the transformed cycles . symplectic transformations act in a complicated way on @xmath44 ; however , since we already fixed @xmath220 and @xmath229 is a symplectic invariant , we deduce that the combined application of ( [ eq : sdual2 ] ) and ( [ eq : sp2 ] ) fixes _ all _ @xmath230 , @xmath231 . summarizing , we have showed how @xmath232 may be used to set all the ns fluxes through the @xmath128 cycles to zero . the transformed axio - dilaton is @xmath233 ; the transformation of @xmath96 will be analyzed shortly . consider next the effect of the geometric transition @xcite @xmath234 where @xmath235 is the projective resolution blowing - up each conifold point in @xmath211 to a @xmath163 ; see figure [ fig : trans ] . we end with @xmath200 copies of _ the same _ 5-brane bound state @xmath201 , wrapping the @xmath7 @xmath163 s. the gauge theory is then @xmath236 where @xmath237 is the number of @xmath201 bound states on the i - th @xmath163 . this is in agreement with our previous bound state reasoning in terms of eigenvalue tunnelling . the 3-cycles @xmath20 do nt collapse in the geometric transition , so in the open string side we still have the fluxes @xmath180 . ( 0,0 ) # 1#2#3#4#5 @font ( 6150,3120)(2176,-5011 ) ( 5551,-2311)(0,0)[lb ] ( 2776,-4561)(0,0)[lb ] ( 8326,-4561)(0,0)[lb ] ( 5101,-2011)(0,0)[lb ] ( 7651,-5011)(0,0)[lb ] ( 2176,-5011)(0,0)[lb ] we do nt know how to prove the duality @xmath238 conjectured in the previous subsection . although the introduction of both rr and ns fluxes through the compact cycles is a natural extension of the dijkgraaf - vafa duality , an open topological string description of @xmath201 5-brane bound states is not available . instead , by computing the effective superpotential for both sides , we shall show that their predictions agree in the ir limit . as a further check , in section [ sec : examples ] we will prove that the gravity and gauge descriptions have the same number of degrees of freedom even in strongly coupled regimes , such as argyres - douglas singularities . @xmath239 fivebranes wrapping an @xmath73 have also been considered in the different context of @xmath240 sym @xcite . consider how the effective flux superpotential ( [ eq : superp3 ] ) transforms under the @xmath232 transformation given by ( [ eq : sdual2 ] ) and ( [ eq : sp2 ] ) : @xmath241 we made explicit the @xmath207 duality transformation in the second term to exhibit the fractional dependence on @xmath242 ; apart from this , @xmath243 are all integers . rename @xmath244 and drop all the primes : @xmath245 here @xmath246 are arbitrary integers and should nt be confused with the original parameters appearing in ( [ eq : superp3 ] ) . let us spell out the holomorphic properties of the gauge theory . six dimensional gauge theories based on @xmath239 5-branes were studied for example in @xcite . the situation here is more complicated , because the bound states are wrapping @xmath163 s , and there is @xmath180 flux through such cycles . given that we have the same bound states @xmath201 in every @xmath163 , it is enough to study a single bound state wrapping a @xmath163 and extending in four space - time dimensions . since @xmath186 and @xmath187 are relatively prime , the s - duality transformation ( [ eq : sdual2 ] ) maps the bound state to a single d5 brane . we denote with tildes the variables after the transformation . the dbi action is @xcite @xmath247 @xmath248^{1/2}\ ] ] @xmath249\,.\ ] ] @xmath250 denotes the @xmath251 gauge field on the d - brane . near the geometric transition point , where the @xmath73 shrinks , the holomorphic gauge coupling is given by @xmath252 the action for the @xmath201 bound state and the properties of its gauge theory follow from ( [ eq : dbi1 ] ) and s - duality : @xmath253 @xmath254 @xmath255 @xmath256 noting that @xmath257 the gauge coupling becomes @xmath258 where we set @xmath259 . this coincides exactly with the fractional holomorphic coupling derived from the flux side , eq . ( [ eq : superp4 ] ) . furthermore , once we map the system of @xmath239 5-branes to d5 branes , the arguments of @xcite may be applied to this n=1 sym theory to deduce that the effective superpotential has precisely the form given in ( [ eq : superp4 ] ) . generalizing to the case of @xmath7 @xmath163s , the gauge theory is @xmath260 , @xmath261 , and each @xmath262 has a holomorphic coupling @xmath263 from our previous construction , it is clear that we did nt fix all the symplectic symmetries . in particular , we can still perform monodromies @xmath264 corresponding to @xmath265 . this implies that @xmath266 is defined only modulo @xmath267 or , equivalently , @xmath268 we thus see that the information in the original brane system is not lost after the @xmath207-duality @xmath269 , but rather it is encoded in the holomorphic gauge couplings of the new theory . it is worth noting that the holomorphic couplings @xmath266 , besides being fractional , they are also independent since we can choose arbitrary integers @xmath96 . equivalently from ( [ eq : renormgeom ] ) , each @xmath262 factor has an independent physical scale @xmath97 . this situation is natural from the dbi action , but it can not arise as the ir limit of the usual @xmath130 @xmath270 sym broken to @xmath0 by the tree level superpotential @xmath271 . let us exhibit a simple generalization that may account for independent @xmath266 s. coming from string theory , we wo nt require this uv gauge theory to be renormalizable , so we look for a modified kinetic term @xmath272 if @xmath273 , the gauge group is not broken and @xmath274 should give a unique gauge coupling . on the other hand , when we turn on the superpotential , the basic property of @xmath275 is that it should be equal to @xmath266 on the subspace @xmath178 . the matrix function that does this is simply constructed from the idempotents of the classical chiral ring : @xmath276 which satisfy @xmath277 . then we may define @xmath278 the nonrenormalizable gauge theory ( [ eq : modgauge ] ) with this choice of @xmath275 gives independent gauge couplings in the infrared . another striking property of this brane system is the appearance of noncommutative dipoles in the uv . this is due to the ns fluxes through the @xmath163 s. such dipole deformations of the gauge theory have been recently considered in @xcite for geometric transitions based on d5 branes . it would be interesting to try to extend this analysis to the case of @xmath279 5-branes , although the supergravity description might be much more involved . to summarize , using @xmath232 in this section we mapped a general flux configuration to a gauge theory , after the geometric transition . all the flux parameters have a natural gauge interpretation ; in particular the fluxes @xmath280 through the 3-cycles , which do nt collapse after the transition , do nt contribute brane degrees of freedom . they combine in a nontrivial way to determine the holomorphic gauge couplings of the different gauge factors . the purpose of constructing a dual gauge theory to count flux vacua is that in such field theories the number of vacua is always finite . the geometric transition preserves this number . in the present section we show from the gauge theory side that @xmath281 is indeed finite . we begin by showing that the number of supersymmetric gauge vacua , i.e. , solutions to @xmath282 from equation ( [ eq : superp4 ] ) , is finite . as discussed before , this is based on the tadpole constraint @xmath283 here @xmath284 ; also recall that @xmath285 , @xmath286 , @xmath287 . we have to sum over all choices of fluxes satisfying ( [ eq : gaugetad ] ) . here we run into the main obstacle . the reason why this could in principle diverge is that there may be flux configurations such that two terms in @xmath137 grow in a correlated way to plus and minus infinity respectively , but keeping @xmath137 finite and positive . this would give an infinite number of allowed flux points ( and hence supersymmetric vacua ) . this is the point where having a gauge theory based on the geometry ( [ eq : cy2 ] ) proves useful . in the gauge theory , @xmath94 is holomorphic in the couplings @xmath202 , so the number of solutions to the equations @xmath288 is invariant under smooth changes of the parameters , being protected by holomorphy . do nt depend on @xmath202 , it is more convenient to take derivatives w.r.t . @xmath289 and not @xmath101 . ] an equivalent statement is that the number of vacua coincides with the dimension of the chiral ring of the theory , and such a quantity is independent of the gauge couplings . this topological behavior was already encountered in the gravity side , when we showed ( section [ subsec : susytadpole ] ) that the number of supersymmetric vacua coincides with the supersymmetric index . we now argue , from a variation of the @xmath202 , that each term in @xmath137 is in fact positive even around singularities . the discriminant locus consists of generic conifold points and higher codimension ad singularities . the later can not be neglected because they have a higher ` weight ' in the counting of degrees of freedom , as measured by @xmath290 . both situations will be exemplified in section [ sec : examples ] . consider a point in moduli space @xmath100 corresponding to the semiclassical limit . this is just the origin @xmath291 of @xmath100 . in this case the geometry is a product of independent conifold - like configurations . the effective superpotential follows from ( [ eq : superp4 ] ) using monodromy arguments @xcite : @xmath292 denoting @xmath293 and @xmath294 , the supersymmetric vacua may be written as @xmath295 then counting vacua in the neighborhood of the conifold limit implies summing over fluxes giving @xmath296 . is some final energy scale associated to @xmath262 . ] clearly this requires @xmath297 , to avoid vacua exponentially far away from the origin . we therefore see that the number of vacua around the semiclassical point is finite because each term in @xmath137 is _ separately _ positive . without loss of generality , we can just take all the fluxes to be positive . the holomorphic dependence of @xmath94 on @xmath202 implies that this is true for the whole moduli space . indeed , every point in moduli space can be connected to the semiclassical limit by such a variation of couplings . of course , strongly coupled limits may have quite complicated superpotentials , but we are interested in the number of vacua , which is a topological invariant . for concreteness , we show this for @xmath298 . the hyperelliptic curve is @xmath299 we only need to worry about singularities in @xmath100 since it is known that @xmath281 is finite around smooth points . there are two types ; the codimension one singularities are conifolds , and correspond to the semiclassical regime where we showed the finiteness of @xmath281 . there is also a codimension two @xmath300 singularity . it corresponds to the singular limit of @xmath28 : @xmath301 three roots coincide at @xmath302 giving two vanishing intersecting cycles , while the last one is fixed at @xmath303 . comparing to ( [ eq : curve1 ] ) , we find the ` double scaling ' limit @xmath304 and , for the couplings , @xmath305 to connect this to the semiclassical point , vary the couplings @xmath306 from their previous double - scaled values to @xmath307 , while keeping the @xmath289 fixed at ( [ eq : doublelimit1 ] ) . clearly , at the new point in @xmath100 the semiclassical approximation is valid . this process is depicted in figure [ fig : changeg ] . ( 0,0 ) # 1#2#3#4#5 @font ( 12012,7449)(676,-7423 ) ( 4400,-2561)(0,0)[lb ] ( 4201,-5386)(0,0)[lb ] ( 11476,-5311)(0,0)[lb ] ( 2476,-1411)(0,0)[lb ] ( 6601,-5236)(0,0)[lb ] ( 6850,-360)(0,0)[lb ] ( 11601,-4200)(0,0)[lb ] ( 676,-2311)(0,0)[lb ] ( 6676,-2311)(0,0)[lb ] ( 4876,-6361)(0,0)[lb ] ( 9451,-6361)(0,0)[lb ] therefore we have shown that any point in @xmath100 can be connected to the conifold limit by a smooth variation of the @xmath202 . in other words , the gauge theory tells us how to do , on every point in moduli space , a change of variables @xmath308 such that : ( i ) each term in @xmath137 is explicitly positive and ( ii ) the number of supersymmetric vacua does nt change . furthermore , since we can work in a regime @xmath309 by tuning @xmath310 , we can always do power - series expansions and hence the change of variables is continuous . this maps compact regions to compact regions , assuring that the number of vacua does nt diverge . the meaning of this transformation becomes transparent if we consider the chiral ring . it is generated by idempotents and nilpotents @xcite . if we move around the moduli space @xmath101 by changing the couplings until we encounter a singularity , the result on the chiral ring is that some idempotents become nilpotents . the total number of generators is conserved in the process . in order to compare with the gravity side result ( [ eq : ivacua ] ) , we next compute the number of supersymmetric gauge vacua around an arbitrary point in @xmath100 . as argued before , holomorphy implies that we can as well compute it around the semiclassical limit . because of the monodromies leading to ( [ eq : valuesbeta ] ) , at fixed @xmath267 , the number of vacua is @xmath312 the @xmath267 satisfy @xmath313 . this is quite different to the result from a standard @xmath0 sym , @xmath314 . ( [ eq : ivacua ] ) includes an integration over a region in moduli space . we need to specify the analogous condition in the gauge side . it is associated to the rg flow of the gauge theory from the cutoff @xmath81 up to some ir energy scale @xmath315 . for concreteness , we compute @xmath281 for the simplest case , namely when each @xmath262 flows up to a scale @xmath316 . in other words , we assume that we are integrating on disks @xmath317 . the renormalization of gauge couplings ( [ eq : renormgeom ] ) applied to the case ( [ eq : superp4 ] ) gives @xmath318 here we set , for simplicity , @xmath319 , @xmath320 . this is possible because in the noncompact model the axio - dilaton is fixed and behaves as a coupling ; therefore @xmath281 can not depend on it . since we are summing the degrees of freedom with @xmath321 , ( [ eq : renormgeom2 ] ) implies @xmath322 replacing in the gauge tadpole condition ( [ eq : gaugetad ] ) , @xmath323 once we fix arbitrary @xmath324 , the dual fluxes @xmath325 are integers satisfying the diophantine equation ( [ eq : gaugetad ] ) . this has solutions iff @xmath326 ; the number of integer solutions is of course infinite , but we argued that @xmath327 . so we take the fluxes to be positive , and multiply the number of vacua by @xmath328 . the number of positive solutions to the tadpole constraint will be denoted by @xmath329 . for large @xmath137 , this number is typically of order 1 . combining all the previous elements , the total number of supersymmetric vacua is @xmath330\ ; \times\ ] ] @xmath331 the notation here is the following . the sum on @xmath201 is over coprime integers . the sum on @xmath324 should be done over inequivalent fluxes with respect to the residual symplectic transformations ; indeed , some generators in ( [ eq : gens ] ) were not fixed by the mapping to the region @xmath332 . also , recall that @xmath329 is the number of positive solutions to the diophantine equation ( [ eq : gaugetad ] ) ; for large @xmath136 , it will give subleading contributions so , to a good approximation , we may set @xmath333 . lastly , @xmath334 specifies the region in flux space over which we are summing vacua . for instance , if we integrate on disks of radius @xmath335 , ( [ eq : gaugetad2 ] ) gives the heaviside function @xmath336 in this section we compare the formulas ( [ eq : ivacua ] ) and ( [ eq : gaugenfinal ] ) for @xmath281 in the gravity and gauge side , respectively . this is done for the conifold and argyres - douglas degenerations . _ gravity side _ we start by considering the case of a single deformed conifold in the closed string side . the total number of vacua for the conifold has been computed in @xcite in the context of f - theory compactifications . here we quickly summarize the result for fixed axio - dilaton . there is only one compact cycle ( a ) , and a dual noncompact one ( b ) . from monodromy arguments , @xmath337 @xmath338 is the complex modulus ( here we do nt use @xmath207 to make clear the distinction between the gravity and gauge side ) and @xmath339 is a constant added for dimensional reasons . it depends on the cutoff necessary to regulate the b - integral . further , the dots refer to analytic terms in @xmath338 . replacing in ( [ eq : kahler ] ) and then in ( [ eq : curv ] ) , @xmath340 for @xmath341 , @xmath342 and hence @xmath343 , in agreement with the deduced result ( [ eq : ivacua ] ) . integrating on @xmath344 , the number of supersymmetric vacua for fixed axio - dilation is @xmath345 the superindex @xmath153 reminds us that this is the result from the closed string side . _ gauge theory side _ next we calculate in detail the result from ( [ eq : gaugenfinal ] ) . from the gauge theory side , the conifold corresponds to the semiclassical limit of the superpotential with @xmath346 : @xmath347 and from ( [ eq : fn ] ) , @xmath348 , where we are setting @xmath349 . there are @xmath350 vacua satisfying @xmath351 and we have to compute the number of vacua with @xmath352 for some final energy scale @xmath353 . from the running of the gauge coupling , @xmath354 the meaning of this formula is that the gauge theory analogue of integrating a given modulus on a disk is the rg flow of the gauge coupling from the uv cutoff up to a final energy scale given by the radius of the disk . the number of vacua for given l is then given by @xmath355 we multiply by 2 since we are considering only @xmath356 . the superindex @xmath357 refers to the open string side . the gravity result @xmath358 arises in the continuum limit @xmath359 . therefore we need to estimate the asymptotic behavior of @xmath360 . we did this with a c++ program that adds coprime numbers ( modulo permutations ) inside a disk of radius @xmath361 and then sums over all the divisors of @xmath137 , according to ( [ eq : nconif2 ] ) . fitting the numerical predictions of @xmath362 for @xmath363 , we deduce an asymptotic dependence @xmath364 . to leading order we find @xmath365 the numerical results and the fit are shown in figure [ fig : conif ] . we do nt completely understand the subleading corrections to the gravity result . even though we fit the numerical formula with @xmath366 , the power of @xmath136 could be smaller . ( 0,200 ) for the conifold , showing both the gravity and gauge side predictions , which agree almost exactly . we chose a scale @xmath367 to simplify the results.,title="fig : " ] let us compare ( [ eq : nconif1 ] ) and ( [ eq : nconif3 ] ) ; we naturally identify @xmath368 and @xmath369 and both results match very well . the power @xmath370 is a good approximation to the gravity result @xmath371 . it turns out to be related to properties of the divisor functions @xmath372 . the agreement is nontrivial , involving very different concepts in the gauge and gravity side . the crucial ingredients from the gauge side are the running of the gauge coupling and the correct tadpole condition . in other words , the gravity side with general fluxes has the same number of degrees of freedom as the sym theory described in section [ sec : dualym ] . next we analyze some aspects of two - parameter models which arise from @xmath298 superpotentials : @xmath373 the novel phenomenon for @xmath374 is the appearance of argyres - douglas points , when three or more roots coincide ; see ( [ eq : curvsing ] ) . when intersecting cycles vanish simultaneously nonlocal dyons become massless . the physics is radically different to that of the conifold , giving rise to an interacting scft @xcite . unfortunately , the complications of the model forbid a straightforward analysis similar to the one done in the previous subsection . from the gravity side , the discriminant locus is a knot - like complex curve @xcite with self - intersections ; integrating over all the moduli space to get the total number of vacua is hence quite involved . on the other hand , in the gauge theory , the combinatorics present in formula ( [ eq : gaugenfinal ] ) are equally complicated . therefore we will only study the vicinity of the ad point and we will show that the number of vacua obtained from @xmath375 has the expected gauge theory scaling behavior . _ _ gravity side _ _ the dynamics around the ad point is controlled by the small complex curve @xmath376 the discriminant locus is @xmath377 which is not smooth ; indeed @xmath378 has solution @xmath379 . this is the argyres - douglas point @xcite . usual monodromy arguments used to construct the periods can not be applied now , since the self - intersection is not normal . therefore we need to blow - up ( [ eq : discr3 ] ) . the general procedure is described in @xcite and has been recently applied to our present situation in @xcite . the normal - crossing variables close to the ad point turn out to be @xmath380 the original discriminant locus corresponds to @xmath381 ; @xmath382 is the scaling variable in the scft . by rescaling @xmath383 , @xmath384 the dependence on @xmath382 disappears ; the dependence on @xmath385 follows from the usual monodromy @xmath386 . we do a symplectic transformation so that the small periods are @xmath387 and the large ones are @xmath388 . the dependence on @xmath385 and @xmath382 is @xmath389 the large periods are analytic in @xmath385 and @xmath382 . replacing these expressions in ( [ eq : metric ] ) and ( [ eq : curv ] ) , the density of vacua ( [ eq : ivacua ] ) around the ad point is @xmath390 we see that the density of vacua is integrable on a disk around @xmath391 . in particular , integrating on @xmath392 gives a total number of vacua @xmath393 this proves that the number of vacua around the ad singularity is finite . the constant @xmath394 depends on analytic data from the long cycles ; in general it can not be computed using monodromy arguments . the result ( [ eq : iad ] ) is of the general form encountered in the analysis of different singularities in @xcite @xmath395 where @xmath396 denotes de discriminant locus ( in normal crossing variables ) . we will now justify this behavior from the field theory point of view . _ gauge side _ this example is quite interesting , since we have to use the map connecting the strongly coupled ad point to the semiclassical regime . the procedure was described in section [ sec : finite ] . we vary @xmath397 from ( [ eq : doublelimit2 ] ) to @xmath398 , while keeping @xmath289 fixed at ( [ eq : doublelimit1 ] ) . the condition that we end in the semiclassical regime is @xmath399 . furthermore , we can set @xmath400 by choosing a perturbation with @xmath401 . indeed , we only want to reproduce the divergence @xmath402 associated to the ` physical ' discriminant component @xmath381 . expanding for @xmath403 large , the expression for @xmath101 in terms of @xmath385 is @xmath404 also , @xmath405 . therefore , to leading order in @xmath403 , @xmath406 and they depend linearly on @xmath385 ; up to a shift by a constant , the integral @xmath407 is hence translated to @xmath408 . in this case , the gauge vacua formula ( [ eq : gaugenfinal ] ) reads @xmath409 @xmath410 a numerical evaluation shows that ( [ eq : nvacad ] ) has the same dependence as ( [ eq : iad2 ] ) : @xmath411 for @xmath412 . subleading corrections should be taken into account , but their general dependence is hard to estimate . this is a nontrivial check for the argument that we can map any complicated singularity to the conifold regime and equivalently count vacua there . moreover , @xmath298 is the smallest genus for which the symplectic transformations @xmath214 come into play to count gauge vacua . in this paper we have shown that the number of supersymmetric vacua @xmath281 around ade singularities of calabi - yau s in type iib flux compactifications is finite . the argument is based on the existence of dual gauge theories , where finiteness may be shown . such singularities can be embedded in the noncompact cy ( [ eq : cy1 ] ) and it is crucial that some of the fields become nondynamical ( couplings ) . the moduli are stabilized by turning on both rr and ns flux through the compact cycles . we then perform a geometric transition to connect this to the open string side . the gauge theory is based on 5-brane bound states wrapping the resolved 2-cycles . its main properties are obtained by applying s - duality to the dbi action on the resolved background . in particular , the theory has fractional gauge couplings @xmath266 ; the couplings are independent and hence can not come from a uv theory which is the usual @xmath0 sym with superpotential @xmath271 . more importantly , the effective superpotential of the field theory depends holomorphically on the couplings @xmath413 . the dimension of the chiral ring ( @xmath281 ) is thus invariant under changes @xmath414 . we used this property to map a generic point in field space @xmath415 to the conifold limit , while preserving the number of vacua . in this semiclassical limit we showed that @xmath281 is finite . finally , we computed explicitly this number for the two simplest singularities , namely the conifold point and the @xmath298 argyres - douglas point . the results from the gravity side and gauge side match . this agreement is nontrivial since it involves quite different concepts on both sides . let us compare both formulas . the gravity formula @xmath99 relates supersymmetric vacua to the geometry of the moduli space . a simple topological interpretation @xcite is that it gives the number of zeroes of the section @xmath416 . clearly , it is explicitly invariant under symplectic transformations . however , the analysis of singularities is not straightforward , in particular because the blow - up procedure becomes very involved as we analyze higher codimension singularities . on the other hand , the physics underlying the gauge theory result is that of fractional instantons , bound states of 5-branes , rg flow of the gauge couplings and combinatorics between the matrix model cuts . the formula is explicitly finite after the mapping to the semiclassical region . as a result , we recognize the exponent @xmath417 in ( [ eq : eguchi ] ) as the degree of the tree - level gauge superpotential @xmath271 in which the singularity may be minimally embedded . we should nevertheless point out that for @xmath418 there remain symplectic generators that have to be fixed by further restricting the fluxes to a fundamental region and this is in general complicated . another issue is that the combinatorics grows very rapidly with @xmath7 and numerical computations become more difficult . a technical point that could be addressed in the future is to understand better the origin of subleading corrections to the gravity formula . these appear because the flux space is in fact a lattice . the gauge theory approach might help in this direction . we should note that the present results are based on the duality between the closed ( deformed ) and open ( resolved ) sides . we havent been able to fully prove this , although we did show that both sectors have the same ir physics . it would be very interesting to continue this , perhaps with a supergravity analysis along the lines of @xcite . if a lift to m - theory is possible , the geometric transition might reduce to a duality between m5 branes , as in the dijkgraaf - vafa context . first i would like to thank my advisor m. r. douglas for suggesting this problem and for his support and guidance throughout all the stages of the project . from the beginning i have greatly benefited from extensive explanations and discussions with f. denef , d .- diaconescu and b. florea . it is also a pleasure to thank g. aldazabal , d. belov , c. d. fosco , s. franco , j. juknevich , s. klevtsov , s. lukic , a. nacif , s. ramanujam , k. van den broek and a. uranga for many interesting discussions and suggestions . this research is supported by rutgers department of physics . 99 m. r. douglas and s. kachru , _ flux compactification _ , hep - th/0610102 . m. r. douglas , _ the statistics of string / m theory vacua _ , jhep 0305:046 , 2003 , hep - th/0303194 . m. r. douglas and b. acharya , _ a finite landscape ? _ , hep - th/0606212 . m. r. douglas and zhiqin lu , _ finiteness of volume of moduli spaces _ , hep - th/0509224 . m. r. douglas and zhiqin lu , _ on the geometry of moduli space of polarized calabi - yau manifolds _ , rims kokyuroku 1487 , 2006 , math.dg/0603414 . s. ashok and m.r . douglas , _ counting flux vacua _ , jhep 0401:060 , 2004 , hep - th/0307049 . r. dijkgraaf and c. vafa , _ a perturbative window into nonperturbative physics _ , hep - th/0208048 . a. klemm , w. lerche , p. mayr , c. vafa and n. warner , _ selfdual strings and n=2 supersymmetric field theory _ , nucl.phys.b477:746-766 , 1996 , hep - th/9604034 . a. lawrence and j. mcgreevy , _ local string models of soft supersymmetry breaking _ , jhep 0406:007,2004 , hep - th/0401034 . f. cachazo , k. intriligator and c. vafa , _ a large n duality via a geometric transition _ , nucl.phys.b603:3-41 , 2001 , hep - th/0103067 . h. farkas and i. kra , _ riemann surfaces _ , springer verlag , new york , 1980 . f. cachazo , n. seiberg and e. witten , _ phases of n=1 supersymmetric gauge theories and matrices _ , jhep 0302:042 , 2003 , hep - th/0301006 . s. gukov , c. vafa and e. witten , _ cft s from calabi - yau four folds _ , nucl.phys.b584:69-108 , 2000 , erratum - ibid.b608:477 - 478 , 2001 , hep - th/9906070 . f. cachazo , n. seiberg and e. witten , _ chiral rings and phases of supersymmetric gauge theories _ , jhep 0304:018 , 2003 , hep - th/0303207 . p. argyres and m. r. douglas , _ new phenomena in su(3 ) supersymmetric gauge theory _ , nucl.phys.b448 : 93 - 126 , 1995 , hep - th/9505062 . a. klemm , w. lerche and s. theisen , _ nonperturbative effective actions of n=2 supersymmetric gauge theories _ , int.j.mod.phys.a11:1929-1974 , 1996 , hep - th/9505150 . s. giddings , s. kachru and j. polchinski , _ hierarchies from fluxes in string compactifications _ , phys.rev.d66:106006 , 2002 , hep - th/0105097 . i. klebanov and m. strassler , _ supergravity and a confining gauge theory : duality cascades and chi sb resolution of naked singularities _ , jhep 0008:052 , 2000 , hep - th/0007191 c. vafa , _ superstrings and topological strings at large n _ , j.math.phys.42:2798-2817 , 2001 , hep - th/0008142 . r. gopakumar and c. vafa , _ on the gauge theory / geometry correspondence _ , adv.theor.math.phys.3:1415-1443 , 1999 , hep - th/9811131 . r. dijkgraaf and c. vafa , _ matrix models , topological strings , and supersymmetric gauge theories _ , nucl.phys.b644:3-20 , 2002 , hep - th/0206255 . f. cachazo , m. r. douglas , n. seiberg and e. witten , _ chiral rings and anomalies in supersymmetric gauge theory _ , jhep 0212:071 , 2002 , hep - th/0211170 . j. polchinski , _ superstring theory _ , volume 2 , cambridge university press , 1998 . d. mumford , _ tata lectures on theta i _ , progress in mathematics , birkhauser , boston , 1982 . diaconescu , r. donagi , r. dijkgraaf , c. hofman and t. pantev , _ geometric transitions and integrable systems _ , nucl.phys.b752:329-390,2006 , hep - th/0506196 . diaconescu , r. donagi and t. pantev , _ geometric transitions and mixed hodge structures _ , hep - th/0506195 . j. polchinski and m. strassler , _ the string dual of a confining four dimensional gauge theory _ , hep - th/0003136 . e. witten , _ new gauge theories in six - dimensions _ , jhep 9801:001 , 1998 , adv.theor.math.phys.2:61-90 , 1998 , hep - th/9710065 . k. dasgupta , m. grisaru , r. gwyn , s. katz , a. knauf and r. tatar , _ gauge - gravity dualities , dipoles and new non - kahler manifolds _ , hep - th/0605201 . p. svrcek , _ chiral rings , vacua and gaugino condensation of supersymmetric gauge theories _ , jhep 0408:036,2004 , hep - th/0308037 . f. denef and m. r. douglas , _ distributions of flux vacua _ , jhep 0405:072 , 2004 , hep - th/0404116 . v. arnold , a. gusein - zade and a. varchenko , _ singularities of differentiable maps , i , ii _ , birkhauser , 1985 . t. eguchi and y. tachikawa , _ distribution of flux vacua around singular points in calabi - yau moduli space _ , jhep 0601:100 , 2006 , hep - th/0510061 . k. dasgupta , k. oh and r. tatar , _ geometric transition , large n dualities and mqcd dynamics _ , nucl.phys.b610:331-346 , 2001 , hep - th/0105066 .

we consider the general conditions of quark droplets formation in high density neutron matter . the growth of the quark bubble ( assumed to contain a sufficiently large number of particles ) can be described by means of a fokker - planck equation . the dynamics of the nucleation essentially depends on the physical properties of the medium it takes place . the conditions for quark bubble formation are analyzed within the frameworks of both dissipative and non - dissipative ( with zero bulk and shear viscosity coefficients ) approaches . the conversion time of the neutron star to a quark star is obtained as a function of the equation of state of the neutron matter and of the microscopic parameters of the quark nuclei . as an application of the obtained formalism we analyze the first order phase transition from neutron matter to quark matter in rapidly rotating neutron stars cores , triggered by the gravitational energy released during the spinning down of the neutron star . the endothermic conversion process , via gravitational energy absorption , could take place , in a very short time interval , of the order of few tens seconds , in a class of dense compact objects , with very high magnetic fields , called magnetars .

astro-ph0403200

since @xcite , following early proposals by @xcite and @xcite , suggested that strange quark matter , consisting of @xmath0- , @xmath1- and @xmath2-quarks is energetically the most favorable state of the matter , the problem of existence of strange quark stars has been intensively investigated in the physical and astrophysical literature . the possibility that some compact objects could be strange stars remains an interesting and intriguing , but still open question . @xcite also proposed two ways of formation of strange matter : the quark - hadron phase transition in the early universe and conversion of neutron stars into strange ones at ultrahigh densities . in the theories of strong interaction quark bag models suppose that breaking of physical vacuum takes place inside hadrons . as a result vacuum energy densities inside and outside a hadron become essentially different , and the vacuum pressure on the bag wall equilibrates the pressure of quarks , thus stabilizing the system . if the hypothesis of the quark matter is true , then some of neutrons stars could actually be strange stars , built entirely of strange matter @xcite ) . however , there are general arguments against the existence of strange stars @xcite . for an extensive review of strange star properties see @xcite . several mechanisms have been proposed for the formation of quark stars . quark stars are expected to form during the collapse of the core of a massive star after the supernova explosion , as a result of a first or second order phase transition , resulting in deconfined quark matter @xcite . the proto - neutron star core , or the neutron star core , is a favorable environment for the conversion of ordinary matter to strange quark matter @xcite . another possibility is that some neutron stars in low - mass x - ray binaries can accrete sufficient mass to undergo a phase transition to become strange stars @xcite . this mechanism has also been proposed as a source of radiation emission for cosmological @xmath3-ray bursts @xcite . some basic properties of strange stars like mass , radius , cooling , collapse and surface radiation have been also studied @xcite . the physical mechanisms of the transition from neutron matter to quark matter in an astrophysical background have been studied within several models . the first is due to @xcite , who used a non - relativistic diffusion model . as such , this is a slow combustion model , with the burning front propagating at a speed of approximately @xmath4 m/@xmath5 . this is determined primarily by the rate at which one of the down quarks inside the neutrons is converted , through a weak decay , to a strange quark : @xmath6 . the second method of describing the conversion process was first suggested by @xcite , and analyzed in detail by @xcite and @xcite , who modelled the conversion as a detonation . in this case the conversion rate is several orders of magnitude faster than that predicted by the slow combustion model . this model is based on the relativistic shock waves and combustion theory . but regardless of the way in which the transformation occurs , an initial seed of quark matter is needed to start the process . real neutron stars have two conserved charges- electric and baryonic . therefore , a neutron star has more than one independent component and in this sense it is a complex " system @xcite . the characteristics of a first - order phase transition , like the deconfinement transition , are very different in the two cases . in the complex " system the conserved charges can be shared by the two phases in equilibrium in different concentrations in each phase . the mixed phase , formed from hadrons and quarks can not exist in a simple body in the presence of the gravity because the pressure in that phase is a constant @xcite . this causes a discontinuity in the density distribution in the star occurring at the radius where gibbs criteria are satisfied . the isospin symmetry energy in neutron rich matter will exploit the degree of freedom of readjusting the charges between hadronic and quark phases in equilibrium so as to reduce the symmetry energy to the extent consistent with charge conservation . regions of hadronic matter will have a net positive charge neutralized by a net negative charge on the quark matter regions @xcite . coulomb repulsion will prevent the regions of like charge to grow too large , and the surface energy will act in the opposite sense in preventing them to become too small . consequently , the mixed phase will form a coulomb lattice so as to minimize the sum of coulomb and surface interface energy at each proportion of phases . as the quark phase becomes more abundant , the droplets merge to form rods , rods merge to form slabs etc . @xcite . hence the actual geometric phase of quark nuggets in a neutron star evolves as a function of the pressure and of the surface energy between dense nuclear matter and quark matter . in order to describe the process of formation and evolution of quark nuclei in the neutron stars one must use nucleation theory . the goal of nucleation theory is to compute the probability that a bubble or droplet of the @xmath7 phase appears in a system initially in the @xmath8 phase near the critical temperature @xcite . homogeneous nucleation theory applies when the system is pure . nucleation theory is applicable for first - order phase transitions when the matter is not dramatically supercooled or superheated . if substantial supercooling or superheating is present , or if the phase transition is second order , then the relevant dynamics is spinodal decomposition @xcite . the nucleation of strange quark matter inside hot , dense nuclear matter was investigated , with the use of zeldovich s kinetic theory of nucleation , by @xcite . by assuming that the newly formed strange quark matter bubble can be described by a simple bag model containing @xmath9 ultrarelativistic quarks , and by assuming that the time evolution of the radius of the bubble is described by an equation of the form @xmath10 , where @xmath11 is the critical radius of the bubble and @xmath12 the weak interaction time scale , the nucleation rate @xmath13 is given by @xmath14 where @xmath15 is the critical quark number , @xmath16 is the surface tension of the bubble and @xmath17 and @xmath18 are the particle number densities in the neutron and quark phase , respectively @xcite . in this way a lower bound for the temperature of the nucleation , @xmath19 , can be obtained . the effects of the curvature energy term on thermal strange quark matter nucleation have been considered by @xcite , who derived , within the same approach , the following expression for the nucleation rate : @xmath20 all the effects of the curvature enter only through the value of @xmath11 . as a general conclusion one obtains the result that even the curvature term acts against strange quark matter nucleation , the physical temperature of a just born proto - neutron star is , in any model , more than enough to drive an efficient boiling of the neutron material ( the observations of the neutrino flux from sn1987a are consistent with an effective temperature of @xmath21 mev ) . the possibility of stable strange quark matter ( both bulk and quasi - bulk ) at finite temperature , and some of its properties have been investigated , within the framework of the dynamical density - dependent quark mass model of confinement , by @xcite . the behavior of the surface tension and the stability of quark droplets at @xmath22 has also been discussed , with reference to strangelet formation in ultra - relativistic heavy - ion collisions . a different approach to the problem of thermal nucleation was taken by @xcite . they used the assumption ( standard in the theory of bubble nucleation in first - order phase transitions ) that bubbles form at a rate given by @xmath23 , with @xmath24 the minimum work required to form a bubble with radius @xmath25 . as a main result it follows that if the bag constant lies in the interval where three - flavor but not two - flavor quark matter is stable at zero pressure and temperature ( @xmath26 mev@xmath27 mev ) , then all or parts of a neutron star will be converted into strange matter , during the first second of its existence . for bag constant above the stability interval , a partial transformation is still possible @xcite . @xcite calculated the rate of formation of quark matter droplets in neutron stars from a combination of bubble formation rates in cold degenerate and high temperature matter , taking into account nuclear matter calculations of the viscosity and thermal conductivity . the droplet formation rate is that of @xcite , given by @xmath28 , with @xmath29 the statistical prefactor , which measures the phase - space volume of the saddle point around @xmath11 that the droplet has to pass , in its way to the lower energy state . @xmath30 is the quark correlation length . the dynamical prefactor determines the droplet growth rate and is given by @xmath31 , where @xmath32 is the enthalpy difference and @xmath33 and @xmath34 are the thermal conductivity and the shear and bulk viscosities , respectively . the droplet formation rate can be expressed as @xcite @xmath35 with @xmath36 is the quark chemical potential in units of @xmath37 mev . the total number @xmath38 of droplets formed by nucleation is the integrated rate over the volume of the neutron star and the time after the nucleation process starts at the moment @xmath39 , @xmath40dt$ ] , where @xmath41 is the radius of the neutron star . the pressure and temperature depend sensitively on the equation of state of the nuclear and quark matter . to convert the core of the neutron star into quark matter at least one droplet must be formed , i.e. , @xmath42 . then , with the use of eq . ( [ eqn ] ) , the condition for the formation of at least one droplet requires @xmath43 mev @xmath44 , where @xmath45 and @xmath46 are the core values of the pressure and temperature , respectively @xcite . the pressure and enthalpy difference depends strongly on the equation of state of nuclear and quark matter and hence the droplet formation rate can not be reliably estimated . the effect of subcritical hadron bubbles on an inhomogeneous first order quark - hadron phase transition was studied in @xcite . the transition from nuclear matter ( consisting of neutrons , protons and electrons ) to matter containing strangeness was considered , within a mean field type model and with the use of langer nucleation theory , by @xcite . an estimate of the time the new phase to appear at various densities and times in the cooling history of a protoneutron star was also given . on the other hand , as a result of an increase of the central density of the star , due for example to accretion or spinning down , a metastable , super - compressed neutron phase can appear , with a star consisting of the new phase surrounded by normal neutron matter @xcite . this situation is similar to a gas undergoing a phase transition to its liquid state if compressed to a volume @xmath47 at fixed @xmath48 . if it is slowly compressed , it may stay in the vapor state even for @xmath49 ( similar examples are liquid water freed of dissolved air which may be heated above boiling temperature without the formation of vapor and cooled below freezing temperature without solidification ) . the transition from the super - compressed neutron phase to the quark phase can take place in a very long time , which in some cases can be longer than the estimated life of the neutron star @xcite . therefore an increase in the density of the neutron star does not automatically lead to the transition to quark matter inside the star . the effect of the magnetic field on the quark star structure and on the nucleation process of the quark bubbles has been studied in @xcite . in the presence of strong magnetic fields the equation of state of strange quark matter changes significantly . the strange stars become more compact , the magnetic field reducing the mass and radius of the star . the surface energy and the curvature term of the quark phase both diverge , with the surface tension diverging logarithmically , while the curvature term diverges much faster . therefore the thermal nucleation of quark bubbles in a compact metastable state of neutron matter is completely forbidden in the presence of a strong magnetic field . these results for the formation of quark bubbles in neutron matter have been obtained by using the csernai and kapusta theory of nucleation @xcite and its extension including the thermal conductivity of dense matter @xcite . in this theory both the hadron and quark materials are considered as substances with zero baryonic number , and are treated by using a relativistic formalism . the principal result in this approach is the suggestion that the prefactor @xmath50 is proportional to the transport coefficients ( viscosity and thermal conductivity ) of the neutron matter , and thus when these coefficients vanish a quark bubble necessarily does not form . however , @xcite argued that the energy flow does not vanish in absence of any heat conduction or viscous damping . since the change of energy density @xmath51 in time is given , in the low velocity limit , by the conservation equation @xmath52 @xcite , where @xmath53 is the enthalpy and @xmath54 the velocity , this implies that the energy flow @xmath55 is always present . therefore , an expression for the prefactor can be derived , which does not vanish in the absence of viscosity . the viscous effects cause only a small perturbation to the prefactor . the differences between the csernai - kapusta ( ck ) @xcite and ruggieri - friedman ( rf ) @xcite results are due to the technical differences in the treatment of the pressure gradients . a generalized approach , following the ck formalism leading to the prefactor in both viscous and non - viscous regimes was developed by @xcite . unlike in @xcite , the linearized relativistic hydrodynamics equations have been solved in all regions . it is the purpose of the present paper to extend the zeldovich nucleation theory for the formation of quark droplets in neutron matter , by taking into account both the effects of the energy flow and of thermal conductivity , shear and bulk viscosities . as a result a more general expression for the quark matter droplet rate formation can be obtained , which is also valid in the limiting case of vanishing conductivity and viscosity coefficients . in the nucleation theory of @xcite , the transition from the neutron to quark state is possible only for a normal matter having viscous properties . for zero bulk and shear viscosities the transition is impossible . this assumption is , however , to restrictive , and the transition can also take place for a perfect ( no viscosity or heat conduction ) neutron matter fluid . we also derive the expression of the rate formation for this case . an alternative point of view of the transformation of neutron matter into strange matter can be developed if one assumes that the conversion process is endothermic . in this case the strange quark matter formation inside a neutron star is triggered by an external source of energy ( accretion from the companion star or spin - down ) . pulsars are born with an enormous angular momentum and rotational energy , which they radiate over a long period of time via electromagnetic radiation and electron - positron pair emission . when rotating rapidly , a pulsar is centrifugally flattened . with decreasing angular velocity the central density of the star is increasing and may attain the critical density necessary for a phase transition . first at the center and then in an expanding region the neutron matter will be converted to a highly compressible quark matter phase @xcite . the conversion of neutron matter to quark matter alters the moment of inertia of the star , and the epoch over which conversion takes place will be signaled in the spin - down characteristics of the pulsar . a measurable quantity , the braking index @xmath56 , where @xmath57 is the angular velocity of the star , could be an observational indicator of the slow transition from neutron to quark phase @xcite . by using the formalism developed for the nucleation of quark matter we reconsider the idea of phase transition from neutron matter to quark matter in rotating compact stellar objects , by assuming that the nucleation and the growth of the quark droplets and the formation of a quark core in a neutron star is a result of the transfer of the gravitational energy to the quark droplets , due to the spinning down of the rapidly rotating neutron star . the present paper is organized as follows . the nucleation kinetics of strange matter bubbles is considered , within the framework of the zeldovich nucleation theory , in section ii . the expressions for the rate of formation of quark matter are derived , in both the perfect and dissipative neutron fluid case , in section iii . the transition of a neutron star to a quark star , due to increase of the central density as a result of the spinning down is considered in section iv . in section v we discuss and conclude our results . we consider that the change from the metastable neutron phase to the stable quark phase occurs as the result of fluctuations in a homogeneous medium , formed of neutrons , in which small quantities of the quark phase ( called bubbles or nuclei ) are randomly generated . since the process of creation of an interface is energetically unfavorable , it follows that when a quark nucleus is below a certain size , it is unstable and disappears again . surface effects disfavor the survival of small bubbles below the radius @xmath11 ( called critical size - the nuclei of this size are called critical nuclei or bubbles ) , which is nothing but the value that extremizes the thermodynamical work @xmath58 necessary to create the bubbles . only nuclei whose size @xmath59 is above the value @xmath11 are stable , and can survive @xcite . the nuclei are assumed to be macroscopic objects containing a large number of particles ( quarks ) . for strongly degenerate neutron matter the thermodynamic work @xmath58 necessary to create a quark bubble is @xcite @xmath60 \frac{4\pi } { 3}r^{3}+4\pi \sigma r^{2}+8\pi \gamma r+e_{c},\ ] ] where @xmath61 are the pressures of the quark matter and neutron matter ( exterior to the bubble , assumed to be spherical ) , @xmath16 is the surface tension , @xmath18 is the particle - number density of quark matter , @xmath62 are the chemical potentials of each phase , @xmath3 is the curvature coefficient and @xmath63 is the coulomb energy of the droplet @xcite . here @xmath64 is the excess charge of the droplet with volume @xmath65 compared with the surrounding medium , and @xmath66 and @xmath67 are the charge densities of the quark and nuclear matter , respectively . therefore the coulomb energy of the droplet is given by @xmath68 . assuming that the quark matter is immersed in a uniform background of electros , and that it is in @xmath69 equilibrium , with @xmath70 , it follows that the total electric charge in the quark phase can be expressed as @xmath71 , where @xmath72 @xcite . generally @xmath73 . the appropriate form of the curvature energy can be obtained from @xcite @xmath74 ^{-1}=\frac{g\mu ^{2}r}{6\pi } \left [ 1+o\left ( t/\mu \right ) ^{2}\right ] , \ ] ] where @xmath75 is the statistical weight and @xmath76 the chemical potential . hence the curvature coefficient is given by @xmath77 mev@xmath78 @xcite . requiring @xmath58 to be an extreme , @xmath79 , gives the following equation for the value of the critical radius @xmath80 : @xmath81 r^{2}+8\pi \sigma r+8\pi \gamma = -\frac{16\pi ^{2}}{3}\left ( \rho _ { q}-\rho _ { n}\right ) ^{2}r^{4}. \label{eqgen}\ ] ] by neglecting the coulomb energy , we obtain for the critical radius @xmath82 of the bubble the expression @xmath83 where @xmath84 and @xmath85 the parameter @xmath86 can also be represented as @xmath87 @xmath88 ^{-2}$ ] @xcite . to find the value of the critical radius for a non - negligible electrostatic energy of the bubble we use an iterative method , by substituting in the right - hand side of eq . ( [ eqgen ] ) @xmath59 by the zeroth order approximation given by eq . ( [ rc0 ] ) . then , in the first order , the critical radius is given by @xmath89 where @xmath90 .\ ] ] for values of @xmath91 mev@xmath92 , @xmath93 mev@xmath94 , @xmath16 in the range @xmath95 , @xmath96 @xmath92 and negligible @xmath67 the critical radius is of the order of @xmath97 fm . the coulomb correction has the effect of increasing the critical radius . the value @xmath80 corresponds to the limit beyond which large quantities of the quark phase begin to be formed . in fact , it is more appropriate to refer not to a limit point @xmath80 , but to a critical range of values of @xmath59 near that point , with width @xmath98 @xcite . the fluctuational development of nuclei in this range can still , with high probability , throw them back into the subcritical range , but nuclei beyond the critical range will inevitably develop into the new quark phase . the minimum critical work required to form a stable quark bubble is therefore given by @xmath99 , \ ] ] where we denoted @xmath100 and @xmath101 .\ ] ] within a purely thermodynamic approach , one can put only the problem of calculating the probability of occurrence in a medium of fluctuational nuclei of various sizes , the medium being regarded as in equilibrium . instead of the thermodynamic probability of nucleation , it is more convenient to use the equilibrium distribution function for nuclei of various sizes existing in the medium , denoted @xmath102 . @xmath103 is the number of nuclei per unit volume of the medium with sizes in the range @xmath104 . according to the thermodynamic theory of fluctuations @xcite , @xmath105 .\ ] ] near @xmath80 the thermodynamic work can be expressed as @xmath106 \nonumber\\ & & -\left [ \frac{12\pi \sigma c^{3}\sqrt{1+b}-32\pi ^{2}\left ( \rho _ { q}-\rho _ { n}\right ) ^{2}\sigma ^{3}\left ( 1+\sqrt{1+b}\right ) ^{3}}{3c^{3}}\right ] \left ( r - r_{c}\right ) ^{2}.\end{aligned}\ ] ] therefore for the equilibrium distribution function we find the expression @xmath107 , \ ] ] where @xmath108 , @xmath109constant . in order to estimate the coefficient @xmath110 of the exponential in @xmath111 we follow @xcite , and assume @xmath112 , with @xmath17 and @xmath18 the particle number densities in the neutron and quark phases , respectively . thus we obtain @xmath113 \right\ } .\ ] ] let @xmath114 be the kinetic size distribution function of the nuclei . the elementary process which changes the size of a nucleus is the attachment to it , or the loss by it , of a quark droplet , and this is to be regarded as a small change , since the nuclei are considered to be macroscopic objects . therefore the growth of the nuclei is described by a fokker - planck equation @xcite @xmath115 where @xmath116 is the flux in the size space . @xmath8 is the nuclear size diffusion coefficient and @xmath7 is connected with @xmath8 by a relationship , which follows from the fact that for an equilibrium distribution @xmath117 . therefore we find @xmath118 . in the case of a continuous stationary phase - transition process we have @xmath119constant . the constant flux is just the number of nuclei passing through the critical range per unit time per unit volume of the medium , i.e. it defines the rate of the process . with the use of the condition of the constant flux we obtain @xmath120 giving @xmath121 the constant in this equation and @xmath122 are found from the boundary conditions for small and large @xmath59 . the fluctuation probability increases rapidly with decreasing size and small nuclei have a high probability of occurrence . this is expressed by the boundary condition @xmath123 as @xmath124 . the boundary condition for large @xmath59 can be established by noting that above the critical range the function @xmath125 increases without limit , whereas the true distribution function @xmath126 remains finite . this situation is expressed by imposing the boundary condition @xmath127 for @xmath128 . the solution which satisfies the above conditions is @xcite @xmath129 in these equations the integrand has a sharp maximum at @xmath80 . by extending the integration with respect to @xmath59 from @xmath130 to @xmath131 , one obtain for the number of viable quark nuclei formed in stationary conditions per unit time and per unit volume the expression : @xmath132 above the critical range , the distribution function is constant : having reached that point , the nucleus becomes steadily larger , with practically no change in the reverse direction . accordingly we can neglect the term containing the derivative @xmath133 in the flux , leading to @xmath134 . from the significance of the flux it follows that the coefficient @xmath7 acts as a velocity in size space , @xmath135 @xcite . therefore we find for @xmath8 : @xmath136 \left ( r - r_{c}\right ) } \left ( \frac{dr}{dt}\right ) _ { macro}.\end{aligned}\ ] ] the rate of growth of the bubble radius near the critical radius is given by @xmath137 , where @xmath50 is the dynamical prefactor @xcite . the prefactor @xmath50 has been evaluated , by solving the equations of relativistic fluid dynamics in all regions , in @xcite . the result ( also taking into account heat conduction ) is @xmath138 , \label{pre}\ ] ] where @xmath139 is a constant ( the velocity of the sound in the medium around the saddle configuration ) and @xmath140 , @xmath141 and @xmath142 are the thermal conductivity and shear and bulk viscosity coefficients of the neutron matter , respectively . the first term in the above equation is the same as obtained by @xcite , corresponding to the case of non - viscous matter . the second term is similar to the results obtained by @xcite and @xcite , except with a minor difference , i.e. instead of @xmath143 there is a factor @xmath144 in the numerator . with the use of eq . ( [ pre ] ) for the prefactor it follows that the rate of formation of quark bubbles in neutron matter is given by @xmath145 \times \nonumber \\ & & \exp \left\ { -\frac{\frac{4\pi \sigma ^{3}}{3c^{2}}f(b)\left [ 1 + 4\pi \left ( \rho _ { q}-\rho _ { n}\right ) ^{2}\frac{\sigma ^{2}}{c^{3}}g(b , b_{0})\right ] } { t}\right\ } . \label{j}\end{aligned}\ ] ] to obtain the number @xmath13 of the net strange quark matter bubbles , the rate @xmath122 must be multiplied by the time interval available for prompt nucleation , @xmath146 and by the volume @xmath147 , where the nucleation can take place in the dense core @xcite . we impose that at least one quark bubble appears ( which would suffice to convert the whole neutron star ) . therefore , with the use of eq . ( [ j ] ) , one obtains the following general condition for a quark bubble formation in a neutron star core : @xmath148 the quark bubble formation rate essentially depends on the enthalpy difference in the two phases . in the density range of interest , all the neutron matter eos can be very well parameterized as polytropes , @xmath149 @xcite . the corresponding energy density in the neutron phase is @xmath150 , leading to @xmath151 . we also assume that the formed quark bubble consist of @xmath0 and @xmath1 quarks in the ratio @xmath152 ; only later weak interactions may change the composition to an energetically more favorable state . the quarks chemical potentials are related by @xmath153 , and , assuming chemical equilibrium across the phase boundary , we also have @xmath154 @xcite . then the pressure in the quark phase is ( assuming a simple bag model ) @xmath155 and @xmath156 @xcite , giving @xmath157 . hence , the enthalpy difference in the two phases can be approximated by @xmath158 assuming that the energy flow is provided by the viscous effects only , one obtains for the prefactor the expression @xcite @xmath159 . \label{k1}\ ] ] in the limit of zero baryon number , @xmath160 and we obtain the result of @xcite . if the matter is baryon - rich , but viscous damping is negligible , @xmath161 , and we obtain the results of @xcite and @xcite . therefore the condition of the formation of a quark bubble is given by @xmath162 \times \nonumber \\ & & \exp \left\ { -\frac{4\pi \sigma ^{3}}{3c^{2}t}f(b)\left [ 1 + 4\pi \left ( \rho _ { q}-\rho _ { n}\right ) ^{2}\frac{\sigma ^{2}}{c^{3}}g(b , b_{0})\right ] \right\ } \delta t\delta v_{0}\geq 1.\end{aligned}\ ] ] the transport properties of dense matter have been intensively investigated in both high energy physics and astrophysical frameworks @xcite . the shear viscosity @xmath163 and thermal conductivity @xmath164 for neutron matter have been derived by @xcite from the uhlenbeck - uehling equation ( see also @xcite and @xcite ) . they are given by @xmath165 and @xmath166 , where @xmath167 is the fermi momentum and @xmath168 are some quantities related to particle - particle cross sections , estimated in @xcite . the contribution of the bulk viscosity and thermal conductivity are negligible , and the main contribution is from the shear viscosity , which is typically of the order of @xmath169 mev/@xmath170 @xcite . but in the case of an ideal neutron gas , with zero viscosity , there will be no bubble growth , and in this case the transition from neutron matter to quark matter in astrophysical objects can not take place . assuming that the energy flow does not vanish in the absence of any heat conduction or viscous damping , and considering that the viscous effects are small and can be neglected , the condition of the formation of a quark bubble inside the dense core of a neutron star becomes : @xmath171 \right\ } \delta tv_{0}\geq 1.\end{aligned}\ ] ] for @xmath172 the time growth of the radius of the quark droplets can be described by the equation @xcite @xmath173 with the general solution @xmath174 , \ ] ] where we have used the initial condition @xmath175 . therefore for a neutron star with radius @xmath41 , the time @xmath176 required for the conversion of the whole star to a quark star can be obtained from @xmath177 .\ ] ] since @xmath178 , with a very good approximation we find @xmath179 neglecting the viscous effects , and by assuming again that the neutron matter is described by a polytropic equation of state , while the quark phase obeys the simple bag model equation of state , one obtains for the conversion time @xmath180 the volume @xmath181 of the neutron matter converted to quark matter after a time interval @xmath182 is given , in this simple model , by @xmath183 if the transition from neutron to quark phase is driven by viscous processes only , then @xmath184 } r^{2},\ ] ] and @xmath185 ^{3/2}t^{3/2}.\ ] ] in the previous sections we have considered the process of nucleation of quark bubbles in neutron star cores , and we have derived the conditions for quark matter formation in an astrophysical context . in the present section we shall apply the results previously obtained to analyze the possibility of the transition from neutron to quark phase during the spinning down of a rapidly rotating neutron star . in this case the rotational energy of the star can be used to trigger the phase transition . the minimum energy density required to create a quark bubble is @xmath186 . \label{1}\ ] ] an estimation of eq . ( [ 1 ] ) for @xmath187 mev@xmath92 , @xmath188 mev@xmath94 , @xmath189 , @xmath96 @xmath92 and @xmath190 gives @xmath191 . we consider that the evolution of the rotating neutron star can be approximated by a sequence of maclaurin spheroids @xcite . the mass of the star will be denoted by @xmath192 , and the major and minor axis of the star by @xmath193 and @xmath194 ( equatorial and polar radius ) . then the eccentricity of the star is defined according to @xmath195 . let @xmath196constant denote the density of the star , corresponding to a given value of the angular velocity @xmath57 and of the equatorial radius @xmath193 . hence the basic equations describing the mass , hydrostatic equilibrium and gravitational energy @xmath197 of the rotating star are @xcite @xmath198 where @xmath199 assuming that the total mass of the star is a constant , one obtain the following relation between the variation @xmath200 and @xmath201 of the equatorial radius , angular velocity , density and eccentricity , respectively : @xmath202 the variation of the eccentricity with respect to the angular velocity is given by @xmath203 where @xmath204 with the use of the above equations the variation of the gravitational energy of the neutron star , corresponding to a simultaneous change in equatorial radius , angular velocity and density , is given by @xmath205 \delta \omega , \ ] ] where we denoted @xmath206 @xmath207 the gravitational energy density released as the result of the slowing down of the neutron star can trigger the endothermic phase transition at the center of the star . inside a sphere of radius @xmath208 the gravitational energy density is @xmath209 \delta \omega .\ ] ] this energy should be greater or equal to the minimum energy necessary for the formation of a quark bubble , given by eq . ( [ 1 ] ) . therefore the gravitational energy converts to the quark phase the neutron matter inside a sphere of radius @xmath210 ^{1/3}\times \nonumber \\ & & \frac{\left ( 1+\sqrt{1+b}\right ) c^{-1/3}}{\left [ 2 + 2\left ( 1+b\right ) ^{3/2}+3b\right ] ^{1/3}}\left ( \delta \omega \right ) ^{1/3}\times 10^{6}\textrm { cm.}\end{aligned}\ ] ] in order the whole neutron star be converted to a quark star the condition @xmath211 must hold . on the other hand , for typical neutron star densities of the order of @xmath212 g/@xmath213 , the term @xmath214 is much smaller than @xmath215 , @xmath216 for small eccentricities @xmath51 the function @xmath217 can be approximated by @xmath218 . therefore we obtain the following condition for the relative change in the central density , necessary to convert a neutron star to a strange star , due to the endothermic gravitational energy transfer in the whole volume of the star : @xmath219 } { \left ( 1+\sqrt{1+b}\right ) ^{3}}\left ( \frac{m}{m_{\odot } } \right ) ^{-2/3}.\ ] ] for a neutron star of around two solar masses , @xmath220 and by assuming @xmath187 mev and @xmath221 , we obtain @xmath222 . of course this result is strongly dependent on the precise numerical values of @xmath223 and @xmath224 , which are generally poorly known . @xmath223 is also a neutron matter equation of state dependent parameter . for example , assuming for @xmath223 a value of @xmath4 mev will lead to @xmath225 . an increase in the mass @xmath192 of the star will also decrease the value of the relative change in the density of the star necessary to convert the neutron star . at present , we have very little observational knowledge about how fast newborn neutron star can rotate . the most well - known young pulsar is that the crab pulsar was born with a rotational period of about @xmath226 ms @xcite . if so , during its lifetime the pulsar its central density only increases about @xmath227 @xcite . currently the fastest rotating pulsar known is psr1937 + 214 , which has a period of 1.55ms or @xmath228 . but it has weak magnetic field and is suggested to be spun - up by accretion @xcite . however , some rapidly spinning millisecond pulsars are suggested to be born by accretion induced collapse from white dwarfs @xcite . therefore it can not ruled out that some pulsars can be born with milliseconds and strong magnetic field . the theoretical investigation of rotating general relativistic objects performed by @xcite shows that for some realistic equations of state of neutron matter this variation of the central density can be achieved during the complete spin - down of the star ( see table i ) . . variation of the central density of the rapidly rotating neutron star for different equations of state @xcite : a @xcite , au @xcite , fps @xcite , l @xcite and m @xcite . [ cols="^,^,^,^,^",options="header " , ] the central density increase due to the spin - down can be easily realized in a short time interval for a special class of stellar type objects , called magnetars . magnetars are compact objects with super strong magnetic fields of the order @xmath229 gauss or even higher @xcite . it is now believed that soft gamma repeaters ( sgr s ) - a small class ( 4 confirmed and one candidate ) of high energy transient discovered through their emission of bright x - ray/@xmath3-ray bursts , which repeat on timescales of seconds to years - are magnetars . there is evidence that the giant sgr flares involve the cooling of a confined @xmath230-photon plasma in an ultrastrong magnetic field @xcite . some authors @xcite have suggested that sgr s are strange stars , motivated in part by the super - eddington luminosities of their giant flares . the magnetars differ from the canonical pulsars ( with low magnetic fields of the order @xmath231 gauss ) in the sense that they spin down much more rapidly . if we assume that the spin - down of the magnetar is completely determined by the torque of its relativistic wind emission , generated via the magnetic dipole radiation , then the time variation of the angular velocity @xmath57 of the star is given by @xcite @xmath232 where @xmath233 is the moment of inertia of the star and @xmath234 is the magnetic dipole moment . when the star is born with a spin period much shorter that the observed one , the age of the star is the spin - down age @xmath235 , where @xmath236 and @xmath237 are the present spin period and its time derivative . if the spin - down is entirely due to the magnetic dipole radiation we obtain @xmath238 by adopting for the moment of inertia and the radius of the magnetar the typical values @xmath239 g@xmath240 and @xmath241 cm , and by assuming that the star was born with a magnetic field of the order of @xmath229 gauss and with an angular velocity of @xmath242 rads@xmath243 , the spin down age is given by @xmath244 s. from eq . ( [ dec ] ) it follows that the decay law of the angular velocity @xmath57 of the star is of the form @xmath245 where @xmath246 is the initial angular velocity of the star . hence , in a time interval of around @xmath247 s and for a magnetic field @xmath248 g the angular velocity of the star decreases from @xmath242 rads@xmath243 to @xmath249 rads@xmath243 . for an initial angular velocity of the order of @xmath250 rads@xmath243 and for a magnetic field of the order of @xmath251 g , @xmath252 s. in a time interval of around @xmath253 s the angular velocity of the star decreases to @xmath254 rads@xmath243 . therefore in a magnetar density changes , due to the spin - down , take place in a short interval of time . hence they can provide enough energy to trigger spontaneous nucleation to the quark phase , once the central density of the star goes above the critical density . the nucleation of quark droplets in the presence of strong magnetic fields essentially depends on the strength of the magnetic field . the energy of a quantum particle changes significantly if the magnetic field is of the order or exceeds the critical value @xmath255 , where @xmath256 is the mass of the particle . for electrons the value of the critical field is @xmath257 g. in order to calculate the value of the critical magnetic field for quarks we have to use the estimations for the quark masses . however , there are many uncertainties in the numerical values of the quark masses , which play an essential role in the calculation of the critical magnetic field and which are generally poorly known . in the case of free quarks , the current quark mass is considered to be in the range of @xmath258 mev . then the corresponding critical magnetic field is @xmath259 g. for the current quark mass , this is the typical strength of the magnetic field at which the cyclotron lines begin to occur . in this limit the cyclotron quantum is of the order or greater than its rest energy . this is equivalent to the requirement that the de broglie wavelength is of the order or greater than the larmor radius of the particle in the magnetic field @xcite . however , from the physical point of view a better description of the quark mass can be obtained by using instead the current quark mass the effective or constitute quark mass , which takes into account the effects of the strong interactions , and which is of the order @xmath260 mev @xcite . the critical magnetic field corresponding to the effective quark mass could be as high as @xmath261 g. the study of the behavior of the surface and curvature energy terms in strong magnetic fields have shown that both quantities diverge for @xmath262 , with the curvature term diverging much faster . consequently , in the presence of strong magnetic fields the rate of stable quark droplet formation per unit volume @xmath263 tends to zero , @xmath264 , and therefore there can not be any thermal nucleation of quark droplets in neutron star cores @xcite . the value of @xmath8 inside a magnetar is not known , and it is not certain if the interior magnetic field is much stronger than the surface field . if in the interior of the magnetar the value of the magnetic field can exceed the critical value @xmath265 g , which represents the most conservative estimation of the critical magnetic field , corresponding to the free quark mass @xmath266 mev , then the rate of droplet formation can be considerably reduced and the transition from neutron matter to quark matter could not take place . on the other hand , by taking into account that the physical mass of the quark is the effective mass , the corresponding critical magnetic field could be of the order @xmath267 g , value which could be higher than the magnitude of the magnetic field inside the magnetars . for magnetic fields smaller than @xmath268 , the endothermic nucleation process can take place in a short time interval during the spin - down of the magnetar . in the present paper we have considered a kinetic theory of strange matter nucleation in neutron stars . we derived general necessary conditions for neutron to quark matter conversion . there have been several assumptions made in the derivation of the results , mostly related to the form of the growth speed of the bubble . the result is valid only if non - linear effects can be ignored , and the linearized hydrodynamic equations are applicable . furthermore , we have assumed that the radii of the bubbles are larger than the correlation length and that heating due to dissipation is slow , causing the temperature to vary slowly across the bubble wall . we also considered that the phase transition is strongly first order , releasing considerably heat . the growth of the critical size quark bubbles nucleated in the first order transition in the neutron star core is governed by the dynamical prefactor @xmath269 . there are several expressions for the prefactor obtained in the physical literature . in the csernai - kapusta approach @xcite , the prefactor essentially depends on the dissipative properties ( shear , bulk viscosity and thermal conductivity ) of neutron matter . therefore , in a non - viscous medium with @xmath270 , the growth of the quark bubbles can not take place . the main physical assumption is that the energy flow is provided by the viscous effects only . on the contrary , one can argue that the energy flow does not vanish in absence of heat conduction or viscous damping @xcite . in this case , and in the limit of zero viscosity , the prefactor depends only on two scale parameters , the correlation length @xmath34 and the critical radius of the quark bubble @xmath11 . for a viscous medium the prefactor can simply be written as the sum of the viscous and non - viscous term , with the viscous term not affecting the growth process significantly . the previous ( viscous ) results can be reobtained by using some assumptions for the velocity of sound in the medium around the saddle configuration . the condition for the formation of at least quark bubble in the neutron matter as a result of the thermal fluctuation is given by eq . ( [ cond1 ] ) . the condition includes both the cases of viscous and ideal neutron matter . it is extremely sensitive with respect to the numerical values of the microscopic parameters characterizing the quark bubble , like the surface tension @xmath16 , the bag constant @xmath8 or the curvature coefficient @xmath271 . in discussing the astrophysical implications of the expression ( [ cond1 ] ) we shall consider separately the two limiting cases , corresponding to viscous and non - viscous neutron matter , respectively . in the case of ideal neutron matter , the condition of the formation of a quark bubble is given by eq . ( [ cond2 ] ) . this equation determines at which temperature the phase transition takes place . for example , in the case of the walecka mean field equation of state @xcite , with @xmath272 and @xmath273 @xcite , the temperature @xmath46 necessary for the formation of a stable quark bubble in a second ( @xmath274 s ) , in a volume @xmath275 with radius @xmath276 km ( @xmath277 km@xmath278 ) , and at the center of the neutron star , is around @xmath279 mev , for @xmath280 mev@xmath281 . here we have also assumed the standard values for the quark matter chemical potential and for the bag constant , @xmath282 mev and @xmath283 mev@xmath92 @xcite . since the temperature of a newborn neutron star is around @xmath4 mev @xcite , this value of the surface tension seems to rule out , for the given equation of state , the possibility of formation of quark nuclei in the neutron matter . however , a small variation in the value of @xmath16 , @xmath284 mev@xmath285 reduces the temperature for quark nuclei formation to @xmath286 mev , a value which does not exclude the possibility that quark nuclei formation can be initiated in the early stages of the neutron star evolution . smaller values of @xmath16 , of the order of @xmath287 mev@xmath281 , can lower even more the phase transition temperature , making it possible even for cold neutron stars . the transition temperature is relatively insensitive to the details of the equation of state of the dense neutron matter . if the temperature at which the phase transition is initiated is strongly dependent by the microscopic model adopted for describing the neutron matter properties , the total transition time of the neutron star to a quark star , given by eq . ( [ conv ] ) , is much less dependent on the details of the equation of state or the exact numerical values of the surface tension or curvature of the bubble . generally , for a neutron star with radius @xmath288 km , the conversion time is very long , of the order of @xmath289 years for @xmath290 mev@xmath281 . for @xmath291 mev@xmath281 or even lower , the conversion time could have values of the order of @xmath292 years , which are still of the same order of magnitude as the lifetime of a neutron star ( @xmath292 years ) . hence , in the present approach , which is limited to the consideration of linear effects only in the hydrodynamic description of the first order phase transitions for an ideal neutron fluid , the growth of the quark nuclei is very slow . therefore the results of the present analysis suggest that even quark nuclei could appear inside the neutron star at a very early stage , the transition to a pure quark phase ends only in the final stage of its evolution . if the neutron matter - quark matter phase transition is mainly driven by viscous processes , like in the csernai - kapusta scenario , then the time necessary for a neutron star to convert to a quark star follows from eqs . ( [ k1 ] ) and ( [ tconv ] ) . in order to numerically estimate the conversion time , we need an estimation of viscosity and conduction coefficients of nuclear matter . we shall also consider two limiting cases for the phase transition . below the critical density @xmath293 g/@xmath294 , cold neutron matter consists of two distinct phases . at low density , neutron star matter consists of nuclei which form a solid lattice for temperatures below the melting temperature and a sea of relativistic electrons . when the density reaches @xmath295 g/@xmath294 , neutrons begin to drip from the neutron - rich nuclei and then , in addition to the relativistic electrons there is a sea of non - relativistic neutrons . at the critical density the neutron - rich nuclei dissolve , leaving seas of degenerate neutrons , protons and electrons . the neutrons in cold neutron star matter are quite likely superfluid , and therefore the transport coefficients are dominated by electrons , with the motion of electron determined by electron - electron and electron - proton scattering @xciteas a result the viscosity is much lower , but the thermal conductivity is only slightly smaller as compared to the case in which both neutrons and protons form normal fluids @xcite . hence , we can neglect the viscosity coefficients with respect to the thermal conductivity and approximate @xmath140 by @xmath296 , with @xmath297 @xcite . a slightly different expression for the leading term in thermal conductivity has been proposed in @xcite , @xmath298 , with @xmath299 @xmath92 and @xmath300 @xmath92 . hence , by using this last expression , we obtain for @xmath176 @xmath301 the transition time is independent of the temperature of the neutron matter . it depends only on the microscopic properties of the quark bubble ( surface tension and curvature coefficient ) , density and the equation of state of the matter inside the star . for @xmath302 mev and @xmath283 mev@xmath92 and for @xmath290 mev@xmath281 eq . ( [ therm ] ) gives , for a neutron star with radius @xmath303 km and for several equations of state values of the conversion time of the order @xmath304 years . in the case of young and hot neutron stars , the main dissipative mechanism which could drive the phase transition is the shear viscosity of the neutron matter . the temperature dependence of the shear viscosity coefficient can be approximated by @xmath305 , with @xmath306 mev@xmath307/@xmath170 @xcite . consequently , by neglecting the heat conduction and the bulk viscosity , the conversion time becomes @xmath308 in fig . 1 we have represented @xmath176 given by eq . ( [ shear ] ) as a function of temperature , for different equations of state of neutron matter . we have considered four equations of state , namely , the equation of state ( eos ) of the free neutron gas @xcite , the bethe - johnson eos @xcite , the lattimer - ravenhall eos @xcite and the walecka eos @xcite . in all cases we calculated the conversion time for a neutron star with radius @xmath309 km . as one can see from the figure , in this case the conversion time is much shorter as compared to the ideal neutron matter case or thermal conduction driven phase transition . @xmath176 is rapidly decreasing with the temperature . for a neutron star with rapidly cooling processes , e.g. direct urca process , kaon condensation etc . , the stellar temperature can decrease to less than @xmath310mev in a time scale of @xmath311 yr @xcite . according to fig . 1 , the conversion can take place in a time scale less than @xmath312 yr if the spin - down time scale is less than this time scale . this can be case if this neutron star is a magnetar . we should point out that the interior of magnetars are not quantum liquid , i.e. superfluid or superconducting , because the strong magnetic field can destroy the nucleon cooper pairs @xcite . on the other hand , if the neutrons and protons are superfluid , then the transport properties are determined completely by electron - electron scattering . in fact in this case the viscosity of the nuclear fluid is given by @xmath313 t^{2}$ ] @xcite , where @xmath314 is the fine structure constant , @xmath315 is the thomas - fermi wave vector , and @xmath316 , @xmath317 and @xmath318 are the electron mass , momentum and energy , respectively . we assume that the electron gas inside the neutron star is extreme relativistic and strongly degenerate . then it follows that @xmath319 @xcite , with @xmath320 the fermi momentum and @xmath321 the electron number density . since @xmath321 can be related to the total particle number @xmath322 by means of the relation @xmath323 , with @xmath324 the mean number of electrons per baryon , it follows that @xmath325 . hence we obtain for the shear viscosity coefficient of the nuclear matter @xmath326 ^{2}+\left [ \frac{m_{e}\left ( 8\pi \right ) ^{1/3}}{\left ( 3y_{e}n_{0}\right ) ^{1/3}\left ( n / n_{0}\right ) ^{1/3}}\right ] ^{4}\right\ } t^{2}.\ ] ] in this case , since the viscosity of the superfluid nuclear matter is much smaller , the conversion time is longer than for a nuclear fluid dominated by the neutron viscosity . however , in this case the phase transition from neutron matter to quark matter can also take place at very low temperatures . consequently , a very short @xmath176 , of the order of hours or a few days , is also allowed in the framework of this model . it is widely accepted that at asymptotic densities the ground state of qcd with @xmath328 is the color - flavor locked ( cfl ) phase @xcite . in this phase , color gauge symmetry is completely broken , as are both chiral symmetry and baryon number . the effective coupling is weak and the low - energy properties can be determined by using methods from the theory of superconductivity . for large gaps @xmath329 , the cfl phase is rigorously electrically neutral , despite the unequal quark masses , and even in the presence of electron chemical potential . the transition from nuclear matter to quark matter via bubble nucleation may be greatly simplified if the transition occurs directly to quark matter in the cfl phase @xcite . for a given baryonic chemical potential @xmath330 , electrically neutral nuclear matter and electrically neutral quark matter have different values of the electronic chemical potential @xmath331 . since @xmath331 must be continuous across any interface , a mixed phase region is formed , within which positively charged nuclear matter and and negatively charged quark matter with the same @xmath332 coexist at any given radius . the growth of the quark droplets is at the expense of the nuclear matter . however , if the quark matter is in the cfl phase , an interface between bulk nuclear matter with non - zero electron number @xmath333 and cfl quark matter with @xmath334 may be stable , as long as @xmath331 satisfies the condition @xmath335 , with @xmath336 the variation of the chemical potential @xcite . therefore , the existence of the cfl phase supports the idea that in neutron stars quark matter and baryonic matter may be continuously connected . then the growth of the interface between the neutron and cfl quark phase could be the result of exterior energy absorption , via an endothermic mechanism . in summary , by using the kinetic nucleation theory we have considered the possibility that the neutron matter - quark matter phase transition could be triggered by the gravitational energy released during the spin - down of a pulsar . this process could take place in a very short time interval , of the order of few tens seconds , in a class of dense compact objects , with very high magnetic fields , called magnetars . however , the percentage of neutron star matter that can be converted into quark matter depends on the initial period , the neutron star mass and on the equation of state of the neutron and quark phases . an important and open issue is how could we determine if such a process is taking place or took already place in some pulsars or magnetars , and what are the observational signatures of the endothermic neutron matter - quark matter transition . in particular , it would be interesting to know if such a process could be a possible gamma - ray burst mechanism , as suggested by @xcite . all of these issues will be considered in a subsequent paper @xcite . alcock , c. & olinto , a. 1989 , phys . d , 39 , 1233 alcock , c. , farhi , e. & olinto , a. 1986 , apj , 310 , 261 alford , m. , rajagopal , k. & wilczek , f. 1998 , phys . lett . b , 422 , 247 alpar , m. a. , cheng , a. f. , ruderman , m. a. & shaham , j. , 1982 , nature , 300 , 728 arons , j. 1983 , nature , 302 , 301 berezhiani , z. , bombaci , i. , drago , a. , frontera , f. & lavagno , a. , 2003 , apj,586 , 1250 bodmer a. r. 1971 , phys . d , 4 , 1601 caldwell , r. r. & friedman , j. r. 1991 , phys . b , 264 , 143 chakrabarty , s. 1993 , phys . rev . d , 48 , 1409 chakrabarty , s. 1995 , phys . d , 51 , 4591 chakrabarty , s. & sahu , p. k. 1996 , phys . d , 53 , 4687 chakrabarty , s. 1996 , phys . d , 54 , 1306 chandrasekhar , s. 1986 , ellipsoidal figures of equilibrium , new york : dover chau , h.f . , cheng , k.s . & ding , k.y . 1992 , apj , 399 , 213 cheng , k.s . & dai , z. g. 1996 , phys . lett . , 77 , 1210 cheng , k.s . & dai , z. g. 1998 , phys . lett . , 80 , 18 cheng , k. s. & harko , t. 2000 , phys . rev . d , 30 , 083001 cheng , k. s. & harko , t. 2003 , apj , 596 , 451 cheng , k. s. , chau , w. y. , zhang , j. l. & chau , h. f. 1992 , apj , 396 , 135 cheng , k.s . , dai , z.g . , wei , d.m . & lu , t. 1998 , science , 280 , 407 cheng , k. s. , dai , z. g. & lu , t. 1998 , int . j. mod . d , 7 , 139 cook , g. b. , shapiro , s. l. & teukolsky , s. a. 1994 , apj , 424 , 823 csernai , l. p. & kapusta , j. i. 1992a , phys . d , 46 , 1379 csernai , l. p. & kapusta , j. i. 1992b , phys . lett . , 69 , 737 cutler , c. , lindblom , l. & splinter , r. j. 1990 , apj , 363 , 603 dai , z. g. , peng , q. h. & lu , t. 1995 , apj , 440 , 815 danielewitz , p. 1984 , phys . b , 146 , 168 duncan , r. c. & thompson , c. 1992 , apj , 392 , l9 flowers , e. & itoh , n. 1976 , apj , 206 , 218 flowers , e. & itoh , n. 1979 , apj , 230 , 847 flowers , e. & itoh , n. 1981 , apj , 250 , 750 glendenning , n. k. 1992 , phys . d , 46 , 1274 glendenning , n. k. 2000 , compact stars , nuclear physics , particle physics and general relativity , new york : springer glendenning , n. k. , pei , s. & weber , f. 1997 , phys . lett . , 79 , 1603 grassi , f. 1998 , apj , 492 , 263 haensel , p. , zdunik , j. l. & schaeffer , r. 1986 , a&a , 160 , 121 hakim , r. & mornas , l. 1993 , phys . c , 47 , 2846 harko , t. & cheng , k. s. 2000 , phys . lett . a , 266 , 249 harko , t. & cheng , k. s. 2002 , a&a , 385 , 947 heiselberg , h. 1995 , quark matter droplet formation in neutron stars , international symposium on strangeness and quark matter , singapore : world scientific , 338 - 342 heiselberg , h. , pethick , c. j. & staubo , e. f. 1993 , phys . lett . , 70 , 1355 horvath , j. e. & benvenuto , o.g . 1988 , phys . b , 213 , 516 horvath , j. e. 1994 , phys . d , 49 , 5590 horvath , j. e. , benvenuto , o. g. & vucetich , h. 1992 , phys . d , 45 , 3865 itoh , n. 1970 , prog . phys . , 44 , 291 kawasaki , k. 1975 , j. stat . , 12 , 365 landau , l. d. & lifshitz , e. m. 1980 , statistical physics , oxford : pergamon press langer , j. s. & turski , l. a. 1973 , phys . rev . a , 8 , 3230 lattimer , j. m. & ravenhall , d. g. 1978 , apj , 223 , 314 lorentz , c. p. , ravenhall , d. g. & pethick , c. j. 1993 , phys . , 70 , 379 lugones , g. & benvenuto , o. g. 1995 , phys . d , 52 , 1276 lugones , g. , benvenuto , o. g. & vucetich , h. 1994 , phys . d , 50 , 6100 lyubarsky , y. , eichler , d. & thompson , c. 2002 , apj , 580 , l69 ma , f. & xie , b. 1996 , apj , 462 , l63 madsen , j. & olesen , m. l. 1991 , phys . d , 43 , 1069 madsen , j. 1993 , phys . lett . , 70 , 391 manchester , r. n. & taylor , j. h. 1977 , pulsars , san francisco : freeman mardor , i. & svetitsky , b. 1991 , phys . d , 44 , 878 ng , c. y. , cheng , k. s. & chu , m. c. 2003 , astropart . phys . , 19 , 171 norsen , t. 2002 , phys . rev . c , 65 , 045805 olesen , m. l. & madsen , j. 1994 , phys . rev . d , 49 , 2698 olinto , a. v. 1987 , phys . b , 192 , 71 paczynski , b. 1992 , acta astron , 42 , 145 pandharipande , v. r. & smith , r. a. 1975 , nucl . a , 237 , 507 pandharipande , v. r. 1971 , nucl . a , 174 , 641 pandharipande , v. r. , pines , d. & smith , r. a. 1976 , apj , 208 , 550 rajagopal , k. & wilczek , f. 2001 , phys . , 86 , 3492 ruggeri , f. & friedman , w. a. 1996 , phys . d , 53 , 6543 sawyer , r. s. 1989 , phys . b , 233 , 412 shapiro , s. l. & teukolsky , s. a. 1983 , black holes , white dwarfs and neutron stars , new york : john wiley & sons shukla , p. & mohanty , a. k. 2001 , phys . c , 64 , 054910 shukla , p. , mohanty , a. k. , gupta , s. k. & gleiser , m. 2000 , phys . c , 62 , 054904 shukla , p. , mohanty , a. k. & gupta , s. k. 2001 , phys . d , 63 , 014012 tang , p.s . , cheng , k.s . & t. harko , 2004 , in preparation thompson , c. & duncan , r. c. 1995 , mnras , 275 , 255 thompson , c. & duncan , r. c. 1996 , apj , 473 , 322 tsuruta , s. , 1998 , phys . rep . , 292 , 1 usov , v. v. 2001 , phys . lett . , 787 , 21101 venugopalan , r. & vischer , a. p. 1994 e , 49 , 5849 walecka , j. d. 1975 , phys . b , 59 , 109 wiringa , r. b. , fiks , v. & fabrocini , a. 1988 , phys . rev . c , 38 , 1010 witten , e. 1984 , phys . d , 30 , 272

we study a bose - hubbard model having on - site repulsion , nearest - neighbor tunneling , and ferromagneticlike coupling between occupation parities of nearest - neighbor sites . for a uniform system in any dimension at zero tunneling , we obtain an exact phase diagram characterized by mott - insulator ( mi ) and pair liquid phases and regions of phase separation of two mis . for a general trapped system in one and two dimensions with finite tunneling , we perform quantum monte carlo and gutzwiller mean - field calculations , both of which show the evolution of the system , as the parity coupling increases , from a superfluid to wedding - cake - structure mis with their occupations jumping by 2 . we also identify an exotic pair superfluid at relatively large tunneling strength . our model ought to effectively describe recent findings in imbalanced fermi gases in two - dimensional optical lattices and also potentially apply to an anisotropic version of bilinear - biquadratic spin systems .

1309.3709

since half a century ago , the hubbard model has been widely studied as a simple model accounting for several nontrivial phenomena in electronic systems , such as the mott insulator , ( anti)ferromagnetism , and novel superconductivity@xcite . its bosonic version , the bose - hubbard ( bh ) model , has been appreciably investigated for more than two decades@xcite . the bh model not only aptly describes the many - body behavior of bosons in lattices , but also serves as a good and simple example for understanding how competition between two common mechanisms , localization and itineracy , can drive matter toward extremely different phases , namely mott - insulator ( mi ) or superfluid . the mi phase , characterized by commensurate occupations , gapped excitations and incompressibility , undergoes a transition to a superfluid , characterized by bose - einstein condensation , gapless excitations and finite compressibility , as the ratio of tunneling to interaction energy increases beyond a density - dependent critical value . differently from the fermionic case , the bosonic nature allows the bh phase diagram of multiple mi regions for all possible integer occupations , surrounded by the superfluid region . various realizations of the bh model have been early suggested in granular systems with an embedded condensate , such as a josephson junction array@xcite or liquid helium in porous media@xcite , and later in lattice bosons , such as bosonic atoms in optical lattices@xcite or exciton - polaritons in an array of microcavities@xcite . recent achievements in cold - atom experiments have successfully created inhomogeneous bh systems and realized the mi - superfluid transition@xcite as well as a spatially separated structure of multiple mis@xcite . these multiple applications have stimulated a broad interest in the bh physics and lead to an active study of a great variety of bh models @xcite . recently , a theoretical investigation we carried out on imbalanced fermionic superfluids in a two - dimensional ( 2d ) array of coupled tubes@xcite [ a setup which has been experimentally realized@xcite in the search for the elusive fulde - ferrell - larkin - ovchinnikov ( fflo ) state@xcite ] has shown an exotic compressible - incompressible quantum phase transition of the tube occupation of unpaired majority fermions ( umfs ) in the system . the phase diagram obtained from the microscopic model of fermions exhibits similarities and contrasts to the structure as a bh phase diagram . on the one hand there is a phase with well defined umf occupation numbers in each tube . on the other hand , the occupations of the incompressible ( mi ) regions are either all even or all odd . the similarities reflect the competition between the ontube energy and cross - tube kinetics of the umfs and hence suggests an effective description of the ( projected ) system by a 2d bh model , while the contrasts suggest the need for an additional term in the effective model that `` orients '' the occupation parities ( see sec . [ sec : model ] below for a detailed discussion ) . motivated by this previous work , we study an extended bh model having a ferromagneticlike coupling between occupation parities of nearest - neighbor sites . this proposed coupling can be pictured as domain - wall energies between nearest - neighbor sites of opposite occupation parities , reminiscent of local magnetization kinks in the ferromagnetic ising model@xcite , resulting in a ground state that favors all the sites having the same occupation parity . the parity coupling shows interesting interplays with the other two original bh ingredients , single - particle tunneling and onsite repulsion . first , it is antagonistic toward the single - particle tunneling process , during which two involved neighbor sites flip the parity and form domain walls between themselves and the surrounding sites [ see illustration in figs . [ fig : f01](a ) and [ fig : f01](b ) ] . second , it is different from the onsite repulsion in that it drives the system to accept additional doping as pairs rather than single particles , to avoid the domain wall energy , while the onsite repulsion does the opposite as a way to reduce the interaction energy [ see fig . [ fig : f01](c ) and [ fig : f01](d ) ] . in this paper , we analyze the system and characterize the effects of the occupation - parity coupling . we will suggest two possible experimental realizations for exploring the model in cold atomic and condensed matter systems . we shall find a rich phase diagram for the uniform case in the zero - tunneling limit , reflecting different doping preferences driven by the parity coupling and onsite repulsion . for the finite - tunneling case , we apply two commonly used treatments , quantum monte carlo@xcite ( qmc ) and gutzwiller mean - field@xcite ( gmf ) methods , on trapped systems , thus successfully describing the state of the system as the parity coupling increases for finite tunneling . in the large - tunneling regime , we identify an exotic pair superfluid state , which emerges with a different mechanism from previously studied ones@xcite . , interaction @xmath0 , and parity coupling @xmath1 of the system . ( a ) a reference state of each site occupied by @xmath2 particles . ( b ) after a single particle tunneling , the change in number of two neighbor sites alters the interaction energy by @xmath3 , while the change in parity of them results in a domain - wall formation ( dashed curve ) of energy cost @xmath4 . ( c ) at small @xmath5 , two doping particles tend to occupy different sites , minimizing the interaction energy . ( d ) at large @xmath5 , the two occupy the same site as a pair doping , avoiding the domain - wall energy.,width=302 ] the paper outline is as follows . in sec . [ sec : model ] , we introduce the model hamiltonian and discuss its possible realizations by studying its emergence in imbalance fermi gases as well as in spin systems . in sec . [ sec : phase ] , we perform a detailed analysis of the competition between the parity coupling and the onsite repulsion , and obtain an exact phase diagram in the zero - tunneling plane . in sec . [ sec : numerical ] , we present results from qmc and gmf calculations that show the evolution of a trapped system and the emergence of the pair superfluid . finally , we summarize our work in sec . [ sec : conclusion ] . the extended bh model with occupation - parity couplings between nearest - neighbor sites is described by the hamiltonian @xmath6 where @xmath7 is a bosonic operator on site @xmath8 , @xmath9 is the number operator , @xmath10 is the number - parity operator , @xmath11 is the local chemical potential and @xmath12 denotes a pair of neighboring sites . the non - negative parameters @xmath13 , @xmath1 , and @xmath0 give the strength of nearest - neighbor tunneling , ferromagnetic - like nearest - neighbor parity coupling and onsite repulsion , respectively . each pair of them show a competition that is reflected in the ground state of the system . at @xmath14 , the hamiltonian returns to the original bh one , in which the domination of itineracy ( large @xmath15 ) or localization ( large @xmath16 ) results in a superfluid or mi , respectively . the presence of @xmath1 terms can be regarded as an energy cost of domain walls between two neighboring sites with different number parities . from this point of view , the competition between @xmath13 and @xmath1 can be described by the picture that a single - particle tunneling changes the parities of two sites and hence pays an energy cost of creating a domain wall surrounding them [ see figs . [ fig : f01](a ) and [ fig : f01](b ) ] . therefore , when @xmath1 dominates , the system favors the minimization of the total number of domain walls , that is , to have all sites with the same occupation parity . the interplay between @xmath0 and @xmath1 yields a rich phase diagram , which will be discussed in detail in sec . [ sec : phase ] . here , we now turn to discuss two possible realizations of our model . first , a recent study in ref . [ ] has suggested that the mechanism in the hamiltonian of eq . ( [ eqn : ham0 ] ) accounts for an exotic quantum phase transition in imbalanced fermionic superfluids in optical lattices . the system has the geometry of a 2d array composed of one - dimensional ( 1d ) tubes that has been realized in experiments@xcite . in such a system it is the behavior of spin imbalance per tube , or the fillings of umfs , that undergoes a transition from a compressible state ( umf move easily across tubes ) to a mi ( umfs are localized on each tube ) , yielding a similar phase diagram to the regular bh one except the filling numbers of the mi phases are either all even or all odd while they can be any integer in the regular case . regarding the projection of this anisotropic three - dimensional fermionic system to an effective 2d lattice model described by eq . ( [ eqn : ham0 ] ) , we find the tunneling @xmath13 and on - site interaction @xmath0 as a result from the interplay of the lattice geometry , on - tube interaction , and on - tube pairing order . the occupation - parity coupling @xmath1 comes from the domain - wall energy between two neighboring tubes having different spatial parities of the oscillatory superconducting order parameter , which is directly related to the occupation parity of umfs . in the appendix [ sec : dbh ] , we present a detailed derivation showing how the physics of umfs is effectively described by eq . ( [ eqn : ham0 ] ) , by starting from the original hamiltonian of the fermionic superfluid system . we remark that the parameter regimes discussed in ref . [ ] are for the large @xmath1 limit . however , the high tunability of lattice geometry and interatomic interaction in this cold atomic system could enable the exploration of sufficiently wide parameter regimes of our model hamiltonian . second , the regular bh model can be mapped to a spin-1 ( @xmath17 ) system around the tip of the mott insulator lobe in the large - occupation limit , in which a three - state truncation applies on each site@xcite . for our model , the occupation parity coupling terms are mapped to spin couplings of the ising - type with quadratic and biquadratic forms . taking the truncation basis of integer occupation states @xmath18 and @xmath19 , we perform the mapping @xmath20 , @xmath21 as well as @xmath22 $ ] on eq . ( [ eqn : ham0 ] ) and obtain a spin hamiltonian , @xmath23 up to a constant offset . here @xmath24 , @xmath25 , @xmath26 with @xmath27 being the coordination number and @xmath28 . one immediately sees the competitions in eq . ( [ eqn : ham1 ] ) : the bilinear coupling @xmath29 aligns spins in the @xmath30-@xmath31 plane , the biquadratic coupling @xmath32 drives spins to @xmath33 states , the non - linear zeeman coupling @xmath34 drives them to @xmath35 state , and the local magnetic field @xmath36 aligns them in the @xmath37 direction . the first line in eq . ( [ eqn : ham1 ] ) can be regarded as an anisotropic version of standard bilinear - biquadratic spin models@xcite , whose experimental realization has been discussed in condensed matter@xcite and cold atomic gases@xcite . the second line commonly appears in nonlinear spin systems@xcite . a recent study on fermionic dipolar gases@xcite also found a route toward the creation of an anisotropic spin system , as a variant of eq . ( [ eqn : ham1 ] ) . due to the connection between the two models , the realization of the spin system could provide an alternative approach for examining a variant of our model bosonic system , and vice versa . in this section we discuss the competition between the onsite interaction @xmath0 and the parity coupling @xmath1 through additional doping of a uniform system in the zero - tunneling regime . we derive exact energy functionals in any dimension for each competing phase and obtain the ground - state phase diagram marking their relatively energetically favorable regions . at @xmath38 , the ground state is a product of each single - site state with integer occupation @xmath39 on site @xmath8 . the configuration of @xmath40 is obtained by minimizing the system s energy . at integer fillings @xmath2 , we have a mi with all @xmath41 [ fig . [ fig : f01](a ) ] , independent of the ratio @xmath5 . now considering the doping of two additional particles in the system , if @xmath0 dominates ( small @xmath5 ) , we expect that the two particles will occupy different sites to avoid higher interaction energy [ fig . [ fig : f01](c ) ] . such two sites with @xmath42 occupation can be arbitrary ones at @xmath14 and should bind together as neighbors at @xmath43 due to the _ domain - wall _ effect . for more than two particles added , they similarly tend to singly dope sites that randomly spread if @xmath14 or cluster if @xmath43 . we call the former case a single - particle doped liquid ( sl ) and the latter a phase separation of two mis ( ps ) . if @xmath1 dominates ( large @xmath5 ) , to avoid the domain wall energies , the two doping particles tend to occupy the same site producing pair doping [ fig . [ fig : f01](d ) ] , thus keeping the parity of all sites unchanged . more than two particles will doubly dope in the same way , up to one doping pair per site due to the interaction effects . we call such state a pair - doped liquid ( pl ) . similarly , when particles are taken out of the system ( in analogy to the hole doping in fermionic systems ) , they leave as single particles at large @xmath0 and as onsite pairs at large @xmath1 . below we calculate the phase boundary between ps and pl in a homogeneous system . in the case of doping @xmath44 particles ( holes ) in a @xmath45-dimensional uniform system of @xmath46 sites with integer filling @xmath2 ( @xmath44 is taken even and @xmath47 ) , the ps state has @xmath44 sites of @xmath48 filling , @xmath49 sites of @xmath2 filling and @xmath50 domain - wall links , while the pl state has @xmath51 sites of @xmath52 filling , @xmath53 sites of @xmath2 filling , and no domain walls . the energies of the ps and pl states can thus be written as @xmath54 and @xmath55 respectively , where @xmath56 is the onsite interaction energy . we let @xmath57 to obtain a critical ratio @xmath58 that the separates ps and pl states , @xmath59 given periodic boundary conditions for the system , the doped domain favors a spherical shape to minimize the surface area or the number of domain - wall links , so we have @xmath60 and @xmath61 , where @xmath62 is the domain radius and @xmath63 ( @xmath64 ) is the volume ( surface area ) of a @xmath45-dimensional unit sphere . substituting these relations in eq . ( [ eqn : vuc ] ) and rewriting the doping in terms of a rescaled coupling @xmath65 and the average filling @xmath66 such that @xmath67 , we obtain @xmath68 where @xmath69 . equation ( [ eqn : vuc2 ] ) for @xmath70 ( respectively , @xmath71 ) describes the ps - pl transition due to particle ( hole ) doping on the @xmath2-filling mi ; except for @xmath72 when there is no pl that is hole doped from the @xmath73 mi , so the phases are determined only by particle doping on the vacuum state , where @xmath74 . and @xmath38 cases . here @xmath75 and @xmath76 with @xmath46 being the volume of the system ( in @xmath45 dimensions ) . in the @xmath14 plane , the system is a mott insulator ( mi , solid section line ) at integer fillings and @xmath77 or is a superfluid ( sf ) otherwise . in the @xmath38 plane , the system is mi ( solid line ) at integer fillings and any @xmath1 . at any fractional filling and finite @xmath1 , the system is phase separated into two mis ( ps ) , or is a pair - doped liquid ( pl ) , distinguished by the critical relation in eq . ( [ eqn : vuc2 ] ) [ dashed ( dotted ) curve in two ( one ) dimensions ] . in the pl regime at @xmath78 , the half - integer fillings ( dash - dotted curve ) separate two types of pl of different parities , pl@xmath79 as well as pl@xmath80 , and meet the ps lobe at a triple point " @xmath81 ( see text for more details ) . in the @xmath82 axis , the fractional - filling region is a single - particle doped liquid ( sl).,width=283 ] figure [ fig : f02 ] shows a combined phase diagram of the @xmath14 ( @xmath82@xmath83 plane , where @xmath75 ) and @xmath38 ( @xmath82@xmath84 plane ) cases in one and two dimensions . at @xmath14 , the system turns to the original bh model : the ground state is a mi ( solid section lines ) at integer fillings and @xmath83 below a critical value @xmath85 or is a superfluid ( sf ) at any fractional filling . here @xmath85 monotonically increases with dimension @xmath45 and decreases with filling @xmath2 . at @xmath38 , the integer fillings are always mis ( the solid lines extend to @xmath86 ) , while at fractional fillings there are lobelike ps regions ( having @xmath2 and @xmath42 mis spatially separated given @xmath87 ) at small @xmath84 , and pl regions outside the lobes . in the pl case , when @xmath88 and goes across half integers ( dash - dotted line ) , the ground state suddenly switches to another subspace of opposite parity and occupations , from a pl being particle doped from the @xmath2-filling mi to one being hole doped from the ( @xmath42)-filling mi [ denoted as pl(@xmath2,@xmath89 ) and pl(@xmath42,@xmath90 ) , respectively ] . ( this continuity in density but discontinuity in parity could invalidate the local - density approximation , which is commonly used to profile a trapped system from the uniform phase diagram@xcite . ) such parity boundary meets the tip of the ps lobes at a triple point @xmath91 , which is a function of dimension but independent of filling . at @xmath92 , there is no such triple point due to the lack of hole - doped pl from the @xmath73 mi . the pl here is always of even parity [ pl(0,2 ) ] . the scaling @xmath65 indicates that the pl will eventually disappear in the limit of @xmath93 , except in the free case ( @xmath94 ) when it is the ps state instead of the pl one that disappears . on the other hand , the pl phase is always allowed in a finite - sized or a trapped system and will exist at the interphase between different mi plateaus as will be discussed below . in addition , the rich structure in the @xmath82@xmath84 diagram implies a nontrivial interplay when all @xmath13 , @xmath0 and @xmath1 are non - zero . in sec . [ sec : numerical ] , we numerically study and present results for a general trapped system using quantum monte carlo and gutzwiller mean - field methods . in this section we apply two commonly used methods to do calculations for a trapped system ( a natural setup in cold - atom experiments ) , at various values of the parity coupling in one and two dimensions . in a harmonic trap , the system has a local chemical potential of the form @xmath96 where @xmath97 is the global chemical potential , @xmath98 is the trap s curvature , and @xmath99 is the position of site @xmath8 from the trap center . first , we use the quantum monte carlo ( qmc ) method and , in particular , the stochastic series expansion ( sse ) algorithm@xcite , which has been successfully applied on a variety of bh model studies@xcite . we plot the density profile @xmath100 from the trap center in 1d lattices [ fig . [ fig : f03](a ) ] and that along the diagonal of 2d square lattices [ fig . [ fig : f03](b ) ] at a given @xmath13 that makes the system a superfluid when @xmath14 . in the original bh case ( @xmath14 ) , the superfluid s density profile monotonically and smoothly decreases to zero from the center to the edge ( solid curve ) , matching the shape of the trap . in both the 1d and 2d cases , with an increase in @xmath1 , the system first develops small mi plateaus at lower integer fillings @xmath101 ( dashed curves ) , and then the @xmath102 one disappears while the @xmath103 one expands , as well as higher integer plateaus @xmath104 develop ( dotted curve ) . finally only the even - integer mi plateaus @xmath105 survive at large @xmath1 ( dot - dashed curve ) . the first development of the integer plateaus is because a local domain with integer occupation costs less parity energy than that with fractional one , while the later survival of only the even plateaus is attributable to a boundary behavior , minimizing the interphase energy between plateaus , plus the fact that the density always decreases to zero ( even parity ) at the edge of the system . we notice that the 2d profile exhibits a rotational symmetry , so at large @xmath1 it does look like a wedding cake " but with the adjacent mi plateaus differing by two units . obtained from the quantum monte carlo ( qmc ) method at various @xmath1 . ( a ) one - dimensional ( 1d ) profiles from the trap center to the edge , at @xmath106 , @xmath107 , @xmath108 , and @xmath109 . the model parameters are @xmath110 , @xmath111 , @xmath112 , where @xmath113 is the lattice spacing , and the temperature @xmath114 . ( b ) two - dimensional ( 2d ) profiles along the diagonal of a square lattice , from the trap center to the edge at @xmath106 , @xmath107 , @xmath115 , and @xmath116 . the model parameters are @xmath117 , @xmath111 , @xmath118 , and @xmath114 . in both panels the data are presented in solid , dashed , dotted , and dot - dashed curves respectively for increasing @xmath1.,width=257 ] , @xmath108 , @xmath109 , and @xmath119 . ( b ) 2d diagonal profiles at @xmath106 , @xmath120 , @xmath121 , and @xmath108 . data are presented in the same convention as in fig . [ fig : f03 ] and the model parameters are also the same except for @xmath122.,width=257 ] and ( b ) the corresponding density profile along the diagonal in 2d trapped lattices at @xmath106 , @xmath108 , @xmath109 , and @xmath119 ( solid , dashed , dotted , and dot - dashed curves respectively ) . data are obtained using gmf with the model parameters being @xmath123 , @xmath124 , @xmath118 , and @xmath122.,width=257 ] to compare with the qmc results , we use a gutzwiller mean - field ( gmf ) method@xcite that has also been widely applied to a variety of lattice - boson systems@xcite . the key assumption of the gmf method is to treat the ground state of the system as a variational product of all single - site states , @xmath125 } \left| { { \rm{vac } } } \right\rangle,\label{eqn : gmf}\end{aligned}\ ] ] where @xmath126 is the upper bound of site occupation , @xmath127 is the amplitude of @xmath2 occupation on site @xmath8 , and @xmath128 denotes the vacuum state . we obtain the solutions of @xmath129 by minimizing the energy functional @xmath130 , subject to the normalization constraint @xmath131 . here @xmath126 is chosen large enough such that we observe convergence of the lowest energy state within the gutzwiller assumption . figures [ fig : f04](a ) and [ fig : f04](b ) show 1d and 2d density profiles obtained from gmf , respectively . we see the same four - stage evolution from the superfluid to the wedding - cake structure of even - occupation mis , qualitatively in agreement with the qmc results . in addition , we investigate a 2d case with larger @xmath13 , in which the system is always a superfluid ( @xmath132 everywhere as long as @xmath133 ) in a range of @xmath1 of interest . we find that with the increase in @xmath1 , the outer superfluid shell develops a strong parity _ preference _ , @xmath134 , while the inner superfluid core remains the same state ( @xmath135 ) as in the @xmath14 case . this parity preference indicates a pair - doped nature analogous to the pl state in the @xmath136 limit ( discussed in sec . [ sec : phase ] ) , while the concurrence of the superfluid order suggests the itineracy of particles as pairs , reminiscent of pair superfluids in multi - species lattice bosons@xcite or dimer superfluids in bh systems with attractive interactions@xcite . we identify such pair - superfluid state by a combined order parameter , @xmath137 notice that the sign of @xmath138 tells the pair superfluid s parity being even ( @xmath139 ) or odd ( @xmath140 ) . in fig . [ fig : f05 ] we plot profiles of @xmath138 [ [ fig : f05](a ) ] and @xmath141 [ [ fig : f05](b ) ] obtained from gmf calculations at a set of increasing @xmath1 values and @xmath13 five times larger than that in fig . [ fig : f04](b ) . at @xmath14 , @xmath138 shows no parity preference in the central region , due to a smooth occupation distribution @xmath142 , and slowly alternates between positive ( even ) and negative ( odd ) when the density becomes dilute near the edge ( because the occupation amplitudes @xmath129 are asymmetrically distributed in @xmath2 , with a natural cutoff by @xmath143 ) . as @xmath1 increases , @xmath138 develops a strong preference of even parity in the outer shell region ( pair superfluid ) , leaving the inner core still parity undefined ( regular superfluid ) . we see that the density of the outer shell drops much faster when it becomes a pair superfluid , indicating that the pair superfluid has higher compressibility ( @xmath144 ) than the regular superfluid . we point out that the pair superfluid here results from the co - tunneling of two particles to avoid the flipping of site s occupation parities . the mechanism is different from the pair superfluids in other systems@xcite , in which two particles are directly bound by attractive interactions . in this paper , we studied an extended bose - hubbard model incorporating a ferromagneticlike coupling between the nearest - neighbor site - occupation parities . this parity coupling generates domain - wall energies that compete with the single - particle tunneling and favors pair doping rather than the single - particle one that is favored instead by the on - site repulsion . the interplay between the parity coupling and on - site repulsion leads to a rich phase diagram of a uniform system at zero tunneling , characterized by ( 1 ) mott - insulator ( mi ) phases at each commensurate filling and ( 2 ) phase separation of two mis when the on - site interaction dominates or ( 3 ) pair - liquid phases when the parity coupling dominates at incommensurate fillings . in the finite - tunneling inhomogeneous case , we have obtained both quantum monte carlo and gutzwiller mean - field results that agreeably show an evolution of a trapped system from a superfluid to a wedding - cake - structure of mis with the same parity ( or the occupation jumping by 2 ) , as the parity coupling increases . in a relatively large tunneling and large - parity coupling regime , the trapped system exhibits a structure with the inner bulk being a single - particle superfluid as usual and the outer shell being an exotic pair superfluid . considering the realization of our model , we have shown that it effectively describes the behavior of the umfs in a system of imbalanced fermionic superfluids in 2d optical lattices of tubes , which has been realized@xcite and is being studied in ongoing experiments@xcite . the transition between a superfluid and mis of the same parity in our model corresponds to that between compressible and incompressible states of umfs in the tubular system with a strong oscillatory pairing order@xcite . such transition can be detected by the response of the umfs to optical - lattice modulation@xcite or by the momentum distribution of umfs in a time - of - flight experiment @xcite . the pair - superfluid state in our model corresponds to the compressible state of itinerant umf pairs in the tubular system , reminiscent of a triplet - pair superfluid . in addition , a truncated version of our model can be mapped to an anisotropic bilinear - biquadratic spin model , in which the occupation - parity coupling is associated with biquadratic spin coupling . such mapping provides a potential realization for exploring our model in spin systems and would also stimulate interest in the physics of these two models . we are grateful to r. g. hulet , michael ma and matthias troyer for useful discussions . the qmc calculations have been performed using the sse algorithm provided by the alps project@xcite . this work is supported by darpa - aro award no . w911nf-07 - 1 - 0464 and by the university of cincinnati . furthermore , c.j.b . would like to acknowledge the hospitality of the aspen center for physics , supported by the nsf under grant no . phys-1066293 . in this appendix we derive the hamiltonian of eq . ( [ eqn : ham0 ] ) as an effective model describing the behavior of umfs in a system of an imbalanced fermionic superfluid in a 2d optical lattice of 1d tubes . the hamiltonian of the tube system comprises three parts@xcite , @xmath145 where the @xmath146 direction is along the tubes axis and @xmath147 denotes tubes indexed in the lattice plane perpendicular to @xmath146 . the operator @xmath148 creates a majority ( minority ) spin at position @xmath37 on tube @xmath149 . the first part @xmath150 contains the kinetic energy in the @xmath146 direction , the hartree field @xmath0 and the spin - dependent chemical potential @xmath151 ( we assume homogeneity in the @xmath152 direction ) , the second part @xmath153 includes the superfluid pairing order @xmath154 , while the third part @xmath155 models the tunneling between nearest - neighbor tubes @xmath156 . the tunneling field @xmath157 describes the effective single - particle tunneling under the influence of the pairing state of the surrounding tubes@xcite . the three coupling fields @xmath0 , @xmath154 , and @xmath157 are introduced after employing a bogoliubove de gennes ( bdg ) treatment@xcite on the original hamiltonian of a quartic form ; which includes , besides the single - particle hamiltonian , a two - body point - contact interaction and ( possible ) pair tunneling . the hamiltonian @xmath158 can be solved in a quasiparticle basis , resulting in a set of @xmath159 that self - consistently describes the equilibrium state of the system ( see detailed discussions in refs . [ ] and [ ] as well as a wide application of this treatment on similar systems @xcite ) . the self - consistent solutions show that in the presence of spin imbalance , the system can accommodate an oscillatory pairing order @xmath154 along the tube axis ( the fflo state@xcite ) . for a single tube or uncoupled tubes , each umf is localized around a node of the pairing order parameter such that the total numbers of nodes and umfs are equal . given the spatially symmetric confinement along the tube axis , e.g. , a harmonic trap , a tube with odd ( even ) imbalance has a pairing node ( antinode ) at the tube center . in particular , the localized umfs can be regarded as occupied andreev bound states@xcite , resulting in an imbalance profile like separated solitons@xcite if the spatial distribution of the pairing nodes is relatively sparse or like a ripple@xcite if it is dense . ( the same signatures have also been found in isotropic three - dimensional hubbard lattices@xcite or coupled hubbard chains@xcite . ) both configurations arise from the same effect of localization and can be characterized as the spatial concurrence of imbalance peaks and pairing nodes . the localization of umfs suppresses the degrees of freedom in the @xmath146 direction and hence allows us to construct an effective 2d lattice model describing the physics of umfs with only the @xmath160 degree of freedom . we remark that even if umfs localize along the @xmath146 direction , they can still be mobile in the@xmath160 direction , constituting the physics of interests as discussed below . in such a 2d model the state in the @xmath146 direction is taken as a hidden degree of freedom of a lattice site . ( this idea has also been applied on bosonic gases in tubular lattices@xcite , in which the multiorbital wave functions in the tube direction are incorporated into on - site degrees of freedom for an effective 2d model across the tubes . ) the hidden degrees of freedom here can be directly related in one - to - one correspondence to the site occupation number , and since more than single occupation per site are possible , this suggests a bosonic description for the modeling . formally , one can still keep the anticommutation relation in the @xmath160 direction by using a combined operator @xmath161 where @xmath162 is a bosonic operator accounting for occupation - related physics and @xmath163 is a majorana fermion operator carrying the statistical property of the original fermions . based on this decoupling , we build our model step by step below . first , for coupled tubes , the pairing nodes tend to line up at the same @xmath37 positions on each tube@xcite . the leading kinetics across the tubes is the transverse tunneling of an umf from a pairing node in a tube to the corresponding one in the nearest neighbor . the one - particle tunneling of minority fermions is suppressed provided they are all paired ( at low temperature ) . because a node is surrounded by other nodes where the pairing is zero , the tunneling field reduces to a free particle tunneling strength@xcite , @xmath164 . therefore , considering the order - parameter configuration in @xmath153 that limits the kinetics in @xmath155 to be only for the umfs , we obtain the tunneling term in the model as @xmath165 second , if the coupling between the tubes is weak enough , the on - site energy is mainly contributed by the on - tube kinetics , hartree energy , pairing energy , and chemical potentials , all of which are determined once the on - site filling ( of umfs ) is given . the self - consistent calculation in ref . [ ] shows that the on - site energy is nonlinear and can be fitted by a repulsive interaction @xmath0 plus a chemical potential @xmath151 . as a result , the on - tube energies @xmath150 and @xmath153 effectively become the on - site terms of our model as @xmath166 where @xmath167 . third , the bdg calculations show that the system energy rises if two nearest - neighbor tubes have different spatial parities of the oscillatory pairing order parameter . such energy increase comes from a drastic mismatch in the order parameters and can be regarded as a cost of domain - wall formation , exhibiting similar physics to that in superconducting @xmath168 junctions@xcite or weakly imbalanced cold atomic gases@xcite . because the spatial parity of the order parameter is equal to the occupation parity of the umfs , one can describe the domain - wall physics with the occupation - parity coupling between nearest - neighbor sites in the bh model , @xmath169 where @xmath170 is the onsite parity operator and @xmath1 is the unit - length energy of domain walls . finally , by collecting all terms in eqs . ( [ eqn : ham2a])([eqn : ham2c ] ) we obtain an effective hamiltonian , @xmath171 } \nonumber\\ & & + \sum\limits_{\bf{r } } { \left [ { \frac{u}{2}{{\hat n}_{\bf{r}}}({{\hat n}_{\bf{r } } } - 1 ) - \mu { { \hat n}_{\bf{r } } } } \right]}.\label{eqn : ham2}\end{aligned}\ ] ] because the physical quantities of interest to be addressed here can be expressed as local static ( or equal - time ) correlations of the ground state , not involving the exchange of particles , the statistics ( majorana sector ) factors out . therefore , by replacing @xmath172 with @xmath173 ( and allowing the chemical potential to vary in space ) , we arrive at the hamiltonian of eq . ( [ eqn : ham0 ] ) . considering a typical experimental setup@xcite , we obtain @xmath174 , @xmath175@xmath176 in the range of optical lattice depth being 123@xmath177 , and @xmath178 ( regime of the wedding - cake structure of even - filling mis ) , where @xmath179 is the binding energy of a pair of fermions and @xmath177 is the recoil energy . one can also expect a high tunability of @xmath0 and @xmath1 via the tuning of the feshbach - resonant@xcite interactions and the overall trapping potential used in the experiments . j. hubbard , http://dx.doi.org/10.1098/rspa.1963.0204[proc . a * 276 * , 238 ( 1963 ) ] . d. belitz and t. r. kirkpatrick , http://dx.doi.org/10.1103/revmodphys.66.261[rev . phys . * 66 * , 261 ( 1994 ) ] . e. dagotto , http://dx.doi.org/10.1103/revmodphys.66.763[rev . * 66 * , 763 ( 1994 ) ] . m. imada , a. fujimori , and y. tokura , http://dx.doi.org/10.1103/revmodphys.70.1039[rev . 70 * , 1039 ( 1998 ) ] . lee , n. nagaosa , and x .- wen , http://dx.doi.org/10.1103/revmodphys.78.17[rev . mod . phys . * 78 * , 17 ( 2006 ) ] . m. p. a. fisher , p. b. weichman , g. grinstein , and d. s. fisher , http://dx.doi.org/10.1103/physrevb.40.546[phys . b * 40 * , 546 ( 1989 ) ] . s. sachdev , _ quantum phase transitions _ , ( cambridge univ . press , cambridge , 1999 ) . h. s. j. van der zant , f. c. fritschy , w. j. elion , l. j. geerligs , and j. e. mooij , http://dx.doi.org/10.1103/physrevlett.69.2971[phys . . lett . * 69 * , 2971 ( 1992 ) ] . a. van oudenaarden and j. e. mooij , http://dx.doi.org/10.1103/physrevlett.76.4947[phys . lett . * 76 * , 4947 ( 1996 ) ] . r. m. dimeo , p. e. sokol , c. r. anderson , w. g. stirling , k. h. andersen , and m. a. adams , http://dx.doi.org/10.1103/physrevlett.81.5860[phys . lett . * 81 * , 5860 ( 1998 ) ] . o. plantevin , b. fk , h. r. glyde , n. mulders , j. bossy , g. coddens , and h. schober , http://dx.doi.org/10.1103/physrevb.63.224508[phys . rev . b * 63 * , 224508 ( 2001 ) ] . m. j. hartmann , f. g. s. l. brando , and m. b. plenio , http://dx.doi.org/10.1038/nphys462[nat . * 2 * , 849 ( 2006 ) ] . a. d. greentree , c. tahan , j. h. cole , and l. c. l. hollenberg , http://dx.doi.org/10.1038/nphys466[nat . phys . * 2 * , 856 ( 2006 ) ] . c. w. lai , n. y. kim , s. utsunomiya , g. roumpos , h. deng , m. d. fraser , t. byrnes , p. recher , n. kumada , t. fujisawa , and y. yamamoto , http://dx.doi.org/10.1038/nature06334[nature ( london ) * 450 * , 529 ( 2007 ) ] . t. byrnes , p. recher , and y. yamamoto , http://dx.doi.org/10.1103/physrevb.81.205312[phys . b * 81 * , 205312 ( 2010 ) ] . m. greiner , o. mandel , t. esslinger , t. w. hansch , and i. bloch , http://dx.doi.org/10.1038/415039a[nature ( london ) * 415 * , 39 ( 2002 ) ] . g. k. campbell , j. mun , m. boyd , p. medley , a. e. leanhardt , l. g. marcassa , d. e. pritchard , and w. ketterle , http://dx.doi.org/10.1126/science.1130365[science * 313 * , 649 ( 2006 ) ] . d. jaksch and p. zoller , http://dx.doi.org/10.1016/j.aop.2004.09.010[ann . phys . * 315 * , 52 ( 2005 ) ] . o. morsch and m. oberthaler , http://dx.doi.org/10.1103/revmodphys.78.179[rev . phys . * 78 * , 179 ( 2006 ) ] . i. bloch , j. dalibard , and w. zwerger , http://dx.doi.org/10.1103/revmodphys.80.885[rev . mod . phys . * 80 * , 885 ( 2008 ) ] . m. a. cazalilla , r. citro , t. giamarchi , e. orignac , and m. rigol , http://dx.doi.org/10.1103/revmodphys.83.1405[rev . phys . * 83 * , 1405 ( 2011 ) ] . n. marzari , a. a. mostofi , j. r. yates , i. souza , and d. vanderbilt , http://dx.doi.org/10.1103/revmodphys.84.1419[rev . 84 * , 1419 ( 2012 ) ] . p. fulde and r. a. ferrell , http://dx.doi.org/10.1103/physrev.135.a550[phys . rev . * 135 * , a550 ( 1964 ) ] . a. i. larkin and yu . n. ovchinnikov , zh . eksp . teor . fiz . * 47 * , 1136 ( 1964 ) [ sov . jetp * 20 * , 762 ( 1965 ) ] . s. giorgini , l. p. pitaevskii , and s. stringari , http://dx.doi.org/10.1103/revmodphys.80.1215[rev . phys . * 80 * , 1215 ( 2008 ) ] . l. radzihovsky and d. e. sheehy , http://dx.doi.org/10.1088/0034-4885/73/7/076501[rep . prog . phys . * 73 * , 076501 ( 2010 ) ] . et al . _ , http://dx.doi.org/10.1143/jpsjs.74s.30[j . jpn . * 74 * , 30 ( 2005 ) ] . a. f. albuquerque _ , http://dx.doi.org/10.1016/j.jmmm.2006.10.304[j . magn . mater . * 310 * , 1187 ( 2007 ) ] . et al . _ , http://dx.doi.org/10.1088/1742-5468/2011/05/p05001[j ( * 2011 * ) , p05001 ] . c. trefzger , c. menotti , and m. lewenstein , http://dx.doi.org/10.1103/physrevlett.103.035304[phys . * 103 * , 035304 ( 2009 ) ] . a. hu , l. mathey , i. danshita , e. tiesinga , c. j. williams , and c. w. clark , http://dx.doi.org/10.1103/physreva.80.023619[phys . a * 80 * , 023619 ( 2009 ) ] . c. menotti and s. stringari , http://dx.doi.org/10.1103/physreva.81.045604[phys . a * 81 * , 045604 ( 2010 ) ] . s. diehl , m. baranov , a. j. daley , and p. zoller , http://dx.doi.org/10.1103/physrevlett.104.165301[phys . 104 , 165301 ( 2010 ) ] ; http://dx.doi.org/10.1103/physrevb.82.064509[phys . rev . b 82 , 064509 ( 2010 ) ] ; http://dx.doi.org/10.1103/physrevb.82.064510[82 , 064510 ( 2010 ) ] . l. bonnes and s. wessel , http://dx.doi.org/10.1103/physrevlett.106.185302[phys . 106 * , 185302 ( 2011 ) ] . k . ng and m .- f . yang , http://dx.doi.org/10.1103/physrevb.83.100511[phys . b * 83 * , 100511(r ) ( 2011 ) ] . y .- c . chen , k .- k . ng , and m .- f . yang , http://dx.doi.org/10.1103/physrevb.84.092503[phys . b * 84 * , 092503 ( 2011 ) ] . h. moritz , t. stferle , k. genter , m. khl , and t. esslinger , http://dx.doi.org/10.1103/physrevlett.94.210401[phys . * 94 * , 210401 ( 2005 ) ] . m. aidelsburger , m. atala , s. nascimbne , s. trotzky , y .- a . chen , and i. bloch , http://dx.doi.org/10.1103/physrevlett.107.255301[phys . lett . * 107 * , 255301 ( 2011 ) ] . m. n. barber and m. t. batchelor , http://dx.doi.org/10.1103/physrevb.40.4621[phys . b * 40 * , 4621 ( 1989 ) ] . g. fth and j. slyom , http://dx.doi.org/10.1103/physrevb.44.11836[phys . b * 44 * , 11836 ( 1991 ) ] ; http://dx.doi.org/10.1103/physrevb.47.872[*47 * , 872 ( 1993 ) ] ; http://dx.doi.org/10.1103/physrevb.51.3620[*51 * , 3620 ( 1995 ) ] . b. a. ivanov and a. k. kolezhuk , http://dx.doi.org/10.1103/physrevb.68.052401[phys . rev . b * 68 * , 052401 ( 2003 ) ] . . legeza and j. slyom , http://dx.doi.org/10.1103/physrevlett.96.116401[phys . * 96 * , 116401 ( 2006 ) ] . a. luchli , g. schmid , and simon trebst , http://dx.doi.org/10.1103/physrevb.74.144426[phys . b * 74 * , 144426 ( 2006 ) ] . z .- liu , yi zhou , h .- h . wen , and t .- k . ng , http://dx.doi.org/10.1103/physrevb.85.195144[phys . b * 85 * , 195144 ( 2012 ) ] . y. a. fridman , o. a. kosmacheva , p. n. klevetsa , http://dx.doi.org/10.1016/j.jmmm.2012.08.027[j . of magn . magn . mater . * 325 * , 125 ( 2013 ) ] . s. k. yip , http://dx.doi.org/10.1103/physrevlett.90.250402[phys . lett . * 90 * , 250402 ( 2003 ) ] . j. j. garci - ripoll , m. a. martin - delgado , and j. i. cirac , http://dx.doi.org/10.1103/physrevlett.93.250405[phys lett . * 93 * , 250405 ( 2004 ) ] . f. d. m. haldane , http://dx.doi.org/10.1016/0375-9601(83)90631-x[phys . lett . * 93a * , 464 ( 1983 ) ] ; http://dx.doi.org/10.1103/physrevlett.50.1153[phys . lett . * 50 * , 1153 ( 1983 ) ] . h. kadowaki , k. ubukoshi and k. hirakawa , http://dx.doi.org/10.1143/jpsj.56.751[j . . jpn . * 56 * , 751 ( 1987 ) ] . i. affleck , http://dx.doi.org/10.1088/0953-8984/1/19/001[j . phys . : condens . matter * 1 * , 3047 ( 1989 ) ] . o. gygi , h. g. katzgraber , m. troyer , s. wessel , and g. g. batrouni , http://dx.doi.org/10.1103/physreva.73.063606[phys . a * 73 * , 063606 ( 2006 ) ] . c. kollath , a. m. luchli , and e. altman , http://dx.doi.org/10.1103/physrevlett.98.180601[phys . lett . * 98 * , 180601 ( 2007 ) ] . s. wessel , http://dx.doi.org/10.1103/physrevb.75.174301[phys . b * 75 * , 174301 ( 2007 ) ] ; http://dx.doi.org/10.1016/j.cpc.2007.02.008[comput . . commun . * 177 * , 166 ( 2007 ) ] ; http://dx.doi.org/10.1103/physrevb.78.075112[phys . b * 78 * , 075112 ( 2008 ) ] . s. guertler , m. troyer , and f .- c . zhang , http://dx.doi.org/10.1103/physrevb.77.184505[phys . b * 77 * , 184505 ( 2008 ) ] . k. p. schmidt , julien dorier , and andreas m. luchli , http://dx.doi.org/10.1103/physrevlett.101.150405[phys . * 101 * , 150405 ( 2008 ) ] . k. yamamoto , s. todo , and s. miyashita , http://dx.doi.org/10.1103/physrevb.79.094503[phys . b * 79 * , 094503 ( 2009 ) ] . a. kalz , a. honecker , s. fuchs , and t. pruschke , http://dx.doi.org/10.1103/physrevb.83.174519[phys . b * 83 * , 174519 ( 2011 ) ] . e. arrigoni , m. knap , and w. von der linden , http://dx.doi.org/10.1103/physrevb.84.014535[phys . b * 84 * , 014535 ( 2011 ) ] . a. albus , f. illuminati , and j. eisert , http://dx.doi.org/10.1103/physreva.68.023606[phys . a * 68 * , 023606 ( 2003 ) ] . l. pollet , s. rombouts , k. heyde , and j. dukelsky , http://dx.doi.org/10.1103/physreva.69.043601[phys . a * 69 * , 043601 ( 2004 ) ] . t. kimura , s. tsuchiya , and s. kurihara , http://dx.doi.org/10.1103/physrevlett.94.110403[phys . * 94 * , 110403 ( 2005 ) ] . v. w. scarola and s. das sarma , http://dx.doi.org/10.1103/physrevlett.95.033003[phys . lett . * 95 * , 033003 ( 2005 ) ] ; http://dx.doi.org/10.1103/physrevlett.98.210403[phys . 98 * , 210403 ( 2007 ) ] . t. miyakawa and p. meystre , http://dx.doi.org/10.1103/physreva.74.043615[phys . a * 74 * , 043615 ( 2006 ) ] . p. buonsante , f. massel , v. penna , and a. vezzani , http://dx.doi.org/10.1103/physreva.79.013623[phys . a * 79 * , 013623 ( 2009 ) ] . m. yamashita and m. w. jack , http://dx.doi.org/10.1103/physreva.79.023609[phys . a * 79 * , 023609 ( 2009 ) ] . k. sun , c. lannert , and s. vishveshwara , http://dx.doi.org/10.1103/physreva.79.043422[phys . a * 79 * , 043422 ( 2009 ) ] . m. iskin , http://dx.doi.org/10.1103/physreva.83.051606[phys . a * 83 * , 051606 ( 2011 ) ] . t. kimura , http://dx.doi.org/10.1103/physreva.84.063630[phys . a * 84 * , 063630 ( 2011 ) ] . t. saito , i danshita , t. ozaki , and t. nikuni , http://dx.doi.org/10.1103/physreva.86.023623[phys . a * 86 * , 023623 ( 2012 ) ] . d. pekker , b. wunsch , t. kitagawa , e. manousakis , a. s. sorensen , and e. demler , http://dx.doi.org/10.1103/physrevb.86.144527[phys . b * 86 * , 144527 ( 2012 ) ] . lin , r. sensarma , k. sengupta , and s. das sarma , http://dx.doi.org/10.1103/physrevb.86.214207[phys . b * 86 * , 214207 ( 2012 ) ] . lhmann , http://dx.doi.org/10.1103/physreva.87.043619[phys . a * 87 * , 043619 ( 2013 ) ] . m. yamashita , s. kato , a. yamaguchi , s. sugawa , t. fukuhara , s. uetake , and y. takahashi , http://dx.doi.org/10.1103/physreva.87.041604[phys . a * 87 * , 041604 ( 2013 ) ] . p. buonsante , l. orefice , and a. smerzi , http://dx.doi.org/10.1103/physreva.87.063620[phys . a * 87 * , 063620 ( 2013 ) ] . t. stferle , h. moritz , c. schori , m. khl , and t. esslinger , http://dx.doi.org/10.1103/physrevlett.92.130403[phys . * 92 * , 130403 ( 2004 ) ] . r. jrdens , n. strohmaier , k. gnter , h. moritz , and t. esslinger , http://dx.doi.org/10.1038/nature07244[nature ( london ) * 455 * , 204 ( 2008 ) ] . t. kinoshita , t. wenger , and d. s. weiss , http://dx.doi.org/10.1126/science.1100700[science * 305 * , 1125 ( 2004 ) ] ; http://dx.doi.org/10.1038/nature04693[nature ( london ) * 440 * , 900 ( 2006 ) ] . m. swanson , y. l. loh , and n. trivedi , http://dx.doi.org/10.1088/1367-2630/14/3/033036[new j. phys . * 14 * , 033036 ( 2012 ) ] . hong lu , l. o. baksmaty , c. j. bolech , and han pu , http://dx.doi.org/10.1103/physrevlett.108.225302[phys . * 108 * , 225302 ( 2012 ) ] . c. j. bolech , f. heidrich - meisner , s. langer , i. p. mcculloch , g. orso , and m. rigol , http://dx.doi.org/10.1103/physrevlett.109.110602[phys . lett . * 109 * , 110602 ( 2012 ) ] . p. g. de gennes , _ superconductivity of metals and alloys _ ( addison - wesley , reading , ma , 1989 ) . t. mizushima , k. machida , and m. ichioka , http://dx.doi.org/10.1103/physrevlett.94.060404[phys . lett . * 94 * , 060404 ( 2005 ) ] . m. m. parish , s. k. baur , e. j. mueller , and d. a. huse , http://dx.doi.org/10.1103/physrevlett.99.250403[phys . lett . * 99 * , 250403 ( 2007 ) ] . x .- ji liu , h. hu , and p. d. drummond , http://dx.doi.org/10.1103/physreva.76.043605[phys . a * 76 * , 043605 ( 2007 ) ] ; http://dx.doi.org/10.1103/physreva.78.023601[phys . a * 78 * , 023601 ( 2008 ) ] . l. o. baksmaty , h. lu , c. j. bolech , and h. pu , http://dx.doi.org/10.1103/physreva.83.023604[phys . a * 83 * , 023604 ( 2011 ) ] ; http://dx.doi.org/10.1088/1367-2630/13/5/055014[new j. phys . * 13 * , 055014 ( 2011 ) ] . k. sun , j. s. meyer , d. e. sheehy , and s. vishveshwara , http://dx.doi.org/10.1103/physreva.83.033608[phys . a * 83 * , 033608 ( 2011 ) ] . n. yoshida and s .- k . yip , http://dx.doi.org/10.1103/physreva.75.063601[phys . a * 75 * , 063601 ( 2007 ) ] . s. brazovskii , http://dx.doi.org/10.1016/j.physb.2008.11.236[physica , b * 404 * , 482 ( 2009 ) ] . r. m. lutchyn , m. dzero , and v. m. yakovenko , http://dx.doi.org/10.1103/physreva.84.033609[phys . a * 84 * , 033609 ( 2011 ) ] . y. l. loh and n. trivedi , http://dx.doi.org/10.1103/physrevlett.104.165302[phys . lett . * 104 * , 165302 ( 2010 ) ] . kim and p. trm , http://dx.doi.org/10.1103/physrevb.85.180508[phys . b * 85 * , 180508 ( 2012 ) ] . g. wirth , m. lschlger , and a. hemmerich , http://dx.doi.org/10.1038/nphys1857[nat . * 7 * , 147 ( 2011 ) ] . x. li , e. zhao , and w. v. liu , http://dx.doi.org/10.1038/ncomms2523[nat . commun . * 4 * , 1523 ( 2013 ) ] . m. lschlger , t. kock , g. wirth , a. ewerbeck , c. morais smith , and a. hemmerich , http://dx.doi.org/10.1088/1367-2630/15/8/083041[new j. phys . * 15 * , 083041 ( 2013 ) ] . f. hbert , zi cai , v. g. rousseau , c. wu , r. t. scalettar , and g. g. batrouni , http://dx.doi.org/10.1103/physrevb.87.224505[phys . b * 87 * , 224505 ( 2013 ) ] . x. li , a. paramekanti , a. hemmerich , and w. v. liu , http://dx.doi.org/10.1038/ncomms4205[nat . commun . * 5 * , 3205 ( 2014 ) ] .

accurate experimental values for the vibrational ground tone or fundamental vibrational energy splitting of h , hd , and d are presented . absolute accuracies of are obtained from doppler - free laser spectroscopy applied in a collisionless environment . the vibrational splitting frequencies are derived from the combination difference between separate electronic excitations from the and vibrational states to a common state . the present work on rotational quantum states extends the results reported by dickenson _ et al . _ on [ phys . rev . lett . 110 ( 2013 ) 193601 ] . the experimental procedures leading to this high accuracy are discussed in detail . a comparison is made with full _ ab initio _ calculations encompassing born - oppenheimer energies , adiabatic and non - adiabatic corrections , as well as relativistic corrections and qed - contributions . the present agreement between the experimental results and the calculations provides a stringent test on the application of quantum electrodynamics in molecules . furthermore , the combined experimental - theoretical uncertainty can be interpreted to provide bounds to new interactions beyond the standard model of physics or _ fifth forces _ between hadrons . astron . astroph . molecular hydrogen , fundamental vibration , uv spectroscopy , test of qed

1404.0184

the first modern explanation of the chemical bond between two neutral hydrogen atoms was put forth in 1927 by heitler and london @xcite , and is one of the earliest applications of quantum theory , specifically that of schrdinger s wave mechanics formulation in 1926 @xcite . heitler and london showed that by accounting for pauli s exclusion principle @xcite in combining atomic hydrogen wavefunctions to construct molecular wavefunctions , the existence of a bound molecular state is explained . despite their calculated binding energy being off by some 30% from the contemporary experimental value , their pioneering quantum mechanical calculation for the stability of molecular hydrogen ushered the era of quantum chemistry . it is interesting to note that the born - oppenheimer approximation @xcite was also proposed in 1927 , and this approach of separating electronic and nuclear motions has largely shaped molecular theory since . the next breakthrough in _ ab initio _ potential calculations for h@xmath0 was achieved by james and coolidge in 1933 in their treatment of the ( @xmath8 ) ground state @xcite . using two - electron wave functions with explicitly correlated electrons , an approach introduced by hylleraas for the helium atom @xcite , they transcended the concept of electrons being in individual states as used in the hartree - fock method . the james - coolidge solution relied on the variational method to determine the correct nonlinear parameters in combining the wave functions . with a set of only 13 of these wave functions , taken as a truncated basis to represent the total hilbert space of infinite dimension , they improved the minimum energy in the born - oppenheimer potential of the @xmath9 state to @xmath10 @xmath2 . this was a substantial improvement of about @xmath11 @xmath2 with respect to the best theoretical values available at the time . over the years improvements on the accuracy has been obtained @xcite , and important methodical reviews can be found in refs . the achievement of the initial studies of james and coolidge @xcite can best be appreciated considering that further improvement in the calculated potential has been only 222 @xmath2 since then , obtained by wolniewicz in 1995 with essentially the same method but with a basis of 883 wave functions @xcite . at present , the born - oppenheimer potential energy can be evaluated to accuracies better than 15 digits using more than 22,000 basis functions @xcite , made possible by developments in numerical procedures and improvements in computing power . the precision of the calculated born - oppenheimer energy may be considered exact for the purpose of comparisons with experiment . corrections beyond the born - oppenheimer approximation need to be evaluated to improve upon the accuracy of the _ ab initio _ values . in addition to adiabatic and nonadiabatic effects comprising the non - relativistic born - oppenheimer corrections , it is also necessary to account for accurate relativistic and radiative or quantum electrodynamic ( qed ) corrections . until up to 2010 , the work of wolniewicz @xcite that included estimates of radiative corrections , had constituted the state - of - the - art for calculations of level energies in the @xmath8 ground state of molecular hydrogen . this led to a calculated energy of the actual ground state ( or equivalently the dissociation limit ) to an accuracy 0.01 @xmath2 . the recent work of pachucki , komasa and co - workers has achieved breakthroughs in the evaluation of nonadiabatic effects @xcite as well as relativistic and radiative corrections @xcite , resulting in accurate level energies of @xmath8 rovibrational levels @xcite . in fig . [ energycontribution ] the different contributions to the level energy of the lowest quantum state ( @xmath12 ) with respect to the dissociation energy of molecular hydrogen are represented graphically to give an impression of the scale of the corrections . ) as corrections to the born - oppenheimer approximation level energy , with respect to the dissociation limit , of the @xmath13 , @xmath14 state for h@xmath15 , hd and d@xmath15 . bo : born - oppenheimer energy ; ad : adiabatic ; nad : nonadiabatic ; rel : relativistic ; qed : radiative corrections . ] theoretical and experimental efforts on the determination of ground state level energies in molecular hydrogen mutually stimulated improvements on both fronts as soon as more accurate values were obtained . as an illustration , consider the dissociation energy of the @xmath8 ground state , a benchmark quantity for the comparison of experiment and theory . the experimental determination of the dissociation energy by witmer @xcite in 1926 already gave results within 3@xmath16 of the modern value , an order of magnitude better than the heitler - london calculations as mentioned . the james - coolidge calculations in 1933 @xcite resulted in a dissociation energy that is within @xmath17 of the present value , matched later by the experimental determination by beutler in 1935 @xcite that was also accurate to within @xmath17 . this lively dynamics continued through the 1960s-1970s , between the experimental efforts of herzberg and co - workers @xcite and theoretical efforts by koos and co - workers @xcite . in the middle of the 1990s , the theoretical result of wolniewicz @xcite for the dissociation energy was expressed in 8 significant digits , although the uncertainty was not explicitly mentioned . eyler and co - workers determined the dissociation limit to an accuracy of 0.01 @xmath2 @xcite in 2004 , improving upon their previous result @xcite using the same method . the most accurate experimental dissociation energy for h@xmath15 was obtained in 2009 by liu _ et al . _ @xcite , and later extended to d@xmath15 @xcite and hd @xcite . remarkably , accurate theoretical values for h@xmath15 and d@xmath15 dissociation energies @xcite as well as hd @xcite were presented a short time thereafter . _ ab initio _ theory can also be tested through a comparison with the experimental determinations of level splittings in the rovibrational manifold of the ground state . herzberg first predicted , in 1938 , that it should be possible to record rovibrational transitions in the ground state manifold @xcite , and later discovered the quadrupole spectrum in 1949 by photographing a total of eight lines in the ( 2,0 ) and ( 3,0 ) bands @xcite . subsequently the quadrupole spectrum including the fundamental ( 1,0 ) band was investigated by several other groups , for example by rank and co - workers @xcite . the measurements by bragg _ et al . _ @xcite greatly improved the accuracy of the spectroscopy of the quadrupole bands and was until recent years considered as the most accurate work on the direct measurement of the vibrational splittings . laser - based direct excitation of the weaker ( 4,0 ) and ( 5,0 ) overtone quadrupole bands was performed in the visible domain @xcite . later investigations using cavity - ring down spectroscopy on the h@xmath15 ( 3,0 ) overtone band were carried out by robie _ et al . _ @xcite using a pulsed source and hu _ et al . _ @xcite using a cw source . campargue and co - workers have recently performed high - resolution determinations of the ( 2,0 ) overtone bands of h@xmath15 @xcite and d@xmath15 @xcite using quantum cascade lasers . maddaloni _ et al . _ @xcite performed precision measurements using cavity - ring down techniques for the fundamental band of d@xmath15 . the quadrupole excitations in the ground electronic state described above have very low transition probabilities . the ground state energy splittings can be determined indirectly from appropriate combinations of dipole - allowed transitions between ground state and excited electronic states . for example , the strongest molecular hydrogen transitions in lyman ( @xmath18 @xmath19 ) and werner ( @xmath20@xmath19 ) bands have been used to derive ground state rovibrational constants . using this approach , stanke _ et al . _ @xcite derived accurate ground state molecular constants based largely on the experimental data of dabrowski @xcite but also including quadrupolar transitions . the natural linewidths of transitions in the lyman and werner bands ultimately limit the accuracy that can be achieved @xcite . in contrast , the rovibrational levels of the lowest - lying excited singlet gerade state @xmath21 of molecular hydrogen have longer natural lifetimes , even up to 150 ns @xcite , since one - photon transitions to the ground state are forbidden . the gerade states can be accessed from the ground state through two - photon spectroscopies , which also allow for more accurate level energy determinations . this first excited singlet gerade state in molecular hydrogen , the @xmath22 state , shown to correspond to a double - well potential @xcite , has been investigated thoroughly over the years . eyler and coworkers performed a number of laser spectroscopic studies of increasing accuracy @xcite . a determination of frequencies of q - branch transitions in the lowest @xmath23 ( 0,0 ) band was performed with improved accuracy by hannemann _ et al . _ the lowest rotational levels in the @xmath22 state derived from the latter study were used as anchor lines , to which a large number of levels in the excited state manifold , obtained from high - resolution fourier - transform studies , were connected to the ground state @xcite . accurate values for level energies of the high rotational states up to @xmath24 in the @xmath25 electronic state were obtained in ref . @xcite using uv two - photon spectroscopy . in this paper , we present accurate experimental and theoretical values for the fundamental vibrational splitting of h@xmath15 , d@xmath15 and hd . this extends a recent report @xcite on the _ rotationless _ ground tone frequencies of hydrogen and its isotopomers , now also including values for @xmath26 and @xmath27 levels . the experimental determination of the fundamental vibrational splitting is based on combination differences of the transition frequencies between the @xmath13 and @xmath21 states , measured by two - photon doppler - free spectroscopy . relies on the measurement of q - line transitions the @xmath23 ( 0,1 ) band obtained in the present study and on the measurement of the q - lines in the @xmath23 ( 0,0 ) band , obtained in ref . the excitation channels are indicated by ( 1 ) and ( 0 ) , respectively . an auxiliary laser beam of 355-nm radiation was used in the rempi detection scheme . the squared moduli of the vibrational wavefunctions are indicated in the inset . ] in this study , high - precision uv two - photon spectroscopy is performed on vibrationally excited molecular hydrogen to determine the transition frequencies of the q(0 ) , q(1 ) and q(2 ) lines in the @xmath28 band for all three isotopomers h@xmath0 , hd and d@xmath0 . in combination with the previous determination by hannemann _ et al . _ @xcite for the three q - branch lines in the @xmath29 band , accurate values of the fundamental ground tone splittings are obtained for @xmath30 and @xmath27 rotational levels . the excitation scheme of the present and previous measurements is drawn in fig . [ potentialenergy ] , with the potential energy curves of the relevant electronic states depicted , and the probed two - photon transitions indicated . the experimental setup is schematically shown in fig . [ setup ] , with blocks representing the narrowband laser source , the frequency calibration setup and the molecular beam machine in which the doppler - free spectroscopy is performed . a schematic representation of the laser system is shown as part of figure [ setup ] . the ti : sa pulsed laser system is based on an injection - seeded oscillator - amplifier scheme . a continuous wave ( cw ) ti : sa laser serves as the injection - seed for the pulse generation , and is also used to lock the length of the oscillator cavity by a hnsch - couillaud scheme . the oscillator cavity is pumped with @xmath31 mj of 5-ns pulses of the 532-nm ( second - harmonic ) output from an injection - seeded nd : yag laser operating at 10-hz repetition rate . the resulting pulse from the oscillator cavity is further amplified in a 9-pass bowtie ti : sa amplifier pumped with @xmath32 mj of the same nd : yag pump laser . typical output pulse energies of the amplifier are approximately 45 mj at the fundamental wavelength of 844 nm . the pulse duration of the fundamental ir pulses is @xmath3320 ns corresponding to a fourier - limited bandwidth of @xmath34 mhz . an extensive description of the laser system and its operation can be found in ref . @xcite , while its performance in the measurements on the @xmath23 ( 0,0 ) band were described in ref . a 1-mm diameter pinhole is used as a spatial filter for the output of the ti : sa oscillator - amplifier system before the subsequent harmonic conversion . spatial filtering selects a smaller portion of the beam thereby reducing frequency chirp effects across the beam profile , which is crucial in subsequent frequency calibrations . the spatially - filtered pulsed output is frequency up - converted in two successive frequency - doubling stages ( using @xmath35-barium borate ( bbo ) crystals ) to yield fourth - harmonic uv radiation , with approximately 350 @xmath36j pulse energies at 211 nm . two - photon doppler - free techniques were combined with resonantly enhanced multi - photon ionization ( rempi ) in the spectroscopic experiment . in an isotropic gas sample , two - photon absorption from two counter - propagating laser beams results in the cancellation of first - order doppler shifts . however , the application in a molecular beam with a defined unidirectional trajectory , results in residual first - order doppler shifts if there is a misalignment between the counter - propagating laser beams . to improve the laser beam alignment , an interferometric scheme is implemented @xcite . the uv probe beam is split in two arms of equal intensity and arranged as part of a sagnac interferometer , as depicted in fig . [ setup ] , with the interference fringes indicating the degree of alignment . narrow bandwidth uv radiation at 211 nm was used to probe the two - photon transitions in h@xmath15 , while hd and d@xmath15 measurements required 209 and 207 nm , respectively . a 355-nm laser pulse is used to further ionize the molecules excited in the @xmath21 , with the ionization laser delayed by 30 ns with respect to the probe laser . the h@xmath37 molecular ions are accelerated by electrostatic lenses and then traverse a field - free time - of - flight ( tof ) region . the ions impinge upon a multi - channel plate ( mcp ) detector attached to a phosphor screen , with the resulting fluorescence collected onto a photomultiplier tube ( pmt ) , for signal registration . the preparation of vibrationally excited molecules in @xmath13 @xmath4 is achieved through electron bombardment of the molecular beam in a discharge source similar to that employed in ref . the discharge source is a ceramic pinhole nozzle with metallic electrodes ( depicted in the top left corner of fig . [ setup ] ) , attached to a pulsed solenoid valve ( general valve series 9 ) operated at 10-hz . the molecular beam pulse is discharged by applying a high voltage pulse to the cathode , with the discharge electrons moving upstream against the molecular beam trajectory . the anode is kept at the same potential as the valve orifice to avoid disturbing the valve operation . sufficient population of vibrationally excited molecules was achieved at a voltage of approximately -750 v applied to the cathode . to reduce ions produced in the same discharge from reaching the detection zone , a pair of deflection plates was installed near the end of the discharge nozzle . the vibrationally excited molecular hydrogen passes through a skimmer ( 2-mm diameter ) before entering the interaction zone where the uv spectroscopy takes place . the fundamental frequency @xmath38 is calibrated by referencing part of the light from the ti : sa cw - seed laser to a ( menlo systems m - comb femtosecond fiber ) frequency comb acting as an optical frequency ruler . the carrier - envelope phase offset frequency @xmath39 and repetition frequency @xmath40 of the frequency comb are locked to a local rubidium - clock that is referenced to the global positioning system . a heterodyne beat note @xmath41 is made between the cw - ti : sa laser and the frequency comb modes on an avalanche photodiode , which is counted electronically . in practice , a number of frequency comb modes participate in the heterodyne process contributing to background noise . thus , the frequency comb spectrum is dispersed with a grating and subsequent spatial filtering of the unwanted modes is implemented to increase the signal - to - noise ratio in the @xmath41 measurement . the optical frequency of the uv laser system @xmath42 can be expressed as @xmath43 where the prefactor of 4 accounts for the harmonic order , and @xmath44 is mode number of the frequency comb component used . the cw - pulse frequency offset @xmath45 of the ti : sa system due to frequency chirp will be discussed below . the mode number determination follows from a coarse calibration of the laser using a burleigh wavemeter accurate to @xmath3330 mhz , which is sufficient for an unambiguous mode assignment since @xmath46 mhz . the sign of the respective frequency contributions to @xmath38 may be positive or negative but can be easily determined in practice . in the frequency calibration procedure , the ti : sa cw - seed optical frequency @xmath38 is determined , while the output of the oscillator - amplifier ti : sa _ pulsed _ laser system is actually used in the spectroscopy , after frequency upconversion to the 4@xmath47 harmonic . any frequency offset @xmath45 between the ti : sa cw - seed frequency and pulsed output frequency needs to be determined , where it is possible that the optical frequency is time - dependent within the laser pulse . the pulse generated in the ti : sa oscillator is subject to cavity - mode pulling effects , as a result of an optically - induced change in refractive index of the ti : sa crystal due to the intense 532-nm pump pulse , contributing to a frequency offset . the actual cavity resonance in the presence of the pump pulse is then frequency - shifted with respect to the locking point of the unpumped cavity . the frequency shift can be separated into contributions from a thermal change of the refractive index @xmath48 and an effect from the population inversion @xmath49 in the ti : sa crystal @xcite . the total effective frequency offset , typically tens of mhz in the fundamental @xmath38 , can be compensated by controlling the lock setpoint of the ti : sa oscillator , e.g. locking at the side of the fringe in the hnsch - couillaud scheme . a related phenomenon is spatial frequency chirp where the pulsed optical frequency varies across the transverse beam profile of the laser beam . this effect occurs in the amplification stage , because subsequent passes sample a different area within the pump beam profile that has a gaussian intensity distribution . the spatial frequency difference from either edges of the beam profile was found to be a few mhz with respect to @xmath38 . to minimize such a spatial frequency offset , a 1-mm pinhole after the ti : sa amplifier acts as a spatial filter while still providing sufficient energy for the frequency upconversion process . the cw - pulse offset frequency @xmath45 is measured by heterodyning the amplified pulsed output of the ti : sa system with part of the cw - seed that is shifted by 250 mhz using an acousto - optic modulator ( aom ) . the resulting beat signal from a fast photo - detector is recorded with an oscilloscope for further frequency chirp analysis @xcite . a single - shot frequency chirp analysis is performed for every ti : sa pulse in order to account for the frequency offset @xmath45 during each measurement point . a more detailed description of the frequency chirp measurements and analysis for the ti : sa pulsed laser system is given in refs . @xcite . ( 0,1 ) q(1 ) two - photon transition in d@xmath15 . a gaussian profile ( blue line ) fitted to the averaged datapoints ( in red ) is shown . ] a spectral recording of the d@xmath15 @xmath50(0,1 ) q(1 ) two - photon transition is shown in fig . [ spect ] . the full - width half - maximum of the spectral lines is @xmath51 mhz and is well - approximated by a gaussian profile , limited by the instrumental linewidth of the uv laser source , and accounting for the two - photon excitation . as a test of robustness , the obtained line positions were also checked when imposing different background levels . the uncertainty contribution is then estimated to be better than 500 khz for most scans of good signal - to - noise ratio . for more noisy spectra , the line fitting uncertainty contribution is difficult to separate from statistical scatter , thus it is subsumed into the statistics entry . several measurements of a specific transition taken over several days , demonstrate reproducibility of within @xmath52 mhz in the transition frequency . an important systematic effect is the ac - stark frequency shift induced by the power density of the probe radiation . the use of a separate ionization laser , delayed with respect to the probe , allows for sufficient detection efficiency even at a reduced probe beam intensity , thereby minimizing the ac - stark shift . the ac - stark measurements are performed in two different beam configurations , referred to as ( i ) and ( ii ) . for most of the measurements , the uv beam is collimated with a beam diameter of 0.5 mm , using a combination of concave mirrors , that also serves to reduce beam astigmatism . this configuration ( i ) is typical for low uv power density . an additional benefit of configuration ( i ) is that it reduces residual first - order doppler shifts compared to the case when using counter - propagating beams with curved wavefronts @xcite . using ( i ) , a 355-nm laser pulse , delayed by 30 ns relative to the probe beam , was employed in order to ionize molecular hydrogen in the @xmath21 state . to aid in the ac - stark shift assessment and to improve the extrapolation to the field - free case , a second configuration is implemented , referred to as ( ii ) . in this configuration , a lens with 1-m focal length is placed before the beam splitter in fig . [ setup ] , thus reducing the beam diameter to @xmath53 @xmath36 m in the interaction region so that the power density is increased by a factor @xmath54 with respect to that of ( i ) . the intensity of the spectroscopy uv beam in ( ii ) is sufficient to induce ionization , so that a separate ionization laser is not required . measurements are performed for different probe beam intensities in both configurations and the results are illustrated in fig . [ stark ] . the measurements for higher power densities ( ii ) are shown in the full graph , while the measurements at low power density ( i ) are enlarged in the inset . the field - free transition frequency at zero intensity is obtained from a weighted linear fit of the combined measurement results of ( i ) and ( ii ) . the horizontal axis represents the relative power density and is derived from a measurement of the probe beam intensity and a measurement of the beam diameter at the interaction region . it is worth noting that for the energy range between 30 and 300 @xmath36j in ( i ) , the ac - stark frequency shifts are within statistical scatter . the estimated error from the ac - stark effect is deduced from the error of the intercept determined in the weighted linear fit , with the extrapolated field - free frequency estimated to be accurate to @xmath55 mhz . this extrapolation procedure was performed for every transition measured . ( 0,1 ) q(0 ) transition in d@xmath15 plotted against the 207-nm probe power density in the interaction zone . blue datapoints were collected using focused probe beams ( conf . ii ) , while red datapoints were taken with collimated probe beams ( conf . i ) as also shown in the inset . ] the dc - stark effect is avoided by pulsing the voltages of the ion extraction plates so that the transitions are probed under dc field - free conditions . however , no measurable change in the transition frequencies was observed when the extraction fields were operated in either pulsed- or dc - mode , and we therefore estimate a contribution of @xmath56 mhz on the systematic uncertainty due to the dc stark effect . the residual first - order doppler shifts estimated from the sagnac interferometer alignment of the counter - propagating probe beams were experimentally verified by purposely misaligning the probe beams , where the resulting shifts were found to be below statistical scatter . further tests were also performed by using mixed samples of molecular hydrogen and krypton to reduce the speed of h@xmath15 . more finely - tuned molecular beam velocities were obtained by varying the delay between the timing of the valve opening and the trigger of the laser pulse , by means of which molecules in the leading , middle and trailing parts of the gas pulse are probed . [ doppler ] shows the frequency measurements of the @xmath23 ( 0,1 ) q(1 ) line for the six different h@xmath15 beam velocities thus obtained . no effect above the statistical uncertainty is observed . q(1 ) line measured at different molecular beam rms speed , obtained from different concentrations of h@xmath15-kr mixtures , and time delay settings between valve opening and laser trigger . ] to assess possible pressure shifts , we use the pressure shift coefficient for rydberg states from ref . @xcite of @xmath57 @xmath2/amagat as an upper limit , noting that for the pure ground electronic state , quadrupole transition shifts are in the order of @xmath58 @xmath2/amagat , see e.g. @xcite . from local gas densities used in the experiment , an upper limit for the pressure shift of @xmath59 mhz is estimated . table [ tab : uncertainty ] summarizes the uncertainty contributions in mhz from different error sources in the determination of the @xmath23 ( 0,1 ) transitions . systematic corrections were applied separately for each spectral recording , for example for the frequency chirp and ac - stark shift . the statistics entries denote the statistical @xmath60 standard deviations of all the measurements after the contributions of various systematic effects have been corrected for . the uncertainties in the determination of the line positions are not indicated as they are already included in the statistical entry , where the averaging is weighted by uncertainty in the line fitting for each recorded spectrum . the data collection for each line covered a period of different days , where the different transitions were remeasured throughout the whole measurement period to confirm the reproducibility of the results . .estimated systematic and statistical uncertainty contributions for the frequency calibrations of the @xmath23 ( 0,1 ) transitions in h@xmath15 , hd and d@xmath15 . the uncertainty values ( unc . ) are given in mhz . [ cols="<,<,^ , > " , ] ac - stark and chirp offsets are corrected for and not indicated in the table . standard deviation based on 63 measurements . standard deviation based on 69 measurements . standard deviation based on 64 measurements . quadrature sum of errors . q - branch transition frequencies in the @xmath23 ( 0,1 ) band are listed in table [ efxtransitions ] for h@xmath15 , hd , d@xmath15 . the q(1 ) and q(2 ) transitions for each isotopomer extend the rotationless transitions reported by dickenson _ et al . _ @xcite . also listed in table [ efxtransitions ] are the corresponding transition frequencies in the @xmath23 ( 0,0 ) band taken from hannemann _ et al . _ the improved uncertainties for the present @xmath23 ( 0,1 ) transitions are primarily due to the reduction in the spatial frequency chirp of the ti : sa laser system output by spatial filtering . the data on the @xmath23 ( 0,1 ) band presented here are in agreement with results of eyler _ et al . _ @xcite , with the present results representing a 100-fold improvement over the previous measurement . [ efxtransitions ] c@c@c@c & q(0 ) & q(1 ) & q(2 ) + + h@xmath15 & 95003.62055(10 ) & 94954.47739(10 ) & 94856.71748(10 ) + hd & 95669.18610(10 ) & 95631.61343(10 ) & + d@xmath15 & 96467.83202(10 ) & 96442.20932(10 ) & 96391.10000(10 ) + + h@xmath15 & 99164.78691(11 ) & 99109.73139(18 ) & 99000.18301(11 ) + hd & 99301.34662(20 ) & 99259.91793(20 ) & + d@xmath15 & 99461.44908(11 ) & 99433.71638(11 ) & 99378.39352(11 ) + _ ab initio _ values for the vibrational energy splittings in the @xmath13 electronic ground state were presented in refs . @xcite for h@xmath15 and d@xmath15 while the rovibrational level energies of the hd ground state were given in ref . these state - of - art calculations are based on nonadiabatic perturbation theory ( napt ) @xcite to obtain the nonrelativistic energy contributions , while the nonrelativistic quantum electrodynamics ( nrqed ) formalism @xcite is used to pertubatively obtain the relativistic and qed energy terms . more accurate theoretical values for the rotationless transitions of h@xmath15 , hd and d@xmath15 were presented in ref . the theoretical vibrational energy splittings are summarized in table [ tab : splittings ] , where the various energy contributions are separately listed , including the estimated uncertainties where available . the most accurate values are for the rotationless transitions from ref . @xcite with accuracies of better than @xmath61 @xmath2 . for the @xmath6 transitions of h@xmath15 and d@xmath15 , accurate values were obtained from the supplementary material of komasa _ et al . _ @xcite , however , only the accuracy of the total level energies was indicated . the obtained fundamental vibrational transition energies are estimated to be accurate to @xmath62 @xmath2 . for the @xmath63 hd transition , level energies from pachucki and komasa @xcite were used , however , the energy contributions were not separately indicated . the estimated theoretical uncertainty for the more accurate rotationless transition energies stems from the better uncertainty of the nonadiabatic corrections . this estimate is aided by available theoretical results for the rotationless vibrational splittings from adamowic and co - workers @xcite , using an independent theoretical methodology based on a variational procedure to obtain the nonadiabatic wave functions . similar calculations for transitions involving @xmath64 , however , are not available . since komasa , pachucki and co - workers obtain transition energies using the same method for the rotationless case as for all @xmath65 quantum numbers , the original uncertainty estimates indicated in refs . @xcite are likely to be overestimated . l [email protected] [email protected] [email protected] h@xmath0 & & & + born - oppenheimer & 4163 & 40350 & 4157 & 4837 & 4145 & 6805 + adiabatic & -1 & 40284 & -1 & 3963 & -1 & 3832 + nonadiabatic & -0 & 83649 & -0 & 8354 & -0 & 8335 + nonrel subtotal ( @xmath66 ) & 4161 & 16416(1 ) & 4155 & 2520 & 4143 & 4638 + relativistic ( @xmath67 ) & 0 & 02341(3 ) & 0 & 0232 & 0 & 0227 + qed ( @xmath68 ) & -0 & 02129(2 ) & -0 & 0212 & -0 & 0211 + hqed ( @xmath69 ) & -0 & 00016(8 ) & -0 & 0002 & -0 & 0002 + rel + qed subtotal & 0 & 00196(9 ) & 0 & 0018 & 0 & 0014 + theory total & 4161 & 16612(9 ) & 4155 & 2538(9 ) & 4143 & 4653(9 ) + hd & & + born - oppenheimer & 3633 & 71956 & + adiabatic & -0 & 93259 & + nonadiabatic & -0 & 62872 & + nonrel subtotal ( @xmath66 ) & 3632 & 15826(1 ) & + relativistic ( @xmath67 ) & 0 & 02093(2 ) & + qed ( @xmath68 ) & -0 & 01863(2 ) & + hqed ( @xmath69 ) & -0 & 00014(7 ) & + rel + qed subtotal & 0 & 00216(8 ) & + theory total & 3632 & 16041(8 ) & 3628&3044(10 ) + d@xmath0 & & & + born - oppenheimer & 2994 & 44084 & 2992 & 3295 & 2988 & 1133 + adiabatic & -0 & 52150 & -0 & 5204 & -0 & 5180 + nonadiabatic & -0 & 30447 & -0 & 3043 & -0 & 3040 + nonrel subtotal ( @xmath66 ) & 2993 & 61487(1 ) & 2991 & 5048 & 2987 & 2913 + relativistic ( @xmath67 ) & 0 & 01771(2 ) & 0 & 0176 & 0 & 0175 + qed ( @xmath68 ) & -0 & 01539(2 ) & -0 & 0154 & -0 & 0153 + hqed ( @xmath69 ) & -0 & 00012(6 ) & -0 & 0001 & -0 & 0001 + rel + qed subtotal & 0 & 00220(7 ) & 0 & 0021 & 0 & 0021 + theory total & 2993 & 61708(7 ) & 2991 & 5070(2 ) & 2987 & 2934(2 ) + ground state @xmath13 energy splittings are obtained from the combination differences of the @xmath23 transition energies and are listed in table [ vibtransitions ] . the indicated uncertainty of the fundamental vibrational energy splittings is the quadrature sum of the uncertainties in the particular @xmath23 transition energies used . the difference between the experimental and theoretical results @xmath70 is also listed in table [ vibtransitions ] , along with the combined uncertainty @xmath71 , from the experimental @xmath72 and theoretical @xmath73 uncertainties . the accuracy of present experimental values for the vibrational transitions is predominantly limited by the @xmath23 ( 0,0 ) results from hannemann _ et al . _ @xcite . for the rotationless transitions the theoretical values are more accurate than the experimental ones . however , for transitions involving @xmath6 the experimental results are 5 times more accurate . for the transitions listed , the difference @xmath70 is statistically consistent with null within the combined uncertainty @xmath74 . ( the @xmath75 deviation for the h@xmath15 @xmath76 comparison is compatible with expectations from statistics . ) the overall comparison demonstrates excellent agreement between the present experimental and theoretical values . [ vibtransitions ] c @ [email protected] @ [email protected] @ [email protected]@ [email protected] & & & & + + h@xmath15 & 4161&16636(15 ) & 4161&16612(9 ) & 0&00024 & 0&00017 + hd & 3632&16052(22 ) & 3632&16041(8 ) & 0&00011 & 0&00023 + d@xmath15 & 2993&61706(15 ) & 2993&61708(7 ) & -0&00002 & 0&00017 + + h@xmath15 & 4155&25400(21 ) & 4155&2538(9 ) & 0&0002 & 0&0009 + hd & 3628&30450(22 ) & 3628&3044(10 ) & 0&0001 & 0&0010 + d@xmath15 & 2991&50706(15 ) & 2991&5070(2 ) & 0&0001 & 0&0003 + + h@xmath15 & 4143&46553(15 ) & 4143&4653(9 ) & 0&0002 & 0&0009 + d@xmath15 & 2987&29352(15 ) & 2987&2934(2 ) & 0&0001 & 0&0003 + various methods have been employed in direct excitations of ground state rovibrational transitions that include among others quadrupole absorption studies , raman spectroscopy and electric - field induced dipole spectroscopy . since the first measurements of the quadrupole spectra of h@xmath15 overtone bands by herzberg @xcite using long - path classical absorption techniques , refinements have been applied through the years , e.g. by using more accurate echelle spectrometers by rank and co - workers @xcite . the improvements continued in the study of bragg _ et al . _ @xcite who employed fourier - transform ( ft ) spectroscopy techniques . since the first measurements of rasetti @xcite and the early investigations of stoicheff @xcite , the raman spectrum has also been measured with ever increasing accuracies . from the first demonstrations by crawford and dagg @xcite of the electric field induced dipole excitation , parallel improvements in accuracy have also been achieved by later investigators . the weak transitions probed necessitated high pressures that led to collisional shifts and broadening effects first investigated by may _ et al . _ numerous other investigations on the h@xmath15 fundamental band at varying accuracies for the transition energies include refs . @xcite . the comparison of the present experimental results to selected previous determinations is shown graphically in fig . [ h2comp ] for h@xmath15 . only those investigations with the highest claimed accuracy for a particular method are included in the figure . the results of bragg and co - workers @xcite stood as the most accurate for decades , making use of fourier transform spectroscopy with long - path absorption samples . for the q(1 ) transition , however , the result from bragg _ _ differs from the present result by several standard deviations . in that study , the collision - induced dipole spectrum adds a broad background signal for the ( 1,0 ) band ( e.g. fig . 2 in ref . @xcite ) that could affect the determination of the line positions . the most accurate electric field induced spectrum was recorded by buijs @xcite who also employed fourier transform spectroscopy . for raman spectroscopy , the most accurate measurements were performed by rahn and rosasco @xcite , using a pulsed laser source based on difference - frequency mixing . the results of rahn and rosasco @xcite and that of buijs @xcite are in fair agreement with the present results . the latter comparison suggests that electric - field induced frequency shifts are less severe than pressure - induced systematic shifts , even after pressure - shift corrections . ground state fundamental band q - branch transitions to previous studies representing the most accurate values from a particular method : from long - path absorption reported by bragg _ et al . _ @xcite ( ) ; from raman spectroscopy by rahn and rosasco @xcite ( ) ; from spectroscopy of electric field induced transitions by buijs @xcite ( ) . the results on empirical fitting of the global h@xmath15 database by stanke _ et al . _ @xcite ( ) are also included , along with the present theoretical results ( ) . ] durie and herzberg @xcite recorded the weak dipole absorption spectrum of the ( 1,0 ) , ( 2,0 ) , ( 3,0 ) and ( 4,0 ) bands in the ground state of hd . stoicheff @xcite performed raman spectroscopy on the fundamental band of hd , while brannon _ et al . _ @xcite measured electric - field induced transitions in the same band . mckellar and co - workers @xcite carried out several long - path absorption investigations covering the fundamental and overtone bands up to ( 6,0 ) band of the hd ground state . in these investigations , the pressure shifts on the transition energies were systemically studied as was the case in ref . the raman study of veirs and rosenblatt @xcite also included the hd fundamental band , as well as the fundamental band for most other molecular hydrogen isotopologues . [ hdcomp ] is a graphical representation of the comparison of the present experimental results for the hd fundamental band to selected previous determinations . rich _ et al . _ @xcite reported the most accurate results with fourier transform spectroscopy on long - path absorption samples . the most accurate electric field induced spectrum was that of brannon _ et al . _ @xcite while raman spectra for hd were obtained by veirs and rosenblatt @xcite . the fundamental band of d@xmath15 was also included in the raman spectroscopic investigations of stoicheff @xcite . later , looi _ et al . _ @xcite improved upon the raman spectroscopy , and in addition investigated pressure dependent energy shifts . electric field induced transitions of d@xmath15 were also measured by brannon _ et al . _ @xcite with pressure shift corrections . mckellar and oka @xcite used a difference - frequency laser system to investigate the fundamental vibrational band of d@xmath15 in a long - path cell . jennings _ et al . _ @xcite performed accurate long - path absorption measurements using ft spectroscopy . a comparison of the d@xmath15 fundamental band transitions is plotted in fig . [ d2comp ] . the older results from long - path absorption and electric - field induced spectra coincide with the present results . the results from the raman investigations deviate considerably from the other studies including the present , suggesting systematic errors in ref . @xcite . ground state fundamental band q - branch transitions to previous studies representing the most accurate values from a particular method : from long - path absorption reported by jennings _ et al . _ @xcite ( ) ; from raman spectroscopy by looi _ et al . _ @xcite ( ) ; from spectroscopy of electric field induced transitions by brannon _ et al . _ @xcite ( ) . the results on empirical fitting of the global d@xmath15 database by bubin _ et al . _ @xcite ( ) are also included , along with the present theoretical results ( ) . note that for this energy scale the error bars for the present experimental and theoretical results are not resolved and thus overlap . ] an indirect approach through empirical fitting of dunham coefficients have been carried out by stanke _ et al . _ @xcite for h@xmath15 . experimental data included in the fit came from dabrowski @xcite for lyman and werner band transitions , but also most ground state studies mentioned above @xcite . similar global fitting analyses based on the dunham relation was also performed by the same group for hd @xcite and d@xmath15 @xcite resulting in similar uncertainties as in the case of h@xmath15 . remarkable accuracy was achieved for these global fits , despite considerable deviations of results from individual experiments used . the excitation of molecular hydrogen in a collisionless environment in the present study is distinct from previous studies , where the low excitation probability required the use of dense gas samples with pressures of a few bars for detection . the recorded lines had doppler - widths on the order of ghz and the collisional perturbation gives rise to pressure shifts amounting to 100 mhz @xcite . these issues are effectively absent in the present study using molecular beam conditions . as a case in point , it is interesting to look at the claimed uncertainty of maddaloni _ et al . _ @xcite , who deployed a high - resolution laser source based on difference - frequency - generation referenced to the cs - clock primary standard using an optical frequency comb synthesizer . nevertheless , a comparison with theory gives differences of @xmath77 and @xmath78 @xmath2 for s(0 ) and s(1 ) in ref . @xcite , respectively , as pointed out in ref . the excellent agreement of the present experimental results with the same _ ab initio _ calculations suggests that the systematic uncertainties in ref . @xcite might be underestimated . as can be seen in table [ vibtransitions ] , the _ ab initio _ values of the rotationless ground tone energies are the most accurate for all isotopomers . for these energy splittings , the relativistic and qed terms exhibit the dominant uncertainty contribution to the theoretical value while the nonrelativistic energy hardly contributes to the uncertainy . by subtracting the theoretical nonrelativistic energy term from the experimental value , one obtains a hybrid experimental - theoretical determination of the relativistic and qed contribution of @xmath79 @xmath2 for h@xmath15 . comparing the latter value to the _ ab initio _ values represents a test of relativistic and qed calculations in molecules on the order of @xmath80 . applying similar arguments for the dissociation limit of the ground electronic state of molecular hydrogen , and using the experimental results of ref . @xcite and theoretical values from ref . @xcite , relativistic and qed calculations in molecules are verified on the order of @xmath81 . the excellent agreement between theory and experiment discussed above , can be exploited to constrain effects beyond the standard model of physics . since molecular ( and atomic ) structure is dominated by the electromagnetic interaction , with the effects of the strong- , weak- and gravitational forces many orders of magnitudes weaker ( at least for light systems as hydrogen , where weak force effects are not yet detectable ) , any discrepancy between measurements and theory points to new physics . new interactions or modifications beyond the standard model are expected as it does not provide explanations for several phenomena , e.g. dark matter @xcite and dark energy @xcite . the possibility that these hypothetical new interactions or fifth forces have subtle effects in atomic or molecular structure is a complementary approach to the searches and tests in particle physics or astronomy . energy resonances in calculable few - body systems provide an ideal search ground . simple systems to search for extra interactions between lepton and hadrons would be atomic hydrogen or the helium ion @xcite , while extra hadron - hadron interactions can be probed in molecular hydrogen and molecular hydrogen ions . we express a general fifth force in terms of a yukawa potential @xmath82 as a function of the distance @xmath83 between hadrons , or the internuclear distance in molecular hydrogen : @xmath84 where @xmath85 is an interaction strength , and @xmath86 is the characteristic range of the interaction . we assume that the long - range effect ( as opposed to the short range of the strong force ) scales with the number of nucleons of each nucleus , @xmath87 and @xmath88 respectively , and we treat protons and neutrons equally . the potential @xmath82 can be considered as a perturbation on the energies of quantum states , leading to a differential energy shift @xmath89 on energy level differences between states @xmath90 and @xmath91 : @xmath92 \nonumber \\ & = & \alpha_5 n_{1 } n_{2 } \delta y_{\lambda } \label{yv5}\end{aligned}\ ] ] where @xmath93 are the wave functions representing the probability of finding the nuclei at a certain separation @xmath83 within the molecule . using accurate wave functions from ref . @xcite , the quantity @xmath94 has been evaluated taking @xmath86 as a parameter in ref . @xmath94 can be considered as the inherent sensitivity coefficient for a specific transition , which becomes larger as the wavefunctions of the two levels involved are more different ( e.g. wavefunctions in fig . [ potentialenergy ] ) . thus transitions with greater @xmath95 have a larger @xmath89 energy shift . the largest inherent sensitivity is expected for the dissociation energy @xmath96 since the shift is @xmath97 . for a particular transition , @xmath94 is greatest for h@xmath15 and least for d@xmath15 since the spatial extent of the wavefunctions between any two levels , with corresponding set of quantum numbers , are more similar for d@xmath15 than in h@xmath15 . however , the effect of nucleon numbers , @xmath87 and @xmath88 , more than compensates for the lower @xmath94 in the heavier isotopomers , so that the expected energy shift @xmath89 is actually greatest for d@xmath15 and least for h@xmath15 for a particular transition . of possible fifth forces relative to the fine structure constant ( left vertical axis ) or the gravitational force ( right axis ) , for different interaction length range @xmath86 obtained from the fundamental vibration of molecular hydrogen . similar bounds from the d@xmath15 dissociation energy @xmath96 as well as rovibrational transitions in hd@xmath98 are also indicated . the yellow area indicates excluded regions based mainly on constraints obtained from hd@xmath98 . ] an upper bound for the interaction strength @xmath85 can be be obtained from the combined experimental - theoretical uncertainty @xmath74 for a particular transition by the relation @xmath99 . [ limits ] shows the constraint for the interaction strength @xmath85 , for certain range @xmath86 of the interaction , obtained from the rotationless fundamental vibrational transitions in h@xmath15 , hd and d@xmath15 , with the shaded area indicating the excluded region . the fifth - force interaction strength is given in terms of the strengths of the electromagnetic force , @xmath100 , and the gravitational force , @xmath101 . more stringent bounds in the same length scale are discussed in ref . @xcite based on spectroscopic determinations on the dissociation energy @xmath102 of molecular hydrogen @xcite as well as rovibrational transitions in hd@xmath98 @xcite . similar constraints have been extended to shorter interaction length at sub - angstrom scales in ref . @xcite based on level splittings in exotic molecules . the interaction length scale @xmath86 can be associated with the mass @xmath103 of a fifth - force carrier particle as indicated in the upper horizontal axis in fig . [ limits ] . this boson mass range is complementary to the mass sensitivity range in particle accelerators on the high end , and the low - mass sensitivity in gravitational test experiments on the other extreme . from the present measurements on level splittings of the fundamental ground tones in the hydrogen molecule and its isotopomers , a constraint on the strength of a possible fifth force between hadrons is determined to be @xmath104 for a force range of 1 . the fundamental vibrational energy splitting of h@xmath0 , hd , and d@xmath0 were determined to absolute accuracies of @xmath1 @xmath2 , or a relative accuracy of a few parts in @xmath105 . the vibrational splitting frequencies are derived from the combination difference between separate electronic excitations from the @xmath3 and @xmath4 vibrational states to a common @xmath5 state . doppler - free laser spectroscopic investigation applied in a collisionless molecular beam environment leads to high accuracy , where pressure effects are negligible in contrast to studies based on gas cells . the excellent agreement between the experimental results and the calculations provides a stringent test on the application of quantum electrodynamics in molecules , and can be used to provide bounds to new interactions . upper bounds derived from molecular hydrogen indicate that the interaction strength of possible fifth forces must be at least 8 orders of magnitude weaker than the electromagnetic strength , for a fifth - force interaction range in the order of typical internuclear distances of @xmath106 . this brings molecular spectroscopy studies again to the forefront of physics , reminiscent of the early days of quantum mechanics . this work was supported by the netherlands astrochemistry program of nwo ( cw - ew ) and by the netherlands foundation for fundamental research of matter ( fom ) through the program `` broken mirrors & drifting constants '' . w. heitler and f. london , 44 ( 1927 ) 619 . e. schrdinger , 28 ( 1926 ) 1049 . w. pauli , 31 ( 1925 ) 765 . m. born and r. oppenheimer , 389 ( 1927 ) 457 . h. j. james and a. s. coolidge , 1 ( 1933 ) 825 . e. a. hylleraas , 54 ( 1929 ) 347 . h. j. james and a. s. coolidge , 49 ( 1936 ) 688 . w. koos and l. wolniewicz , 41 ( 1964 ) 3663 . w. koos and l. wolniewicz , 43 ( 1965 ) 2429 . w. koos and l. wolniewicz , 45 ( 1966 ) 515 . w. koos and l. wolniewicz , 49 ( 1968 ) 404 . l. wolniewicz , 99 ( 1993 ) 1851 . w. koos and c. j. j. roothaan , rev . 32 ( 1960 ) 205 . w. koos and l. wolniewicz , rev . 35 ( 1963 ) 473 . w. koos , adv . quantum chem . 5 ( 1970 ) 99 . d. m. bishop and l. m. cheung , adv . quantum chem . 12 ( 1980 ) 1 . j. rychlewski , adv . quantum chem . 31 ( 1999 ) 173 . l. wolniewicz , 103 ( 1995 ) 1792 . k. pachucki , 82 ( 2010 ) 032509 . k. pachucki and j. komasa , 129 ( 2008 ) 034102 . k. pachucki and j. komasa , 130 ( 2009 ) 164113 . k. pachucki , 71 ( 2005 ) 012503 . k. pachucki , 76 ( 2007 ) 022106 . k. piszczatowski , g. ach , m. przybytek , j. komasa , k. pachucki and b. jeziorski , 5 ( 2009 ) 3039.k . pachucki and j. komasa , 12 ( 2010 ) 9188 . j. komasa , k. piszczatowski , g. lach , m. przybytek , b. jeziorski and k. pachucki , 7 ( 2011 ) 3105 . e. e. witmer , 28 ( 1926 ) 1223 . h. beutler , z. phys . 29 ( 1935 ) 315 . g. herzberg , 23 ( 1969 ) 1081 . g. herzberg , 33 ( 1970 ) 147 . g. herzberg and c. jungen , 41 ( 1972 ) 425 . y. p. zhang , c. h. cheng , j. t. kim , j. stanojevic , and e. e. eyler , 92 ( 2004 ) 203003 . e. e. eyler and n. melikechi , 48 ( 1987 ) r18 . j. liu , e. j. salumbides , u. hollenstein , j. c. j. koelemeij , k. s. e. eikema , w. ubachs and f. merkt , 130 ( 2009 ) 174306 . j. liu , d. sprecher , c. jungen , w. ubachs and f. merkt , 132 ( 2010 ) 154301 . d. sprecher , j. liu , c. jungen , w. ubachs and f. merkt , 133 ( 2010 ) 111102 . g. herzberg , 87 ( 1938 ) 428 . g. herzberg , nature 163 ( 1949 ) 170 . u. fink , t. a. wiggins and d. h. rank , 18 ( 1965 ) 384 . j. v. foltz , d. h. rank and t. a. wiggins , 21 ( 1966 ) 203 . s. l. bragg , w. h. smith and j. w. brault , 263 ( 1982 ) 999 . ferguson , k.n . rao , m.e . mickelson and l.e . larson , 160 ( 1993 ) 315 . d. c. robie and j. t. hodges , 124 ( 2006 ) 024307 . f . cheng , y. r. sun , h. pan , j. wang , a. w. liu , a. campargue and s .- m . hu , 85 ( 2012 ) 024501 . s .- hu , h. pan , c .- f . cheng , y. r. sun , x. f. li , j. wang , a. campargue and a. w. liu , 749 ( 2012 ) 76 . s. kassi , a. campargue , k. pachucki and j. komasa , 136 ( 2012 ) 184309 . a. campargue , s. kassi , k. pachucki and j. komasa , 14 ( 2012 ) 802 . p. maddaloni , p. malara , e. de tommasi , m. de rosa , i. ricciardi , g. gagliardi , f. tamassia , g. di lonardo and p. de natale , 133 ( 2010 ) 154317 . m. stanke , d. kedziera , s. bubin , m. molski and l. adamowicz , 128 ( 2008 ) 114313 . i. dabrowski , 62 ( 1984 ) 1639 . j. philip , j. p. sprengers , th . pielage , c. a. de lange , w. ubachs and e. reinhold , 82 ( 2004 ) 713 . w. ubachs and e. reinhold , 92 ( 2004 ) 101302 . d. w. chandler and l. r. thorne , 85 ( 1986 ) 1733 . e. r. davidsson , 33 ( 1960 ) 1577 . e. e. eyler , j. gilligan , e. f. mccormack , a. nussenzweig and e. pollack , 36 ( 1987 ) 3486 . j.m . gilligan and e.e . eyler , 46 ( 1992 ) 3676 . d. shiner , j. m. gilligan , b. m. cook , and w. lichten , 47 ( 1993 ) 4042 . a. yiannopoulou , n. melikechi , s. gangopadhyay , j. c. meiners , c. h. cheng and e. e. eyler , 73 ( 2006 ) 022506 . s. hannemann , e. j. salumbides , s. witte , r. t. zinkstok , e .- van duijn , k. s. e. eikema and w. ubachs , 74 ( 2006 ) 062514 . e. j. salumbides , d. bailly , a. khramov , a. l. wolf , k. s. e. eikema , m. vervloet and w. ubachs , 101 ( 2008 ) 223001 . d. bailly , e. j. salumbides , m. vervloet and w. ubachs , 108 ( 2010 ) 827 . e. j. salumbides , g. d. dickenson , t. i. ivanov and w. ubachs , 107 ( 2011 ) 043005 . g. d. dickenson , m. l. niu , e. j. salumbides , j. komasa , k. s. e. eikema , k. pachucki and w. ubachs , 110 ( 2013 ) 193601 . s. hannemann , e .- j . van duijn and w. ubachs , 78 ( 2007 ) 103102 . s. hannemann , e. j. salumbides and w. ubachs , 32 ( 2007 ) 1381 . d. zhao , n. wehres , h. linnartz and w. ubachs , 501 ( 2011 ) 232 . m. s. fee , k. danzmann and s. chu , 45 ( 1992 ) 4911 . k. s. e. eikema , w. ubachs , w. vassen and w. hogervorst , 55 ( 1997 ) 1866 . w. e. caswell and g. p. lepage , , 167 ( 1986 ) 437 . k. pachucki , 74 ( 2006 ) 022512 . m. stanke , s. bubin , m. molski and l. adamowicz , 79 ( 2009 ) 032507 . s. bubin , m. stanke , m. molski and l. adamowicz , 494 ( 2010 ) 21 . d. h. rank , b. s. rao , p. sitaram , a. f. slomba and t. a. wiggins , 52 ( 1962 ) 1004 . f. rasetti , 34 ( 1929 ) 367 . b. p. stoicheff , 35 ( 1957 ) 730 . m. f. crawford and i. r. dagg , 91 ( 1953 ) 1569 . a. d. may , v. degen , j. c. stryland and h. l. welsh , 39 ( 1961 ) 1769 . a. d. may , g. varghese , j. c. stryland and h. l. welsh , 42 ( 1964 ) 1058 . e. c. looi , j. c. stryland and h. l. welsh , 56 ( 1978 ) 1102 . d. k. veirs and g. m. rosenblatt , 121 ( 1987 ) 401 . g. j. germann and j. j. valentini , 92 ( 1988 ) 3792 . l. a. rahn and g. j. rosasco , 41 ( 1990 ) 3698 . h. l. buijs and h. p. gush , 49 ( 1971 ) 2366 . r. a. durie and g. herzberg , 38 ( 1960 ) 806 . p. j. brannon , c. h. church and c. w. peters , 27 ( 1968 ) 44 . a. r. w. mckellar , 51 ( 1973 ) 389 . a. r. w. mckellar , w. goetz , and d. a. ramsay , 207 ( 1976 ) 663 . n. h. rich , j. w. c. johns , and a. r. w. mckellar , 95 ( 1982 ) 432 . s. nazemi , a. javan and a. s. pine , 78 ( 1983 ) 4797 . a. r. w. mckellar and t. oka , 56 ( 1978 ) 1315 . d. e. jennings , a. weber and j. w. brault , 25 ( 1986 ) 284 . g. bertone , d. hooper and j. silk , phys . 405 ( 2005 ) 279 . p. j. e. peebles and b. ratra , 75 ( 2003 ) 559606 . s. g. karshenboim , 82 ( 2010 ) 073003 . w. ubachs , w. vassen , e. j. salumbides and k. s. e. eikema , ann . ( berlin ) 575 ( 2013 ) a113 . e. j. salumbides , j. c. j. koelemeij , j. komasa , k. pachucki , k. s. e. eikema , and w. ubachs , 87 ( 2013 ) 112008 . j. c. j. koelemeij , b. roth , a. wicht , i. ernsting , and s. schiller , 98 ( 2007 ) 173002 . j. c. j. koelemeij , d. w. e. noom , d. de jong , m. a. haddad and w. ubachs , appl . b 107 ( 2012 ) 1075 . u. bressel , a. borodin , j. shen , m. hansen , i. ernsting and s. schiller , 108 ( 2012 ) 183003 . e. j. salumbides , w. ubachs and v. i. korobov , submitted to ( 2014 ) .

although silicon is a promising material for quantum computation , the degeneracy of the conduction band minima ( valleys ) must be lifted with a splitting sufficient to ensure formation of well - defined and long - lived spin qubits . here we demonstrate that valley separation can be accurately tuned via electrostatic gate control in a metal - oxide - semiconductor quantum dot , providing splittings spanning 0.3 - 0.8 mev . the splitting varies linearly with applied electric field , with a ratio in agreement with atomistic tight - binding predictions . we demonstrate single - shot spin readout and measure the spin relaxation for different valley configurations and dot occupancies , finding one - electron lifetimes exceeding 2 seconds . spin relaxation occurs via phonon emission due to spin - orbit coupling between the valley states , a process not previously anticipated for silicon quantum dots . an analytical theory describes the magnetic field dependence of the relaxation rate , including the presence of a dramatic rate enhancement ( or hot - spot ) when zeeman and valley splittings coincide .

1302.0983

silicon is at the heart of all modern microelectronics . its properties have allowed the semiconductor industry to follow moore s law for nearly half a century , delivering nowadays billions of nanometre - scale transistors per chip . remarkably , silicon is also an ideal material to manipulate quantum information encoded in individual electron spins @xcite . this is a consequence of the weak spin - orbit coupling and the existence of an abundant spin - zero isotope , which can be further enriched to obtain a `` semiconductor vacuum '' in which an electron spin can preserve a coherent quantum superposition state for exceptionally long times @xcite . + in order to define a robust spin-@xmath0/@xmath1 qubit hilbert space , it is necessary that the energy scale of the two - level system is well separated from higher excitations . in this respect , a major challenge for the use of silicon is represented by the multi - valley nature of its conduction band . in a bulk silicon crystal the conduction band minima are six - fold degenerate , but in a two - dimensional electron gas ( 2deg ) , the degeneracy is broken @xcite into a two - fold degenerate ground state ( @xmath2 valleys ) and a four - fold degenerate excited state ( @xmath3 valleys ) , due to vertical confinement of electrons with different effective mass along the longitudinal and transverse directions , respectively . furthermore , the @xmath2 valley degeneracy is generally lifted by a sharp perpendicular potential @xcite and the relevant energy separation is termed the valley splitting ( vs ) . + the valley splitting depends on physics at the atomic scale @xcite ( e.g. roughness , alloy and interface disorder ) , and so it is not surprising that experiments have revealed a large variability of splittings among devices , ranging from hundreds of @xmath4ev @xcite up to tens of mev in exceptional cases @xcite . at present , the lack of a reliable experimental strategy to achieve control over the vs is driving an intense research effort for the development of devices that can assure robust electron spin qubits by minimising multi - valley detrimental effects @xcite , or even exploit the valley degree of freedom @xcite for new types of qubits . + another crucial parameter to assess the suitability of a physical system to encode spin - based qubits is the relaxation time of spin excited states ( @xmath5 ) . spin lifetimes have been measured for gate - defined si quantum dots ( qds ) @xcite , si / sige qds @xcite and donors in si @xcite , reporting values which span from a few milliseconds to a few seconds . furthermore , the dependence of the spin relaxation rate ( @xmath6 ) on an externally - applied magnetic field ( @xmath7 ) has been investigated . different mechanisms apply to donors and qds , accounting for observed @xmath8 and @xmath9 dependencies @xcite , respectively . in principle , @xmath10 depends on the valley configuration and the details of the excited states above the spin ground state . however , until now , no experimental observation of the effects of a variable vs on the relaxation rate has been reported . + here , we demonstrate for the first time that the valley splitting in a silicon qd can be finely tuned by direct control of an electrostatic gate potential . we find that the dependence of the vs on vertical electric field at the si / sio@xmath1 interface is strikingly linear , and show that its tunability is in excellent agreement with atomistic tight - binding predictions . we demonstrate accurate control of the vs over a range of about 500 @xmath4ev and use it to explore the physics of spin relaxation for different qd occupancies ( @xmath11=1 , 2 , 3 ) . we probe both the regime where the vs is much larger than the zeeman splitting at all magnetic fields and that where the valley and spin splittings are comparable . we observe a dramatic enhancement of the spin decay rate ( relaxation hot - spot ) when spin and valley splittings coincide . to our knowledge , such hot - spots have been predicted for relaxation involving orbital states @xcite ( not valley states ) , but these are yet to be observed . we develop an analytic theory that explains the @xmath7-field dependence of the relaxation rates and the details of the relaxation hot - spot in terms of admixing of spin - valley states . this mechanism is seen to be significantly more prominent than the conventional spin - orbit hybridization @xcite . ) is sensitive to the qd charge state due to the qd / set capacitive coupling ( @xmath12 ) . ( c ) device cross - sectional schematic . an electron layer is formed underneath the positively biased gates : r1 and r2 define the qd reservoir ; p controls the qd population ; and st the sensor s island . the sio@xmath1 layer ( in purple ) thickness and plunger gate width are indicated . ( d ) energy diagram showing qualitatively the conduction band profile in the device . electrons accumulate wherever the gate bias lowers the conduction band below the fermi level , @xmath13 . ( e ) charging energy as a function of electron number . spikes corresponding to complete 2d shell filling are observed . ( f ) schematic of electron filling for two - valley 2d fock - darwin states . each state can hold two electrons of antiparallel spin and is identified by a pair of quantum numbers ( @xmath14 ) and its valley occupancy ( @xmath15 ) . ] our device is fabricated using a multi - level gated metal - oxide - semiconductor ( mos ) technology @xcite , and its architecture is depicted in fig . [ device](a)-(d ) . a quantum dot is formed under gate p by applying a positive bias voltage to induce an electron accumulation layer . strong planar confinement for the dot s potential well is achieved by negatively biasing gates b , c1 and c2 . a 2deg reservoir is also induced by positively biasing gates r1 and r2 , and the qd occupancy can be modified by inducing electrons to tunnel between this reservoir and the dot . the remaining gates , namely sb1 , sb2 , st , are employed to define a single - electron transistor ( set ) , capacitively coupled to the qd and used as a read - out device . the high flexibility of our design would allow us to use the same device also as a ( single - lead ) double dot structure by rearranging the gate bias ( e.g. dots can be formed under gates b and c2 ) . however , in this work , we only present results relevant to the single - dot configuration . + in order to characterize the addition spectrum of the qd , we make use of a technique previously developed for gaas - based systems which combines charge detection and gate pulsing @xcite . there is no direct transport through the single - lead qd and , therefore , addition / removal of charge is only detected via modifications in the set current . in particular , charge transitions are detected as current peaks in the set signal whenever the qd energy eigenstates come into resonance with the reservoir s fermi level . note that the set - qd coupling is merely capacitive ( via @xmath12 ) and electrons do not tunnel between them . in order to maximize charge sensitivity in the detector , we employ dynamic voltage compensation @xcite on different gates which makes our read - out signal virtually unaffected by slow charge drifts and random charge rearrangements . a comprehensive discussion of the charge stability measurements can be found in the supplementary note 1 . 1(e ) illustrates the addition energy spectrum for the first fourteen electron additions to the qd . there is very little variation of charging energy ( @xmath16 ) for high occupancies ( @xmath17 11 mev for @xmath18 ) . however , by decreasing the electron number , the charging energy steadily increases , as expected when the dot size is significantly affected by the electron number . this evidently indicates that the few - electron regime has been achieved . most interestingly , the energy spectrum shows peaks for the addition of the 5th and 13th electrons . the extra addition energy needed for those transitions can be attributed to complete filling of the first and second orbital shells . as illustrated in fig . 1(f ) , this is consistent with the energy spectrum of two - valley 2d fock - darwin states @xcite , where the first and second orbital shells hold 4 and 8 electrons respectively . this confirms that we can probe the occupancy until the last electron . to our knowledge , such a clear manifestation of two - dimensional shell structure has been observed before only in ingaas dots @xcite and in si / sige dots @xcite . in order to measure the spin state of individual electrons in the qd , we use an energy - selective readout technique @xcite . the readout protocol consists of three phases clocked by a three - level pulsed voltage applied to gate p , which directly controls the dot s electrochemical potential [ see fig . [ t1vsb ] ( a ) ] . firstly , an electron of unknown spin is loaded into the dot causing a sudden decrease in the sensor current . next , the potential of the dot is lifted so that the fermi level of the reservoir lies between the spin - up and spin - down states of the dot , meaning that a spin - up electron can tunnel off the dot while a spin - down electron is blocked . this is the read phase , during which the presence of a spin - up state would be signalled by a current transient ( spin - up tunnels out and then spin - down tunnels in ) whereas a spin - down electron would lead to no current modification . finally , the dot s potential is further lifted to allow the electron to tunnel off , regardless of its spin orientation . in fig . [ t1vsb ] ( b ) single - shot traces for both spin - up ( in blue ) and spin - down ( in green ) detection are plotted . the longer the system is held in the load phase before performing a read operation , the more likely it is for the spin - up excited state to decay to the spin - down ground state . thus , by varying the length of the load phase and monitoring the probability of detecting a spin - up electron , we can determine @xcite the spin lifetime , @xmath5 . in our experiments the @xmath7-field is directed along the [ 110 ] crystallographic axis . a comprehensive discussion of both the spin - up fraction measurements and the fitting procedure to evaluate @xmath5 is included in the supplementary note 2 . as shown in fig . [ t1vsb ] ( d ) , we observe a wide range of spin lifetimes as a function of magnetic field , with lifetimes as long as 2.6s at the lowest fields studied , @xmath7=1.25 t . these are some of the longest lifetimes observed to date in silicon quantum dots @xcite . + a key focus of our experiment was to electrostatically tune the valley energy separation and measure relaxation rates in different valley configurations and qd electron occupancies . _ as we show below , our data definitively indicate that excited valley states play a critical role in the spin relaxation processes . _ we develop a theory to explain how changes in the valley splitting affects the spin - valley state mixing and leads to the observed relaxation times . + as we detail in section [ vs ctrl ] , we have attained accurate gate control of the valley splitting , allowing us to tune it over a range of hundreds of @xmath4ev . this permits us to conduct experiments in regimes where the valley splitting ( @xmath19 ) is either larger or smaller than the zeeman spin splitting ( @xmath20 ) , depending on the magnitude of the magnetic field [ see fig . [ t1vsb ] ( c ) ] . [ t1vsb ] ( d ) presents measurements of spin relaxation rates as a function of magnetic field for two valley splitting values at a fixed dot population of @xmath21 . we start by examining a configuration where the valley separation is larger than the spin splitting at all fields ( green data set ) . in other words , we operate in a regime for which @xmath22 where @xmath23 is the electron gyromagnetic ratio , @xmath24 is the bohr magneton and @xmath7 is the applied in - plane magnetic field . + for @xmath19=0.75 mev [ green data in fig . [ t1vsb ] ( d ) ] , we observe a monotonic increase of the rate with respect to @xmath7 that becomes increasingly fast as @xmath20 approaches @xmath19 . in our experimental conditions ( @xmath7-field parallel to [ 110 ] ) , the @xmath8 dependences for known bulk - like mechanisms in silicon @xcite should not apply , while predicted @xcite rates @xmath25 do not explain the experimental data . + by decreasing the valley separation to @xmath260.33 mev , we can achieve the condition where the zeeman splitting matches or exceeds the valley splitting . the red data in fig . [ t1vsb ] ( d ) illustrate the situation where inequality ( 1 ) only holds for @xmath27 2.8 t . when @xmath28 ( i.e. for @xmath7=2.8 t ) , a spike in the relaxation rate occurs . relaxation hot - spots have been previously predicted to occur for spin relaxation involving orbital states in single and coupled qds @xcite._to our knowledge , this is the first experimental observation of such a phenomenon . _ + in order to understand the relaxation mechanisms , we have developed a model that takes into account the perturbations in pure spin states due to spin - orbit coupling ( soc ) , yielding eigenstates which are admixtures of spin and valley states . the four lowest spin - valley states [ see fig . [ t1vsb](c ) ] are defined as @xmath29 , @xmath30 , @xmath31 , @xmath32 . these states are considered to be only very weakly affected by higher excitations , such as orbital levels which are at least 8 mev above the ground state in our device @xcite . in the supplementary note 3 we detail how mixing to a 2@xmath33-like orbital state leads to a @xmath9 dependence in @xmath6 and is , therefore , important mainly for high @xmath7-fields ( above the anticrossing point ) . at lower fields , the prominent mechanism is the spin - valley admixing , which we now discuss in detail . + the relaxation between pure spin states is forbidden because the electron - phonon interaction does not involve spin flipping . however , in the presence of interface disorder , soc can mix states that contain both the valley and spin degrees of freedom , thus permitting phonon - induced relaxation . indeed , in the non - ideal case of qds with a disordered interface , roughness can perturb the envelope function of both valleys ( otherwise identical for ideal interfaces ) and allows one to assume non - zero dipole matrix elements connecting the valley states ( see supplementary note 3 ) , such as @xmath34 , @xmath35 , @xmath36 ( for ideal interfaces these are non - zero only due to a strongly suppressed umklapp process ) . by means of perturbation theory , we define renormalized excited states @xmath37 and @xmath38 that can relax to the ground state @xmath39 , as they have an admixture of the state @xmath40 of the same spin projection [ see fig . [ t1vsb](c ) ] . the details of the soc hamiltonian , @xmath41 , and perturbation matrix are reported in the supplementary note 3 . the leading order wavefunctions are given by : @xmath42 @xmath43 where @xmath44^{1/2}$ ] , @xmath45^{1/2}$ ] , and @xmath46 is an expression involving the detuning from the anticrossing point , @xmath47 , and the energy splitting at the anticrossing : @xmath48 where @xmath49 ( @xmath50 ) is the dresselhaus ( rashba ) soc parameter , @xmath51 is the reduced planck s constant and @xmath52 is the transverse effective electron mass . + by evaluating the relaxation rate via the electron - phonon deformation potentials ( proportional to the deformation potential constants , @xmath53 ) , we obtain the rate below the anticrossing as : @xmath54 where the pure valley relaxation rates are ( for longitudinal and transverse phonons ) : @xmath55 = \frac{\delta e_{v'v}^5}{4\pi \rho \hbar^6}\ , \frac{r^2}{v_{\sigma}^7}\ , i^{(\sigma ) } \label{valley - relax - gam0}\ ] ] where @xmath56 is the silicon mass density , @xmath57 is the speed of sound in silicon , @xmath58 $ ] , @xmath59 are the angular integrals , and @xmath60 and * _ r _ * are the energy difference and the dipole matrix element relevant to the transition , respectively ( see also supplementary note 3 ) . the experimental condition for which the hot - spot occurs ( i.e. @xmath61 ) is modelled as an anticrossing point for the mixed states @xmath37 and @xmath38 . at that point , spin relaxation is maximized and @xmath62 approaches the valley relaxation rate , as @xmath63 in eq . ( [ rate21 ] ) . + above the anticrossing ( i.e. @xmath64 ) , the relevant relaxation transitions are @xmath65 and @xmath66 ( the subsequent decay @xmath67 is in the form of a fast inter - valley transition and is , therefore , neglected ) . the analytical formulations of these contributions read : @xmath68 @xmath69\ ] ] + the dashed lines in fig . 2(d ) show the calculated relaxation rates relevant to the two experimental values of @xmath19 discussed , including also @xmath9 contribution from soc mixing with the higher orbital state ( see supplementary note 3 ) . we use dipole matrix elements as a single free parameter by assuming @xmath70 . a least - square fit to the experimental data is performed by fixing the soc strength to ( @xmath71)@xmath72 m / s ( justified by the high electric field @xmath73v / m , see refs . [ @xcite ] ) . the fit then extracts a dipole size @xmath741 - 2 nm for both values of @xmath19 . + the good agreement between the calculations and experiment , as well as the presence of a hot - spot at the point of degeneracy between zeeman and valley splitting , provide strong evidence of our ansatz that the spin relaxation is predominantly due to a _ new mechanism _ : that of mixing with the excited valley states via rashba / dresselhaus - like soc in the presence of interface disorder . + both the splitting at the anticrossing , eq.([delta_a - b-(110 ) ] ) , and the intervalley relaxation , eq.([valley - relax - gam0 ] ) , depend crucially on the size of the dipole matrix element , @xmath75 , predicting a fast phonon relaxation of @xmath76 for @xmath77 , at the hot - spot of fig . [ t1vsb](d ) . this confirms our core findings that when spin - valley states anticross , the inter - valley rates are fast for these samples , with the only available relaxation mechanism being the inter - valley decay . we point out that these relaxation rates are expected to be sample / material - dependent , given the effect of interface disorder on valley mixing . + we now examine the case @xmath78 electrons , and investigate the dependence of the relaxation rate on the magnetic field at a fixed valley splitting ( @xmath19=0.58 mev ) . we note that the energy levels accessible for loading the second electron in the dot , when the @xmath21 spin - down ground state is already occupied , are either the singlet ( @xmath79 ) or the two lower triplets ( @xmath80 ) , while the higher triplet ( @xmath81 ) would require a spin - flip and is , therefore , not readily accessible [ see fig . [ multi_n](a ) ] . in general , for triplet states , the antisymmetry of the two - electron wavefunction requires one electron to occupy a higher energy state . for our multi - valley qd [ fig . [ device ] ( f ) ] , this requirement is fulfilled when the two electrons occupy different valley states [ see fig . [ multi_n ] ( a ) ] . for low fields , the ground state is @xmath79 and the triplets have higher energies . this results in excited states ( triplets ) that extend over two valleys and relax to single - valley ground state ( singlet ) . + as the magnetic field is increased , @xmath79 and @xmath82 undergo an avoided crossing ( @xmath83 ) , and then @xmath82 becomes the ground state . we adjust the levels of our pulsed readout protocol so that , during the load phase only @xmath79 and @xmath82 are below the reservoir s fermi energy , while , during the read phase , the fermi energy is positioned within the singlet - triplet ( st ) energy gap . as a consequence , for @xmath84 ( @xmath85 ) a @xmath82 ( @xmath79 ) state would be signalled with a current transient , and relaxation rates can be extracted as for the @xmath21 occupancy . the experimental relaxation rates in fig . [ multi_n](b ) show a strongly non - monotonic behaviour , approaching an absolute minimum at the anticrossing point ( @xmath86=5 t ) . the trend is strikingly symmetric , as can be appreciated when @xmath6 is plotted against the detuning energy , as shown in the right inset . this symmetry is reflected in the qd energy spectrum [ fig . [ multi_n](a ) ] , as far as the detuning @xmath87 is concerned . for @xmath88 , the st energy gap decreases with increasing @xmath7 , resulting in slower relaxation rates . by contrast , for @xmath89 , the st energy gap increases for increasing field , and so does the relaxation rate . + as opposed to the 1-electron case , we note that the 2-electron eigenstates anticrossing leads to a _ minimum _ in the relaxation rate ( _ cold - spot _ ) , defined by a splitting at the anticrossing , @xmath90 , of the same order as that of eq.([delta_a - b-(110 ) ] ) ( see fig . [ multi_n](a ) and supplementary note 3 ) . the occurrence of this minimum does not strictly depend on the nature of the states involved in the decay ( spin - like , valley - like , orbital - like or admixtures ) . it is due to the fact that the avoided crossing takes place between the ground and the first excited state , while for the case @xmath11=1 it involves the first and the second excited states without affecting the ground state . + to model the 2-electron case , we build the wavefunctions for @xmath79 and @xmath82 from the single - particle states by considering the coulomb interaction as a perturbing averaged field . the corresponding states are defined as @xmath91 , @xmath92 . next , the additional perturbation given by soc leads to renormalized eigenstates which are admixtures of singlet and triplet : @xmath93 @xmath94 being similar forms to eq . ( [ 2correct ] ) ( [ 3correct ] ) . as we show in the supplementary note 3 , by evaluating the electron - phonon hamiltonian matrix element for the transition between these states , one finds that it coincides in its form with its 1-electron counterpart for @xmath95 . therefore , we can conclude that the corresponding relaxation rate , @xmath96 , has the same functional form as those derived in eq . ( [ rate32 ] ) , although the matrix elements for the two cases will be different ( see supplementary note 3 ) . dashed lines in fig . [ multi_n ] ( b ) represent the calculated rates which are fitted to the experimental data similarly to the case @xmath21 . once again , the model convincingly reproduces the main features of the experimental trend , in particular the rates for fields away from the anticrossing together with the symmetry of the characteristics with respect to @xmath86 . further work may be needed to improve the fit in the vicinity of the anticrossing point . + we also measured the relaxation rates for @xmath97 electrons . when the qd occupancy is set at @xmath11=2 , the lower valley is fully occupied and for low @xmath7-fields the ground state is a singlet . in this condition , the readout protocol is adjusted to probe spin relaxation within the upper valley upon loading / unloading of the third electron . by keeping @xmath98mev and using the same methodology described before , we measure relaxation rates for the third electron spin state . we find that there is no significant difference between the spin relaxation rates for @xmath97 and @xmath21 , as shown in the left inset of fig . [ multi_n ] ( b ) . two main conclusions can be drawn from this . firstly , we can infer that the effect of electron - electron interactions on the multi - valley spectrum may be negligible @xcite , which is plausible . indeed , for valley 3-electron states , two electrons are just `` spectators '' , so that the remaining electron establishes the same energy level structure as in fig . [ t1vsb ] ( c ) , and the coulomb corrections should not affect the valley splitting . secondly , as we report in ref . [ @xcite ] , in small qds for higher occupancies a significantly reduced energy separation between the ground state and the first excited orbital state is observed . this would introduce a non - negligible perturbation on the relaxation if this were affected by the orbital degree of freedom . hence , the similarities in behaviour in terms of decay rates are a further indication that for our qd the dominant relaxation mechanism resides in the degree of spin - valley admixing , as opposed to the spin - orbit admixing relevant for other semiconductor systems @xcite . we now turn to the experimental demonstration of accurate control of the valley splitting , @xmath19 , via electrostatic gating . to determine @xmath19 , we use two different experimental approaches . one utilizes the rapid increase in spin relaxation at the hot - spot , and is applicable in the low magnetic field regime . the other is based on magnetospectroscopy , and is relevant for high fields . + the first technique stems from the fact that the hot - spot can be reliably detected by monitoring the spin - up probability as a function of magnetic field . in fig . [ fig_vs ] ( a ) , we show measurements of the spin - up probability performed with the same method as the one used to evaluate spin lifetimes ( see supplementary note 2 ) . we see that the probability of detecting a spin - up electron decreseas significantly at some magnetic fields . a sudden drop of the spin - up fraction in a narrow range of field identifies the increase in relaxation rate associated with the hot - spot . given that valley and zeeman splittings coincide at the hot - spot , one can extract the valley separation as @xmath99 , where @xmath100 is defined as the field at which the hot - spot is observed . for varying gate voltage configurations , we scan @xmath7 in the range 2.8t@xmath1015 t , and identify @xmath100 by setting an arbitrary probability threshold [ green shaded area in fig . [ fig_vs](a ) ] below which the hot - spot is assumed to occur . the use of this technique is limited to @xmath102 t because the lifetime drop at the hot - spot can be therein confidently assessed . at higher fields the relaxation becomes increasingly fast and its enhancement at the hot - spot is indistinguishable within our measurement bandwidth ( @xmath103 10 khz ) . + in order to evaluate @xmath19 at higher magnetic fields , we use a more conventional magneto - spectroscopic approach , as shown in fig . [ fig_vs ] ( b ) . by employing the same gate - pulsed technique used for the charge stability experiments ( see supplementary note 1 ) , we focus on the singlet - triplet ground state transition as we load the second electron into the dot ( i.e. @xmath104 transition ) in the range [email protected] t . this is clearly identified as the point where the @xmath79 ( light grey feature ) and @xmath82 ( dark grey feature ) states cross . here , @xmath105 , as seen in fig . [ multi_n](a ) . + the data points in fig . [ fig_vs ] ( c ) represent the measured valley separation as a function of @xmath106 , obtained by means of the aforementioned techniques . the solid line fit shows remarkable consistency between the two sets of data and reveals that @xmath19 depends linearly on the gate voltage over a range of nearly 500 @xmath4ev , with a slope of 640 @xmath4ev / v . in order to keep constant the dot s occupancy and tunnelling rates for different @xmath106 , a voltage compensation is carried out by tuning gates c1 and b accordingly . we note that we previously reported valley splittings of comparable magnitude ( few hundreds of @xmath4ev ) in devices realized with the same technology @xcite . however , to our knowledge , this is the first demonstration of the ability to accurately tune the valley splitting electrostatically in a silicon device . + a linear dependence of the valley splitting with respect to the vertical electric field has been predicted for 2deg systems via effective mass theory @xcite . a similar dependence for mos - based qds @xcite has also been reported by employing atomistic tight - binding calculations @xcite . in order to compare our experimental finding with the theoretical predictions , we simulate the vertical electric field ( @xmath107 ) in the vicinity of the dot for the range of gate voltages used in the experiments . we employ the commercial semiconductor software ise - tcad @xcite to model the device electrostatic potential , and thereby the electric fields in the nanostructure . for this purpose , tcad solves the poisson equation with an approximation of newton s iterative method @xcite to obtain convergence at low temperatures . + the spatial extent of the dot is identified by regions where the calculated conduction band energy drops below the fermi level [ red area in the top inset of fig . 4(c ) ] . note that our calculations are performed on a three - dimensional geometry identical to the real device with the only free parameter being the amount of offset interface charge . this is adjusted to match the experimental threshold voltage of the device ( @xmath108=0.625v ) , as explained in the supplementary note 4 . + the computed variation of interface electric field with gate voltage @xmath106 is used to determine the valley splitting according to both the atomistic @xcite and effective mass @xcite predictions . dashed lines in fig . [ fig_vs ] ( c ) depict the trends for both approaches , with both exceeding by more than 1 mev the measured values . despite this offset , the atomistic calculations give a tunability of the valley splitting with gate voltage , @xmath109 , in good agreement with the experiments . the calculated value of 597 @xmath4ev / v agrees with the measured value to within less than 7% . the value of 541 @xmath4ev / v calculated using the effective mass approach reveals a larger deviation ( @xmath110% ) from the experiments . the presence of an offset in the computed valley splitting may be due to the contribution of surface roughness that is not accounted for in the models , and is thought to be responsible for a global reduction of @xmath19 @xcite . we emphasize , however , that the gate tunability would remain robust against this effect , which is not dependent on electric field . in this work , we have shown that the valley splitting in a silicon device can be electrostatically controlled by simple tuning of the gate bias . we used this valley splitting control , together with spin relaxation measurements , to explore the interplay between spin and valley levels in a few - electron quantum dot . + the relaxation rates for a one - electron system exhibit a dramatic hot - spot enhancement when the spin zeeman energy equals the valley splitting , while for a two - electron system the rates reach a minimum at this condition . we found that the known mechanisms for spin relaxation , such as the admixing of spin and @xmath33 orbital states , were unable to explain the key features of the experimental lifetime data , and so introduced a novel approach based on admixing of valley and spin eigenstates . our theory , which showed good agreement with experiment , implies that spin relaxation via phonon emission due to spin - orbit coupling can occur in realistic quantum dot systems , most likely due to interface disorder . + our results show that by electrical tuning of the valley splitting in silicon quantum dots , it is possible to ensure the long lifetimes ( @xmath111 s ) required for robust spin qubit operation . despite this , the excited valley state will generally be lower than orbital states in small quantum dots , placing an ultimate limit on the lifetimes accessible in very small dots , due to the spin - valley mixing described above . + electrical manipulation of the valley states is also a fundamental requirement to perform coherent valley operations . however , the experimental relaxation rate at the observed hot - spot was found to be fast ( @xmath112 khz ) for our devices , implying a fast inter - valley relaxation rate . + finally , in the context of realizing scalable quantum computers , these results allow us to address questions of device uniformity and reproducibility with greater optimism . indeed , our work suggests that issues related to the wide variability of the valley splitting observed in silicon nanostructures to date can shift from the elusive atomic level ( surface roughness , strain , interface disorder ) to the more accessible device level , where gate geometry and electrostatic confinement can be engineered to ensure robust qubit systems . the samples fabricated for these experiments are silicon mos planar structures . the high purity , near intrinsic , natural isotope silicon substrate has n+ ohmic regions for source / drain contacts defined via phosphorous diffusion . high quality sio@xmath1 gate oxide is 8 nm thick and is grown by dry oxidation at 800@xmath113c . the gates are defined by electron - beam lithography , al thermal evaporation and oxidation . three layers of al / al@xmath1o@xmath114 are stacked and used to selectively form a 2deg at the si / sio@xmath1 interface and provide quantum confinement in all the three dimensions . + measurements are carried out in a dilution refrigerator with a base temperature t@xmath115 40mk . flexible coaxial lines fitted with low - temperature low - pass filters connect the device with the room - temperature electronics . in order to reduce pick - up noise , the gates are biased via battery powered and opto - isolated voltage sources . the set current is amplified by a room - temperature transimpedance amplifier and measured via a fast digitizing oscilloscope and a lock - in amplifier for the single - shot and energy spectrum experiments , respectively . gate voltage pulses are produced by an arbitrary wave - function generator and combined with a dc offset via a room - temperature bias - tee . the authors thank d. culcer for insightful discussions , and f. hudson and d. barber for technical support . this work was supported by the australian national fabrication facility , the australian research council ( under contract ce110001027 ) , and by the u.s . army research office ( under contract w911nf-13 - 1 - 0024 ) . the use of nanohub.org computational resources operated by the network for computational nanotechnology funded by the us national science foundation under grant eec-0228390 is gratefully acknowledged . the authors declare no competing financial interests . c.h.y . carried out the measurements . n.s.l . designed and fabricated the device . c.h.y . , a.r . analyzed the data . c.h.y . , a.r . , , g.k . , a.m. , a.s.d . discussed the results . , c.t . modelled the relaxation rates . f.m . simulated the electric field profiles . conceived and planned the project . a.r . wrote the manuscript with input from all coauthors . the addition spectrum of the qd is characterized via a technique widely used for gaas - based systems , which employs both charge detection and gate pulsing@xmath116 . a train of square voltage pulses is applied to gate p , in addition to its dc voltage , shifting the energy levels of the dot up and down . this pulse train modulates the set sensor current with the same frequency of the pulse due to cross - capacitance effects . the resulting current at this frequency is measured with a lock - in amplifier . full details of the measurement set - up are reported elsewhere@xmath117 . supplementary figure s1 ( a ) shows the measured stability map as a function of plunger gate voltage , @xmath106 , and barrier gate voltage , @xmath118 ( see inset ) . current peaks in the lock - in signal indicate the occurrence of charge transitions due to tunnelling between the qd and the reservoir . we have converted this signal into an average change of occupancy@xmath117 , @xmath119 , which is shown in the grey - scale plot . for each transition , we can also probe the first orbital excited state whenever a slow decrease in the detection signal is followed by a steep rise@xmath117 . we observe charge additions for the first fourteen electrons in the dot . in order to translate the voltage dependence of these features into a spectroscopy of the dot s charging energy ( @xmath16 ) , we determine the voltage to energy conversion factor , @xmath120 , for gate p via temperature dependence measurements and we find @xmath121 0.16 ev / v . the dashed horizontal line indicates that the addition spectrum shown in the main article has been obtained for @xmath118=0.2v . here , we describe the procedure utilized to measure spin lifetimes for varying @xmath7 . the probability of observing a spin - up electron , @xmath122 , decreases when the loading time , @xmath123 , is increased . a threshold detection method@xmath124 is used to determine the fraction of shots that contribute to the spin - up count for each @xmath123 at different @xmath7 . supplementary figure s1 ( b ) shows that the probability is well fitted by a single exponential time decay , @xmath125 , from which @xmath126 is evaluated . + by modifying the pulse sequence and recording readout traces for different voltage levels of the read phase , it is possible to extract the single - particle zeeman splitting@xmath124 . indeed , by stepping the read voltage from a level where both spin states are lifted above the fermi energy of the reservoir to a level where both spin states are pushed below the reservoir s energy , the separation between the two spin states can be evaluated . supplementary figure s1 ( c ) shows the qd occupancy averaged over 256 single - shot traces for varying read levels at @xmath7=3 t ( note that the energy scale on the y - axis is the effective shift of the dot s potential ) . as explained in the main article , the detection of a spin - up electron would be signalled by a transient in the detector s current . this transient is seen to extend for about 350@xmath4ev , as expected from the zeeman equation , @xmath127 , by assuming @xmath128 . the field dependence of the measured @xmath129 also shows good agreement with the zeeman equation . this proves that we can reliably load / unload electrons onto spin - split single - particle states . + we consider a small quantum dot ( qd ) defined electrostatically in a mos si / sio@xmath1 heterostructure , as described in the main text . for a qd occupied by a single electron the first excited orbital state is some 8 mev above the ground state and it only weakly influences the ground state shell ( see however below ) . due to strong electric field at the heterostructure , the 6-fold degeneracy of the conduction band electrons is lifted : only two lowest energy valleys remain relevant , at momentums @xmath130 $ ] , with @xmath131 ; the remaining two - fold degeneracy is lifted via the sharp interface potential , leading to a valley splitting , @xmath19 , of the order of few hundred @xmath132 , and linearly proportional to the electric field at the interface , @xmath133 . this forms the ground state shell of the two lowest valley states , @xmath134 and @xmath135 ; ( these approach states of definite parity , @xmath136 in strong field , as per notation in the main article ) . in the effective mass approximation the valley wave functions are generally written as @xmath137 where @xmath138 are the kohn - luttinger valley functions , with the periodic part @xmath139 , and @xmath140 ( @xmath141 ) is the position of the valley minima . here , @xmath142 are the envelope functions corresponding to the orbital ground state ( @xmath143-like ) . in the ideal qd ( ideal interface , symmetric circular dot , etc . ) the envelopes are separable , e.g. : @xmath144 ; generally , the index of the state , @xmath134 or @xmath135 , distinguishes different envelopes , originating from valley - orbit coupling in the presence of a non - ideal interface , also separability may be lost . the relations : @xmath145 and @xmath146 are maintained for each state by time reversal . for the states of definite parity above , the valley populations are : @xmath147 , @xmath148 . by switching on a magnetic field , the states are zeeman split and one gets four lowest spin - valley states denoted as @xmath149 , @xmath150 , @xmath151 , @xmath152 . the corresponding ( unperturbed ) energies are @xmath153 , @xmath154 , @xmath155 , @xmath156 , where levels 2 and 3 cross at @xmath28 , and @xmath157 is the zeeman splitting . since electron - phonon interaction does not flip spin , phonon relaxation between @xmath158 and @xmath159 is forbidden , while @xmath160 to @xmath159 is allowed as a pure valley ( non - spin flip ) relaxation . it is the mixing between valley states @xmath158 and @xmath160 due to spin - orbit coupling ( soc ) that causes a renormalized state @xmath161 to have a spin - down component , that allows phonon transition @xmath162 to take place . the soc interaction forms an anticrossing point for the states @xmath161 , @xmath163 , with a splitting @xmath164 , where the spin flip is maximal and so is the @xmath162 relaxation , reaching the order of pure phonon - valley relaxation , which is observed as a relaxation `` hot - spot '' ( see fig . 2 ( c , d ) of main text ) . both the anticipated non - zero splitting at the anticrossing of the two valley states , and the phonon relaxation mechanism requires non - zero valley dipole matrix elements : @xmath165 , @xmath166 , and @xmath167 . in the ideal case , however , they will take a non - zero value only from umklapp ( intervalley ) contributions , which are highly suppressed in a qd . while a complete theory of the valley wave functions in the presence of a roughness / steps is still under development , some gross effects of the experiment can be explained just by _ postulating _ non - zero dipole matrix elements . the intravalley contribution is generally given by ( it is zero in the ideal case ) : @xmath168 and similarly for @xmath169 , @xmath170 . numerical fit to the experimental data presented in this paper consistently reveal values of @xmath171 the order of few nm ( these may depend on the applied electric field at the interface , and also on the qd occupancy ) . to calculate the splitting at anticrossing of the spin - valley states , @xmath150 , and @xmath151 , we use the spin - orbit coupling ( soc ) . in two dimensions the well known soc hamiltonian reads : @xmath172 , with the dresselhaus and rashba terms , related to conduction band electrons , confined in a @xmath173 $ ] 2deg : @xmath174 here @xmath175 , @xmath176 are the pauli matrices along the principal crystal axes , and @xmath177 is the generalized momentum . related to current experiment , we consider an in - plane magnetic field , which is parallel to the @xmath178-direction , and its vector potential in the symmetric gauge reads : @xmath179 . using the eigenoperators @xmath180 , ( @xmath181 ) , and hamiltonian commutation relations one can express the perturbation matrix via dipole matrix elements , e.g. : @xmath182 while @xmath183 . estimation for the splitting gives : @xmath184 , where , we have used transverse mass @xmath185 , and @xmath186 is the dipole size ( @xmath187 ) . we assume a soc strength @xmath188 for an interface field of @xmath189 ( implying it may reach @xmath190 higher values in the current experiment ) . note , that in general , @xmath164 depends linearly on the valley splitting and the dipole matrix element . to calculate the phonon relaxation rate between the lowest spin - valley states we take into account the hybridization of the levels @xmath158 and @xmath160 due to soc [ see fig . the relevant matrix elements @xmath191 \label{relaxat - me32}\end{aligned}\ ] ] are expressed via the admixture of the state @xmath192 , to the states @xmath161 , @xmath163 ( eqs . ( 2),(3 ) of the main text ) and via the phonon transition matrix element between the valley states . calculation of the valley phonon relaxation rate is performed via the electron - phonon deformation potential interaction @xcite in the approximation when only intravalley contributions are taken into account , and using the non - zero dipole matrix element , eq.([dipole - me ] ) . for the relaxation below the anticrossing ( @xmath162 ) we obtain @xmath193 . above anticrossing , the experimentally observable transitions are @xmath194 and @xmath195 , see fig . 2 ( c ) of main text . ( in the second case , a subsequent decay of @xmath162 above anticrossing is purely valley and therefore very fast ) . the rates are : @xmath196 and @xmath197 , respectively . in all three cases above , the valley phonon relaxation rate @xmath198 is of the same functional form ( we used dipole approximation ) : @xmath199 = \frac{1}{4\pi \rho \hbar}\ , \frac{\omega_{v'v}^5}{v_{\sigma}^7 } \ , \int_{-1}^1 dx \ , d(x,\bm{r}_{12 } ) \ , \xi^{(\sigma)}(x ) \qquad \label{valley - relax - gam}\\ & & d(x,\bm{r}_{12 } ) \equiv \frac{x_{12}^2 + y_{12}^2 } { 2 } ( 1-x^2 ) + z_{12}^2\,x^2 , \ \ x\equiv \cos(\theta ) \label{valley - relax - d } , \end{aligned}\ ] ] and depends on the actual transition energy @xmath60 ; here @xmath200 depends quadratically on the dipole components @xmath201 . note , that for the transition @xmath195 one have to substitute @xmath201 by the difference dipole matrix element @xmath202 . the integrand , that is proportional to the deformation potential constants , @xmath203 , is : @xmath204 for longitudinal phonons , and @xmath205 for transverse phonons . using eq.([valley - relax - gam ] ) , one can see that the dependence of the relaxation on magnetic field is @xmath206 far below anticrossing , since the spin admixture coefficient @xmath207 is only weakly @xmath7-dependent for @xmath208 . at anticrossing , the mixing between spin - up and spin - down states is maximal , and relaxation reaches a fast ( pure valley ) transition rate ; e.g. , for @xmath209 t , @xmath210 for @xmath77 . numerical fits to the experimental data show that the intervalley soc mixing mechanism alone , is not enough to describe the actual @xmath7-dependence above anticrossing ( except close vicinity of the hot - spot ) . an additional mechanism of virtual excitation of the first orbital @xmath211-state with overall @xmath9 dependence becomes important at high magnetic fields ( see below ) . _ spin - valley relaxation mechanism in the 2 electron - case . _ + in considering the two - electron case one replaces the coulomb interaction by a mean field : @xmath212 , and the 2-electron wave functions are constructed from the corresponding single - particle states , @xmath213 , eqs . ( [ v1 - 1e]),([v2 - 1e ] ) . [ the single - particle states in the 2-electron case should include corrections due to the mean field potential , neglected in zeroth - order approximation ] . the corresponding spin - valley states are the spin - singlet and spin - triplet : @xmath214 \label{2e - singlet}\\ & & |v_2',{t_- } \rangle = \frac{1}{\sqrt{2}}\ , \left [ \varphi_1(\bm{r}_1 ) \varphi_2(\bm{r}_2 ) - \varphi_1(\bm{r}_2 ) \varphi_2(\bm{r}_1 ) \right ] \otimes \label{2e - triplet } .\end{aligned}\ ] ] for the energy difference in the 2-electron case , one have to take account of the coulomb interaction in the two configurations : @xmath215 . since the valley states are @xmath143-like ( and using mean field approximation ) one can show that the 2-electron valley splitting coincides with the 1-electron case : @xmath216 . similar to the 1-electron case , one calculates the splitting at anticrossing for the levels @xmath217 and @xmath218 . for magnetic field along the @xmath178-direction the non - diagonal matrix element coincides with the single - particle one , eq.([b-(110 ) ] ) : @xmath219 and the splitting is the same as well : @xmath220 . thus , the hybridized states ( in the presence of soc ) , are admixtures of @xmath221 , @xmath222 , with the same admixture coefficients @xmath223 , @xmath224 , as for the single - electron case . the electron - phonon matrix element can be calculated : @xmath225 $ ] and shown to coincide with the 1-electron matrix element of the @xmath226 transition : @xmath227 . therefore , the corresponding rates also coincide : @xmath228 $ ] , as the energy difference is the same for both transitions . numerical fit to the data just above anticrossing in the 1-electron case for @xmath229 mev , and a fit to the data in the 2-electron case for @xmath98 mev ( see fig . 2(d ) and fig . 3(b ) of the main text ) are consistent with dipole matrix element @xmath202 of few nm . therefore , the experiment inevitably shows that the difference dipole matrix element is non - zero , @xmath230 . _ including @xmath9-corrections due to virtual transitions to orbital 2p - states . 1-electron case _ + in the one electron case we consider orbital excitations above the ground state @xmath231 , split by @xmath232 . for an in - plane magnetic field with weak magnetic confinement , we have @xmath233 and these states are separable@xmath234 : @xmath235 . the corresponding non - zero dipole matrix elements are @xmath236 . soc makes the qubit states , @xmath159 , @xmath158 , to mix with these upper orbital states . calculation in the first order of pt is performed ( similarly to ref.[38 ] ) , while we neglect small corrections due to hybridization of the states @xmath158 , @xmath160 . the resulting rate has the @xmath9 dependence on the magnetic field ( expected for a @xmath237 virtual transition ) : @xmath238 \frac{m_t^2 ( \beta_d - \alpha_r)^2 e_{v'v}^5}{4\pi \rho \hbar^8}(ta+la)\ , \left [ y_{0m_1}^4 + x_{0m_1}^4 \right ] \label{2p-1e - case } ; \ ] ] here @xmath239 accounts for contribution of transverse and longitudinal acoustic phonons . numerical fit to the 1-electron experimental data for @xmath240 shows that this relaxation mechanism has to be taken in to account in order to describe the data above anticrossing . the fit is consistent with @xmath241 dipole matrix elements about twice bigger than for a parabolic qd , of the order of @xmath242 . _ 2-electron case @xmath243 corrections _ + an analogous calculation in the 2-electron case was performed as well . we examined virtual excitations to the whole set of excited states that includes the triplet valley states , the upper singlet valley state , and the singlet and triplet states , involving orbital single - electron states . we found that the only relevant state ( capable to produce a relaxation rate correction of @xmath243 behavior ) is the valley - spin - orbital state : @xmath244 \otimes |{t_{- } } \rangle$ ] , where one of the electrons is occupying @xmath245 state , while the other is orbitally excited . we have derived a contribution to the relaxation rate , @xmath246 , similar to the 1-electron case . however , including this contribution does not help to fit better the 2-electron data for reasonable parameters . _ best fit parameters _ + for the best fit to the 1-electron and 2-electron experimental data we use : + ( 1 ) soc strength @xmath247 , which is justified , since the electric field in the experiment is @xmath248 . + ( 2 ) we take ( for each @xmath19 ) that @xmath249 ( justified by common sense and preliminary calculations ) . + ( 3 ) splitting to the first orbital excited state , @xmath250 ( 1-electron case ) and @xmath251 ( 2-electron case ) taken from experimental values . + ( 4 ) in the 1-electron fit , the @xmath241 dipole matrix elements are @xmath252 ( a plausible assumption for a non - parabolic dot ) , and in the 2-electron case we take the same values for these matrix elements . least - square fit to the data then reveals the valley dipole matrix elements for each set of data , as follows : @xmath253 ( @xmath229 mev ) , @xmath254 ( @xmath255 mev ) in the 1-electron case , and @xmath256 ( @xmath98 mev ) in the 2-electron case , consistent with our expectations . the tcad model of the real device geometry is shown in supplementary figure s2 ( a ) . as mentioned in the main article , in our simulations we use the same voltage range as in the experiments and the only free parameter is the si / sio@xmath1 interface charge , @xmath257 . this has been chosen to match the experimental threshold voltage ( @xmath108=0.625 v ) . in supplementary figure s2 ( b ) the dependence of @xmath108 on @xmath257 is reported . this has been evaluated by integrating the charge density in a region of the 2deg near the interface as all the gate voltages are simultaneously swept ( see inset ) . the value of choice is @xmath258 @xmath259 . the vertical electric field profile in the ( @xmath260 ) plane at @xmath106=1.4v is shown in supplementary figure s2 ( c ) . as highlighted in the main article , the qd is formed in the region where the conduction band falls below the fermi energy , so that the vertical interface field therein is the one relevant to calculate the valley splitting . finally , with the same methodology illustrated in the paper , the dependence of the valley splitting ( at @xmath261 = 1.4v ) on the interface charge density has been estimated with both the atomistic and the effective mass theories , and is shown in supplementary figure s2 ( d ) . it is of note that the variation of @xmath19 with interface charges is relatively small and can not account for the large offset between computed and experimental values . hence , we believe that the main factors that reduce valley splitting in this nanostructure are due to interface effects ( as discussed in the main article ) . kane , b. e. a silicon - based nuclear spin quantum computer . _ nature _ * 393 * , 133 - 137 ( 1998 ) . maune , b. m. _ et al . _ coherent singlet - triplet oscillations in a silicon - based double quantum dot . _ nature _ * 481 * , 7381 ( 2012 ) . pla , j. j. _ et al . _ a single - atom electron spin qubit in silicon . _ nature _ * 489 * , 541 - 545 ( 2012 ) . tyryshkin , a. m. _ et al . _ electron spin coherence exceeding seconds in high - purity silicon . _ nature mat . _ * 11 * , 143 ( 2012 ) . ando , t. , fowler , a. b. & stern , f. electronic properties of two - dimensional systems . _ phys . _ * 54 * , 437 ( 1982 ) . sham , l. j. & nakayama , m. effective - mass approximation in the presence of an interface . b _ * 20 * , 734 ( 1979 ) . saraiva , a. l. _ et al . _ intervalley coupling for interface - bound electrons in silicon : an effective mass study . b _ * 84 * , 155320 ( 2011 ) . saraiva , a. l. _ et al . _ physical mechanisms of interface - mediated intervalley coupling in si . b _ * 80 * , 081305(r ) ( 2009 ) . boykin , t. b. _ et al . _ valley splitting in strained silicon quantum wells._appl . * 84 * , 115 ( 2004 ) . friesen , m. , chutia , s. , tahan , c. & coppersmith s. n. valley splitting theory of sige / si / sige quantum wells . _ b _ * 75 * , 115318 ( 2007 ) . ando , t. valley splitting in the silicon inversion layer : misorientation effects . b _ * 19 * , 3089 ( 1979 ) . friesen , m. , eriksson , m. a. & coppersmith , s. n. magnetic field dependence of valley splitting in realistic si / sige quantum wells._appl . lett . _ * 89 * , 202106 ( 2006 ) . goswami , s. _ et al . _ controllable valley splitting in silicon quantum devices . _ nature phys . _ * 3 * , 41 - 45 ( 2007 ) . lansbergen , g. p. _ et al . _ lifetime - enhanced transport in silicon due to spin and valley blockade . lett . _ * 107 * , 136602 ( 2011 ) . roche , b. _ et al . _ detection of a large valley - orbit splitting in silicon with two - donor spectroscopy . * 108 * , 206812 ( 2012 ) . takashina , k. , ono , y. , fujiwara , a. , takahashi , y. & hirayama , y. valley polarization in si(100 ) at zero magnetic field . lett . _ * 96 * , 236801 ( 2006 ) . culcer , d. , cywiski , . , li , q. , hu , x. & das sarma , s. realizing singlet - triplet qubits in multivalley si quantum dots . b _ * 80 * , 205302 ( 2009 ) . li , q. , cywiski , . , culcer , d. , hu , x. & das sarma , s. exchange coupling in silicon quantum dots : theoretical considerations for quantum computation . _ phys . rev . b _ * 81 * , 085313 ( 2010 ) . smelyanskiy , v. n. , petukhov , a. g. & osipov , v. v. quantum computing on long - lived donor states of li in si . b _ * 72 * , 081304 ( 2005 ) . culcer , d. , saraiva , a. l. , koiller , b. , hu , x. & das sarma , s. valley - based noise - resistant quantum computation using si quantum dots . lett . _ * 108 * , 126804 ( 2012 ) . xiao , m. , house , m. g. & jiang , h. w. measurement of the spin relaxation time of single electrons in a silicon metal - oxide - semiconductor - based quantum dot . _ * 104 * , 096801 ( 2010 ) . simmons , c. b. _ et al . _ tunable spin loading and @xmath5 of a silicon spin qubit measured by single - shot readout . lett . _ * 106 * , 156804 ( 2011 ) . hayes , r. r. _ et al . _ lifetime measurements ( t1 ) of electron spins in si / sige quantum dots . preprint at http://arxiv.org/abs/0908.0173 ( 2009 ) . morello , a. _ et al . _ single - shot readout of an electron spin in silicon . _ nature _ * 467 * , 687 ( 2010 ) . zwanenburg , f. a. _ et al . _ silicon quantum electronics . preprint at http://arxiv.org/abs/1206.5202 ( 2012 ) . bulaev , d. v. & loss , d. spin relaxation and anticrossing in quantum dots : rashba versus dresselhaus spin - orbit coupling . b _ * 71 * , 205324 ( 2005 ) . raith , m. , stano , p. & fabian , j. theory of single electron spin relaxation in si / sige lateral coupled quantum dots . _ b _ * 83 * , 195318 ( 2011 ) . hanson , r. _ et al . _ spins in few - electron quantum dots . _ rev . phys . _ * 79 * , 1217 ( 2007 ) . angus , s. j. , ferguson , a. j. , dzurak , a. s. & clark , r. g. gate - defined quantum dots in intrinsic silicon . _ nano lett . _ * 7 * , 2051 ( 2007 ) . elzerman , j. m. , hanson , r. , willems van beveren , l. h. , vandersypen , l. m. k. & kouwenhoven , l. p. excited - state spectroscopy on a nearly - closed quantum dot via charge detection . lett . _ * 84 * , 23 ( 2004 ) . yang , c. h. , lim , w. h. , zwanenburg , f. a. & dzurak , a. s. dynamically controlled charge sensing of a few - electron silicon quantum dot . _ aip advances _ * 1 * , 042111 ( 2011 ) . kouwenhoven , l. p. , austing , d. g. & tarucha , s. few - electron quantum dots . _ * 64 * , 701 ( 2001 ) . tarucha , s. , austing , d. g. , honda , t. , van der hage , r. j. & kouwenhoven , l. p. shell filling and spin effects in a few electron quantum dot . * 77 * , 3613 ( 1996 ) . borselli , m. g. _ et al . _ measurement of valley splitting in high - symmetry si / sige quantum dots . _ * 98 * , 123118 ( 2011 ) . elzerman , j. m. _ et al . _ single - shot read - out of an individual electron spin in a quantum dot . _ nature _ * 430 * , 431 ( 2004 ) . glavin , b. a. & kim , k. w. spin - lattice relaxation in si quantum dots . b _ * 68 * , 045308 ( 2003 ) . tahan , c. , friesen , m. & joint , r. decoherence of electron spin qubits in si - based quantum computers . b _ * 66 * , 035314 ( 2002 ) . tahan , c. silicon in the quantum limit : quantum computing and decoherence in silicon architectures . preprint at http://arxiv.org/abs/0710.4263 ( 2005 ) . stano , p. & fabian , j. theory of phonon - induced spin relaxation in laterally coupled quantum dots . _ * 96 * , 186602 ( 2006 ) . stano , p. & fabian , j. orbital and spin relaxation in single and coupled quantum dots . b _ * 74 * , 045320 ( 2006 ) . yang , c. h. _ et al . _ orbital and valley state spectra of a few - electron silicon quantum dot . b _ * 86 * , 115319 ( 2012 ) . wilamowski , z. , jantsch , w. , malissa , h. & rossler , u. evidence and evaluation of the bychkov - rashba effect in sige / si / sige quantum wells . b _ * 66 * , 195315 ( 2002 ) . nestoklon , m. o. , golub , l. e. & ivchenko , e. l. spin and valley - orbit splittings in sige / si heterostructures . b _ * 73 * , 235334 ( 2006 ) . wang , l. , shen , k. , sun , b. y. & wu , m. w. singlet - triplet relaxation in multivalley silicon single quantum dots . _ b _ * 81 * , 235326 ( 2010 ) . lim , w. h. , yang , c. h. , zwanenburg , f. a. & dzurak , a. s. spin filling of valley - orbit states in a silicon quantum dot . _ nanotechnology _ * 22 * , 335704 ( 2011 ) . lee , s. doctoral dissertation , purdue university ( 2011 ) . klimeck , g. _ et al . _ atomistic simulation of realistically sized nanodevices using nemo 3-d - part ii : applications _ ieee trans . dev . _ * 54 * , 2090 ( 2007 ) . integrated system engineering , dessis v10 manual , zurich ( 2004 ) . bank , r. e. & rose , d. j. global approximate newton methods . _ numerische mathematik _ * 37 * , 279 ( 1981 ) . culcer , d. , hu , x. & das sarma , s. interface roughness , valley - orbit coupling , and valley manipulation in quantum dots . b _ * 82 * , 205315 ( 2010 ) . friesen , m. & coppersmith , s. n. theory of valley - orbit coupling in a si / sige quantum dot . b _ * 81 * , 115324 ( 2010 ) . yu , p. & cardona , m. _ fundamentals of semiconductors _ ( springer , berlin , 2010 ) .

[ multiblock footnote omitted ] algebraic methods for the construction , design and analysis of series of convolutional codes are developed . the methods use row or block structures of matrices . new codes and infinite series of codes are presented . codes with specific properties are designed and analysed and examples of various types are given . free distances and lower bounds on free distances are proved algebraically and these can be at or near the maximum free distances attainable for their types . methods for designing series of ldpc ( low density parity check ) convolutional codes are derived . design methods for self - dual and dual - containing convolutional codes are included .

1412.1695

convolutional codes are error - correcting codes which are used extensively in many applications including digital video , radio , mobile communication , and satellite / space communications . there is a huge and expanding literature on convolutional codes ; basic constructions and properties may be found in @xcite , @xcite , @xcite , @xcite and many others . the paper @xcite is an often quoted source of information on convolutional codes wherein is mentioned the lack of algebraic methods for their construction and that many of the existing ones have been found by search methods , which limits their size and availability . algebraic implementations , considerations and properties of convolutional codes are our main concerns here . for a nice detailed analysis of the different definitions of convolutional codes found in the literature see @xcite . we are interested in algebraic formulations . a rate @xmath0 convolutional code @xmath1 over a field @xmath2 with parameters @xmath3 is a submodule of @xmath4^n$ ] generated by a reduced basic matrix @xmath5 = ( g_{ij } ) \in f[z]^{r\ti n}$ ] where @xmath6 is the length , @xmath7 is the degree where @xmath8 , and @xmath9 is the _ memory_. the parameters @xmath10 will also be used for such codes when @xmath11 denotes the free distance and @xmath12 will be used to denote a code with free distance @xmath13 . a convolutional code may equivalently and conveniently be described as follows ; details may be found in @xcite and @xcite . a convolutional code @xmath14 of length @xmath6 and dimension @xmath15 is a direct summand of @xmath16^n$ ] of rank @xmath15 . here $ ] is the polynomial ring over @xmath17 and @xmath16^n = \{(v_1 , v_2 , \ldots , v_n ) : v_i \in \f[z]\}$ ] . suppose @xmath18 is a submodule of @xmath16^n$ ] and that @xmath19^n$ ] forms a generating set for @xmath18 . then @xmath20^r\}$ ] where @xmath5= \begin{pmatrix } v_1 \\ \vdots \\ v_r \end{pmatrix } \in \f[z]_{r\times n}$ ] . this @xmath5 $ ] is called a _ generating matrix _ of @xmath18 . a generating matrix @xmath5 \in \f[z]_{r\times n}$ ] having rank @xmath21 is called a _ generator _ or _ encoder matrix _ of @xmath14 . a matrix @xmath22_{n\times(n - k)}$ ] satisfying @xmath23^n : vh = 0 \}$ ] is said to be a _ control matrix _ or _ check matrix _ of the code @xmath14 . the maximum distance attainable by an @xmath24 linear code is @xmath25 and this is known as the _ singleton bound _ of the linear code , @xcite . a linear code @xmath26 attaining the bound is called an _ mds _ ( maximum distance separable ) code . by @xcite the maximum free distance attainable by an @xmath27 code is @xmath28 . ( the singleton bound corresponds to the case @xmath29 , that is , to a system with zero memory ; a linear code . ) the bound , @xmath28 , on convolutional codes will be referred to as the _ generalised singleton bound _ , and is denoted by gsb . for a given generator matrix @xmath5 \in f[z]_{r\ti n}$ ] , a codeword is of the form @xmath30 $ ] with @xmath31^r$ ] . now @xmath32 will be referred to as the _ inputted word _ to the codeword . then @xmath33 for vectors @xmath34 ; the @xmath35 are the _ components _ of @xmath32 . the _ support _ of @xmath32 is the number of non - zero ( vector ) components of @xmath32 . in @xcite a non - constructive proof is given that for any prime @xmath36 and positive integers @xmath37 there exists a rate @xmath38 convolutional code of degree @xmath39 over some suitably big field whose free distance is equal to the gsb . in @xcite explicit examples over suitably big fields are given . attributes of the methods here include : * the constructions are relatively easy to describe and implement . * the free distances and lower bounds on the free distances may often be proven algebraically . * codes are constructed in which the distance of the codeword increases with the support of an inputted word . * codes may be explicitly designed over fields of prime order in which cases modular arithmetic may be used . * a number of different codes with the same parameters can be derived from a single unit scheme type . * the methods allow the design of series of ` good ' convolutional codes . * for given @xmath6 and prime @xmath40 there exists a unit scheme over a finite field of characteristic @xmath40 from which ` good ' convolutional codes of length @xmath41 may be designed . * convolutional codes may be decoded by known techniques , such as viterbi decoding or sequential decoding , @xcite chapter 11 , and algebraic decoding algorithms are known to exist , see @xcite , @xcite , and @xcite for some of those introduced here . * self - dual and dual - containing convolutional codes may be designed and analysed by choosing the unit systems to have particular properties . dual - containing codes are important for the construction of quantum codes , @xcite . * series of ldpc ( low density parity check ) convolutional codes may be designed and analysed by the methods . these may be designed so that the check / control matrices have no short cycles . section [ set - up ] gives further background material . section [ rows ] describes the general methods and proves results on these . section [ cheb1 ] discusses the chebotarv property for matrices , showing how using such matrices leads to good free distances for constructed codes . section [ examples ] gives fundamental examples which have intrinsic interest and can also serve as prototypes for longer length schemes ; this section could now be consulted in order to see the scope of the constructions . section [ convol ] provides details on the designs given in section [ examples ] . section [ blocks ] generalises the row structure design method to a block design structure method . section [ ldpc ] describes the design methods for ldpc convolutional codes . section [ selfdual ] deals with the design and resulting properties of certain self - dual and dual - containing codes . some of the block cases overlap some of those in @xcite . so as to reduce the length of the paper some proofs are omitted , particularly some of the algebraic proofs of the distances attained . background on algebra may be found in @xcite . recall the following from @xcite ; further details may be found in expanded book form within @xcite . let @xmath42 denote the ring of @xmath43 matrices with entries from @xmath44 , a ring usually a field but not restricted to such . suppose @xmath45 in @xmath42 . taking any @xmath21 rows of @xmath46 as a generator matrix defines an @xmath24 code and the check matrix is obtained by deleting the corresponding columns of @xmath18 . let such a code be denoted by @xmath47 . when @xmath44 is a field and the matrix @xmath18 has the property that the determinant of any square submatrix of @xmath18 is non - zero then any such code @xmath47 is an mds @xmath26 linear code , see @xcite for details . convolutional codes have been constructed in @xcite using @xmath45 in which @xmath48 are polynomial or group rings and further details on these may be obtained in book format at @xcite . one of the methods in @xcite for the construction of series of mds linear codes essentially choose rows of certain fourier matrices where the fourier matrices , over a finite or an infinite field in which it exists , have the property that the determinant of any square submatrix is non - zero . finite fields over which fourier matrices exist with this property are determined in @xcite . [ bigdist ] suppose @xmath45 in @xmath49 where @xmath2 is a field and that the determinant of any square submatrix of @xmath18 is non - zero . then a code @xmath47 formed using any @xmath21 rows of @xmath46 has distance @xmath25 ; it is thus an @xmath26 mds code . the proof follows for example from theorem 3.2.2/corollary 3.2.3 of @xcite as the check matrix is obtained using certain columns of @xmath18 , see @xcite , and any square submatrix of @xmath18 has non - zero determinant . choosing any @xmath21 rows of @xmath46 allows the construction of many different @xmath24 codes from an @xmath50 unit system @xmath51 and when @xmath18 satisfies the conditions as in theorem [ bigdist ] each code is mds . in section [ rows ] and in section [ blocks ] convolutional codes are constructed from a system @xmath45 . when @xmath18 has this subdeterminant property the convolutional codes constructed will have good free distances . suppose that @xmath45 in @xmath49 . denote the rows of @xmath46 by @xmath52 and the columns of @xmath18 by @xmath53 . then @xmath54 , the kronecker delta . choose any @xmath21 rows of @xmath46 to form an @xmath55 matrix @xmath56 which then has rank @xmath21 . any order on the rows may be chosen with which to construct the @xmath56 and there are @xmath57 ways of choosing @xmath21 rows . suppose @xmath58 are @xmath59 such @xmath55 matrices . define @xmath60= e_0+e_1z+\ldots + e_sz^s .\ ] ] consider @xmath5 $ ] to be a generating matrix for a convolutional code . not every choice for the @xmath61 will be suitable or useful . it is desirable to choose the @xmath61 so that @xmath5 $ ] is _ noncatastrophic_. for definition of noncatastrophic see @xcite or @xcite or others . here it is sufficient to note , see @xcite , that @xmath5 $ ] is noncatastrophic if it has a polynomial right inverse , that is if there exists a polynomial @xmath62 matrix @xmath63 $ ] of rank @xmath21 such that @xmath5h[z]=i_r$ ] , the identity @xmath64 matrix . for a given @xmath43 system @xmath45 and a given @xmath65 the method allows the construction of codes of rate @xmath0 . say @xmath66 if @xmath61 contains the row @xmath67 and @xmath68 to mean that @xmath61 does not contain the row @xmath67 . define @xmath69 and @xmath70 to be the transpose of @xmath71 with @xmath72 replaced by @xmath73 . for example if @xmath74 then @xmath75 . when @xmath76 is formed from @xmath77 none of which are contained in any of the other @xmath78 then the resulting matrix will be noncatastrophic ; this is done by showing it has a one - sided polynomial inverse . [ jerome ] suppose @xmath5=e_0+e_1z+\ldots + e_sz^r$ ] and that @xmath79 implies @xmath68 for @xmath80 . then @xmath5 $ ] is a noncatastrophic generator matrix for a convolutional @xmath24 code . by @xcite if there exists a polynomial matrix @xmath63 $ ] such that @xmath5h[z]= i_r$ ] , the identity @xmath81 matrix , then @xmath5 $ ] is noncatastrophic and it must then have rank @xmath21 . now @xmath82 and thus @xmath83 as no row of @xmath61 is a row of @xmath76 for @xmath80 . i then follows that @xmath5e_0^*=i_r$ ] . we can do better : [ first ] suppose there exists a row @xmath84 such that @xmath85 but @xmath86 for any @xmath87 . then @xmath88 $ ] is a noncatastrophic generator matrix for a convolutional code . we shall assume that @xmath76 has its first row entry which is not in any other @xmath78 . the proof for any other cases is similar . thus @xmath89 . here @xmath90 where each @xmath61 has size @xmath55 . we show there exists a @xmath63 $ ] such that @xmath5h[z]= i_{r\ti r}$ ] . consider @xmath5(f , f_{i_1 } , \ldots , f_{i_t})$ ] and then construct @xmath63 $ ] as follows . let @xmath91=(f , f_{i_1 } , \ldots , f_{i_t})$ ] . suppose @xmath92 $ ] has been constructed for @xmath93 . let @xmath94 occur in @xmath95 at rows @xmath96 respectively . for each @xmath97 let @xmath98=\al_iz^{m_i}$ ] where @xmath35 is the @xmath62 vector with columns consisting of @xmath99 except for the @xmath100 column which is @xmath101 . then let @xmath102= \sum_{i=1}^wt_i[z]$ ] and @xmath103=h_{k-1}[z]-t_k[z]$ ] . define @xmath63= h_t[z]$ ] . then @xmath5h[z ] = i_{r\ti r}$ ] . thus @xmath5 $ ] is a noncatastrophic generator matrix for a convolutional code . ( in short order the proof is as follows . consider @xmath5(f , f_{i_1 } , \ldots , f_{i_t})$ ] and where @xmath104 appears in a @xmath71 for @xmath105 use the @xmath101 corresponding to the unique @xmath85 in a position corresponding to the position of the @xmath104 in @xmath71 together with zero matrices to give a correct size matrix which will offset the multiplication of @xmath71 with @xmath106 . ) a control matrix for such a noncatastrophic generator matrix may explicitly be derived . here is an illustrative example . suppose @xmath107 satisfies @xmath108 where @xmath109 and @xmath110 are the row and column structures of @xmath111 respectively . consider @xmath5= \begin{pmatrix}e_0 \\ e_1\end{pmatrix}+ \begin{pmatrix } e_2 \\ e_3 \end{pmatrix}z + \begin{pmatrix}e_4 \\ e_5 \end{pmatrix}z^2 $ ] . then @xmath5 $ ] is a noncatastrophic generator matrix of rank @xmath112 for a convolutional code by proposition [ jerome ] . set @xmath63 = ( f_2,f_3,f_4,f_5,f_6)-(f_0,f_1 , \un{0},\un{0},\un{0})z -(\un{0},\un{0 } , f_0,f_1,\un{0})z^2 $ ] where @xmath113 denotes the zero @xmath114 vector . then @xmath63 $ ] is a @xmath115 matrix with @xmath5h[z]= 0_{2\ti 5}$ ] . a left inverse for @xmath63 $ ] is easily constructed and so in particular it has rank @xmath116 . there are other cases besides those in theorem [ first ] for which @xmath5 $ ] is noncatastrophic . permuting the rows in ( [ eq : geneq ] ) gives generator matrices for _ different _ convolutional codes . some of these are equivalent but not many ; unless the same permutation is made on each of the @xmath61 non - equivalent codes are obtained . suppose it is desired for given @xmath117 with @xmath118 to design a @xmath119 code . find a suitable @xmath43 system @xmath111 with @xmath45 . choose @xmath120 where each @xmath61 consists of @xmath21 rows of @xmath46 . to ensure that the resultant generating matrix is noncatastrophic it is enough by theorem [ first ] to reserve one of the rows to be in @xmath76 and not in any other @xmath61 . these @xmath61 can be different choices of @xmath21 rows of @xmath46 if @xmath121 . if @xmath122 then choose the @xmath61 for @xmath123 to be different choices then permutations of the rows of an @xmath61 ( @xmath87 ) already chosen m are used and if ( in the unlikely case that these are not sufficient @xmath61 ) repeats of the choices of the @xmath61 constructed may then be used . eventually @xmath124 will be selected . form @xmath5= e_0+e_iz + \ldots e_\mu z^\mu$ ] and @xmath5 $ ] is a noncatastrophic generator matrix for a convolutional @xmath125 code where @xmath126 . an algebraic proof of the bounds on the free distances may sometimes be given . proofs for the free distances depend on theorem [ distancemds ] and in particular on its corollary [ distance ] . matrices , necessarily invertible , with additional properties such as being unitary or orthogonal may also be chosen to design suitable convolutional codes with specific properties ( such as for example being _ ldpc _ ) ; see section [ blocks ] below for developments . most of the convolutional codes in use and available have been constructed on a case by case basis and/or by computer ; computer techniques are now often beyond the range of computers . see also section [ design ] for specific design methods using matrices satisfying the _ chebotarv property_. these have good distance properties . so find your favourite matrix , fourier or otherwise , satisfying the chebotarv property or not or with specific suitable properties and begin designing ( good ) convolutional codes by the methods . consider cases where @xmath45 and @xmath18 has the _ chebotarv property_. this property is defined in section [ four ] below . certain fourier matrices have this property . then the matrix formed from any @xmath21 rows of @xmath46 is a generator matrix for an mds @xmath26 ( linear ) code , @xcite . using @xmath111 where @xmath18 has the chebotarv property with which to construct convolutional codes by the method of section [ rows ] will give good free distances . the free distances and/or lower bounds on the free distances may often be proven algebraically and the codes are relatively easy to implement and simulate . suppose a primitive @xmath127 root of unity , @xmath128 , exists in a field @xmath129 . the fourier @xmath43 matrix @xmath130 over @xmath129 is @xmath131 . defining @xmath132 then @xmath133 which gives a formula for the inverse of @xmath130 . as @xmath6 must then divide @xmath134 , the inverse of @xmath128 exists in @xmath129 . it is clear that @xmath135 is a primitive @xmath136 root of @xmath137 and the matrix @xmath138 is also a fourier matrix over @xmath129 . now @xmath130 may be normalised by dividing by @xmath139 , when it exists , to give a unitary matrix but this is not necessary for our purposes . chebotarv s theorem is proved for fourier matrices over the complex numbers . it has been proved and reproved a number of times , see for example @xcite , @xcite , @xcite or @xcite . it has been termed _ an uncertainty principle _ in @xcite . [ cheb](chebotarv ) suppose that @xmath140 is a primitive @xmath141 root of unity where @xmath36 is a prime . let @xmath18 be the fourier matrix with @xmath142-entry equal to @xmath143 , for @xmath144 . then all square submatrices of @xmath18 have nonzero determinant . say a matrix @xmath145 over the ring @xmath129 has the _ chebotarv property _ if the determinant of any square submatrix is non - zero . it is clear that if @xmath145 has the chebotarv property then so does @xmath146 for any non - zero scalar @xmath128 . in @xcite it is shown , using chebotarv s theorem , that the fourier matrix @xmath130 over a field @xmath129 of characteristic @xmath147 has the chebotarv property for a prime @xmath6 . using results of @xcite , fourier matrices with the chebotarv property over fields of characteristic @xmath148 are developed in @xcite . these are used in @xcite to construct series of mds ( maximal distance separable ) linear codes over these fields . the following is a corollary of theorem [ bigdist ] . [ distancemds ] suppose @xmath46 is an @xmath43 matrix with rows @xmath149 with @xmath51 and that @xmath18 satisfies the chebotarv property . then the code generated by any @xmath21 of the rows of @xmath46 is an mds @xmath26 linear code . [ distance ] a non - zero linear combination of @xmath21 rows of @xmath46 has distance (= number of non - zero entries ) greater than or equal to @xmath150 . in another paper @xcite efficient decoding techniques for many of the convolutional codes derived here , and for the mds codes of @xcite , are developed . the decoding algebraic methods are based on techniques of error location as for example in @xcite . any collection of @xmath21 rows of @xmath46 may be used to generate an mds @xmath26 code when @xmath45 and @xmath18 satisfies the chebotarv property . hence there are @xmath151 mds @xmath26 different codes derived from the single unit . in @xcite fourier matrices over finite fields which satisfy the chebotarv property are constructed . let @xmath152 be unequal primes where the order of @xmath40 mod @xmath36 is @xmath153 , and ( hence as shown in @xcite ) @xmath154 is irreducible over @xmath155 . then the fourier matrix @xmath156 over @xmath157 exists and satisfies the chebotarv property . there are clearly many such examples and particular ones are given in the various sections of @xcite . a germain prime @xmath36 is a prime such that @xmath158 is also a prime . in such a case the prime @xmath158 is called the ( corresponding ) _ safe prime_. let @xmath36 be germain prime so that @xmath159 is also a prime . it is shown in @xcite that the fourier matrix @xmath156 exists in @xmath160 and satisfies the chebotarv property . algebra in matrices over @xmath161 is modular arithmetic . suppose now for given @xmath6 and given prime @xmath40 it is desired to construct good convolutional codes of length @xmath162 over a field of characteristic @xmath40 . let @xmath36 be the first prime @xmath162 for which for which the order of @xmath163 is @xmath164 . then construct the fourier matrix @xmath156 over @xmath157 . now @xmath156 satisfies the chebotarv property and may be used to construct convolutional codes of length @xmath36 and of varying rates . for example suppose it is required to construct convolutional codes of length @xmath165 over a field of characteristic @xmath112 . now the order of @xmath166 is @xmath167 and not @xmath168 but the order of @xmath169 is @xmath170 . hence construct @xmath171 over @xmath172 and then construct convolutional codes of length @xmath173 of varying rates over @xmath172 . it is unknown if there exist infinitely many germain primes but it is conjectured that this is the case . ` big ' germain primes are known . suppose it is desired to design convolutional codes of length @xmath162 over a field of prime order . let @xmath36 be the smallest germain prime @xmath162 if such exists . then construct the fourier matrix @xmath156 over @xmath174 and from this design convolutional codes of varying rates . the following are examples of the types of series of convolutional codes that can be constructed using fourier matrices satisfying the chebotarv property : 1 . length @xmath175 : series of @xmath176 codes attaining the gsb . series of @xmath177 codes where inputted words of support @xmath178 have distance @xmath179 ; the gsb here is @xmath116 . the fields here can be @xmath180 or @xmath181 see @xcite . length @xmath116 : 1 . series of @xmath182 codes . the gsb for such codes is @xmath183 . for an inputted word of support @xmath178 the distance of the codeword is @xmath184 ; indeed for a word of support @xmath185 the code word then has distance @xmath186 . series of @xmath187 codes . the gsb for such codes is @xmath188 . 3 . series of @xmath189 codes . the gsb for such codes is @xmath167 ; a codeword for an inputted word of support @xmath178 has distance 4 . series of @xmath190 codes which attain the gsb . + the fields here can be @xmath191 or @xmath192 , see @xcite . length @xmath193 : 1 . series of @xmath194 codes . the gsb for such codes is @xmath195 . an inputted word of support @xmath196 has corresponding codeword of distance @xmath186 . 2 . series of @xmath197 codes . the gsb for such codes is @xmath198 . 3 . series of @xmath199 codes . the gsb for such codes is @xmath200 . 4 . series of @xmath201 codes . the gbs is attained but the rate is not good . + the fields here can be @xmath202 or @xmath203 , see @xcite . length @xmath204 : 1 . series of @xmath205 codes ; the gsb for such codes is @xmath206 . inputted words of support @xmath196 have distance @xmath207 . 2 . series of @xmath208 codes . series of @xmath209 codes ; the gsb here is @xmath173 . 3 . series of @xmath210 codes ; the gsb for such codes is @xmath211 . 4 . series of @xmath212 codes ; the gsb for such is @xmath213 . general @xmath6 with memory @xmath137 : suppose the fourier matrix @xmath130 exists over @xmath129 and satisfies the chebotarv property ; see @xcite for examples of such @xmath6 and such fields . . then series of @xmath215 codes are constructed . an inputted word of support @xmath196 has codeword of distance @xmath216 . the gsb for such @xmath217 codes is @xmath218 . the rate is approximately a half . large examples in modular arithmetic : use for example germain primes ; these are primes @xmath36 such that @xmath158 is also a prime . fourier @xmath219 matrices over @xmath220 satisfying the chebotarv property are constructed in @xcite . let @xmath221 and @xmath222 . here @xmath223 is a germain prime and @xmath224 is the corresponding safe prime . then explicit series of @xmath225 codes over @xmath226 and others may be designed . the gsb used in the above is calculated using @xmath227 ) . these are _ some _ examples of the possible constructions . more details on these are given in section [ convol ] . an initial case is to take @xmath228 in equation ( [ eq : geneq ] ) and use all the rows of a fourier matrix satisfying chebotarv property when this is available . thus define @xmath5 = e_0+e_1z+\ldots + e_{n-1}z^{n-1}$ ] where @xmath130 has rows @xmath77 and satisfies the property . this is a @xmath229 convolutional code with degree and memory @xmath230 . the rate is not very good particularly for large @xmath6 but indeed the maximum free distance is attained . the maximal free distance attainable by such a code is by @xcite @xmath231 . [ general1 ] the free distance of the code generated by @xmath5 $ ] above is @xmath232 . the proof is omitted . note that each @xmath77 generates a @xmath233 code and that any non - zero combination of @xmath21 of the @xmath77 has distance @xmath234 . the general proof then depends on corollary [ distance ] and is done by showing directly that @xmath235g[z]$ ] has distance @xmath236 for a polynomial @xmath235 $ ] . by proposition [ jerome ] or theorem [ first ] @xmath5 $ ] is noncatastrophic . it is fairly easy to show this directly and we give an independent proof and produce the control / check matrix . [ gen2 ] let @xmath5 = e_0+e_1z+\ldots + e_{n-1}z^{n-1}$ ] as above . \(i ) define @xmath237 . then @xmath5h(z ) = 1 $ ] . \(ii ) define @xmath238= ( e_1^*,e_2^ * , \ldots , e_{n-1}^ * ) - ( e_0^ * , \un{0 } , \ldots , \un{0 } ) - ( \un{0},e_0^ * , \un{0 } , \ldots , \un{0})z - \ldots - ( \un{0 } , \un{0 } , \ldots , \un{0 } , e_0^*)z^{n-1}$ ] where @xmath113 is the @xmath239 zero matrix . then @xmath5k[z ] = 0_{1\ti ( n-1)}$ ] and @xmath238 $ ] has rank @xmath240 . these follow by direct multiplication on noting @xmath241 ( kronecker delta ) . [ gen1 ] @xmath5 $ ] is noncatastrophic . by permuting the @xmath77 we may obtain @xmath242 such different codes . suppose the fourier @xmath243 matrix @xmath244 exists over @xmath129 and satisfies the chebotarv property . as pointed out in @xcite cases of such @xmath129 are @xmath245 , @xmath246 , @xmath247 . denote the rows of @xmath244 by @xmath248 . define @xmath5=e_0+e_1z+e_2z^2 $ ] . then @xmath5 $ ] is a noncatastrophic generator matrix for a @xmath249 convolutional code which is mds . for example when @xmath250 then @xmath112 is a primitive @xmath251 root of @xmath137 . and get @xmath252 with entries which are integers modulo @xmath193 . the roles of @xmath248 can be interchanged to give @xmath253 such convolutional @xmath254 codes . let @xmath255 be a fourier matrix over a field @xmath129 such that @xmath255 satisfies the chebotarv property . such fields and @xmath255 are given explicitly in @xcite as for example @xmath256 and @xmath257 . in such a field let @xmath258 be primitive @xmath259 root of unity . consider @xmath260 . here note that @xmath261 is irreducible over @xmath262 and if @xmath128 is a primitive element then choose @xmath263 . in @xmath264 it is seen that the order of @xmath112 is @xmath183 with @xmath265 and let @xmath266 to get a primitive @xmath259 root of unity . in all cases @xmath267 and its inverse is such that the determinant of every submatrix non - zero . explicitly when for example @xmath268 then @xmath269 is a fourier matrix over @xmath264 which has the chebotarv property . the entries are elements of @xmath270 . let the rows of @xmath255 be denoted by @xmath271 . by theorem [ general1 ] the code generated by @xmath5= e_0+e_1z+e_2z^2+e_3z^3 + e_4z^4 $ ] is a @xmath190 code attaining the gsb and is noncatastrophic . consider @xmath272 and let @xmath5= e_0+e_1z$ ] . this is a @xmath273 with @xmath274 which can have at best free distance @xmath275 by @xcite . \(i ) the matrix @xmath5 $ ] is noncatastrophic . \(ii ) the code generated by @xmath5 $ ] has @xmath276 . \(iii ) for any inputted word @xmath277 of support @xmath178 the codeword @xmath278 $ ] has distance @xmath184 . \(i ) @xmath279 ( ii ) for @xmath280 vectors @xmath35 , @xmath235=(\al_0 + \al_1z+ \ldots + \al_rz^r)(e_0+e_1z)= \al_0e_0 + ... + \al_re_1z^{r+1}$ ] . now @xmath281 both generate @xmath282 codes and thus free distance of @xmath235 $ ] is @xmath283 . it is easy to show that the actual distance @xmath167 is attained by considering @xmath284 $ ] . \(iii ) the proof of this is similar and the general proof is omitted . it is noted that a linear code with generator matrix consisting of rows @xmath285 is a @xmath286 code and thus a non - zero linear combination of these has distance @xmath178 . a zero combination of these puts restrictions on the vectors which increases the distances at other components . consider a fourier @xmath287 matrix @xmath288 over a field @xmath129 which satisfies the chebotarv property . examples of such fields are given in @xcite . denote the rows of @xmath288 by @xmath289 . setting @xmath5= e_0+e_1z+e_2z^2 + \ldots + e_6z^6 $ ] gives a @xmath290 by theorem [ general1 ] . setting @xmath291 for @xmath292 and then letting @xmath5= e_0+e_1z+e_2z^2 $ ] gives a @xmath293 code . the maximum free distance of such a @xmath294 code is @xmath295 . by permuting the @xmath77 we may obtain @xmath296 such codes . how many of these are equivalent ? suppose we wish to obtain a memory @xmath175 code from this structure . let @xmath297 . note here that @xmath298 have @xmath299 as common . set @xmath5 = e_0+e_1z+e_2z^2+e_3z^3 $ ] . then @xmath5 $ ] is noncatastrophic follows from theorem [ first ] but as easily seen @xmath5e_1^*= i_{2\ti 2}$ ] . it may be shown that this is a @xmath300 code . the maximum free distance of such a @xmath301 code is by @xcite @xmath302 . setting @xmath303 and @xmath5 = e_0+e_1z$ ] gives a noncatastrophic matrix generator of a @xmath304 code . the maximum free distance of such a code is @xmath305 . also note that @xmath278 $ ] has distance @xmath306 for any @xmath277 of support @xmath178 . permuting the @xmath77 gives @xmath296 such codes . setting @xmath307 and then @xmath5 = e_0+e_1z+e_2z^2 $ ] , gives a @xmath308 code . @xmath5 $ ] is noncatastrophic from theorem [ first ] . the free distance is @xmath309 . the maximum free distance of such a code is @xmath310 . the next case taken is that of a fourier matrix over a finite field in which the entries are elements of @xmath312 . now @xmath204 is a germain prime with safe prime @xmath313 and so by @xcite the fourier matrix @xmath314 over @xmath312 exists and satisfies chebotarv property . denote the rows of the fourier matrix by @xmath315 . then 1 . @xmath5= e_0+e_1z+ \ldots + e_{10}z^{10}$ ] is a noncatastrophic matrix for a @xmath316 code which has the mds free distance @xmath317 . there are @xmath318 similar such codes obtained by permuting the order of @xmath319 . 2 . let @xmath320 . + define @xmath5=e_0+e_1z$ ] . then @xmath5 $ ] is a noncatastrophic generator matrix for a @xmath321 convolutional code which has free distance @xmath188 . the maximum distance for such a @xmath322 convolutional code is @xmath323 . an inputted word of memory @xmath324 has distance at least @xmath168 . an inputted word of memory @xmath178 has distance at least @xmath206 . now @xmath318 such codes of this type . 3 . define @xmath325 . define @xmath5= e_0+e_1z+e_2z^2 $ ] . then @xmath5 $ ] is a noncatastrophic generator matrix for a @xmath326 convolutional code . its free distance is @xmath200 . the maximum free distance for such an @xmath327 code is @xmath328 . now @xmath318 such codes may be constructed by permuting the order of @xmath329 . 4 . define @xmath291 for @xmath330 . define @xmath5= e_0+e_1z+e_2z^2+e_3z^3+e_4z^4 $ ] . then @xmath5 $ ] is a noncatastrophic generator matrix for a @xmath331 code and it has free distance @xmath332 . the maximum distance for such a @xmath331 code is @xmath333 . if the inputted word has memory @xmath324 then the distance is @xmath334 . in @xmath335 a primitive element is @xmath116 and so @xmath336 is an element of order @xmath204 from which the fourier matrix @xmath314 over @xmath335 can be constructed . this gives @xmath337 . consider @xmath272 and let @xmath5= e_0+e_1z$ ] . this is a @xmath273 with @xmath274 which can have at best free distance @xmath338 . \(i ) the code @xmath5 $ ] is noncatastrophic . \(ii ) the code @xmath5 $ ] has @xmath276 . \(iii ) for any inputted word @xmath277 of support @xmath178 the corresponding codeword has distance @xmath339 . let @xmath5 = e_0+e_1z+e_2z^2 $ ] . the free distance of the code generated by @xmath5 $ ] is @xmath200 . note that a non - zero linear combination of any three of @xmath340 has distance at least @xmath341 , a non - zero linear combination of any @xmath342 of these has distance at least @xmath342 and a non - zero linear combination of @xmath341 of these has distance at least @xmath175 by corollary [ distance ] . consider @xmath343 . by @xcite the fourier matrix @xmath344 exists over @xmath343 and satisfies the chebotarv property since @xmath223 is a germain prime with matching safe prime @xmath345 . this for example enables the construction of @xmath346 ( different ) mds @xmath347 codes over @xmath226 . the number @xmath348 is of order @xmath349 ( order @xmath350 ) . [ deg1 ] let @xmath351 be mds linear @xmath352 codes . then @xmath5= a + bz$ ] is generator matrix for a convolutional code of free distance @xmath353 . consider @xmath63 = \al_0 + \al_1z+\ldots + \al_tz^t$ ] for @xmath354 vectors @xmath35 and assume @xmath355 . then @xmath63g[z]= \al_0a + ( r[z ] ) + \al_{t}bz^{t+1}$ ] where @xmath356 $ ] is a polynomial of degree at least 1 and at most @xmath185 . now thus @xmath358g[z]\geq ( n - r+1)+(n - r+1)= 2(n - r+1)$ ] . thus memory 1 codes rate ( nearly ) 1/2 with good free distances may be formed from the rows @xmath359 of matrix whose inverse satisfies the chebotarv condition as for example certain fourier matrices as follows : set @xmath360 . let @xmath76 be @xmath55 formed by having as rows @xmath21 of @xmath361 and let @xmath362 be formed by taking @xmath21 other different of @xmath363 as rows . e_0+e_1z$ ] . by [ first ] @xmath5 $ ] is the generator matrix of a noncatastrophic convolutional code of type @xmath24 . its rank is @xmath21 and its degree @xmath39 and memory @xmath364 are @xmath21 . the mds of such a code is @xmath365 . the code is less than its possible mds by @xmath366 . it is clear that many cases and examples may be determined from the general constructions . the methods using _ rows _ of matrices in order to construct convolutional codes are now generalised to using _ blocks _ of ( invertible ) matrices . this leads in particular to the algebraic construction of _ ldpc convolutional codes _ and _ self - dual _ and _ dual - containing _ convolutional codes . we begin with an illustrative example . let @xmath111 be @xmath367 matrices with @xmath45 and @xmath111 have block representations as follows : @xmath368 where @xmath351 are block @xmath369 matrices and @xmath370 are block @xmath371 matrices . then @xmath45 implies @xmath372 and hence hence @xmath5 $ ] is of rank @xmath6 with left inverse @xmath374 and so is a noncatastrophic generator matrix for a @xmath375 convolutional code @xmath1 . also the check / control matrix is @xmath376 which is also of rank @xmath6 as may be shown by producing a right inverse for it . now if @xmath18 is a low density matrix we have constructed a low density convolutional code ; see section [ ldpc ] below for relevant definitions of _ low density_. let @xmath377 denote the distance of a linear code generated by @xmath378 . suppose that @xmath45 in @xmath380 and that @xmath111 have block structures @xmath381 and @xmath382 where the @xmath383 are @xmath384 matrices and the @xmath385 are @xmath386 matrices . then @xmath387 . choose any @xmath21 of these blocks of @xmath46 to form an @xmath388 matrix @xmath56 which then has rank @xmath389 ; any order on the blocks may be chosen with which to construct the @xmath56 . there are @xmath390 ways of choosing @xmath21 blocks . suppose @xmath58 are @xmath59 such @xmath388 matrices . define @xmath391= e_0+e_1z+\ldots + e_sz^s.\ ] ] consider @xmath5 $ ] to be the generating matrix for a convolutional code . say that @xmath392 if @xmath393 occurs as a block in forming @xmath56 and @xmath394 if @xmath393 does not occur as a block in @xmath56 . there are other cases where the @xmath5 $ ] is also noncatastrophic . we illustrate the method for a unit with @xmath175 blocks . suppose @xmath45 are @xmath399 matrices with @xmath400 where @xmath401 are @xmath402 matrices and @xmath403 are @xmath404 matrices . this then gives : define @xmath5 = a + bz+cz^2 $ ] . then @xmath5d = i_{n\ti n } , g[z](e - dz)=0 $ ] . thus @xmath5 $ ] is a noncatastrophic generator matrix a @xmath408 convolutional code @xmath409 with check matrix @xmath410 . let @xmath411 denotes the distance of the linear code with generator matrix @xmath412 . use @xmath5 = \begin{pmatrix } a \\ b \end{pmatrix}+ \begin{pmatrix } b\\ c \end{pmatrix}z + \begin{pmatrix } c \\ b \end{pmatrix}z^2 $ ] to construct a @xmath418 convolutional code . suppose now @xmath45 where @xmath18 is of _ low density_. say @xmath18 has low density if and only if @xmath18 has a small number of elements , compared to its size , in each row and column . then convolutional codes constructed by the row method of section [ rows ] or the block method of section [ blocks ] must necessarily be ldpc ( low density parity check ) convolutional codes . by ensuring that @xmath18 has no short cycles , which can be done by methods of @xcite , the ldpc convolutional codes constructed will have no short cycles in their check matrices . it is known that ldpc codes with no short cycles in their check matrices perform well . the paper @xcite gives methods for constructing classes of matrices @xmath111 of arbitrary size over various fields with @xmath45 where @xmath18 is of low density . such matrices may for example be obtained , @xcite , from group ring elements @xmath419 with @xmath420 in which the support of @xmath421 as a group ring element is small ; corresponding matrices may be obtained by mapping the group ring into a ring of matrices as per @xcite . matrices of arbitrary size and over many fields including @xmath422 satisfying the conditions may be obtained in this manner . it may also be ensured in the construction that @xmath18 has no short cycles thus ensuring the codes obtained do not have short cycles in their check matrices . since low density implies the length must comparatively be long , to actually write out examples explicitly is more difficult in a research paper but as shown in @xcite many such constructions may be formulated . the support of @xmath421 is @xmath341 and is a low density group ring element . as shown in @xcite @xmath421 has no short cycles . its inverse @xmath429 may be easily found but has large support and so is not written out . the matrices corresponding to @xmath419 are denoted @xmath111 ( see @xcite ) , which in this case are circulant - by - circulant , and have the forms @xmath430 where @xmath383 are @xmath431 matrices and @xmath385 are ( low density ) @xmath432 matrices with @xmath433 . these blocks may now be used to construct ldpc convolutional of various types . for example @xmath5 = a_0 + a_1z + a_2z^2+a_3z^3 $ ] is a noncatastrophic generator matrix for an ldpc convolutional @xmath434 code with check low density matrix , with no short cycles , @xmath63 = ( b_1,b_2,b_3)+ ( b_0,\un{0},\un{0})z + ( \un{0 } , b_0 , \un{0})z^2 + ( \un{0},\un{0 } , b_0)z^3 $ ] where @xmath113 is the zero @xmath432 matrix . also @xmath5 = \begin{pmatrix } a_0\\ a_1 \\ a_2 \end{pmatrix } + \begin{pmatrix } a_1\\ a_2 \\ a_3 \end{pmatrix}z$ ] gives an ldpc @xmath435 code with check low density matrix @xmath63 = b_3+b_2z+b_1z^2+b_0z^3 $ ] . similarly @xmath436 ldpc convolutional codes may be produced . also any permutation of @xmath437 may be used in @xmath5h[z]$ ] and rows of @xmath46 and corresponding columns of @xmath18 may also be permuted for further constructions . higher memory may also be produced but make sure the conditions of theorem [ first1 ] are satisfied in the selection . consider as before @xmath45 with @xmath438 . then @xmath5= a+bz$ ] has check matrix @xmath439 as @xmath5(d - cz)=0_{n\ti n}$ ] . we require @xmath376 to be some @xmath440 or @xmath441 multiple of @xmath5\t$ ] . suppose now @xmath46 is an orthogonal matrix and then get @xmath442 . now @xmath443 and then @xmath444 . in characteristic @xmath112 , @xmath445 and in this case @xmath446\t$ ] is a check matrix ( of rank @xmath6 ) for the code and hence the code is self - dual . methods are developed in @xcite in which to construct orthogonal matrices of arbitrary long sizes over various fields including fields of characteristic @xmath112 . the methods involves constructing such elements in a group ring and then finding the corresponding matrices as per the embedding @xcite of the group ring into a ring of matrices . here is an example . consider @xmath448 where @xmath426 is the cyclic group generated by @xmath449 . let @xmath450 . then @xmath451 , and @xmath452 . thus taking the corresponding matrix , @xcite , @xmath453 . then @xmath5=a + bz$ ] is the generator matrix of a self - dual convolutional @xmath455 code . it is easy to check that the free distance of the code is @xmath456 ; both @xmath351 are generator matrices for @xmath457 linear codes . further examples may be generated similar to methods in @xcite by finding group ring elements with @xmath466 and then going over to the matrix representation as per @xcite . using for example @xmath467 will lead to a @xmath468 self - dual convolutional code . now @xmath5\ { d^*-c^*z + b^*z^2-a^*z^3\ } = 0_{3n\ti n}$ ] as may easily be verified . let @xmath476 = a^ * - b^*z^{-1}+c^*z^{-2}-d^*z^{-3}$ ] . then @xmath476z^{3 } = d^*-c^*z+b^*z^2-a^*z^3 $ ] . thus @xmath476 $ ] is a check matrix for the code also . to show that code with generator @xmath5 $ ] is dual containing it is necessary to show that the code generated by @xmath63 $ ] is contained in the the code @xmath1 generated by @xmath5 $ ] . suppose the characteristic of the field is @xmath112 . then @xmath63 = \{(i_n , \un{0},\un{0 } ) + ( \un{0},\un{0 } , i_n)z^2\}g[z]$ ] and hence code generated by @xmath63\subseteq \mathcal{c}$ ] . thus @xmath1 is dual - containing . here is a specific example . consider @xmath478 in @xmath479 where @xmath480 is generated by @xmath449 . then @xmath481 . thus @xmath482 is orthogonal matrix and is symmetric . the matrix of @xmath483 has the form @xmath484 and @xmath46 is orthogonal and symmetric . then @xmath5= \begin{pmatrix } a \\ b \\ c \end{pmatrix } + \begin{pmatrix } b \\ c \\ d \end{pmatrix } z$ ] determines a dual - containing @xmath485 convolutional code . here + @xmath486 + @xmath487 + @xmath488 + @xmath489 . we may also construct self - dual codes from this set - up . again @xmath46 is an orthogonal @xmath469 matrix @xmath470 with @xmath471 and @xmath472 for blocks @xmath473 of size @xmath474 . define @xmath5=\begin{pmatrix}a \\b \end{pmatrix}+\begin{pmatrix}c \\ . then @xmath5\{(c^*,d^ * ) - ( a^*,b^*)z\ } = 0_{2n\ti 2n}$ ] and @xmath5 $ ] is a generator matrix for convolutional code which is self - dual in characteristic @xmath112 . by permuting the @xmath477 other different self - dual convolutional codes may be obtained . note that @xmath477 must be different as @xmath46 is invertible and this this gives @xmath490 self - dual convolutional codes . the distances of the codes depend on the distances of of the linear codes generated by @xmath477 . here are examples of dual - containing code in characteristic @xmath175 . suppose @xmath46 with rows @xmath491 is an orthogonal matrix . then @xmath51 where @xmath18 has columns @xmath492 . consider @xmath5 = \begin{pmatrix } e_0 \\ e_1 \end{pmatrix}+ \begin{pmatrix}e_1 \\ e_2 \end{pmatrix}z $ ] . then @xmath5\{e_2^*- e_1^*z+e_0^*z^2\}=0_{2\ti 1}$ ] and thus @xmath63= e_0-e_1z+e_2z^2 $ ] satisfies @xmath5h[z]\t=0_{2\ti 1}$ ] . now @xmath493= e_0+e_1z+e_1z+e_2z^2=e_0-e_1z+e_2z^2 $ ] in characteristic @xmath175 . thus the code generated by @xmath63 $ ] is contained in the code generated by @xmath5 $ ] . thus the code generated by @xmath5 $ ] is dual - containing . larger rate dual - containing codes may also be constructed . here we indicate how this can be constructed for rate @xmath494 . the process may be continued . let @xmath46 be an orthogonal @xmath495 matrix @xmath470 with @xmath496 and @xmath497 for blocks @xmath383 of size @xmath498 . define then @xmath5\{a_7^*-a_6^*z + \ldots + a_0^*z^7\}= 0_{7n\ti n}$ ] . it may then be shown that @xmath5 $ ] is a generator matrix for a convolutional @xmath500 code which is dual - containing when the characteristic is @xmath112 . specifically for example consider @xmath501 , with @xmath2 of characteristic @xmath112 and @xmath502 is generated by @xmath449 . . then @xmath466 . the matrix @xmath46 corresponding to @xmath429 as per the isomorphism in @xcite is circulant and has the form @xmath504 for @xmath505 matrices @xmath383 . the resulting @xmath5 $ ] in equation [ kris ] is a dual - containing convolutional @xmath506 code ; the rate is @xmath507 . by taking pairs of the @xmath383 together and forming for example @xmath5= \begin{pmatrix } a_0 \\ a_1 \\ \hline a_2 \\ a_3 \\ \hline a_4\\ a_5\end{pmatrix } + \begin{pmatrix } a_2\\ a_3 \\ \hline a_4 \\ a_5\\ \hline a_6\\ a_7 \end{pmatrix}z$ ] give @xmath508 dual - containing convolutional codes . richard e. blahut , _ algebraic codes for data transmission _ , cambridge university press , 2003 . calderbank , e.m . rains , p.m. shor and n.j.a . `` quantum error correction via codes over @xmath509 '' , ieee transactions on information theory 44(4 ) , pp . 1369 - 1387 , 1998 . p. e. frenkel , `` simple proof of chebotarv s theorem on roots of unity '' , arxiv : math/0312398 . gluesing - luerssen , heide & schmale , wiland , `` on cyclic convolutional codes '' , acta applicandae mathematicae , vol . 82 , no . 2 , 2004 , 183 - 237 . h. gluesing - luerssen , u. helmke , j. i. iglesias curto , `` algebraic decoding for doubly cyclic convolutional codes '' , arxiv:0908.0753 . daniel goldstein , robert m. guralnick , and i. m. isaacs , `` inequalities for finite group permutation modules '' , trans . soc . , 10 , 4017 - 4042 , 2005 . paul hurley and ted hurley , `` module codes in group rings '' , isit2007 , nice , 2007 , 1981 - 1985 . barry hurley and ted hurley , `` systems of mds codes from units and idempotents '' , discrete math . , 335 , 81 - 91 , 2014 . ted hurley , `` group rings and rings of matrices '' , inter . j. pure & appl . , 31 , no.3 , 2006 , 319 - 335 . ted hurley , `` convolutional codes from units in matrix and group rings '' , int . j. pure appl . , 50 , no . 3 , 431 - 463 , 2009 . ted hurley , `` self - dual , dual - containing and related quantum codes from group rings '' , arxiv:0711.3983 . rolf johannesson and kamil zigangirov , _ fundamentals of convolutional coding _ , wiley - ieee press , 1999 . mceliece , _ theory of information and coding _ , 2nd ed . , cambridge university press , 2002 . r. j. mceliece , `` the algebraic theory of convolutional codes '' , in _ handbook of coding theory , volume i _ , north holland , elsevier science , 1998 . csar milies & sudarshan sehgal , _ an introduction to group rings _ , klumer , 2002 . jessica oshaughnessy , `` convolutional codes from group rings '' , to appear intl . j. of information and coding theory . see also jessica oshaughnessy , thesis , national university of ireland 2011 . j. rosenthal & r. smarandache , `` maximum distance separable convolutional codes '' , appl . algebra engrg . comput . 10 ( 1 ) , 15 - 37 , 1999 . rosenthal , j. , `` connections between linear systems and convolutional codes '' , pp . 39 - 66 in : marcus , b ; rosenthal , j. _ codes , systems , and graphical models _ , minneapolis , new york , 1999 . j. rosenthal , `` an algebraic decoding algorithm for convolutional codes '' , in g. picci and d.s . gilliam , editors , dynamical systems , control , coding , computer vision : new trends , interfaces , and interplay , pages 343 - 360 . birkhuser , boston - basel - berlin , 1999 . paul hurley and ted hurley , `` codes from zero - divisors and units in group rings '' , int . j. inform . and coding theory , 1 , 57 - 87 , 2009 . paul hurley and ted hurley , `` block codes from matrix and group rings '' , chapter 5 , 159 - 194 , in _ selected topics in information and coding theory _ i. woungang , s. misra , s.c . misma , world scientific 2010 . paul hurley and ted hurley , `` ldpc and convolutional codes from matrix and group rings '' , chapter 6 , 195 - 239 , in _ selected topics in information and coding theory _ i. woungang , s. misra , s.c . misma , world scientific 2010 . ted hurley , paul mcevoy and jakub wenus , `` algebraic constructions of ldpc codes with no short cycles '' , intl . j. of inform . and coding theory , vol 1 , issue 3 , 285 - 297 , 2010 . ted hurley , `` algebraic decoding techniques for certain linear and certain convolutional codes '' , preprint . f. pakovich , `` a remark on the chebotarv theorem about roots of unity '' , integers : elec . j. of combinatorial number theory , 7 , # a18 , ( 2 pages ) , 2007 . r. pellikaan , `` on decoding by error location and dependent sets of error positions '' , discrete math . , vol . 106/107 , 369 - 381 , 1992 . , vols i & ii , v. c. pless and w. c. huffman eds . , elsevier , 2011 . r. smarandache , h. gluesing - luerssen , j. rosenthal , `` constructions for mds - convolutional codes '' , ieee trans . theory , vol . it-47 , 2045 - 2049 , 2001 . p. stevenhagen and h. w. lenstra , jr . , `` chebotarv and his density theorem '' , math . 18 , 26 - 37 , 1996 . terence tao , `` an uncertainty principle for cyclic groups of prime order '' , math . , 12 , no . 1 , 121 - 127 , 2005 .

we constrain several models of the early universe that predict a statistical anisotropy of the cosmic microwave background ( cmb ) sky . we make use of wmap9 maps deconvolved with beam asymmetries . as compared to previous releases of wmap data , they do not exhibit the anomalously large quadrupole of statistical anisotropy . this allows to strengthen the limits on the parameters of models established earlier in the literature . in particular , the amplitude of the special quadrupole is constrained as at c.l . ( at c.l . ) independently of the preferred direction in the sky . the upper limit is obtained on the total number of-folds in anisotropic inflation with the maxwellian term nonminimally coupled to the inflaton , namely at c.l . ( at c.l . ) for . we also constrain models of the ( pseudo)conformal universe . the strongest constraint is obtained for spectator scenarios involving a long stage of subhorizon evolution after conformal rolling , which reads at c.l . , in terms of the relevant parameter . the analogous constraint is much weaker in dynamical models , e.g. , galilean genesis .

1311.3272

the statistical isotropy of the cosmic microwave background ( cmb ) is one of the central pillars in modern cosmology . manifesting the direction independence of the primordial spectrum at the onset of the hot era , it is renowned for its robustness against modifications in the bulk of inflationary models . simple reasoning usually invokes the cosmic no - hair conjecture @xcite which states the rapid isotropization of the universe in the presence of a positive constant energy density . scalar fluctuations ( e.g. , carried by the inflaton ) evolve in the rotationally invariant metric and acquire a direction - independent spectrum , giving rise to the statistical isotropy of cmb temperature fluctuations . clearly , a violation of this property observed in the cmb sky would indicate a nontrivial extension of currently conventional cosmology . one way to break statistical isotropy is by introducing vector fields with nonvanishing vacuum expectation values @xcite . for example , in the ackermann carroll wise ( acw ) model @xcite , the idea is to add a massive vector with a fixed space - like norm . this allows for the anisotropic evolution of the universe which introduces the direction dependence into the power spectrum of the primordial curvature perturbation @xmath9 . for future convenience , we write it in the generic form , @xmath10 here @xmath11 is the direction associated with the cosmological wavevector @xmath12 ; the @xmath13 s are spherical harmonics , the @xmath14 s are coefficients parametrizing the statistical anisotropy , and @xmath15 is the direction - independent amplitude . in the acw model , the amplitude @xmath15 is constant , i.e. , it can be tuned to unity without loss of generality . furthermore , the predicted statistical anisotropy is of the quadrupole type , i.e. , only the coefficients @xmath16 survive , which are not independent . in particular , by an appropriate choice of the reference frame , one can tune all coefficients @xmath16 , except for @xmath17 , to zero . the @xmath18 axis of this reference frame is associated with the preferred direction in the sky . hereafter , we use the term _ special quadrupole _ for this type of statistical anisotropy . soon afterwards , it was realized that the acw model is unstable @xcite . this is due to the longitudinal component of the massive vector that propagates as a ghost in the inflationary background . similarly , all vector models of inflation violating gauge @xmath19 symmetry share the same problem @xcite . this problem does not arise in the models of ref . @xcite that introduce the maxwellian term modified by the explicit coupling to the inflaton . remarkably , for quite a wide range of coupling functions and inflaton potentials , one can achieve the anisotropic expansion of the universe @xcite in a ghost free manner @xcite . ( see also refs . @xcite for reviews . ) the outcome of anisotropic inflation is still the direction dependence of the acw type , while the preferred direction is associated with the electric component of the electromagnetic field . interestingly , the amplitude of the special quadrupole relies on the overall number of @xmath4-folds during inflation @xcite . ( see also ref . @xcite . ) in turn , this means that the duration of inflation in those models can be strongly constrained with the cmb data . statistical anisotropy may follow naturally from alternative frameworks , e.g. , the ( pseudo)conformal universe @xcite . minimal models of this type incorporate at least two fields : one which we call @xmath20with the conformal weight @xmath21 , and the zero - weighted conformal field @xmath22 . in particular , the conformal rolling scenario @xcite and galilean genesis @xcite two known incarnations of the pseudo(conformal ) universe so far deal with the weight @xmath23 field @xmath20 . we focus on this case in the present paper . the field @xmath20 is then assumed to have the time - dependent solution @xmath24 spontaneously breaking conformal group @xmath25 down to the de sitter subgroup @xmath26 . the constant @xmath27 is the only relevant parameter of the conformal rolling scenario or galilean genesis . the symmetry breaking pattern @xmath28 fixes the phenomenological properties of the weight-0 field perturbations @xmath29 evolving in the background created by the field @xmath20 . in particular regardless of the precise details of the microscopic physics one ends up with the scale - invariant power spectrum of the perturbations @xmath29 @xcite . thus , they may serve as the source of primordial fluctuations in the standard matter at the onset of the rd stage . the setup of the ( pseudo)conformal universe leads unavoidably to a nonzero statistical anisotropy @xcite . this originates from the interaction between weight-0 and weight-1 perturbations . there are two possible types of predictions depending on the state of cosmological perturbations at times when conformal symmetry becomes irrelevant . if they are already superhorizon at these times , the resulting direction dependence is of the quadrupole type akin to anisotropic inflation @xcite . the structure of the statistical anisotropy is particularly rich in the situation with subhorizon cosmological perturbations @xcite . in that case , all the coefficients @xmath14 with even @xmath30 are generically nonzero in eq . . in both cases , the direction dependence is governed by the constant @xmath27 . this gives a simple idea of how to constrain the models of interest from the nonobservation of statistical anisotropy in the cmb sky . our main goal in the present paper is to constrain models of the early universe that predict statistical anisotropy . we do this in the same manner as in our previous paper @xcite , where the conformal rolling scenario was constrained by making use of the wmap7 data . namely , we apply quadratic maximum likelihood ( qml)-based estimators @xcite to the cmb data and establish upper limits on model parameters . in the present paper , we turn to the wmap9 maps . with the latter , there is a strong reason to anticipate much tighter constraints . we expect that the wmap9 data lack the anomalously large quadrupolar statistical anisotropy observed in the @xmath31 band of the 5-year release @xcite . the anomaly was first interpreted as a hint towards the acw model . however , the preferred direction of the signal was found to be highly aligned with poles of the ecliptic plane @xcite . its frequency dependence was also suspicious : the signal so prominent in the @xmath31 band was much weaker in the @xmath32 band @xcite . finally , results from alternative searches , i.e. , large scale structure surveys , favor statistically isotropic primordial perturbations @xcite . these three points strongly indicate the systematic origin of the signal detected . in ref . @xcite , it was suggested that beam asymmetries not accounted for in the previous analysis could strongly bias primordial statistical anisotropy . indeed , upon the inclusion of beam asymmetries into the computational scheme , the large quadrupole statistical anisotropy vanishes . this was first shown in ref . @xcite for the @xmath31 band of the wmap7 data . to deal with this effect in the 9-year final data release , wmap collaboration has produced the beam - symmetrized temperature maps @xcite . this paper is organized as follows . in sec . 2 we review vector models of inflation predicting statistical anisotropy . we focus on the particular class of scenarios with the maxwellian term nonminimally coupled to the inflaton . we review models of the ( pseudo)conformal universe in sec . 3 . 4 we establish the estimators which are most appropriate for our constraining purposes . in sec . 5 we apply the estimators to models of interest and obtain constraints on their parameters . we compare our constraints with similar bounds obtained from the planck data in sec . in this section we briefly discuss inflationary scenarios which lead to statistical anisotropy . we focus on a particular class of slow roll inflation that incorporates abelian gauge fields with @xmath19 gauge symmetry . the healthy extension of inflation in terms of vectors is achieved by the following modification of the standard maxwellian term @xcite : @xmath33 here @xmath34 is some function of the inflaton @xmath35 , and @xmath36 is the field strength of the vector field , @xmath37 . the main statement of ref . @xcite is that for quite a wide range of kinetic gauge functions @xmath34 there is a chance of obtaining a prolonged anisotropic evolution . that is , the space - time metric tends to the attractor solution @xmath38 \ ; .\ ] ] here @xmath39 is the parameter which measures the deviation from the rotational invariance . furthermore , the evolution of inflaton fluctuations in this `` hairy '' background leads to a directional dependence of the acw type in the primordial power spectrum . here we write it in the conventional form , @xmath40 with @xmath41 being the angle between the wave vector @xmath12 and the direction @xmath42 of the electric field @xmath43 generated during the inflationary stage . the amplitude @xmath44 is sourced by the electric field , i.e. , @xmath45 , as we discuss in more detail in what follows . the relationship between the power spectrum and the generic one is given by the relations @xmath46 which are valid in the approximation of the small amplitude @xmath47 , which is somewhat different from the one given by eq . . the difference , however , is the direction - independent piece , which can be absorbed into the redefinition of the spectrum @xmath48 . ] . note that @xmath49 in eq . . to avoid confusion in the future , let us make one remark here . though we use the terms `` electric '' and `` magnetic '' , our definitions of the corresponding fields are different from the conventional ones . namely @xcite , @xmath50 where @xmath51 denotes the expectation value of the function @xmath34 , and a prime denotes the derivative with respect to conformal time . with these definitions , the electromagnetic energy density is given by the standard expression @xmath52 at all times . let us discuss briefly conditions at which statistical anisotropy is generated . in the situation with the standard maxwellian term , i.e. , provided that @xmath53 , electric and magnetic fields fall down rapidly with time , as does the electromagnetic energy density @xmath54 , which redshifts away as @xmath55 . this standard prediction can be avoided given the nontrivial structure of the function @xmath34 . a constant electric field is generated for the class of functions @xmath34 defined by @xcite @xmath56 where @xmath57 is the inflaton potential . in ref . @xcite , the case @xmath58 is argued to be the critical one , i.e. , for @xmath59 the universe is statistically isotropic and so is the primordial power spectrum . on the other hand , for @xmath60 one finds the nontrivial attractor solution for the electric field and the anisotropic metric of the form . the deviation from the rotational invariance is measured by @xcite @xmath61 where @xmath62 is the slow - roll parameter . the amplitude of statistical anisotropy is related to the violation of rotational invariance , i.e. , @xmath63 . in fact , even for the case @xmath58 , one finds the anisotropic power spectrum @xcite . this is due to the enhancement of infrared fluctuations of the gauge field by the standard inflationary mechanism . these quantum fluctuations behave like the classical field after they exit the horizon . consequently , modes which exit the horizon before the time @xmath64 in the number of @xmath4-folds act as an additional anisotropic background for inflaton fluctuations . to account for this new effect ( pointed out in ref . @xcite ) , one should study the evolution of inflaton fluctuations in the overall classical electric field , @xmath65 here @xmath66 follows from the equation of motion ( e.o.m . ) , while @xmath67 originates from quantum fluctuations , which get enhanced and classicalize before the time @xmath68 when inflaton perturbations leave the horizon ( hence the subscript `` ir '' ) . note that @xmath67 is a random gaussian field characterized by zero mean and variance @xcite , @xmath69 here @xmath70 is the number of @xmath4-folds from the beginning of inflation and until the time @xmath68 , i.e. , @xmath71 , with @xmath72 being the overall number of @xmath4-folds during inflation . taking into account both sources of statistical anisotropy , one writes the amplitude @xmath44 as follows @xcite : @xmath73 ( see also ref . @xcite . ) quite unexpectedly , one observes a large magnitude of statistical anisotropy @xcite . so , the order one amplitude @xmath74 is obtained already for the ratio @xmath75 as tiny as @xmath76 . the point is that one expects a much larger energy density @xmath54 on rather general grounds . indeed , according to its quantum origin , we note that the factor @xmath70 in eq . is considered to be a large number in conventional inflationary scenarios . hence , the naturally large amplitude @xmath44 . to handle the situation , one assumes a tuned duration of inflation , i.e. , @xmath77 . we reiterate this statement in the future , when imposing constraints on model parameters . furthermore , the purely classical field @xmath66 is related to the potential @xmath57 by @xmath78 substituting this into eq . , one observes that one needs extremely tuned value of the constant @xmath79 . namely , the order - one amplitude @xmath44 is obtained provided that @xmath80 . to summarize , the special quadrupole predicted in the anisotropic inflation has a twofold origin translating into the twofold treatment of the amplitude @xmath44 . if the purely classical effect is most relevant , the quantity @xmath44 is directly related to the intrinsic parameters of the model . there is no direct matching provided that the amplitude @xmath44 is sourced by the random field @xmath67 . barring fine - tuning , we focus on these two situations in what follows . namely , + * i * ) @xmath81 , and , consequently , @xmath82 is achieved in the formal limit @xmath83 . + in this case , by substituting eq . into eq . , we obtain for the amplitude @xmath84 * ii * ) @xmath85 , and , consequently , @xmath86 , which occurs for @xmath87 . + we write the corresponding amplitude as follows : @xmath88 where we made use of the slow - roll relations @xmath89 the subscript 1 denotes the quantity @xmath90 formally calculated for the total number of @xmath4-folds @xmath91 . the convenience of the vector @xmath92 introduced in eq . is that its components @xmath93 obey gaussian statistics ( unlike the amplitude @xmath44 ) . namely , they have zero means and variances given by @xmath94 considering the cases @xmath95 and @xmath96 separately is natural , since fields @xmath97 and @xmath98 are largely unrelated to each other barring scenarios where two electric fields accidentally cancel each other with a high accuracy . with this qualification case @xmath95 results into the constraint on the parameter @xmath79 , while case @xmath96 leads to an upper limit on the duration of inflation . equations , and will be the starting point of our discussion in sec . 5 , when constraining anisotropic inflation . generating statistical anisotropy at inflation requires strong assumptions about the inflationary stage , at least in the model we have discussed . on the other hand , a violation of statistical isotropy may arise naturally in alternative frameworks , e.g. , the ( pseudo)conformal universe . in secs . 3.1 and 3.2 , we briefly summarize some basic features inherent to this cosmological picture , while referring to refs . @xcite for detailed discussions . predictions for statistical anisotropy are reviewed in secs . 3.3 and 3.4 . in the ( pseudo)conformal universe , the observed flatness of the primordial power spectrum is due to conformal symmetry at very early stages of the universe . in more detail , there are several conditions to be satisfied @xcite : * the space - time is described by the nearly minkowski metric at very early times . * the matter in the universe is in the conformal field theory state . * among the field content of the universe there are at least two scalars : one with the conformal weight @xmath21 and another with the weight @xmath99 . * classical equations of motion admit the nontrivial time - dependent solution of the field with @xmath21 . * the action is invariant under the shift of the weight-0 field , @xmath100 . given these conditions , weight-0 field perturbations evolving in the background created by the @xmath21 conformal field acquire a scale - invariant power spectrum . on the other hand , relaxing one or more conditions above may lead to the small scalar tilt @xcite , as required by the experimental data @xcite . specifying to known realizations of the ( pseudo)conformal universe , we choose the nonzero conformal weight to be equal to 1 . then , the generic action of the ( pseudo)conformal universe can be written in the form @xmath101 where @xmath102 is the action for gravity and some matter pre - existing in the early universe . the third term on the r.h.s . describes the minimal conformal coupling of the weight-0 field @xmath22 to the field @xmath20 , while the second term encodes dynamics of the field @xmath20 , which it has on its own . by assumption , the action @xmath103 allows for the classical solution of the field @xmath20 given by eq . , where the time dependence is fixed by conformal invariance . we recall that the parameter @xmath27 entering eq . is the dimensionless constant originating from the action @xmath103 , while @xmath104 is an arbitrary constant of integration which has the meaning of the end - of - roll time . the solution spontaneously breaks conformal group @xmath25 down to the de sitter subgroup @xmath26 @xcite . interestingly , this symmetry breaking pattern uniquely determines phenomenological properties of perturbations of the field @xmath22 . namely , independently of the details of the microscopic physics , the field @xmath22 acquires the power spectrum of the harrison zeldovich type @xcite . to show this explicitly , one introduces the notation @xmath105 . from eq . , one derives the e.o.m . for the field @xmath106 , @xmath107 ( the unit scale factor @xmath108 is assumed here ) . we observe that with the classical background @xmath109 as in eq . , the e.o.m . coincides with that of the massless scalar field in the de sitter background . hence , the result is the same , namely , a scale - invariant power spectrum which reads in terms of @xmath22 perturbations @xcite @xmath110 we reiterate that the scale invariance as well as the results summarized in the next subsections are largely independent of the details of the microscopic physics . so far , two concrete models have been proposed . these are the conformal rolling scenario @xcite and galilean genesis @xcite . the former represents perhaps the simplest realization of the ( pseudo)conformal universe . there the field @xmath20 is a scalar with the standard kinetic term rolling down the negative quartic potential . the action for this rolling field is given by @xcite @xmath111 \ ; .\ ] ] it is straightforward to show that the classical field @xmath20 has an attractor solution given by eq . . the conformal rolling scenario is natural from both the dynamical and spectator prospectives . in the former situation , the universe driven by the field @xmath20 undergoes a slow contraction @xcite akin to ekpyrotic scenarios @xcite . alternatively , one treats the field @xmath20 as a spectator . in that case , the minkowski metric can be imposed `` by hands '' . the other possibility discussed in the original proposal of the conformal rolling scenario @xcite is to conformally couple the spectator field @xmath20 to gravity . the background evolution of the universe is then allowed to be arbitrary during the conformal phase . in what follows , we assume the minimal coupling to gravity with minimal loss of generality . galilean genesis is the other example of ( pseudo)conformal universe model @xcite . its action is given by @xmath112 \ ; , \ ] ] where the field @xmath113 , the galileon , is defined by @xmath114 . despite the higher derivative structure of the galilean genesis action , the e.o.m . is second order in derivatives of the field @xmath113 . moreover , the e.o.m . admits for the solution of the form . in fact , the correspondence between galilean genesis and the conformal rolling scenario is much deeper : the predictions of the two models are the same modulo the replacement @xmath115 note that the galileon @xmath113 is naturally treated as a dynamical field . driven by the field @xmath113 , the universe is nearly static at very early times , but slowly expands . thus , the first condition outlined in the beginning of this section is satisfied automatically , at least at early times . the interaction between weight-0 and weight-1 field perturbations sources nontrivial phenomenology in ( pseudo)conformal universe models @xcite . in particular , it gives rise to some amount of statistical anisotropy . perturbations of the field @xmath20 have a red power spectrum whose form is fixed by the symmetry breaking pattern @xmath28 . as discussed in refs . @xcite , they can be absorbed into the redefinition of the end - of - roll time @xmath104 , i.e. , @xmath116 here @xmath117 and the shift of time @xmath118 is the random field with the red power spectrum @xmath119 where @xmath120 are wave numbers characterizing the fourier modes of the field @xmath20 . clearly , the shift @xmath121 as it stands is irrelevant from the physical point of view , since it can be absorbed into the redefinition of the end - of - roll time @xmath104 . interesting effects appear , once we consider the spatial variation of @xmath122 . it is convenient to introduce the notation @xmath123 while keeping the standard notation for the second derivative , i.e. , @xmath124 . the field @xmath125 is a random gaussian field with zero mean and variance , @xmath126 here @xmath127 is the present hubble rate ; @xmath128 is an infrared cutoff . to account for the interaction with long ranged radial perturbations , one studies the evolution of weight-0 perturbations in the inhomogeneous background . this calculation was done in ref . @xcite including corrections of the orders @xmath129 and @xmath130 . the result for the perturbations @xmath29 in the late - time regime @xmath131 is @xmath132 here @xmath133 , and the expansion to order @xmath130 is understood ; @xmath134 is the annihilation operator for the perturbations @xmath29 . note that perturbations of the weight-0 field @xmath22 remain frozen out until the end of conformal rolling . after conformal symmetries get broken , the form of the solution is not protected anymore . still , there is the option that cosmological modes of interest are already superhorizon at this time , and the perturbations @xmath29 remain unchanged until the rd stage . in the nomenclature of ref . @xcite , this is `` sub - scenario a '' . note that sub - scenario a is natural from both the dynamical and spectator prospectives . alternatively , the cosmological perturbations @xmath135 may be subhorizon by the end of the roll . in that case , perturbations @xmath29 evolve before the beginning of the hot era . this option is not particularly natural in dynamical models , but can be well accomodated in spectator versions of the ( pseudo)conformal universe . we call this option `` sub - scenario b '' . the two sub - scenarios lead to drastically different predictions for statistical anisotropy , as we discuss below . in sub - scenario a , the primordial power spectrum of scalar perturbations is derived directly from eq . . for reasons which will become clear shortly , we write it up to the quadratic order in the constant @xmath27 , @xmath136 \ ; .\ ] ] here @xmath137 is the leading order contribution , which is nonzero already in the linear order in @xmath27 @xcite , @xmath138 this encodes a statistical anisotropy of the general quadrupole type in contrast to the inflationary predictions of sec . 2 . the third term in the square brackets in eq is given by @xcite , @xmath139 the direction dependence present here is of the special quadrupole type akin to inflation with vector fields . note its quadratic dependence on the `` velocity '' , which implies the suppression by the additional power of the constant @xmath27 as compared to the contribution . this , however , does not mean that the special quadrupole can be ignored in the data analysis . moreover , if the constant @xmath27 is not particularly small , the term effectively makes larger imprint on the cmb sky than the general quadrupole with the amplitude decreasing as @xmath140 . at the level of cosmological measurements , the latter property translates into the suppression at large cmb multipole number @xmath141 . as a result , we have low statistics of multipoles relevant in the analysis , and hence a very weak constraint on the parameter @xmath27 from the nonobservation of the general quadrupole @xcite . we will return to this discussion in sec . 5 , when constraining sub - scenario a. let us rewrite eq . in the conventional form , @xmath142 here @xmath16 are random gaussian quantities with zero means and variances , @xmath143 while the direction - independent amplitude @xmath15 is given by @xmath144 on the other hand , the special quadrupole corresponding to the subleading - order statistical anisotropy is characterized by the scale - independent amplitude @xmath145 and the preferred direction is associated with the unit random vector @xmath146 . remarkably , this prediction is very similar to the statistical anisotropy following from case * ii * of anisotropic inflation , sourced by the gaussian random vector @xmath67 . formally equatting amplitudes and , and using eqs . and , we conclude that the duality holds up to the replacement @xmath147 we exploit this duality when constraining sub - scenario a in sec . 5 . if cosmological modes are subhorizon by the end of the roll , they proceed to evolve at the so - called intermediate stage @xcite which ends as modes of interest leave the horizon . in the conformal rolling scenario , the end of the roll is realized by relaxing the form of the potential in eq . , so that it has a minimum at some large field value @xmath148 . after the field @xmath20 reaches its minimum , the field @xmath22 evolves as the massless scalar minimally coupled to gravity . we claim that the evolution during the intermediate stage is long enough , i.e. , @xmath149 , where @xmath150 is the time when the perturbations @xmath29 get frozen out ( in the conventional sense ) . second , cosmological evolution at the intermediate stage must be described by the nearly minkowski metric . otherwise , the flat spectrum of perturbations generated by the end of the conformal phase would be grossly modified . with these assumptions , the dynamics at the intermediate stage is fairly nontrivial , and the final expression for the perturbations @xmath29 is quite complicated . here we simply write down the result for the power spectrum of primordial scalar perturbations @xcite , @xmath151 \ ; .\ ] ] remarkably , the direction dependence present here is nonzero already at the linear order in the constant @xmath27 . moreover , it encodes the statistical anisotropy of all even multipoles starting from the quadrupole of the general type , i.e. , all the coefficients @xmath14 with even @xmath30 are nonzero in eq . . they are random gaussian variables with zero means and variances given by @xcite @xmath152 remarkably , the amplitude @xmath15 does not depend on the wave number @xmath153 , i.e. , @xmath49 . the prediction of sub - scenario b for statistical anisotropy is in sharp contrast to that of inflationary scenarios . as the first step , we discuss estimators for the amplitudes @xmath14 entering the primordial power spectrum . following ref . @xcite , we make use of the qml methodology . this techniques was argued to be in good agreement with exact likelihood methods applied to the search for the statistical anisotropy with the wmap5 data @xcite . in our previous paper , we implemented qml - based estimators to the 7-year release of the wmap and constrained the conformal rolling scenario from the nonobservation of statistical anisotropy . let us recall the main ideas behind the qml estimator . one starts with the log - likelihood @xmath154 of the observed sky @xmath155 with respect to the coefficients @xmath14 of statistical anisotropy . assuming that these parameters are small enough , one expands the log - likelihood up to the quadratic order in @xmath14 s , @xmath156 where the subscript 0 denotes that corresponding quantities are calculated in the absence of statistical anisotropy . in eq . we replaced the second derivative of the log - likelihood by its expectation value . the qml estimator for @xmath157 is obtained by setting the derivative of the quadratic log - likelihood to zero with respect to @xmath158 , @xmath159 here @xmath160 is the fisher matrix defined as @xmath161 the equality here follows from the normalization condition for the likelihood . to concretize the form of the estimator , we assume gaussian temperature fluctuations . the log - likelihood then reads @xmath162 where @xmath163 denotes the covariance matrix incorporating the theoretical covariance as well as the instrumental noise , @xmath164 . the first derivative of the log - likelihood is given by @xmath165 the vector @xmath166 represents the collection of cmb temperature coefficients filtered with the inverse isotropic covariance , @xmath167 here @xmath168 is the theoretical covariance calculated in the absence of statistical anisotropy , @xmath169 with @xmath170 s representing the standard angular power spectrum . the derivative of the covariance with respect to coefficients @xmath171 is given by @xmath172 where @xmath173 and @xmath174 is a transfer function . for the particular case of @xmath175 and constant amplitude @xmath49 , coefficients @xmath176 reduce to the angular power spectrum @xmath170 . equation follows from the expression for the theoretical covariance @xmath177 which we write here for the future references @xmath178 it takes the diagonal form for a statistically isotropic power spectrum @xmath179 . the straightforward way to evaluate the fisher matrix entering eq . is to average the product of two log - likelihood derivatives over the large number of statistically isotropic monte carlo ( mc ) maps . a good forecast , however , is given in terms of the analytic fisher matrix calculated in the homogeneous noise approximation . only diagonal elements of the fisher matrix survive in that case , i.e. , @xmath180 where @xmath181 ; the prefactor @xmath182 is an unmasked fraction of the sky . the formula completes the derivation of the estimators for the coefficients @xmath14 . out of the amplitudes @xmath14 , one further reconstructs coefficients @xmath183 defined in the standard manner , @xmath184 these can be used to test the cmb statistical anisotropy in a model - independent way . to constrain the statistical anisotropy of the special quadrupole type , we slightly modify the above procedure . our first goal is to construct the estimator for the amplitude @xmath44 given some fixed preferred direction @xmath185 . for this purpose , we consider the log - likelihood as the function of the unique parameter @xmath44 , i.e. , @xmath186 . then , following the same steps as outlined above , we obtain @xmath187 so , the estimate for the amplitude @xmath44 is reproduced immediately from estimates for the coefficients @xmath16 . furthermore , the estimator is unbiased and has a minimal variance . recall that the early universe models generically do not predict the preferred direction of statistical anisotropy . to estimate the amplitude @xmath44 in a universal way , we exploit the second relation in eq . , @xmath188 though this estimator has an intuitively clear form , it is `` blind '' to the sign of the amplitude @xmath44 . the other disadvantage of the estimator is that it does not discriminate between the special and general types of quadrupoles . consequently , we expect somewhat weaker constraints than in the case with the specified preferred direction . this is the price we pay for our ignorance about the latter . generically , statistical anisotropy is described by the infinite number of _ random _ parameters @xmath14 . first , we treat the case of gaussian coefficients @xmath14 as in sub - scenario b of the ( pseudo)conformal universe . in this situation , the likelihood of the observed sky @xmath155 is naturally considered as a function of the parameter @xmath189 . the corresponding estimator has been derived in our previous paper @xcite . let us briefly recall the main idea of the calculation . we write the likelihood of interest as the product of two likelihoods integrated over all possible sets of coefficients @xmath190 , i.e. , @xmath191 here @xmath192 , and @xmath154 is given by eq . ; @xmath193 denotes the likelihood of the particular realization @xmath157 for a given value of the parameter @xmath189 . upon using the approximation , we obtain the simple gaussian form for the integrand in eq . . then the integral is evaluated in a straightforward manner . finally , setting to zero the derivative of the joint likelihood with respect to the parameter @xmath189 , we end up with the estimator @xcite @xmath194 here the @xmath195 s are defined by @xmath196 , and the @xmath197 s are given by eq . . the estimator is simplified considerably in the case of the quadrupole statistical anisotropy with the gaussian @xmath16 s , @xmath198 where we omitted the irrelevant constant prefactor . we exploit this estimator in sec . 5 when constraining the quadrupole of the general type predicted in the leading order of sub - scenario a of the ( pseudo)conformal universe . finally , let us discuss estimators for statistical anisotropy of the special quadrupole type governed by the _ non - gaussian _ amplitude @xmath44 . this type of direction dependence , we recall , arises in sub - scenario a of the ( pseudo)conformal universe ( in the subleading order ) as well as in anisotropic inflation ( case * ii * ) . in that case , the discussion above is not applicable . still , at the price of optimality , we choose to work with simple quadratic estimators built of estimators for the coefficients @xmath16 . namely , @xmath199 in the present section , we constrain early universe models by implementing estimators , , , and to the wmap9 data . the 9-year data set includes a new product : a set of beam - symmetrized maps , produced by deconvolution procedure to eliminate the effects of asymmetric beam @xcite . the latter effects were responsible for the strong bias in the measurements of the statistical anisotropy @xcite . one may treat the deconvolved map as a map measured by a hypothetical wmap - like satellite with a symmetric beam transfer function . therefore , a simple relation between real and observed signals is assumed in deconvolved maps @xmath200 where @xmath201 is a symmetric part of the beam transfer function and @xmath202 is a noise . the same relation was used in the previous analyses @xcite . with this said , we can estimate coefficients @xmath14 and @xmath183 literally following the techniques in ref . the first and most costly step here is to provide the inverse - variance filtering . for the purpose of the numerical computations , we rewrite eq . in the equivalent form @xmath203{\bf s}^{i } \bar{{\bf \theta}}= \tilde{{\bf y}}^{\dagger } { \bf n}^{-1 } \hat{{\bf \theta } } \ ; .\ ] ] here @xmath204 is a matrix which relates harmonic space covariance and the observed map , @xmath205 to solve the system , we make use of the multigrid preconditioner proposed in ref . @xcite . having inversed - filtered temperature anisotropies for both real data and a large number of mc maps , we evaluate estimators for the coefficients @xmath14 by using eq . and substituting eqs . , , , and . we compute the integral over three spherical harmonics in eq . using slatec @xcite and gsl @xcite libraries , and the coefficients @xmath176 by running camb @xcite . the summation over the multipole number @xmath206 in eq . is performed up to @xmath207 . the second term in eq . is calculated by avegaring over the large number of mc maps . we evaluate the fisher matrix using the analytical expression . the wmap9 _ kq85 _ temperature analysis mask is applied to both data and mc maps leaving @xmath208 of the sky unmasked . in fig . [ modelindep ] , we present coefficients @xmath183 reconstructed from @xmath32 and @xmath31 bands of the wmap9 maps . as is clearly seen , wmap9 data favor statistically isotropic primordial perturbations . this is to be compared with the analogous results from the 5- and 7-year releases revealing the anomalously large quadrupole @xcite . now , given the absence of the anomaly , we expect a substantial tightening of constraints on the early universe models . estimated from the @xmath32 ( left ) and @xmath31 ( right ) bands of deconvolved wmap9 maps as a function of the direction in the sky . the plot is in the galactic coordinates with @xmath209 on the left.,title="fig : " ] estimated from the @xmath32 ( left ) and @xmath31 ( right ) bands of deconvolved wmap9 maps as a function of the direction in the sky . the plot is in the galactic coordinates with @xmath209 on the left.,title="fig : " ] we start with constraining the amplitude @xmath44 of the quadrupole of the special type . this is interesting from the viewpoint of scenarios where @xmath44 is uniquely defined by the model parameters , as in the case * i * of anisotropic inflation . first , we apply the estimator in order to constrain the amplitude for some particular preferred directions . the results of the estimation are presented in fig . [ gstar ] . the constraining procedure is as follows : * we calculate the set @xmath210 starting from some fixed value of @xmath44 and preferred direction @xmath185 . * for the set @xmath211 , we generate a number of anisotropic maps and estimate the amplitude @xmath44 from the latter . we make use of eq . . * we compare the values of @xmath44 derived from anisotropic maps with the wmap9 estimate . we request that not more than @xmath1 of them exceed ( are smaller than ) the real estimate in the case of positive ( negative ) @xmath44 fixed in the beginning . the generation of anisotropic maps is perhaps the most nontrivial step in this chain . we do this in line with appendix a of ref . @xcite ( see also our previous paper @xcite ) . the idea is to consider the temperature anisotropy of the form @xmath212^{-1}]\right)^{1/2 } { \bf \theta}^{i } \ ; , \ ] ] where @xmath213 . here @xmath177 and @xmath214 are the anisotropic and isotropic theoretical covariances given by eqs . and , respectively . it is straightforward to check that the temperature anisotropy corresponds to the anisotropic covariance @xmath177 . assuming small statistical anisotropy , we expand it to the linear order in @xmath215 , @xmath216 { \bf \theta}^i \ ; .\ ] ] we multipy the latter by the symmetric beam , convert into the pixel space , add the noise and apply the mask . the anisotropic maps constructed are then analyzed in the same way as the data maps . we followed the above procedure for the three fixed directions in the sky . the directions form an orthogonal basis with the first one aligned with the poles of the ecliptic plane . constraints on the amplitude @xmath44 are presented in table [ constr1 ] . we also establish @xmath3 c.l . limits in the case of the direction aligned with the ecliptic poles , @xmath217 these limits are obtained from the @xmath32 band of the wmap9 data . they are to be compared with the analogous constraints derived from the nonobservation of quadrupolar statistical anisotropy in the data by the sloan digital sky survey which read @xmath218 at @xmath3 c.l . as is clearly seen , our bounds are consistent with previous ones but demonstrate an improvement by a factor of 2 . .wmap9 @xmath32 band 95% c.l . constraints on the amplitude of the special quadrupole @xmath44 for particular preferred directions in the sky . [ cols="^,^",options="header " , ] the amplitude @xmath44 is random in a number of realistic models and the constraints above are not applicable there . for example , this is the case of anisotropic inflation for the amplitude @xmath44 sourced by the quantum excitations of the electric field , @xmath67 . to set limits on this class of models we slightly modify our constraining scheme . that is , we fix some value of the @xmath4-fold number @xmath219 in the beginning . using eq . with @xmath220 we generate the collection of vectors @xmath92 . each vector @xmath221 uniquely defines the amplitude @xmath44 and the set of coefficients @xmath222 . we generate anisotropic maps for each set . finally , we compare the value of the @xmath4-fold number @xmath70 estimated from the anisotropic maps with the wmap9 estimate . we make use of the estimator . limits on the relative @xmath4-fold number @xmath223 are given in table [ constrearly ] at the @xmath1 c.l . the @xmath3 c.l . limit is the strongest one for the @xmath32 band , @xmath224 a similar constraint was obtained in ref . @xcite from the nonobservation of the trispectrum non - gaussianity in the planck data @xcite ; see the discussion in sec . now , let us turn to the models of the ( pseudo)conformal universe . we start with sub - scenario a. as discussed in sec . 3.3 , it predicts the quadrupolar statistical anisotropy of both the general and special types . the latter appears in the nonlinear order ( nlo ) in the constant @xmath27 , while the former is nonzero already in the linear order ( lo ) . still , the general quadrupole makes a weaker imprint on the cmb sky than the special quadrupole . indeed , the former is characterized by the decreasing amplitude @xmath225 , which translates into the additional suppression by the cmb multipole number @xmath226 . consequently , we have effectively low statistics for the multipoles useful in the analysis and , hence , a very weak constraint on the parameter @xmath189 . as a proof , the formal constraint @xmath227 was obtained from the @xmath32 band of the wmap7 data at the @xmath1 c.l . of course , we can not trust so large upper limit , since it violates the assumption of small statistical anisotropy made in the beginning . rather the number 190 demonstrates the low sensitivity of the wmap7 data towards the signal predicted . qualitatively , the same story repeats at the level of wmap9 maps . to show this explicitly , we estimate the coefficients @xmath14 and @xmath183 , but now with the decreasing amplitude @xmath228 in eqs . and . corresponding results are shown in fig . [ modelk ] . for the value of @xmath189 we use the estimator . next , we generate 100 sets of coefficients @xmath16 using eq . out of some fixed value of the parameter @xmath189 , and construct anisotropic maps for each set . we compare values of the estimator for the parameter @xmath189 obtained from the real data and mc - simulated anisotropic maps . the final results are presented in table [ constrearly ] . much tighter bounds are expected from the nonobservation of statistical anisotropy of the special type . remarkably , one can derive them immediately from the upper limits on the relative @xmath4-fold number @xmath70 established in the previous subsection . this is due to the correspondence between predictions of two different setups , i.e. , sub - scenario a of the ( pseudo)conformal universe and case * ii * of anisotropic inflation . the duality holds modulo the presence of the general quadrupole , which anyway produces the negligible effect for @xmath229 . by exploiting eq . , we obtain @xmath1 c.l . limits on the parameter of interest , i.e. , @xmath230 ; see table [ constrearly ] . from eq . , we also derive the @xmath3 c.l . limit , @xmath231 finally , using the duality between the conformal rolling scenario and galilean genesis , we convert the derived constraints into the bounds on the parameter space of the latter model ; see table [ constrearly ] . we conclude with constraining sub - scenario b of the ( pseudo)conformal universe . we recall that statistically anisotropic effects in this case are nonzero already in the first order in the constant @xmath27 . given also the constant amplitude @xmath49 , we anticipate rather strong constraints from the nonobservation of the signal predicted . we use eq . to estimate the parameter @xmath189 . the results plotted in fig . [ h2plot ] are in excellent agreement with expectations from the isotropic hypothesis . particular values of the parameter @xmath189 estimated at @xmath232 read @xmath233 ( @xmath32 band ) and @xmath234 ( @xmath31 band ) . to construct anisotropic maps , we generate a large number of sets of the coefficients @xmath14 out of some fixed value of the parameter @xmath189 . we make use of eq . . the constraints given in table [ constrearly ] demonstrate a roughly one - order - of - magnitude improvement as compared to our wmap7 result . let us discuss our constraints in light of the planck data . the latter have the obvious advantage due to the larger number of multipoles useful in the analysis , namely @xmath235 as compared @xmath236 relevant for the wmap9 data . qualitatively , this implies @xmath237 growth of the sensitivity to statistical anisotropy of the special quadrupole type . indeed , the fisher matrix scales as @xmath238 with the maximal multipole number @xmath239 . on the other hand , the error bars for the amplitude @xmath44 are roughly measured by @xmath240 . this explains the estimate for the growth of sensitivity . looking at eq . , we would thus expect constraints on the amplitude @xmath44 at @xmath241 level in case the signal is not observed . recently , @xmath3 c.l . limits appeared @xcite based on the planck 143 ghz data at the level @xmath242 independently of the preferred direction in the sky . these planck constraints are stronger than the limits derived from the wmap9 data by a factor 2 - 3 . the first planck limits were obtained , however , using only one frequency band and a suboptimal estimator . based on the above , we anticipate somewhat stronger constraints with the future analysis of planck data . interesting consequences are expected for the duration of inflation in models with the nonminimal maxwellian term : in particular , we expect extremely tuned number of @xmath4-folds in those models , i.e. , @xmath243 . planck data are also promising for constraining sub - scenario b of the ( pseudo)conformal universe . indeed , the sensitivity of the data to the parameter @xmath189 grows as @xmath244 . qualitatively , this implies a bound on @xmath189 that is stronger by a factor of @xmath245 in the case where the signal of interest is not observed . a much weaker improvement , i.e. , by a factor @xmath246 , is expected for the parameter @xmath247 of sub - scenario a. we end up with a few remarks . the types of statistical anisotropy studied in this paper cover most predictions that exist in the literature , but not all . for example , the model of ref . @xcite predicts a special quadrupole characterized by an increasing amplitude , i.e. , @xmath248 . this originates from the axial coupling between the inflaton and vector fields . the peculiar form of statistical anisotropy with a vanishing quadrupole term follows from inflation involving scalars with nonminimal kinetic terms @xcite . in both cases , a separate data analysis is required . on the other hand , constraints presented in table [ constrearly ] are easily converted into limits on the parameters of scenarios with more conventional predictions , e.g. , statistical anisotropy of the acw type . these include scenarios based on the noncommutative geometry @xcite , @xmath249-forms @xcite , etc . although we focused on the particular prediction of statistical anisotropy , the scenarios discussed in this paper may have other interesting signatures . indeed , anisotropic inflation gives rise to some amount of non - gaussianities at both the bispectrum @xcite and trispectrum @xcite levels . these are sourced by infrared fluctuations of the electric field , and thus are highly sensitive to the duration of the inflationary phase . in ref . @xcite , this simple observation has been used to establish the upper limit @xmath250 also given that the planck constraint for the trispectrum parameter @xmath251 at @xmath1 c.l . , one concludes that the constraint is fairly similar to our @xmath3 c.l . limit . note , however , that the random nature of the field @xmath67 was not accounted for in the derivation of eq . . thus , the latter is less conservative by its definition and may become weaker if the randomness is included . other predictions of inflation with the nonminimal maxwellian term include the anisotropy in the tensor power spectrum and the cross correlation between curvature and tensor perturbations @xcite . models of the ( pseudo)conformal universe may also lead to potentially large non - gaussianities @xcite . in particular , the trispectrum @xcite governed by the parameter @xmath189 , can be used to constrain the latter from nonobservation in the planck data . this approach appears to be especially promising in a view of the fact that sub - scenario a predicts rather weak signal of statistical anisotropy . it is a pleasure to thank valery rubakov , diego chialva , alessandro gruppuso , maxim libanov , and federico urban for many useful comments and stimulating discussions . we are grateful to d. hanson for kindly providing the code for inverse variance filtering . s. r. is supported by the belgian science policy ( iap vii/37 ) . s. r. is indebted to lund university for warm hospitality . the work of g. r. is supported in part by the rfbr grants 12 - 02 - 00653 , 12 - 02 - 31776 , 13 - 02 - 01293 , by the dynasty foundation , by the grants of the president of the russian federation ns-5590.2012.2 , mk-1170.2013.2 , by the russian federation governement grant no.11.g34.31.0047 . we acknowledge the use of the legacy archive for microwave background data analysis ( lambda ) , part of the high energy astrophysics science archive center ( heasarc ) . heasarc is a service of the astrophysics science division at the nasa goddard space flight center . the numerical part of the work was done at the cluster of the theoretical division of inr ras . 99 r. m. wald , phys . d * 28 * ( 1983 ) 2118 . l. ackerman , s. m. carroll and m. b. wise , phys . d * 75 * ( 2007 ) 083502 [ erratum - ibid . d * 80 * ( 2009 ) 069901 ] [ astro - ph/0701357 ] . a. golovnev , v. mukhanov and v. vanchurin , jcap * 0806 * ( 2008 ) 009 [ arxiv:0802.2068 [ astro - ph ] ] . s. yokoyama and j. soda , jcap * 0808 * ( 2008 ) 005 [ arxiv:0805.4265 [ astro - ph ] ] . k. dimopoulos , m. karciauskas , d. h. lyth and y. rodriguez , jcap * 0905 * ( 2009 ) 013 [ arxiv:0809.1055 [ astro - ph ] ] . + k. dimopoulos , m. karciauskas and j. m. wagstaff , phys . b * 683 * ( 2010 ) 298 [ arxiv:0909.0475 [ hep - ph ] ] . + k. dimopoulos , int . j. mod . d * 21 * ( 2012 ) 1250023 [ erratum - ibid . d * 21 * ( 2012 ) 1292003 ] [ arxiv:1107.2779 [ hep - ph ] ] . a. a. abolhasani , r. emami and h. firouzjahi , arxiv:1311.0493 [ hep - th ] . + r. emami and h. firouzjahi , jcap * 1310 * ( 2013 ) 041 [ arxiv:1301.1219 [ hep - th ] ] . + r. emami , h. firouzjahi , s. m. sadegh movahed and m. zarei , jcap * 1102 * ( 2011 ) 005 [ arxiv:1010.5495 [ astro-ph.co ] ] . b. himmetoglu , c. r. contaldi and m. peloso , phys . d * 79 * ( 2009 ) 063517 [ arxiv:0812.1231 [ astro - ph ] ] . + b. himmetoglu , c. r. contaldi and m. peloso , phys . d * 80 * ( 2009 ) 123530 [ arxiv:0909.3524 [ astro-ph.co ] ] . m. -a . watanabe , s. kanno and j. soda , phys . * 102 * ( 2009 ) 191302 [ arxiv:0902.2833 [ hep - th ] ] . + m. -a . watanabe , s. kanno and j. soda , prog . * 123 * ( 2010 ) 1041 [ arxiv:1003.0056 [ astro-ph.co ] ] . a. e. gumrukcuoglu , b. himmetoglu and m. peloso , phys . d * 81 * ( 2010 ) 063528 [ arxiv:1001.4088 [ astro-ph.co ] ] . + t. r. dulaney and m. i. gresham , phys . d * 81 * ( 2010 ) 103532 [ arxiv:1001.2301 [ astro-ph.co ] ] . j. soda , class . * 29 * ( 2012 ) 083001 [ arxiv:1201.6434 [ hep - th ] ] . a. maleknejad , m. m. sheikh - jabbari and j. soda , phys . * 528 * ( 2013 ) 161 [ arxiv:1212.2921 [ hep - th ] ] . n. bartolo , s. matarrese , m. peloso and a. ricciardone , phys . d * 87 * ( 2013 ) 023504 [ arxiv:1210.3257 [ astro-ph.co ] ] . d. h. lyth and m. karciauskas , jcap * 1305 * ( 2013 ) 011 [ arxiv:1302.7304 [ astro-ph.co ] ] . m. libanov , s. ramazanov and v. rubakov , jcap * 1106 * ( 2011 ) 010 [ arxiv:1102.1390 [ hep - th ] ] . p. creminelli , a. joyce , j. khoury and m. simonovic , jcap * 1304 * ( 2013 ) 020 [ arxiv:1212.3329 [ hep - th ] ] . s. r. ramazanov and g. i. rubtsov , jcap * 1205 * ( 2012 ) 033 [ arxiv:1202.4357 [ astro-ph.co ] ] . d. hanson and a. lewis , phys . d * 80 * ( 2009 ) 063004 [ arxiv:0908.0963 [ astro-ph.co ] ] . n. e. groeneboom and h. k. eriksen , astrophys . j. * 690 * ( 2009 ) 1807 [ arxiv:0807.2242 [ astro - ph ] ] . n. e. groeneboom , l. ackerman , i. k. wehus and h. k. eriksen , astrophys . j. * 722 * ( 2010 ) 452 [ arxiv:0911.0150 [ astro-ph.co ] ] . a. r. pullen and c. m. hirata , jcap * 1005 * ( 2010 ) 027 [ arxiv:1003.0673 [ astro-ph.co ] ] . d. hanson , a. lewis and a. challinor , phys . d * 81 * ( 2010 ) 103003 [ arxiv:1003.0198 [ astro-ph.co ] ] . c. l. bennett _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 208 * , 20 ( 2013 ) [ arxiv:1212.5225 [ astro-ph.co ] ] . a. maleknejad and m. m. sheikh - jabbari , phys . d * 85 * ( 2012 ) 123508 [ arxiv:1203.0219 [ hep - th ] ] . m. osipov and v. rubakov , jetp lett . * 93 * ( 2011 ) 52 [ arxiv:1007.3417 [ hep - th ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5076 [ astro-ph.co ] . j. -l . lehners , phys . rept . * 465 * ( 2008 ) 223 [ arxiv:0806.1245 [ astro - ph ] ] . m. libanov , s. mironov and v. rubakov , phys . d * 84 * ( 2011 ) 083502 [ arxiv:1105.6230 [ astro-ph.co ] ] . + m. libanov , s. mironov and v. rubakov , prog . * 190 * ( 2011 ) 120 [ arxiv:1012.5737 [ hep - th ] ] . k. m. smith , o. zahn and o. dore , phys . d * 76 * ( 2007 ) 043510 [ arxiv:0705.3980 [ astro - ph ] ] a. lewis , a. challinor and a. lasenby , astrophys . j. * 538 * ( 2000 ) 473 [ astro - ph/9911177 ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5084 [ astro-ph.co ] . c. armendariz - picon , jcap * 0709 * ( 2007 ) 014 [ arxiv:0705.1167 [ astro - ph ] ] . e. di grezia , g. esposito , a. funel , g. mangano and g. miele , phys . rev . d * 68 * ( 2003 ) 105012 [ gr - qc/0305050 ] . + e. akofor , a. p. balachandran , s. g. jo , a. joseph and b. a. qureshi , jhep * 0805 * ( 2008 ) 092 [ arxiv:0710.5897 [ astro - ph ] ] . c. germani and a. kehagias , jcap * 0903 * ( 2009 ) 028 [ arxiv:0902.3667 [ astro-ph.co ] ] . + t. s. koivisto and n. j. nunes , phys . b * 685 * ( 2010 ) 105 [ arxiv:0907.3883 [ astro-ph.co ] ] . + t. s. koivisto and n. j. nunes , phys . d * 80 * ( 2009 ) 103509 [ arxiv:0908.0920 [ astro-ph.co ] ] . + c. germani and a. kehagias , jcap * 0911 * ( 2009 ) 005 [ arxiv:0908.0001 [ astro-ph.co ] ] . + t. s. koivisto , d. f. mota and c. pitrou , jhep * 0909 * ( 2009 ) 092 [ arxiv:0903.4158 [ astro-ph.co ] ] . + f. r. urban , phys . d * 88 * ( 2013 ) 063525 [ arxiv:1307.5215 [ astro-ph.co ] ] . n. barnaby , r. namba and m. peloso , phys . d * 85 * ( 2012 ) 123523 [ arxiv:1202.1469 [ astro-ph.co ] ] . + m. shiraishi , e. komatsu , m. peloso and n. barnaby , jcap * 1305 * ( 2013 ) 002 [ arxiv:1302.3056 [ astro-ph.co ] ] . + a. a. abolhasani , r. emami , j. t. firouzjaee and h. firouzjahi , jcap * 1308 * ( 2013 ) 016 [ arxiv:1302.6986 [ astro-ph.co ] ] .

we used the n body code of to build a dozen cuspy ( ) triaxial models of stellar systems through dissipationless collapses of initially spherical distributions of particles . we chose four sets of initial conditions that resulted in models morphologically resembling e2 , e3 , e4 and e5 galaxies , respectively . within each set , three different seed numbers were selected for the random number generator used to create the initial conditions , so that the three models of each set are statistically equivalent . we checked the stability of our models using the values of their central densities and of their moments of inertia , which turned out to be very constant indeed . the changes of those values were all less than 3 per cent over one hubble time and , moreover , we show that the most likely cause of those changes are relaxation effects in the numerical code . we computed the six lyapunov exponents of nearly 5,000 orbits in each model in order to recognize regular , partially and fully chaotic orbits . all the models turned out to be highly chaotic , with less than 25 per cent of their orbits being regular . we conclude that it is quite possible to obtain cuspy triaxial stellar models that contain large fractions of chaotic orbits and are highly stable . the difficulty to build such models with the method of should be attributed to the method itself and not to physical causes . [ firstpage ] galaxies : elliptical and lenticular , cd galaxies : kinematics and dynamics methods : numerical physical data and processes : chaos .

1204.5428

self - consistent models of spherical and disklike stellar systems are relatively simple to build using standard textbook methods @xcite , but special techniques are necessary to obtain triaxial models adequate to represent elliptical galaxies . one of the most popular of those techniques is due to @xcite : one chooses a reasonable mass distribution for the system , obtains the corresponding potential and computes a large library of orbits in that potential selecting suitable initial conditions ; weights are then assigned , according to the time spent on each orbit and in every region of space , and used to set up a system of linear equations linking the mass density in each region to the fractions of the different types of orbits , which can be finally obtained solving the system . a different method , which we will refer to as the n body method , was due to @xcite : adopting an initial distribution of mass points , one integrates their equations of motion with an n body code until an equilibrium configuration is reached ; the potential is then fixed and fitted with an adequate smooth approximation that allows the computation of orbits using the positions and velocities of the bodies as initial conditions . the initial distribution of bodies can be chosen in different ways to obtain a triaxial system : slow deformation of a spherical system @xcite , cold collapse of a spherical system @xcite , merging of stellar systems @xcite , and so on . nevertheless , with the n body method it is not possible to have the degree of control over the final configuration that one has with schwarzschild s method . in the end , both methods yield the same result ( i.e. , a self - consistent stellar system and an analysis of its orbital content ) , but the method of @xcite starts with a chosen mass distribution , obtains orbits and finds which fractions of those orbits are needed to have a self - consistent system with that mass distribution , while the n body method begins obtaining a self - consistent system and then performs the orbital analysis . chaotic motions are frequent in stellar systems @xcite , and @xcite realized that triaxial systems should include chaotic orbits but , when he introduced them in his models , the models evolved on time scales of the order of a hubble time and were not truly stable . this problem was aggravated after it became clear that cuspy models ( where near the centre the density , @xmath2 , is proportional to @xmath3 , where @xmath4 is the radius and @xmath5 ) were needed for elliptical galaxies @xcite . nevertheless , it is perfectly possible to obtain stable triaxial models , even cuspy ones , that contain large fractions of chaotic orbits using the n body method ( @xcite ; @xcite ; @xcite ; @xcite ) , so that several of those authors argued that the difficulty to build such systems with the method of @xcite was due to that method itself and not to a physical cause . clearly , chaotic orbits are difficult to deal with when using the method of @xcite because they cover different regions at different times . worse still , in cuspy triaxial potentials chaotic orbits tend to be extremely sticky @xcite , and sticky orbits behave like regular orbits for long periods of time and chaotically at other intervals . let us assume , as an example , that chaotic orbits occupy part of the time an elongated region of space and another part of the time a nearly spherical region . therefore , the orbits that occupy mostly an elongated region during the integration time used to prepare the library of orbits will have large weights in that region and low weights outside it ; conversely , the orbits that occupy mostly the nearly spherical region during the same integration time will have a more or less even distribution of weight values over this region . now , when these weights are used to establish the fractions of orbits that make up the triaxial system , the orbits of the former group will be strongly favoured and the equilibrium system yielded by the method of @xcite will include many of them and few of the other group . if we then let the system evolve , the former orbits will tend to fill in a more spherical region as time goes by and , conversely , the latter orbits will adopt a flatter distribution . but , since the model included less orbits that originally had a more spherical distribution , the model will become rounder , as shown by the table 6 of @xcite for most of his models . the use of longer integration times , as @xcite advocated , does not guarantee the success of the method either , because the average behaviour of a chaotic orbit over a certain interval , no matter how long , does not necessarily coincide with its behaviour over a different interval , or even a subinterval , of the integration time . to make matters worse , the number of orbits typically computed for schwarzschild s method is not very large , from several hundreds in the works of schwarzschild himself up to about @xmath6 in more recent work like that of @xcite , so that they do not provide a strong enough statistical basis . on the other hand , the n body method uses large numbers of bodies ( of the order of @xmath1 in our own more recent works ) guaranteeing good statistics and the model is created by the evolution of the system itself , that is , the dynamics take care of favoring certain orbits at the expense of others so as to reach a self - consistent state . is that state a stable one ? if all the orbits are regular it certainly is because , once reached that state , the orbits will continue filling in the same regions of space for ever , in other words , we will have a static equilibrium . if the model includes chaotic orbits , however , it is conceivable that , once in equilibrium , when part of the chaotic orbits evolves to occupy a rounder space region , another part of the chaotic orbits which had a rounder distribution evolves towards a more elongated one and fill in the phase space regions left vacant by the former , that is , we might have a dynamic rather than static equilibrium . while all this is concievable there is no guarantee that it will actually happen , but the past investigations that resulted in stable models that contained high fractions of chaotic orbits undoubtedly support this view . therefore , we want to investigate whether it is possible to obtain self - consistent models of cuspy triaxial sytems that are stable . that is , we will deal with the type of model that is most likely to contain high fractions of chaotic orbits and most difficult to obtain with schwarzschild s method . we have already obtained such models with the n body method @xcite but , in order to compensate for the softening needed by the n body code of l.a . aguilar @xcite we had to introduce an additional potential . here we will use the code of l. hernquist @xcite that does not need softening and that uses an expansion of the potential based on the model of @xcite that is particularly adequate for cuspy models . the next section gives a description of how we obtained our models and their main properties . section 3 presents our results on the stability and chaoticity of the models and section 4 summarizes our conclusions . we built our models following the recipe of @xcite , just as we have done in our previous investigations ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for example ) : we randomly created a spherical distribution of @xmath1 particles with a density distribution inversely proportional to the distance to the centre and a gaussian velocity distribution , and we let it collapse following the evolution with the code of @xcite ; due to the radial orbit instability , the result is a triaxial system . the gravitational constant , @xmath7 , the radius of the sphere and the total mass are all set equal to @xmath8 and the collapse time ( the one needed to reach the maximum potential energy ) turns out to be also very close to unity . we followed the initial collapse for 7 time units ( t.u . , hereafter ) and , then , we eliminated the particles with positive energy , we determined the principal axes of the inertia tensor of the @xmath9 per cent most tightly bound particles and we rotated the system so that its major , intermediate and minor axes coincide , respectively , with the x , y and z axes of coordinates . subsequently , we let the system evolve for another 300 crossing times ( @xmath10 , hereafter , see table 1 below ) , eliminated the less bound particles that had not yet reached equilibrium ( less than @xmath11 or @xmath12 per cent of the particles in all cases ) and aligned it again with the system of coordinates . we then performed a last run of @xmath13 @xmath10 to , first , obtain the final model after the initial @xmath14 @xmath10 and , second , check its stability over the final @xmath15 @xmath10 . we must recall that we are only interested in obtaining models morphologically similar to elliptical galaxies and not in creating them through a realistic process . besides , we want to make sure that our models have reached the equilibrium state before testing them for possible evolutionary effects . as we will show later , one hubble time is of the order of @xmath16 @xmath10 for our models , so that the integration times used here are indeed much larger than that time and , as indicated , do not correspond to a realistic process of galaxy formation . after several trials , we chose an integration step of @xmath17 which yielded integration errors in the energies of the individual bodies negligibly small compared to the changes in those energies due to the relaxation effects of the code @xcite . with that choice , the total energy is conserved within about @xmath18 per cent during most of the evolutions , except during the initial collapse when it is conserved only within about @xmath8 or @xmath11 per cent . @xcite investigated the approximation of n body realizations of models of @xcite for @xmath19 with a generalization of the method of @xcite and they found that the choice of the radial basis functions seriously affects the results of the fractions of chaotic orbits or the distribution of the lyapunov characteristic exponents . the model of @xcite corresponds to the @xmath20 case of the models of @xcite so that the method of @xcite , based on the former , should be expected to yield good results for models with @xmath20 and , in fact , that is one conclusion of the work of @xcite . therefore , it is important that we stick to models with @xmath20 , not only because they are fairly cuspy , but also because the n body method we are using works best for such models . notice that even our initial distribution ( an sphere with density proportional to the inverse of the radius ) obeys that @xmath21 value . the code of @xcite allows one to choose the number of terms in the angular and radial expansions ( @xmath22 and @xmath23 , respectively ) , so that we performed tests with different numbers of terms in the ranges @xmath24 and @xmath25 @xcite . for every ( @xmath22 , @xmath23 ) pair we run three simulations that differed only in the seed number used to randomly generate the initial positions and velocities , i.e. , the three were statistically equivalent . the pairs @xmath26 , @xmath27 , @xmath28 , @xmath29 , @xmath30 , @xmath31 and @xmath32 yielded models that were only moderately cuspy , with @xmath33 , so that they were discarded . it might seem odd that , despite the use of a basis of radial functions derived from the hernquist model , which has @xmath20 , the obtention of collapse models with such cuspiness is not guaranteed but , although the zero order radial function is cuspy with @xmath34 , higher order terms are not and in all likelihood their contribution flattens the cusp . most of the remaining ( @xmath22 , @xmath23 ) pairs might have been acceptable , as the reasons to prefer one to another were not as compelling as the low @xmath21 values that made us reject those mentioned before . pairs @xmath35 and @xmath36 yielded @xmath21 values that departed moderately ( less than 0.1 ) from @xmath8 , and the results from their different statistical realizations exhibited larger dispersions than other pairs . of the three last pairs , all with @xmath37 , the one with @xmath38 showed the most consistent results among the different statistical realizations , so that we finally decided to use those numbers of terms in our expansions . following @xcite we took the square roots of the mean square values of coordinates of the @xmath9 per cent most tightly bound particles as the semiaxes of the system ; the major , intermediate and minor axes will be dubbed @xmath39 , @xmath40 and @xmath41 , respectively , hereafter . we built several models adopting different values of the dispersion for the velocity distribution , obtaining less elongated models for larger dispersion values . finally , we adopted four dispersion values that resulted in models with @xmath42 values close to @xmath43 , @xmath44 , @xmath45 and @xmath46 , respectively ( i.e. , corresponding to elliptical galaxies between hubble types e2 and e5 ) . the velocity dispersion for the e5 model was essentially zero , so that more elongated models could not be obtained . a similar result had been obtained in our previous work @xcite , where we could not obtain models more flattened than e6 and we suggested that this might hint that mergers , rather than collapses , are needed to obtain the most elongated ellipticals . for each selected value of the velocity dispersion , three different models ( dubbed @xmath39 , @xmath40 and @xmath41 ) were obtained using three different seed numbers for the random number generator when creating the initial distribution of particles . in brief , we prepared a grand total of 12 models , divided in 4 groups ; the models in each group resemble e2 , e3 , e4 and e5 galaxies , respectively , and the three models within each group differ from each other only from a microscopic point of view , being essentially identical from the point of view of their macroscopic properties . as in our previous work @xcite , several models display figure rotation around the minor axis , i.e. , they rotate very slowly even though their total angular momentum is zero . @lccccccc@ @xmath47 & @xmath48 & @xmath10 & @xmath49 & @xmath50 & @xmath51 & @xmath21 & @xmath52 + e2a & 0.990 & 0.721 & 0.250 & 0.894 & 0.73 & 1.052 @xmath53 0.021 & 0.0003 @xmath53 0.0001 + e2b & 0.990 & 0.723 & 0.253 & 0.897 & 0.78 & 0.997 @xmath53 0.021 & 0.0012 @xmath53 0.0004 + e2c & 0.990 & 0.721 & 0.250 & 0.902 & 0.70 & 1.006 @xmath53 0.021 & 0.0016 @xmath53 0.0004 + + e3a & 0.979 & 0.659 & 0.220 & 0.934 & 0.65 & 1.056 @xmath53 0.021 & -0.0004 @xmath53 0.0001 + e3b & 0.979 & 0.659 & 0.221 & 0.933 & 0.68 & 1.062 @xmath53 0.021 & 0.0002 @xmath53 0.0001 + e3c & 0.978 & 0.657 & 0.219 & 0.929 & 0.66 & 0.987 @xmath53 0.021 & 0.0010 @xmath53 0.0001 + + e4a & 0.911 & 0.526 & 0.182 & 0.957 & 0.62 & 1.053 @xmath53 0.019 & 0.0001 @xmath53 0.0003 + e4b & 0.905 & 0.518 & 0.180 & 0.974 & 0.63 & 1.027 @xmath53 0.021 & -0.0002 @xmath53 0.0001 + e4c & 0.908 & 0.521 & 0.180 & 0.963 & 0.62 & 1.073 @xmath53 0.022 & -0.0002 @xmath53 0.0002 + + e5a & 0.906 & 0.463 & 0.159 & 0.997 & 0.46 & 0.985 @xmath53 0.022 & 0.0020 @xmath53 0.0007 + e5b & 0.908 & 0.466 & 0.160 & 0.986 & 0.48 & 1.000 @xmath53 0.019 & 0.0059 @xmath53 0.0004 + e5c & 0.907 & 0.465 & 0.158 & 0.970 & 0.46 & 1.017 @xmath53 0.020 & 0.0164 @xmath53 0.0003 + table 1 summarizes the global properties of our models : mass , crossing time , effective radius , central radial velocity dispersion , triaxiality , @xmath21 and angular velocity of figure rotation . the effective radius was obtained from the @xmath54 projection and , accordingly , the central radial velocity dispersion was computed from the y components of the velocities of the @xmath6 particles closer to the centre on that projection . triaxiality was evaluated from the semiaxes obtained from the @xmath9 per cent most tightly bound particles as @xmath55 . @xmath21 was obtained as the slope of the @xmath56 vs. @xmath57 line for the innermost @xmath6 particles binned in @xmath14 particle bins . the angular velocity was computed from the angles formed by the major axis with its original position at different times of the final @xmath15 @xmath10 integration . the table clearly shows that , for a given model , the different realizations obtained changing the seed number in the random number generator have essentially the same global properties . the only exception is the angular velocity which displays significant differences , particularly among models e5a , b and c. nevertheless , rotation is very slow in all cases and , except for the e5c case , lower than the 0.00975 value of the investigation by @xcite , which revealed only very small differences between the fractions of chaotic orbits in the rotating and non - rotating system . the present e4 and e5 models are much more triaxial than the corresponding non - cuspy models of @xcite ( @xmath58 and @xmath59 , respectively ) , and the present model e4 is much more triaxial than the cuspy model e4c of @xcite ( @xmath60 ) . as in our previous work ( see * ? ? ? * ; * ? ? ? * for example ) , we chose galaxies ngc1379 and ngc4697 @xcite , whose mass - to - light ratio gradients are zero , to obtain some estimate of the equivalence between our units and those of real galaxies , comparing their observed values of @xmath49 ( @xmath61 and @xmath62 , respectively ) and @xmath50 ( @xmath63 and @xmath64 , respectively ) with those from our table [ tab1 ] , we conclude that values between about @xmath65 and @xmath66 can be used as our length unit and values between about @xmath67 and @xmath68 as our time unit . then , the hubble time can be estimated as between @xmath69 and @xmath70 , and we will adopt a value of @xmath71 , hereafter . @lcccccc@ @xmath47 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & @xmath76 & @xmath77 + @xmath78 & a & 0.068 & 0.117 & 0.172 & 0.300 & 0.578 + e2a & b / a & 0.754 & 0.789 & 0.837 & 0.877 & 0.930 + @xmath78 & c / a & 0.596 & 0.684 & 0.768 & 0.826 & 0.909 + + @xmath78 & a & 0.068 & 0.118 & 0.175 & 0.297 & 0.574 + e2b & b / a & 0.742 & 0.777 & 0.820 & 0.870 & 0.925 + @xmath78 & c / a & 0.602 & 0.687 & 0.763 & 0.829 & 0.906 + + @xmath78 & a & 0.067 & 0.116 & 0.173 & 0.294 & 0.578 + e2c & b / a & 0.767 & 0.801 & 0.845 & 0.882 & 0.937 + @xmath78 & c / a & 0.605 & 0.691 & 0.769 & 0.826 & 0.913 + @xmath78 & a & 0.062 & 0.115 & 0.164 & 0.294 & 1.097 + e3a & b / a & 0.753 & 0.723 & 0.781 & 0.814 & 0.852 + @xmath78 & c / a & 0.597 & 0.564 & 0.642 & 0.694 & 0.810 + + @xmath78 & a & 0.065 & 0.117 & 0.165 & 0.294 & 1.139 + e3b & b / a & 0.710 & 0.701 & 0.761 & 0.802 & 0.840 + @xmath78 & c / a & 0.554 & 0.555 & 0.634 & 0.692 & 0.803 + + @xmath78 & a & 0.065 & 0.115 & 0.165 & 0.294 & 1.144 + e3c & b / a & 0.714 & 0.715 & 0.776 & 0.808 & 0.852 + @xmath78 & c / a & 0.559 & 0.559 & 0.635 & 0.689 & 0.807 + @xmath78 & a & 0.056 & 0.104 & 0.152 & 0.233 & 1.146 + e4a & b / a & 0.731 & 0.697 & 0.735 & 0.769 & 0.775 + @xmath78 & c / a & 0.581 & 0.516 & 0.546 & 0.581 & 0.674 + + @xmath78 & a & 0.056 & 0.102 & 0.150 & 0.231 & 1.160 + e4b & b / a & 0.728 & 0.698 & 0.733 & 0.764 & 0.792 + @xmath78 & c / a & 0.575 & 0.517 & 0.549 & 0.583 & 0.706 + + @xmath78 & a & 0.055 & 0.103 & 0.152 & 0.232 & 1.180 + e4c & b / a & 0.749 & 0.706 & 0.737 & 0.765 & 0.797 + @xmath78 & c / a & 0.588 & 0.515 & 0.539 & 0.575 & 0.708 + @xmath78 & a & 0.052 & 0.095 & 0.148 & 0.216 & 0.482 + e5a & b / a & 0.824 & 0.809 & 0.815 & 0.814 & 0.893 + @xmath78 & c / a & 0.557 & 0.508 & 0.504 & 0.515 & 0.563 + + @xmath78 & a & 0.051 & 0.095 & 0.147 & 0.222 & 0.479 + e5b & b / a & 0.846 & 0.810 & 0.802 & 0.803 & 0.879 + @xmath78 & c / a & 0.596 & 0.509 & 0.506 & 0.508 & 0.555 + + @xmath78 & a & 0.051 & 0.094 & 0.145 & 0.222 & 0.496 + e5c & b / a & 0.847 & 0.825 & 0.810 & 0.810 & 0.883 + @xmath78 & c / a & 0.583 & 0.512 & 0.500 & 0.506 & 0.569 + we obtained the projected distributions of the particles on the @xmath79 plane and , following @xcite , we adopted fixed axial ratios equal to those of the @xmath9 per cent most tightly bound particles for each model ( see below ) , and computed the surface density of particles in elliptical shells containing @xmath6 particles per bin . just as @xcite had found for their models , ours follow very closely a de vaucouleurs law and , again , differences among the different random realization of the same model are negligibly small . as an example , figure [ fig : leyvauce5 ] shows the results for our models e5 . plot for the e5 models , width=317 ] figure [ fig : densvsrade5 ] shows , as an example , the logarithm of the density versus the logarithm of the radius for the central part of the e5 models ; different symbols correspond to models with different seed numbers and a straight line with slope @xmath20 is also shown for comparison . in this case the density was computed in spherical shells containing 100 particles each , so that the poisson relative error of each point on the graph is about @xmath80 . besides , figure [ fig : densvsrade5c ] presents , for the innermost region of model e5c , the logarithm of the density versus the logarithm of the radius for both the adopted model ( circles ) as for that same model evolved @xmath71 , or about a hubble time , ( plus signs ) and we note that the slope of the central cusp is very well conserved . similar results were obtained for the other models . table [ tab2 ] gives for each model the values of the major semiaxis and of the axial ratios for the @xmath81 , @xmath82 , ... and @xmath14 per cent most tightly bound particles . again , the results obtained changing the randomly generated inital conditions for a given model are essentially the same . as in our previous models , there is a general trend towards larger axial ratios at larger distances from the centre , but the trend is weaker for the more flatened models and breaks closer to the centre , probably due to the influence of the cusp . we used the results of the @xmath15 @xmath10 long final evolution to check the stability of the central density and the semiaxes of the stellar systems . for this purpose , we computed at @xmath14 @xmath10 intervals the central density from the @xmath6 particles closer to the centre of each system , and the moments of inertia of the @xmath9 per cent most tightly bound particles . these quantities changed almost linearly with time , so that we obtained their variations over a hubble time from the corresponding best fitting straight lines and the results are presented in table [ tab3 ] . @lcccc@ @xmath47 & @xmath83 & @xmath84 & @xmath85 & @xmath86 + e2a & -0.61 @xmath53 0.29 & -0.63 @xmath53 0.09 & 0.64 @xmath53 0.13 & 0.20 @xmath53 0.09 + e2b & -0.36 @xmath53 0.43 & -0.48 @xmath53 0.07 & 0.51 @xmath53 0.08 & 0.41 @xmath53 0.11 + e2c & -0.83 @xmath53 0.59 & -0.45 @xmath53 0.10 & 0.49 @xmath53 0.10 & 0.40 @xmath53 0.13 + + e3a & -0.76 @xmath53 0.53 & -1.01 @xmath53 0.05 & 1.36 @xmath53 0.11 & 0.78 @xmath53 0.10 + e3b & -0.62 @xmath53 0.15 & -1.08 @xmath53 0.07 & 1.44 @xmath53 0.09 & 0.86 @xmath53 0.15 + e3c & -0.56 @xmath53 0.18 & -1.04 @xmath53 0.11 & 1.57 @xmath53 0.16 & 0.87 @xmath53 0.11 + + e4a & -1.05 @xmath53 0.65 & -1.12 @xmath53 0.06 & 1.16 @xmath53 0.04 & 1.68 @xmath53 0.12 + e4b & -0.86 @xmath53 0.86 & -1.21 @xmath53 0.09 & 1.19 @xmath53 0.08 & 1.65 @xmath53 0.17 + e4c & -0.73 @xmath53 0.56 & -1.20 @xmath53 0.10 & 1.36 @xmath53 0.23 & 1.70 @xmath53 0.18 + + e5a & -1.07 @xmath53 0.78 & -2.17 @xmath53 0.11 & 2.12 @xmath53 0.14 & 2.65 @xmath53 0.21 + e5b & 0.92 @xmath53 0.83 & -1.96 @xmath53 0.13 & 2.00 @xmath53 0.08 & 1.95 @xmath53 0.08 + e5c & -1.23 @xmath53 0.17 & -2.02 @xmath53 0.09 & 1.73 @xmath53 0.21 & 2.31 @xmath53 0.06 + although most of the variations of the central density are not significant at the @xmath87 level it is suggestive that , except for that of model e5b ( also not significant ) , they are all negative . almost all the variations of the moments of inertia are highly significant and indicate a general decrease of the major axes as well as general increases of the intermediate and minor axes . but the changes are very small indeed , all of them being smaller than @xmath12 per cent over one hubble time . interestingly , both the amounts and the senses of the changes are similar to those found in our previous works @xcite , where we attributed them mainly to relaxation effects of the multipolar code @xcite . since in those works we had used the n body code of aguilar and we are now using that of hernquist and ostriker , we decided to perform the same checks we had made before using the models e5 , i.e. , those that exhibit the largest changes . the first check was to eliminate self - consistency , letting the systems evolve again for @xmath15 @xmath10 but keeping the coefficients of the expansion of the potential fixed at their initial values , and the results are shown in table [ tab4 ] . the central density changes , although somewhat lower than the corresponding values of table [ tab3 ] , are again not significant at the @xmath87 level , but the changes of the moments of inertia are much lower and generally significant at the same level . this is what could be expected from changes due to relaxation effects of the n body code , because they would be suppressed when turning off the self - consistency . alternatively , relaxation effects should increase when the number of bodies decreases , and that was our second check . we took at random @xmath65 per cent of the particles of each one of the models e5 , increased their masses 10 times , and we run these new models self - consistently first for @xmath88 @xmath10 to let them relax , and then for another @xmath15 @xmath10 to analyze their stability . the corresponding changes are shown in table [ tab5 ] and , except for those of the central density which remain not significant , they are substantially larger than the equivalente ones of table [ tab3 ] . thus , we may conclude that even the small variations shown in the latter table , are mainly due to relaxation effects of the n body code . [ cols="<,^,^,^,^",options="header " , ] we randomly selected between @xmath89 and @xmath90 particles from each model and adopted their positions and velocities as the initial values to obtain the orbits and investigate their chaoticity . the potentials were fixed , keeping constant the coefficients of their expansions at their final values , and the integrations were carried out in fixed coordinate systems , in those cases where the rotation velocity was not significant ( i.e. , less than three times the mean square error ) , or in systems rotating with the corresponding velocity in those cases where it was significant . we proceeded in this way to be consistent with the models obtained , but even the significant velocities are so small that their effect is in all likelihood negligible @xcite . since the potentials of our systems were fixed , all our orbits obey the energy integral ( or the jacobi integral in the case of systems with significant rotational velocity ) . therefore , regular orbits have to obey two additional isolating integrals , but we can have two kinds of chaotic orbits : partially chaotic orbits obey only one additional integral besides energy , and fully chaotic orbits have no isolating integrals other than energy . we have shown before ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * for example ) that partially and fully chaotic orbits have different distributions , so that it is important to separate them in orbital structure studies . a simple , albeit computing time demanding , way of classifying regular , partially and fully chaotic orbits is through the use of the six lyapunov exponents . since phase space volume is conserved , the exponents come in three pairs of the same absolute value and opposite sign . due to energy conservation , one of those pairs is always zero in our case , and each additional isolating integral makes zero another pair . thus , regular orbits have all their lyapunov exponents equal to zero , partially chaotic orbits have one non - zero pair and fully chaotic orbits have two . the numerical equivalent of the lyapunov exponents ( which demand to integrate the orbit over an infinite time interval ) are the finite time lyapunov characteristic numbers ( hereafter ft - lcns ) and , as in our previous works , we computed them using the liamag subroutine @xcite , kindly provided by d. pfenniger , and that we adapted with the assistance of h.d . navone to include the potential , accelerations and variational equations corresponding to the method of hernquist and ostriker . the liamag uses double precision arithmetic and a runge - kutta - fehlberg integrator of order 7 - 8 with variable time steps . as in our previous works , we adopted the integration and normalization intervals as @xmath91 and @xmath92 , respectively , and we requested a precision of @xmath93 for the step size regulation , which resulted in an energy conservation better than @xmath94 at the end of integration in all cases . we will refer to the largest ft - lcn of a given orbit as @xmath95 and the second largest one as @xmath96 , hereafter . since the ft - lcns are obtained from numerical integrations over a finite time interval ( i.e. , @xmath91 here ) , rather than the infinite one required to obtain lyapunov exponents , they can not reach zero value , but only a limiting minimum value , @xmath97 . applying the definition of the lyapunov exponents to equation ( 7 ) of @xcite , one can estimate that value as @xmath98 , where t is the integration interval ( cincotta , private communication ) , that is , a limiting value of about @xmath99 for our @xmath100 interval . this is an order of magnitude estimate only , however , and it is better to derive a more accurate value from the ft - lcns themselves using , for example , plots of the low end of the @xmath96 versus @xmath95 distribution ( i.e. , the region around that occupied by the representative points of the regular orbits ) . as an example we show in figure [ fig : lya5abc ] the corresponding plot for model e5c . the representative points of the regular orbits are concentrated on the blob at the left and those of chaotic orbits extend towards the right ( the sharp envelope of the identity line is merely due to the fact that , by definition , @xmath101 ) . clearly , any limit that attempts to separate regular from chaotic orbits can have a statistical value only , because some regular orbits may have ft - lcns slightly larger than that limit , while some chaotic orbits may have somewhat lower ft - lcns . with that caveat , using plots equivalent to that of figure [ fig : lya5abc ] for all our models we adopted @xmath102 . versus @xmath96 for model e5c . only the region of low values , around the region corresponding to regular orbits is shown.,width=317 ] now , that @xmath97 corresponds to a lyapunov time of @xmath103 , which is equivalent to about 5 or 6 hubble times for our models . is it reasonable to use such a low @xmath97 to separate regular from chaotic orbits ? would it not be more sensible to adopt a value of @xmath104 which corresponds to a lyapunov time of the same order of the hubble time ? those questions have been considered in our previous work ( see * ? ? ? * ; * ? ? ? * for example ) and we will repeat here the same analysis done there . first , we separated the orbits of each model into three groups : a ) those with @xmath105 , i.e. , those that are classified as regular for both choices of @xmath97 ( regreg , hereafter ) ; b ) those with @xmath106 , i.e. , those that are classified as regular for @xmath107 , but as chaotic for @xmath102 ( regchao , hereafter ) ; c ) those with @xmath108 , ie , those that are classified as chaotic for both elections of @xmath97 ( hereafter chaochao ) . then we considered , for each orbit , 11 @xmath109 orbital positions separated by intervals of @xmath110 , that is , over a total interval of @xmath111 , and , for each model and for each type of orbit , we computed the mean square value of each coordinate . table [ tab6 ] gives the square roots of the ratios of the @xmath112 and @xmath113 mean square values to the @xmath114 mean square value . @lcccc@ @xmath115 & @xmath116 & @xmath117 & @xmath118 & @xmath119 + y / x & e2a & 1.030 @xmath53 0.015 & 0.868 @xmath53 0.013 & 0.865 @xmath53 0.015 + @xmath78 & e2b & 1.023 @xmath53 0.014 & 0.868 @xmath53 0.013 & 0.875 @xmath53 0.015 + @xmath78 & e2c & 1.003 @xmath53 0.014 & 0.847 @xmath53 0.014 & 0.856 @xmath53 0.014 + + @xmath78 & e3a & 0.816 @xmath53 0.022 & 0.851 @xmath53 0.018 & 0.856 @xmath53 0.018 + @xmath78 & e3b & 0.815 @xmath53 0.023 & 0.824 @xmath53 0.019 & 0.829 @xmath53 0.017 + @xmath78 & e3c & 0.773 @xmath53 0.025 & 0.866 @xmath53 0.018 & 0.877 @xmath53 0.016 + + @xmath78 & e4a & 0.520 @xmath53 0.040 & 0.745 @xmath53 0.026 & 0.874 @xmath53 0.019 + @xmath78 & e4b & 0.510 @xmath53 0.044 & 0.788 @xmath53 0.028 & 0.841 @xmath53 0.019 + @xmath78 & e4c & 0.568 @xmath53 0.039 & 0.765 @xmath53 0.023 & 0.830 @xmath53 0.019 + + @xmath78 & e5a & 0.892 @xmath53 0.028 & 0.877 @xmath53 0.025 & 0.908 @xmath53 0.018 + @xmath78 & e5b & 0.965 @xmath53 0.026 & 0.833 @xmath53 0.025 & 0.899 @xmath53 0.018 + @xmath78 & e5c & 0.934 @xmath53 0.029 & 0.759 @xmath53 0.025 & 0.839 @xmath53 0.016 + z / x & e2a & 0.910 @xmath53 0.016 & 0.915 @xmath53 0.013 & 0.905 @xmath53 0.015 + @xmath78 & e2b & 0.937 @xmath53 0.015 & 0.903 @xmath53 0.013 & 0.874 @xmath53 0.015 + @xmath78 & e2c & 0.879 @xmath53 0.015 & 0.928 @xmath53 0.014 & 0.891 @xmath53 0.014 + + @xmath78 & e3a & 0.676 @xmath53 0.024 & 0.840 @xmath53 0.018 & 0.832 @xmath53 0.019 + @xmath78 & e3b & 0.712 @xmath53 0.027 & 0.812 @xmath53 0.018 & 0.771 @xmath53 0.018 + @xmath78 & e3c & 0.668 @xmath53 0.028 & 0.851 @xmath53 0.018 & 0.823 @xmath53 0.017 + + @xmath78 & e4a & 0.424 @xmath53 0.053 & 0.719 @xmath53 0.028 & 0.760 @xmath53 0.020 + @xmath78 & e4b & 0.353 @xmath53 0.044 & 0.660 @xmath53 0.028 & 0.767 @xmath53 0.019 + @xmath78 & e4c & 0.401 @xmath53 0.049 & 0.710 @xmath53 0.026 & 0.769 @xmath53 0.020 + + @xmath78 & e5a & 0.309 @xmath53 0.033 & 0.501 @xmath53 0.031 & 0.759 @xmath53 0.021 + @xmath78 & e5b & 0.361 @xmath53 0.040 & 0.527 @xmath53 0.035 & 0.699 @xmath53 0.020 + @xmath78 & e5c & 0.319 @xmath53 0.043 & 0.473 @xmath53 0.032 & 0.699 @xmath53 0.018 + the results in table [ tab6 ] show that , as in our previous work , most of the axial ratios of the regchao orbits are significantly different from those of the regreg orbits and we may conclude that , despite their low ft - lcn values implying lyapunov times longer than the hubble time , orbits with @xmath120 have a spatial distribution different from that of regular orbits . in other words , the time scale for the exponential divergence of orbits ( measured by the lyapunov time ) is not much relevant for the spatial distribution of those orbits . since we are here interested in that distribution , the sensible approach is thus to adopt @xmath121 . therefore , we classify orbits as regular if @xmath122 , as partially chaotic if @xmath123 and as fully chaotic if @xmath124 . table [ tab7 ] gives the percentages of regular , partially and fully chaotic orbits in our models . as in our previous works , the statistical errors have been estimated from the binomial distribution . the agreement among results of the a , b and c cases of each model is excellent and within the @xmath87 level in all cases , except for the regular and fully chaotic orbits of the e3c case where the differences with the e3a and e3b cases are of the order of @xmath125 . clearly , all our models harbor the lowest fractions of regular orbits in triaxial systems we know about , lower even than the @xmath126 per cent found by @xcite for their model e6c . interestingly , the fraction of regular orbits does not decrease monotonically from type e2 through type e5 , but falls first to rise again ; a somewhat similar , but less pronounced , trend was found in the e4 through e6 models of @xcite . the decreasing fractions of partially chaotic orbits when going from type e2 through e5 show the same trend found in those two previous works of ours . two further tests were performed to check the accuracy of our results . on the one hand , it is interesting to check the influence that the use of a particular n - body snapshot as a model might have . on the other hand , since we are using only about 0.5 per cent of the orbits in each model to investigate chaoticity , it might be worthwile to see if the use of a different sample yields different results . we adopted the e4b model for both tests . for the first one , we took a new model from the snapshot evolved @xmath127 more than the one whose results are presented in table [ tab7 ] , and we obtained fractions of @xmath128 , @xmath129 and @xmath130 , respectively for the regular , partially and fully chaotic orbits . the first two fractions coincide with those of the original model at the @xmath87 level , but the last one differs by @xmath131 . not only are these differences small , but they are probably at least partially due to the axial changes of the models as they evolve described in subsection 3.1 . the @xmath127 evolution changes the triaxiality of the e4b model in the direction of that of the e5 models by about @xmath132 of the triaxiality difference between the e4 and e5 models . since the difference between the fractions of fully chaotic orbits of those models is about @xmath133 we might crudely estimate that about @xmath134 of the change from the original model to the one evolved @xmath127 might be attributed to the axial changes , in which case the remaining difference is of @xmath135 only . nevertheless , the agreement between the original and the @xmath127 evolution results is somewhat poorer than those among statistically different realizations of the same model , except for the already mentioned differences for models e3 . the second test , instead , yielded no significant differences . the new sample of orbits from e4b model had fractions of @xmath136 , @xmath137 and @xmath138 , respectively of regular , partially and full chaotic orbits , all of them well within the @xmath87 level of differences from the original results . @lccc@ @xmath116 & @xmath139 & @xmath140 & @xmath141 + e2a & 22.48 @xmath53 0.59 & 15.30 @xmath53 0.51 & 62.22 @xmath53 0.69 + e2b & 21.35 @xmath53 0.58 & 15.52 @xmath53 0.51 & 63.13 @xmath53 0.69 + e2c & 22.24 @xmath53 0.59 & 13.64 @xmath53 0.49 & 64.12 @xmath53 0.68 + + e3a & 14.21 @xmath53 0.50 & 13.19 @xmath53 0.48 & 72.60 @xmath53 0.64 + e3b & 14.04 @xmath53 0.50 & 13.57 @xmath53 0.49 & 72.39 @xmath53 0.64 + e3c & 10.63 @xmath53 0.44 & 12.90 @xmath53 0.48 & 76.47 @xmath53 0.61 + + e4a & 13.05 @xmath53 0.50 & 10.85 @xmath53 0.46 & 76.10 @xmath53 0.63 + e4b & 12.72 @xmath53 0.50 & 10.58 @xmath53 0.46 & 76.70 @xmath53 0.63 + e5c & 12.67 @xmath53 0.49 & 10.11 @xmath53 0.45 & 77.22 @xmath53 0.62 + + e5a & 21.71 @xmath53 0.61 & 9.69 @xmath53 0.44 & 68.60 @xmath53 0.69 + e5b & 23.06 @xmath53 0.63 & 9.38 @xmath53 0.43 & 67.96 @xmath53 0.69 + e5c & 20.92 @xmath53 0.60 & 8.86 @xmath53 0.42 & 70.22 @xmath53 0.68 + the values of the ft - lcns are larger than those we found for non - cuspy models , in agreement with the results of @xcite and @xcite . figure [ fig : ervmaxlye5abc ] shows , for our model e5c , the plot of @xmath95 versus the reduced energy ( i.e. , the orbital energy , e , divided by the potential energy at the galactic centre , @xmath142 ) , which can be compared with fig . 3 of @xcite . in the present case the largest ft - lcns are close to @xmath143 , while for the non - cuspy model of @xcite they only reached about @xmath144 . besides , in the non - cuspy system there were no chaotic orbits for @xmath145 values close to @xmath146 ( i.e. , orbits very close to the centre of the system ) , because the potential near the centre was essentially that of a three dimensional harmonic oscillator . but , here , the presence of the cusp strongly limits the central region where the potential can be approximated with an integrable potential and , therefore , one can find chaotic orbits near the centre of the system . ) versus reduced orbital energy plot for model e5c , width=317 ] for each model we took eleven orbital positions at intervals of @xmath110 , i.e. , over a total interval of @xmath71 , to obtain the mean square values of each coordinate separately for each type of orbit . table [ tab8 ] gives the @xmath147 and @xmath148 axial ratios , computed from the square roots of those quadratic mean values . all the results for the a , b and c realizations of a same model show very good agreement at the @xmath87 level , except for the @xmath147 ratio difference between models e4b and e4c for the fully chaotic orbits , which is somewhat larger . the differences between the distributions of the partially and fully chaotic orbits are significant at the @xmath87 level only for @xmath148 ratio of the e5 models and the b case of models e4 . this result agrees with those obtained for the models we studied before @xcite in the sense that the differences between the distributions of partially and fully chaotic orbits , on the one hand , increased for more flattened systems and , on the other hand , diminished for cuspy systems . @lcccc@ @xmath115 & @xmath116 & @xmath149 & @xmath150 & @xmath151 + y / x & e2a & 1.030 @xmath53 0.015 & 0.903 @xmath53 0.016 & 0.828 @xmath53 0.014 + @xmath78 & e2b & 1.023 @xmath53 0.014 & 0.899 @xmath53 0.016 & 0.841 @xmath53 0.014 + @xmath78 & e2c & 1.003 @xmath53 0.014 & 0.891 @xmath53 0.016 & 0.813 @xmath53 0.014 + + @xmath78 & e3a & 0.816 @xmath53 0.022 & 0.878 @xmath53 0.020 & 0.800 @xmath53 0.024 + @xmath78 & e3b & 0.815 @xmath53 0.023 & 0.863 @xmath53 0.021 & 0.765 @xmath53 0.027 + @xmath78 & e3c & 0.773 @xmath53 0.025 & 0.915 @xmath53 0.020 & 0.799 @xmath53 0.025 + + @xmath78 & e4a & 0.520 @xmath53 0.040 & 0.804 @xmath53 0.032 & 0.712 @xmath53 0.032 + @xmath78 & e4b & 0.510 @xmath53 0.044 & 0.781 @xmath53 0.035 & 0.814 @xmath53 0.034 + @xmath78 & e4c & 0.568 @xmath53 0.039 & 0.894 @xmath53 0.030 & 0.652 @xmath53 0.029 + + @xmath78 & e5a & 0.892 @xmath53 0.028 & 0.869 @xmath53 0.029 & 0.903 @xmath53 0.018 + @xmath78 & e5b & 0.965 @xmath53 0.026 & 0.868 @xmath53 0.030 & 0.871 @xmath53 0.017 + @xmath78 & e5c & 0.934 @xmath53 0.029 & 0.790 @xmath53 0.029 & 0.813 @xmath53 0.016 + z / x & e2a & 0.910 @xmath53 0.016 & 0.905 @xmath53 0.016 & 0.919 @xmath53 0.013 + @xmath78 & e2b & 0.937 @xmath53 0.015 & 0.870 @xmath53 0.016 & 0.917 @xmath53 0.013 + @xmath78 & e2c & 0.879 @xmath53 0.015 & 0.930 @xmath53 0.016 & 0.903 @xmath53 0.014 + + @xmath78 & e3a & 0.676 @xmath53 0.024 & 0.825 @xmath53 0.021 & 0.867 @xmath53 0.024 + @xmath78 & e3b & 0.712 @xmath53 0.027 & 0.791 @xmath53 0.021 & 0.831 @xmath53 0.026 + @xmath78 & e3c & 0.668 @xmath53 0.028 & 0.845 @xmath53 0.022 & 0.850 @xmath53 0.025 + + @xmath78 & e4a & 0.424 @xmath53 0.053 & 0.648 @xmath53 0.037 & 0.798 @xmath53 0.034 + @xmath78 & e4b & 0.359 @xmath53 0.045 & 0.573 @xmath53 0.037 & 0.789 @xmath53 0.033 + @xmath78 & e4c & 0.401 @xmath53 0.049 & 0.699 @xmath53 0.036 & 0.730 @xmath53 0.031 + + @xmath78 & e5a & 0.309 @xmath53 0.033 & 0.374 @xmath53 0.033 & 0.733 @xmath53 0.020 + @xmath78 & e5b & 0.361 @xmath53 0.040 & 0.484 @xmath53 0.045 & 0.675 @xmath53 0.019 + @xmath78 & e5c & 0.319 @xmath53 0.043 & 0.398 @xmath53 0.038 & 0.671 @xmath53 0.018 + the method of @xcite is clearly better than that of aguilar @xcite to build cuspy triaxial stellar models , since the former needs no softening and uses a radial expansion which is particularly adequate to fit n body distributions with @xmath0 . as a side benefit , the coefficients of both the radial and angular expansions are provided by the method itself and there is no need of additional fittings to obtain them . the only caveat is that , as shown by @xcite , the method of hernquist and ostriker is adequate for cases with @xmath0 only and , for other slopes of the cusp , other related but different methods should be preferred . the models obtained using the same parameters and changing only the seed number of the random number generator show that our results are very robust , indeed . those models exhibit essentially the same global properties ( except for the angular velocity of figure rotation ) and , with very few exceptions , the same fractions and distributions of regular , partially and fully chaotic orbits . the different angular velocities we obtained here for statistically equivalent models further complicates the little we know about the phenomenon of figure rotation ( i.e. , with zero angular momentum ) in this kind of models . on the one hand , the reality of the rotation has been checked using both the aguilar @xcite and aarseth @xcite codes by @xcite and here with the hernquist code and , on the other hand , such triaxial systems with figure rotation are just the stellar equivalent of the riemann ellipsoids of fluid dynamics , so that there is nothing misterious about their existence . nevertheless , except for the increase of the angular velocity when one goes to flatter systems , the rotation does not seem to correlate with other properties of the systems and , moreover , now it turns out that statistically equivalent systems have different velocities . even though , as already indicated , the angular velocities are so low that they have very little effect on orbital chaoticity , this phenomenon is interesting and warrants further investigation . the high stability of our models over time scales of the order of a hubble time is very well stablished . not only are the detected variations very small , but they are most likely due to relaxation effects of the n body code . besides , it should be stressed that we have checked the stability running the models self - consistently , while checks on the stability of models resulting from the method of @xcite are usually done on fixed potentials @xcite . as we have shown , when the potential is fixed , the changes in our models are an order of magnitude smaller than those found self - consistently . although , considering the results of @xcite , @xcite and @xcite , high fractions of chaotic orbits could be expected in our cuspy triaxial models , the fractions found are extremely high , indeed , exceeding @xmath152 per cent for our e2 and e3 models . it is important to emphasize that those high fractions of chaotic orbits are not merely due to the low @xmath97 adopted : had we adopted the @xmath153 limit , those models would still have had between @xmath154 an @xmath155 per cent of chaotic orbits . the main conclusion of our investigation is that it is perfectly possible to build self - consistent stellar models of cuspy triaxial elliptical galaxies that are very stable despite containing high percentages of chaotic orbits . this result confirms and extends previous work by others @xcite and ours @xcite . it also supports the suggestion by @xcite in the sense that the difficulties to obtain such models with the method of schwarzschild should be attributed to the method itself and not to physical causes . it is true that , while regular orbits always occupy the same region of space , chaotic orbits may spend a long interval occupying certain region , only to switch to a different one later on , with sticky orbits being an extreme example . but , although such behaviour conspires against obtaining stable models with high fractions of chaotic orbits with the method of schwarzschild , it does not necessarily prevents the existence of those models . all one needs is that , as some chaotic orbits abandon a region of space to explore a different one , orbits in the latter region move on to replace those in the former . in other words , rather than having a static equilibrium , with each orbit covering always the same zone , we have a dynamic equilibrium where switches from one zone to another are present but balance each other on average . our models show that such state of affairs is perfectly possible . we are very grateful to lars hernquist and daniel pfenniger for allowing us to use their codes , to c. efthymiopoulos for useful comments , to h.d . navone for his assistance with programming , to r.e . martnez and h.r viturro for their technical assistance and to an anonymous referee for his comments on the original version of the present paper . this work was supported with grants from the consejo nacional de investigaciones cientficas y tcnicas de la repblica argentina , the agencia nacional de promocin cientfica y tecnolgica , the universidad nacional de la plata and the universidad nacional de rosario . 99 aarseth s.j . , 2003 , gravitational n body simulations , cambridge university press , cambridge aguilar l.a . , merritt d. , 1990 , apj , 354 , 33 aquilano r.o . , muzzio j.c . , navone h.d . , . , 2007 , celestial mech . dyn . astr . , 99 , 307 binney j. , tremaine s.j . , 2008 , galactic dynamics , princeton university press capuzzo dolcetta r. , leccese l. , merritt d. , vicari a. , 2007 , apj , 666 , 165 cincotta p.m. , giordano c.m . , sim c. , 2003 , physica d , 182 , 151 contopoulos g. , 2004 , order and chaos in dynamical astronomy , springer dehnen w. , 1993 , mnras , 265 , 250 forbes d.a . , ponman t.j . , 1999 , mnras , 309 , 623 hernquist l. , 1990 , apj , 356 , 359 hernquist l. , barnes j. , 1990 , apj , 349 , 562 hernquist l. , ostriker j.p . , 1992 , apj , 386 , 375 holley bockelmann k. , mihos j.c . , sigurdsson s. , hernquist l. , 2001 , apj , 549 , 862 jesseit r. , naab t. , burkert a. , 2005 , mnras , 360 , 1185 kalapotharakos c. , efthymiopoulos c. , voglis n. , 2008 , mnras , 383 , 971 kalapotharakos c. , voglis n. , 2005 , celestial mech . dyn . astr . , 92(13 ) , 157 kandrup h.e . , sideris i.v . , 2002 , celestial mech . 82 kandrup h.e . , siopis c. , 2003,mnras , 345 , 727 merritt d. , fridman t. , 1996 , apj , 460(1 ) , 136 muzzio j.c . , 2006 , celestial mech . astr . , 96(2 ) , 85 muzzio j.c . , carpintero d.d . , wachlin f.c . , 2005 , celestial mech . , 91(1 - 2 ) , 173 muzzio j.c . , navone h.d . , zorzi a.f . , 2009 , celestial mech . astr . , 105(4 ) , 379 napolitano n.r . et al . , 2005 , mnras , 357 , 691 schwarzschild m. , 1979 , apj , 232 , 236 schwarzschild m. , 1993 , apj , 409 , 563 siopis c. , kandrup h.e . , 2000 , mnras , 319 , 43 sparke l.s . , sellwood j.a . , 1987 , mnras , 225 , 653 udry s. , pfenniger d. , 1988 , a&a , 198(12 ) , 135 voglis n. , kalapotharakos c. , stavropoulos i. , 2002 , mnras , 337(2 ) , 619 white s.d.m . , 1983 , apj , 274 , 53 zorzi a. f. , muzzio j. c. , 2009 , memorias del ii congreso de matemtica aplicada , computacional e industrial , rosario ( 2009 )

we discuss theoretically quantum interface between light and a spin polarized ensemble of atoms with the spin based on an off - resonant raman scattering . we present the spectral theory of the light - atoms interaction and show how particular spectral modes of quantum light couple to spatial modes of the extended atomic ensemble . we show how this interaction can be used for quantum memory storage and retrieval and for deterministic entanglement protocols . the proposed protocols are attractive due to their simplicity since they involve just a single pass of light through atoms without the need for elaborate pulse shaping or quantum feedback . as a practically relevant example we consider the interaction of a light pulse with hyperfine components of line ofrb . the quality of the proposed protocols is verified via analytical and numerical analysis .

quant-ph0611228

light - matter quantum interface is a basic element of any quantum information network aiming at long distance quantum communication , cryptography protocols or quantum computing , see @xcite . light is a natural carrier of quantum information and a macroscopic atomic system can be efficiently used for its storage . critical ingredients of the quantum interface are the quantum memory and entanglement protocols , which allow high - fidelity interchange ( transfer , storage and readout ) of quantum states between the light and relatively long - lived atomic subsystems . a number of promising theoretical proposals for the high - fidelity memory and entanglement protocols has been put forward , which can be classified as based on off - resonant interaction , such as raman interaction @xcite , quantum non - demolition ( qnd ) interaction with quantum feedback - qndf @xcite , multiple qnd interactions ref.@xcite , and on resonant interaction using electromagnetically induced transparency - eit @xcite . in spite of a large number of proposals , experimental realization of the complete storage plus retrieval quantum memory with the fidelity higher than classical is yet to be achieved . high fidelity storage ( but not complete retrieval ) has been demonstrated via the qndf approach @xcite and as light - to - atoms teleportation @xcite . low fidelity storage and retrieval of the coherent and single photon pulses based on eit @xcite and raman @xcite processes have been recently demonstrated . all experiments on light - atoms quantum interface to date are conducted with alkali atoms . previously the off - resonant quantum interface protocols @xcite utilized light with the detuning higher than the hyperfine splitting of the excited state . under this condition magnetic sublevels of a hyperfine state of an alkali atom can be effectively reduced to a spin - half system . as shown in the present paper eliminating this restriction and thus using the complete magnetic multipole system of the alkali atom ground state opens up new possibilities for quantum memory and entanglement generation with ensembles of such atoms . the quantum description of correlations in coherent raman scattering in the heisenberg formalism has a long history @xcite . however in quantum information applications the raman process is often discussed in terms of formal @xmath3-scheme configuration . in contrast in our treatment we describe this process in terms of polarization multipoles for multilevel ground state alkali atoms : gyrotropy ( orientation ) , linear birefringence ( alignment ) and higher multipole components . the field subsystem is described by a set of polarization stokes components . this approach allows for complete description of the mean values and fluctuations of the light and optically thick atomic medium . although examples treated in this paper concern atoms initially pumped to one level corresponding to a single @xmath3-scheme ( as shown in figures [ fig.1]and [ fig.2 ] ) , our formalism also allows to treat more complex initial states consisting of several coupled @xmath3-schemes . the formalism is also applicable beyond the heisenberg equations in a linearized form . finally , the multipole expansion facilitates systematic and compact calculation of the coupling constants in the effective hamiltonian taking into account the full hyperfine structure of the relevant atomic levels @xcite . we shall discuss the raman quantum memory scheme first considered in @xcite . we shall add the retrieval step to the protocol and analyze the complete procedure for a realistic model , taking full account of the multilevel hyperfine and magnetic sublevel structure of an alkali atom . we derive the polarization - sensitive coupled wave - type dynamics of an optically thick atomic ensemble and optical field . using spectral mode decomposition for light and spatial mode picture for the atomic ensemble we show that the quantum states of light and the quasi - spin of atoms can be effectively swapped or entangled . the general mathematical formalism for such atomic system with angular momentum higher or equal than one , has been developed in refs.@xcite . an important advantage of the memory and entanglement schemes in such a scenario is in that they can be realized in a single pass of light through atoms and without any feedback channel , which makes experimental realizations more feasible . as an elementary carrier of the quantum information we consider a squeezed state of light . the relevance of quantum memory for these states for quantum information processing has been illustrated in proposals @xcite . we introduce stokes operators for a light pulse consisting of a circularly polarized strong classical mode and the quantum squeezed light in the orthogonal circularly polarized mode . we show that for the quantum stokes variables there is a convenient symmetric interaction with the alignment tensor components of atoms , whose spin angular momenta are equal or greater than one . in the quantum memory scenario the quantum information of the light subsystem can be effectively mapped into the alignment subsystem of atoms . in the entanglement scenario the excitation of atomic spins with coherent pulse generates a parametric - type interaction process , which results in creation of the strong correlations between the quantum fluctuations of the stokes components of the transmitted light and the alignment components in the spin subsystem of atoms . we consider the interaction via the @xmath1 line of @xmath2rb , as an important example , where a convenient spin - one system exists in the lower hyperfine sublevel of its ground state . for verification of the memory protocol we discuss its figure of merit for the mapping of the input squeezed state and for its retrieval after reading the quantum copy out from the spin subsystem with a second coherent light pulse . we discuss how the fidelity for the proposed quantum memory channel could be defined and compared with the respective classical benchmark based on direct measurement of the squeezed state parameters . we show that in an optimal configuration the quantum fidelity is always higher than the limit for the optimal competing classical channel . throughout our analysis we neglect coupling of the ground state atomic degrees of freedom to other variables than the forward propagating light field . this approximation is justified on time scales which ar short compared to dephasing time of the ground state coherence . neglecting coupling to light modes propagating in other than the forward direction can only be a good approximation in the limit of high optical depth as detailed in section [ iiia ] . finally , the influence of the atomic motion capable of washing out spatial spin patterns is not taken into account , which restricts our analysis to cold atomic samples . in this paper we will consider two alternative experimental situations respectively shown in figures [ fig.1 ] and [ fig.2 ] . for both schemes a 100% right - hand circular polarized classical light pulse interacts with an ensemble of ultracold spin - oriented atoms . the principal difference is in the direction of the collective spin of the atomic ensemble , which can be oriented either along ( fig.[fig.1 ] ) or opposite ( fig.[fig.2 ] ) to the propagation direction of light . consider first the process shown in figure [ fig.1 ] . for an off - resonant right - hand polarized pulse , such that incoherent scattering losses do not frustrate the spin polarization , there will be no interaction between the light and atoms . however if a small portion of a left - hand polarized quantum informative light prepared in an unknown squeezed vacuum state is admixed to the classical coherent pulse , the coherent scattering channel will be open . the portion of the weak quantum light will be coherently scattered into the strong classical mode . in this coherent raman process the polarization quantum subsystem of the probe light and the spin subsystem of atoms can effectively swap their quantum states . the quantum state can be mapped into the alignment - type fluctuations of the spin subsystem and further stored in the form of a certain standing spin wave for relatively long time . it can be readout on demand with a second probe light pulse . in the experimental situation shown in figure [ fig.2 ] , for a transparent medium a small portion of the left - hand polarized photons will emerge as a result of the coherent elastic raman scattering of the strong light pulse . this process creates entanglement the left - hand polarized quantum modes of light and the alignment - type quantum fluctuations in the atomic spin subsystem . this kind of deterministic entanglement can be interesting for implementing a quantum repeater protocol between remote atomic systems . the processes shown in figures [ fig.1 ] and [ fig.2 ] have practically identical theoretical description . thus we will discuss them in parallel paying attention to the differences in their physical nature . we begin by introducing the relevant set of the field and atomic variables involved in these processes . the polarization states of the light subsystem can be specified in terms of the following flux - type variables for different stokes components . the heisenberg operators for the total photon flux considered at the spatial point @xmath4 and at the time moment @xmath5 are given by @xmath6 where @xmath7 are the positive / negative frequency components of the electric field heiseberg operators . we assume the quasi - monochromatic and forward propagating probe light such that the spectral envelope of the modes has a carrier frequency @xmath8 , and the light beam has a cross area @xmath9 . the stokes components responsible for three alternative polarization types are given by @xmath10% \nonumber\\% \hat{\xi}_2(z , t)&=&% \frac{s_0\,c}{2\pi\hbar\,\bar{\omega}}\,% \left[\hat{e}_r^{(-)}(z , t)\,\hat{e}_r^{(+)}(z , t)\right.% \nonumber\\% & & \phantom{\frac{s_0\,c}{2\pi\hbar\,\bar{\omega}}\,}% \left.-\,\hat{e}_l^{(-)}(z , t)\,\hat{e}_l^{(+)}(z , t)\right]% \nonumber\\% \hat{\xi}_3(z , t)&=&% \frac{s_0\,c}{2\pi\hbar\,\bar{\omega}}\,% \left[\hat{e}_x^{(-)}(z , t)\,\hat{e}_x^{(+)}(z , t)\right.% \nonumber\\% & & \phantom{\frac{s_0\,c}{2\pi\hbar\,\bar{\omega}}\,}% \left.-\,\hat{e}_y^{(-)}(z , t)\,\hat{e}_y^{(+)}(z , t)\right]% \label{2.2}%\end{aligned}\]]these components subsequently define imbalance in the photon fluxes with respect to the @xmath11 cartesian basis ( @xmath12 ) , to the @xmath13 cartesian basis rotated at @xmath14-angle ( @xmath15 ) , and to the @xmath16 basis of circular polarizations ( @xmath17 ) , see @xcite . the stokes variables ( [ 2.2 ] ) obey the following commutation relations @xmath18\;=\;% 2i\varepsilon_{\mathrm ijk}\,c\,\delta(z - z')\hat{\xi}_k(z , t)% \label{2.3}%\ ] ] where @xmath19 depending on the order of the indices @xmath20 . the @xmath21-function on the right - hand side of this commutation relation is an approximation valid if only a finite bandwidth of the field continuum is important for the correct description of the low frequency fluctuation spectrum . this assumption is in accord with the rotating wave approximation which we will further use in the description of the atom - field interaction . for the processes shown in figures [ fig.1 ] and [ fig.2 ] the commutation relations ( [ 2.3 ] ) can be simplified to @xmath22\;=\;% 2i\,c\,\delta(z - z')\bar{\xi}_2% \label{2.4}%\ ] ] where for the case of small fluctuations the operator on the right - hand side is replaced by its expectation value . this stokes component is approximately conserved in the linear evolution , and because of the conservation law for the number of photons in the coherent process one has @xmath23 . the component @xmath24 is the integral of motion in our model and @xmath25 , where @xmath26 is the interaction time , gives the average number of photons participating in the process . the angular momentum polarization of @xmath27-th atom can be conveniently described in the formalism of the irreducible tensor operators @xmath28with these expansions the set of atomic dyadic operators originally defined in the subspace of the zeeman states @xmath29 , where @xmath30 is the total ( spin + nuclear ) angular momentum of atom and @xmath31 is its zeeman projection , is transformed into the set of the relevant irreducible tensor operators . here @xmath32 are the clebsh - gordan coefficients , with the rank and the projection @xmath33 which respectively vary in the intervals : @xmath34 and @xmath35 , see ref.@xcite . the heisenberg dynamics of the irreducible components ( [ 2.5 ] ) preserves the following commutation relations @xmath36&=&% \delta_{ab}\,[(2k+1)(2k'+1)]^{1/2}\,% \nonumber\\% & & \hspace{-0.8in}\times \sum_{k''}\;[1-(-1)^{k+k'+k''}]\,% \left\{\begin{array}{ccc}% k & k ' & k''\\ f_0 & f_0 & f_0% \end{array}\right\}\,% \nonumber\\% & & \hspace{-0.8in}\times\,(-1)^{2f_0+k''}\,c^{k''q''}_{kq\ , k'q'}\,% \hat{t}_{k''q''}^{(a)}(t)% \label{2.6}%\end{aligned}\]]for the interaction processes shown in figures [ fig.1 ] and [ fig.2 ] only the following two combinations of the alignment operators and the longitudinal orientation component contribute to the dynamics of the atomic spin fluctuations @xmath37% \nonumber\\% \hat{t}_{\xi\eta}^{(a)}(t)&=&% -\frac{1}{2i}\left[\hat{t}_{2 - 2}^{(a)}(t)\,-\,% \hat{t}_{22}^{(a)}(t)\right]% \nonumber\\% \hat{f}_z^{(a)}(t)&=&\frac{1}{\sqrt{3}}\left[f_0(f_0 + 1)(2f_0 + 1)\right]^{1/2}\hat{t}_{10}^{(a)}(t)% \label{2.7}%\end{aligned}\]]where @xmath38 is the heisenberg operator of the angular momentum projection for @xmath27-th atom on @xmath4-axis . for this set of operators the commutation relation ( [ 2.6 ] ) \delta_{ab}\,ic_{1}\,\hat{f}_{z}^{(a)}(t)+\delta_{ab}\,ic_{3}\hat{t}_{30}^{(a)}(t)% \nonumber\\% \left[\hat{f}_{z}^{(a)}(t ) , \hat{t}_{xy}^{(b)}(t)\right]&=&% \delta_{ab}\,2i\hat{t}_{\xi\eta}^{(a)}(t)% \nonumber\\% \left[\hat{f}_{z}^{(a)}(t ) , \hat{t}_{\xi\eta}^{(b)}(t)\right]&=&% -\,\delta_{ab}\,2i\hat{t}_{xy}^{(a)}(t)% \label{2.8}%\end{aligned}\]]the coefficients @xmath40 and @xmath41 are given by @xmath42^{1/2}}% { \left[7f_0(f_0 + 1)(2f_0 - 1)(2f_0 + 1)(2f_0 + 3)\right]^{1/2}}% \nonumber\\% \label{2.8'}%\end{aligned}\]]and the higher - rank irreducible component @xmath43 weighted with coefficient @xmath41 contributes only for atoms with @xmath44 . the operators @xmath45 and @xmath46 have clear physical meaning and define the components of alignment tensor with respect to either @xmath47 or @xmath48 cartesian frames , which were earlier introduced with definition of the polarization stokes components of light . these operators can also be related with the transverse components of a quasi - spin , which could be defined in the system of two zeeman states coupled via the @xmath3-type excitation channels shown in figures [ fig.1 ] and [ fig.2 ] . however in context of our discussion it is more important to emphasize the tensor nature of these components , which are responsible for the fluctuations of linear birefringence of the sample initiated by either stimulated or spontaneous raman processes shown in these figures . the important step in further description of the collective dynamics of the atomic spins consists in assumption that the original discrete distribution of the point - like atomic spins can be smoothed with procedure of the mesoscopic averaging . then for a mesoscopically thin layer located between @xmath4 and @xmath49 one can define @xmath50where the sum is extended on the atoms located inside the layer . these expressions define the smoothed spatial distribution of the alignment components for which the first line of commutation relation ( [ 2.8 ] ) transforms to @xmath51=% \delta(z - z')\,i\bar{c}_{13}\,\bar{\mathcal f}_{z}}% \nonumber\\% & & \bar{c}_{13}=\frac{15}{f_0(f_0 + 1)(2f_0 + 1)(2f_0 + 3)}% \label{2.10}%\end{aligned}\]]where in the case of small fluctuation the right - hand side is replaced by an expectation value for the density of atomic angular momentum @xmath52 . as in the previous section we assume that this projection of the collective spin is approximately conserved in the raman process such that the total angular momentum @xmath53 , where @xmath54 is the sample length and @xmath55 is the number of atoms . since the incoherent losses are ignored @xmath55 remains the integral of motion in this model . the master equation governing the dynamics of atoms - field variables in the heisenberg picture can be derived similarly to the way it was done in ref . we apply the effective hamiltonian derived in that paper and neglect the dissipation channels caused by incoherent scattering . we will assume that the squeezed probe radiation for the excitation scheme , shown in figure [ fig.1 ] , is characterized by a degeneracy parameter ( i.e. by a mean number of photons in the coherence volume of the quantum radiation ) higher than unity . as was shown in ref.@xcite there is a special scenario when the @xmath3-system is probed by extremely weak spdc light source consisting of strongly correlated photon pairs , which should be cooperatively scattered by all atoms of the ensemble . this process can not be described by the hamiltonian of ref.@xcite and should be discussed separately . omitting the derivation details we present the following heisenberg wave - type equations describing the temporal and spatial dynamics of the stokes components and the alignment components @xmath56% \hat{\xi}_1(z , t)&=&% \kappa_1\,\hat{\xi}_3(z , t)\;-\;% 2\epsilon\,\bar{\xi}_2\,\hat{\mathcal t}_{xy}(z , t)% \nonumber\\% \left[\frac{\partial}{\partial z}+\frac{1}{c}% \frac{\partial}{\partial t}\right]% \hat{\xi}_3(z , t)&=&% -\kappa_1\,\hat{\xi}_1(z , t)\;+\;% 2\epsilon\bar{\xi}_2% \hat{\mathcal t}_{\xi\eta}(z , t)\,% \nonumber\\% \frac{\partial}{\partial t}\hat{\mathcal t}_{xy}(z , t)&=&% -\bar{\omega}\hat{\mathcal t}_{\xi\eta}(z , t)\;+\;% \bar{c}_{13}\epsilon\,\bar{\mathcal f}_{z}\,% \hat{\xi}_1(z , t)\,% \nonumber\\% \frac{\partial}{\partial t}\hat{\mathcal t}_{\xi\eta}(z , t)&=&% \phantom{+}\bar{\omega}\,\hat{\mathcal t}_{xy}(z , t)\;-\;% \bar{c}_{13}\epsilon\,\bar{\mathcal f}_{z}\,% \hat{\xi}_3(z , t)\,% \nonumber\\% \label{2.11}%\end{aligned}\ ] ] these equations are written for the physical obsrevables and have a rather clear physical structure . they show that in a linearized regime the modification of the light and atomic polarization in a forward passage results in a combine action of gyrotropy and linear birefringence . in the discussed case the former manifests itself as an averaged interaction between atoms and light and the latter actually creates the quantum fluctuation interface between these subsystems . let us point out that in a formal description based on a @xmath3-type configuration the fluctuations of birefringence ( alignment ) components are usually treated as an atomic coherence created by raman process in a quasi - spin subsystem and the importance of the phase - matching gyrotropy can be underestimated . we show how this effect can be properly taken into consideration by direct solution of eqs.([2.11 ] ) . equations ( [ 2.11 ] ) introduce the following important parameters : @xmath57 , @xmath58 and @xmath59 . the first two describe the changes in the mean classical parameters of light and atoms . @xmath57 describes the gyrotropy of the sample caused by the average angular momentum orientation of the atomic spins . @xmath60 is the light shift between the zeeman sublevels @xmath61 and @xmath62 . it is additionally assumed that the atoms interact with an external magnetic field and @xmath63 is the relevant zeeman splitting for the neighboring sublevels . the third parameter @xmath59 is the most important and is responsible for the coupled dynamics of the fluctuations of the polarization components of light and of the atomic alignment components . the microscopic expressions for all these characteristics for the case of hyperfine transitions of alkali atoms are summarized in appendix [ a ] before we discuss the solutions of these equations we turn again to the physics of the processes shown in figures [ fig.1 ] and [ fig.2 ] . the difference between these processes is only indicated by different signs of @xmath52 in the second pair of equations ( [ 2.11 ] ) . for positive @xmath64 ( fig.[fig.1 ] ) these equations yield the normal wave - type solution leading to the swapping of the quantum states between the light and spin subsystems . then the memory protocol can develop in the following way . if the quantum field is prepared in an unknown left - hand polarized squeezed - vacuum state its momentum / position quadrature components can be expressed as @xmath65 where the phase @xmath66 determining the orientation of the squeezed and anti - squeezed quadratures in the phase plane as well as their variances are the unknown parameters of the state . if such an unknown squeezed state is superimposed with a classical coherent mode one straightforwardly gets that @xmath67the crucial point for the atoms - field dynamics described by equations ( [ 2.11 ] ) and for the memory protocol is that the direction of the axes @xmath47 and @xmath48 remains _ _ c__ompletely unknown in the experiment , see figure [ fig.1 ] . thus the unknown squeezed state transforms into the unknown polarization squeezing such that the @xmath15 component is squeezed and @xmath12 is anti - squeezed . the dynamics of these components is further developing in accordance with equations ( [ 2.11 ] ) and under certain conditions the squeezed state of light can be converted into the squeezed state of the alignment components of the atomic spin angular momenta . for the negative sign of @xmath52 ( fig.[fig.2 ] ) there is no normal wave solution of the equations ( [ 2.11 ] ) . as we will show in this case there is an exponential enhancement of the quantum fluctuations of the stokes components and of the atomic alignment . such an anomalous spin polariton wave describes an entangled state of these subsystems . after interaction the low frequency modes of the polarization and intensity spectrum of the outgoing light will be entangled with the alignment - type standing wave modes of the atomic spins . such process can be for example utilized in the quantum repeater protocol with remote atomic systems . the heisenberg equations ( [ 2.11 ] ) can be solved via the method of laplace transform similar to how it was done in ref.@xcite . the solution and verification of the commutation relations are given in appendix [ b ] . in this section we discuss the physical consequences which are mainly important from the quantum information point of view . for the sake of simplicity and without loss of generality , see ref.@xcite , we will ignore the retardation effects and restrict our discussion to the low frequency fluctuations such that in equations ( [ 2.11 ] ) we neglect time derivatives of the field variables . we also restrict our consideration to the special case when the relevant pair of zeeman sublevels , which is either @xmath68 and @xmath69 ( fig.[fig.1 ] ) or @xmath70 and @xmath71 ( fig.[fig.2 ] ) , is always degenerate such that @xmath72 during the interaction and @xmath73 without the interaction . this non - critical but convenient simplification allows us to consider storage of quantum states of light in the time independent standing spin wave without any zeeman - type oscillations , see appendix [ b ] . the qualitative character of the cooperative dynamics of the atomic and field variables can be illustrated by the dispersion relation ( [ b.6 ] ) for the spin polariton modes and light modes defined in the semi - infinite medium and for infinite interaction time . then temporal and spatial dynamics of the process can be expressed by the respective laplace modes @xmath74 and @xmath75 introduced by expansion ( [ b.4 ] ) . the laplace variables for the inverse transform can be parameterized as @xmath76 . then the inverse transform reveals a spectral fourier expansion over the set of relevant temporal and spatial modes . the wave - type dynamics makes these modes coupled via the following dispersion law @xmath77 this dispersion relation reflects the coupling of low frequency temporal fluctuations of light to fine scale spatial fluctuations of atoms and of fast temporal fluctuations of light to long scale fluctuations of atoms respectively . for an atomic medium of the length @xmath54 and a light pulse of the duration @xmath26 , the combination @xmath78 , characterizing the overall coupling strength between light pulse and atomic ensemble , quantifies this scale relation in natural units . in the following we apply the input / output relations derived in appendix [ b ] to identify the suitable parameters for writing and retrieving to and from a quantum memory . for similar systems the optimization of a quantum memory has been analyzed very thoroughly in the single and few - mode situations @xcite , while here we are interested in a multi - mode situation considering pulses of squeezed light of a duration much longer than the inverse bandwidth of squeezing . prior studies of squeezed state storage in @xcite considered a steady state regime , while here we discuss a temporal sequence of writing , storage and retrieval . consider the experimental scheme shown in figure [ fig.1 ] for a pulse - type excitation of the system with a portion of quantum light with the following correlation properties @xmath79where @xmath80 denotes the anti - commutator of two observables . these correlation functions describe the output generated by the intra - cavity subthreshold degenerate parametric amplifier , see ref.@xcite . the freely propagating light with such correlation properties can be associated with a squeezed state of a harmonic oscillator where @xmath81 component is squeezed ( @xmath82 ) and @xmath83 component is anti - squeezed ( @xmath84 ) such that @xmath85 . the spectral bandwidth of the informative part of the quantum light is limited by the longest of two correlation times @xmath86^{-1}% \nonumber\\% \tau_3&=&\left[\frac{\gamma_c}{2}-\kappa_d\right]^{-1}% \label{3.4}%\end{aligned}\ ] ] where @xmath87 is the cavity loss rate through the output mirror and @xmath88 is the efficiency of the downconversion process . for high level of squeezing one has @xmath89 . this inequality indicates that the minimal duration @xmath26 of the quantum light pulse should be considerably longer than the longest time @xmath90 . the general solution ( [ b.8 ] ) , given in the appendix [ b ] , can be straightforwardly rewritten in the following form @xmath91^{1/2}\right)\,\hat{\xi}_1^{\mathrm{in}}(t)+\ldots% \nonumber\\% \nonumber\\% \lefteqn{\hat{{\mathcal t}}_{\mathrm{i\!i\!i}}^{\mathrm{out}}(z)\equiv% \cos\kappa_1z\,\hat{{\mathcal t}}_{\xi\eta}(z , t)-\sin\kappa_1z\,\hat{{\mathcal t}}_{xy}(z , t)=}% \nonumber\\% & & -c_{13}\epsilon\bar{\cal f}_z\int_0^t\!dt\,% j_0\!\left(2[-a(t - t)z]^{1/2}\right)\,\hat{\xi}_3^{\mathrm{in}}(t)+\ldots% \nonumber\\% \label{3.5}%\end{aligned}\ ] ] which shows how the input field operators are transferred into space - dependent atomic spin operators . the dots in the right hand side stand for the contributions of the atomic operators responsible for reproduction of input atomic state , which make this transformation imperfect . the principle condition , which makes the memory protocol feasible , is that for the low - frequency spatial fluctuations on the left hand side of eqs.([3.5 ] ) the contribution of the input spin fluctuations on the right hand side is suppressed . in the spectral expansion of the operators @xmath92 and @xmath93 such fluctuations belong to the spectral domain @xmath94^{1/2}$ ] . in accordance with the dispersion law ( [ 3.1 ] ) , which is asymptotically valid for @xmath95 , the transform ( [ 3.5 ] ) performs mapping of the field fluctuations within the spectral interval @xmath96 onto the spatially dependent spin fluctuations within the spatial spectral domain @xmath97 . for reliable mapping of the field quantum state onto the atomic alignment subsystem the latter domain should essentially overlap with the spectral domain where the input spin fluctuations are suppressed . for the efficient memory protocol the following inequalities should be fulfilled @xmath98 these inequalities imply that the contribution of the input spin fluctuations are suppressed and ensure that the spatial fluctuations of @xmath92 and @xmath93 correctly reproduce the temporal dynamics of the input field fluctuations ( [ 3.3 ] ) in the relevant part of their spectrum . however the dispersion relation ( [ 3.1 ] ) has a singular behavior for the function @xmath99 near the point @xmath100 such that the shot noise part of the input fluctuation spectrum is reproduced at the zero point of spatial spectrum . thus near the spectral point @xmath101 , associated with an integral collective mode of the standing spin waves in the limit @xmath102 , the informative part of the fluctuation spectrum will be invisible . however , for the sample with a finite length @xmath54 the squeezed state can be mapped onto the integral collective modes of @xmath92 and @xmath93 if @xmath103 i.e. for a spectrally broad incoming quantum light . for retrieval of the quantum state back onto light the atomic ensemble should be probed with another strong coherent light pulse . optimal retrieval occurs when @xmath104 then following ( [ b.7 ] ) the output field operators are given by @xmath105^{1/2}\right)\hat{{\mathcal t}}_{\mathrm{i}}^{\mathrm{out}}(z)+\ldots% \nonumber\\% \hat{\xi}_{3}^{\mathrm{out}}(t)\!\!&=&\!\!% \phantom{+}2\epsilon\bar{\xi}_2'\!\int_0^l\!dz\,% j_0\!\left(2[-a'(l - z)t]^{1/2}\right)\hat{{\mathcal t}}_{\mathrm{i\!i\!i}}^{\mathrm{out}}(z)+\ldots% \nonumber\\% \label{3.8}%\end{aligned}\ ] ] where the initial state of the atoms is @xmath92 and @xmath93 modified in the write - in stage of the protocol . the dots on the right hand side again indicate the imperfection of the transform because of the presence of the input field operators associated with the vacuum modes . the noise contribution caused by the vacuum fluctuation of the input field can be suppressed in the low frequency domain @xmath106^{1/2}$ ] , see eq.([b.7 ] ) . at the retrieval stage of the protocol the spin fluctuations within the spectral interval @xmath107 are mapped onto the time dependent field fluctuations in the frequency spectral domain @xmath108 , where the correlation length is given by @xmath109 . at the retrieval stage of the memory protocol the following inequalities should be fulfilled @xmath110 which , after the obvious replacement of temporal and spatial parameters , are symmetric to the inequalities ( [ 3.6 ] ) and have similar physical meaning . thus the time dynamics of the output field operators @xmath111 and @xmath112 correctly reproduce the spatial distribution of spin fluctuations in the relevant part of the spectrum . basic requirements for the proposed memory protocol can be summarized by combining the inequalities ( [ 3.6 ] ) and ( [ 3.9 ] ) where parameters @xmath113 and @xmath114 are given by eq.([3.1 ] ) . the following limitation should be imposed on the number of atoms @xmath55 and on the numbers of photons in the coherent strong field @xmath115 and @xmath116 participating in the process at the write - in and retrieval stages respectively @xmath117 the substitution of @xmath118 from ( [ 3.6 ] ) into ( [ 3.9 ] ) yields @xmath119 from evaluation of incoherent losses we obtain the following conditions on the numbers of atoms and photons @xmath120 where @xmath121 is the cross section of the incoherent scattering for a left - hand polarized light from the atoms populating the zeeman sublevel @xmath122 and @xmath123 is the cross section of the incoherent scattering for a right - hand polarized light from the atoms populating the @xmath124 zeeman sublevel ; @xmath9 is the beam / sample cross area . the first line indicates that the atomic medium should be transparent for the informative quantum radiation . the second line indicates that the incoherent scattering should have negligible influence on the dynamics of the spin coherence during both the write - in and the retrieval cycles . these requirements lead to the following basic demand on the density of atoms @xmath125 , resonance radiation wavelength @xmath126 and the sample length @xmath54 : @xmath127 , i.e. the on resonance optical depth of the sample needs to be large optimization of the inequalities ( [ 3.11 ] ) and ( [ 3.12 ] ) requires that @xmath128 . to explain this choice let us again consider the coherent raman scattering as the physical background for the memory protocol . in the write - in stage of the protocol the most desirable process is the annihilation of the photons of the quantum modes as a result of the coherent scattering into the classical mode followed by the transfer of their quantum state onto the alignment of atoms . the swapping process can be easier done for the photons in the wings of the spectral distribution ( [ 3.3 ] ) and is more difficult to realize for the resonance photons . that is why the process runs more efficiently with a stronger coherent field with @xmath129 . at the retrieval stage the coherent raman scattering takes photons from the strong mode into the quantum modes . the spin fluctuations of the atomic alignment tensor transfer back into the outgoing modes of the quantum light . to make this process more efficient , i.e. to increase the number of scattering events , it is desirable that the sample is extended and contains a large number of the atoms such that @xmath130 . in figure [ fig.3 ] we show how the spectral variances of the polarization components of light and atoms are modified after sending the broadband squeezed light ( @xmath131 ) through the atomic sample . the graphs clearly display the swapping mechanism at the write - in stage of the memory protocol . the graphs are normalized to the vacuum state variance and the deviations from it are expressed by the respective mandel parameters @xmath132 for @xmath133 , which depend on either frequency @xmath134 ( for light ) or wave number @xmath135 ( for atoms ) . the input squeezed state is described by two spectrally independent parameters @xmath136 and @xmath137 . the suppression of the correlations in the low frequency domain of the temporal spectrum of light is compensated by the enhancement of the correlations in the atoms for the low frequency part of the spatial spectrum . the dominant role of the collective modes in this process is a result of the approximation @xmath131 . ( squares ) and @xmath138 ( triangles ) . the black solid lines indicate the original fluctuation spectra in the light and spin subsystems . the gray lines in each of the panels indicate the original spectra for a complementary system . ] the data , presented in these graphs , corresponds to the red detuning of light near the @xmath1-line of @xmath2rb for atoms in the hyperfine level @xmath139 and for two selected values of the cooperative interaction parameter @xmath140 . the first value corresponds to the detuning @xmath141mhz from @xmath142 resonance , where @xmath143 ( no gyrotropy effect ) and for approximately ten percent losses estimated from eqs.([3.12 ] ) . the second number is achievable for the detuning of a few thousand mhz in the red wing of the @xmath142 transition and for the same level of losses . figure [ fig.4 ] shows the spectral variances of the polarization components for the atoms and light at the retrieval stage of the protocol . the figure shows how well the recovered state can reproduce the input . the parameters are chosen such that @xmath144 corresponding to @xmath145 and @xmath146 corresponding to @xmath138 . let us point out once more that for the best overall efficiency , the number of photons in the strong coherent pulse at the retrieval stage should be smaller than the number of photons applied at the write - in stage of the protocol and smaller than the number of atoms . as follows from these results the retrieved quantum state of light can reproduce the input state only in certain parts of the fluctuation spectrum . further optimization should allow for identification of the best temporal mode , where the retrieval of the original squeezed state would be optimal . . the values of cooperative parameter @xmath144 ( squares ) and @xmath146 ( triangles ) are coordinated with the data of fig.[fig.3 ] . the black solid lines indicate the original fluctuation spectra in the spin and light subsystems . the gray lines in each of the panels indicate the original spectra for the complementary system . ] figures [ fig.5 ] and [ fig.6 ] show how the spectra of the light and atoms are modified at the write - in and retrieval stages with the finite - bandwidth squeezed light as the input . the dependencies plotted in these graphs illustrate the importance of the dispersion relation between the temporal and spatial modes participating in the protocol . the ratio of the interaction time to the correlation time of the squeezing is taken to be @xmath147 . as follows from the displayed results , for the finite - bandwidth squeezed light the integral collective modes are not optimal in both storage and retrieval steps of the protocol . a certain optimization procedure is necessary for the best encoding of the quantum information into particular domains and modes of the fluctuation spectra for the light and atoms . another important observation which follows from these graphs is that the finite bandwidth squeezing is more difficult to store and retrieve than the broadband input state , the fact already established for the storage phase in @xcite . this is the direct consequence of the imperfection of the swapping mechanism for mapping the low frequency fluctuations , as follows from the dispersion law discussed above . , but for the finite - bandwidth squeezed light . the ratio of interaction time to the correlation time of squeezing for these graphs is @xmath147 . ] , but for retrieval of the finite - bandwidth squeezed state mapped onto the spin subsystem , see fig.[fig.5 ] . the original spectrum for the squeezed light in the lower panel relates to the case of @xmath148 . ] fidelity is the fundamental criterion for any quantum memory or teleportation scenario . it can be relatively simply defined for a single mode situation and for a pure input state , and is more subtle for a general case . in the following we show qualitatively that in the present case the quantum scheme has always a better fidelity than for what can be called a classical memory / measurement protocol . we assume that the classical memory scheme utilizes a balance homodyne detection of the input squeezed state for the measurement of its parameters . the measurement should yield the degree of squeezing and the directions of the linear polarizations shown in figure [ fig.1 ] . such a procedure is displayed in figure [ fig.7]a as an overlap of the wigner function associated with the input state with its measured counterpart . this overlap in the limit of high level of squeezing can be expressed by the following fidelity @xmath149 where @xmath150 is half of the variance for the anti - squeezed @xmath83 component , which is assumed to be reliably defined , and @xmath151 is the angular uncertainty remaining after @xmath152 measuring attempts to identify direction of @xmath47 or @xmath48 axes . following ref.@xcite the following constraint should be fulfilled in order to make the classical protocol efficient @xmath153 where @xmath154 is the duration of each measuring attempt as a fraction of the total measuring time @xmath26 . it is clear that for @xmath155 the number of measurements @xmath152 should grow up to infinity such that for any limited time @xmath26 the constraint ( [ 3.12 ] ) can never be fulfilled . on the contrary , in the quantum memory protocol it is only necessary to effectively reproduce the variances associated with the squeezed and anti - squeezed polarization components . the directions of the @xmath47 and @xmath48 axes remain completely unknown in the interaction process . this is shown in figure [ fig.7]b with a more effective overlap than in case of figure [ fig.7]a . the fidelity can be conveniently written in terms of mandel parameter for the input and output states @xmath156^{1/2}}% \label{2.12'''}\ ] ] where `` out '' characteristics should be associated with optimally defined spatial mode . comparing ( [ 3.12 ] ) and ( [ 2.12 ] ) for the same ratio @xmath157 and for the high level of squeezing one can expect that in the optimized case the fidelity for the quantum protocol will be always higher than the relevant classical benchmark . the entanglement between the light and the atomic alignment subsystems created in the process shown in figure [ fig.2 ] is attained in a result of the transformations ( [ b.12 ] ) , ( [ b.13 ] ) under the constraint ( [ 3.7 ] ) . the entanglement can be written in the following canonic form @xmath158the alignment spin waves @xmath92 and @xmath93 are defined by the upper lines of eqs.([3.5 ] ) . the standard entanglement between the quadrature components of certain collective field and atomic spin modes can be recognized in eqs.([3.13 ] ) , see also relations ( [ 2.12 ] ) and ( [ 2.13 ] ) . formally the temporal mode @xmath159 and the spatial mode @xmath160 can be found by solving the following integral equations @xmath161^{1/2}% i_1\left(2\left[al(t'-t)\right]^{1/2}\right)\,h(t')&&% \nonumber\\% - \bar{c}_{13}\epsilon\bar{\mathcal{f}}_z% \int_0^l dz\,i_0\left(2\left[a(t - t)z\right]^{1/2}\right)\,g(z)\to 0\phantom{(3 ... 14)}&&% \nonumber\\% \nonumber\\% g(z)+ \int_z^l dz'\,\left[\frac{at}{z'-z}\right]^{1/2}% i_1\left(2\left[at(z'-z)\right]^{1/2}\right)\,g(z')&&% \nonumber\\% + 2\epsilon\bar{\xi}_2% \int_0^t dt\,i_0\left(2\left[a(l - z)t\right]^{1/2}\right)\,h(t)\to 0\phantom{(3 ... 14)}&&% \label{3.14}%\end{aligned}\]]due to the presence of non - symmetric integral operators in the upper lines of these equations their actual solution does not responsibly exist in terms of real functions . one can search for a pair of real functions @xmath159 and @xmath160 which minimize the absolute value of the left hand side of ( [ 3.14 ] ) . this circumstance is reflected by the `` limit to zero '' in the right hand side , which can be approximately approached for certain optimal observation conditions . it is intuitively clear that this can be achieved for a system consisting of a large number of atoms ( @xmath162 ) and of a large number of coherent photons ( @xmath163 ) . using the common properties of the fredholm - type equations the solutions for the spatial and temporal modes of ( [ 3.13 ] ) can be found in terms of the eigenfunctions of the integral equations ( [ 3.14 ] ) . we discuss the polarization sensitive interaction between the stokes components of light and the alignment components of the atomic ensemble as a resource for quantum information interface . our model accurately describes the quantum nature of interaction between the spin subsystem of ultracold alkali atoms and a plain light wave propagating through the sample , in the absence of losses . the method has two important advantages . first , the process does not require any special manipulations like a feedback or adjustment of the driving classical light . this creates good outlooks for its experimental implementation . second , the model yields an analytical solution under rather general assumptions . this allows for a convenient and clear interpretation of the heisenberg dynamics of atoms and field at every step of the discussed quantum information protocols . we consider two basic protocols : quantum memory and quantum entanglement . the memory protocol is an example of the swapping mechanism between the atomic and field subsystems . the quantum information , which is originally encoded in the polarization degrees of freedom of the light wave , can be mapped onto the spin standing wave associated with atomic alignment components . however because of the multimode nature of the interaction process the quantum information can be spread among a number of spatial spectral modes . we show how the protocol , particularly at the final retrieval stage , can be optimized . as an important step of such optimization we have comprehensively discussed the imperfection of the quantum memory channel and made qualitative comparison of its figure of merit with a competing classical memory / measurement scheme . the second protocol is generation of entanglement in the spin oriented atomic ensemble probed with a circularly polarized coherent light mode . forward scattered raman photons appear strongly correlated with an alignment - type coherence of atomic scatterers . when the process becomes extended in space as well as in time it creates pairs of entangled temporal and spatial modes . we introduced the system of two integral equations , whose solution defines the structure of these modes . we would like to thank dr . igor sokolov for fruitful discussion . the work was supported by the russian foundation for basic research ( rfbr-05 - 02 - 16172-a ) , by intas ( project i d : 7904 ) and by the european grants within the networks covaqial and qap . o.s.m . would like to acknowledge the financial support from the charity foundation `` dynasty '' . d.v.k . would like to acknowledge financial support from the delzell foundation , inc . the wave - type heisenberg equations ( [ 2.11 ] ) straightforwardly follow from the commutation relations between the operators of the respective quantum observables with an effective hamiltonian responsible for the process of coherent forward scattering . the latter was derived in ref.@xcite . the exact equations are linearized with assumption of small fluctuations when @xmath164 and @xmath165 are approximately conserved and interpreted as classical values . then the gyrotropy constant is given by @xmath166^{1/2}}% \label{a.1}\ ] ] here @xmath167 is the dimensionless orientation income into the polarizability tensor for the @xmath168 hyperfine transition @xmath169where @xmath170 is the reduced dipole moment and @xmath171 is the transition frequency . the light shift is given by @xmath172^{1/2}}% \label{a.3}\ ] ] this expression is similar to definition of the gyrotropy constant ( [ a.1 ] ) because both the parameters come from the same faraday - type interaction term of the effective hamiltonian . the coupling constant @xmath59 responsible for the alignment - type interaction is given by @xmath173 where @xmath174 is the dimensionless alignment contribution into the polarizability tensor @xmath175all the parameters are the spectrally dependent values such that @xmath176 , @xmath177 , @xmath178 and this dependence can be very important in practical calculations . we derive the solution for those physical conditions when only low frequency temporal fluctuations are considered as a quantum information carrier . that lets us completely ignore the retardation effects associated with a finite sample size . practically this means that we can neglect the time derivations in the first two lines of the system ( [ 2.11 ] ) . as a first step we make the following local rotational transformations for the stokes components of the field subsystem @xmath179 and for the alignment components of the atomic subsystem @xmath180 where @xmath181 . without losing of generality the parameter @xmath182 can be set as @xmath183 to ignore any non - principle freely precession of atomic spins associated with external magnetic field and light shift . it is also convenient to set the sample length @xmath54 as @xmath184 , then the pair of stokes variables @xmath185 , @xmath186 and @xmath187 , @xmath188 will coincide at the output of the sample . applying the orthogonal transformations ( [ b.1 ] ) and ( [ b.2 ] ) to the system ( [ 2.11 ] ) the latter can be rewritten in the following form @xmath189 then its solution can be found by the method of laplace transform . to show this we can define the laplace images of the space - time dependent stokes components of the probe light and of the collective alignment components of atoms @xmath190with @xmath191 and @xmath192 . then equations ( [ b.3 ] ) is transformed to the system of algebraic equations , which in turn can be straightforwardly solved similar to how it was done in appendix c of ref.@xcite . the important parameter for this procedure is the determinant of the system , which is given by @xmath193 ^ 2% \label{b.5}%\]]and its pole @xmath194 with @xmath195 , can be linked with the dispersion law for the spin polariton wave propagating through the sample , see eq.([3.1 ] ) and discussion around it . @xmath197^{1/2}% j_1\left(2\left[-al(t - t')\right]^{1/2}\right)\hat{\xi}_{\mathrm{i}}^{\mathrm{in}}(t')% \nonumber\\% & & -2\epsilon \bar{\xi}_2\int_0^l\!dz\,% j_0\left(2\left[-a(l - z)t\right]^{1/2}\right)% \hat{{\mathcal t}}_{\mathrm{i}}^{\mathrm{in}}(z)% \nonumber\\% \hat{\xi}_{\mathrm{i\!i\!i}}(l , t)&=&% \hat{\xi}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(t)-\int_0^t\!dt'% \left[\frac{-al}{t - t'}\right]^{1/2}% j_1\left(2\left[-al(t - t')\right]^{1/2}\right)\hat{\xi}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(t')% \nonumber\\% & & + 2\epsilon \bar{\xi}_2\int_0^l\!dz\,% j_0\left(2\left[-a(l - z)t\right]^{1/2}\right)% \hat{{\mathcal t}}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(z)% \label{b.7}%\end{aligned}\ ] ] and for the alignment components of the spin subsystem @xmath198^{1/2}% j_1\left(2\left[-at(z - z')\right]^{1/2}\right)\hat{\mathcal t}_{\mathrm{i}}^{\mathrm{in}}(z')% \nonumber\\% & & + \bar{c}_{13}\epsilon \bar{\mathcal f}_z% \int_0^t\!dt\,j_0\left(2\left[-a(t - t)z\right]^{1/2}\right)% \hat{\xi}_{\mathrm{i}}^{\mathrm{in}}(t)% \nonumber\\% \hat{\mathcal t}_{\mathrm{i\!i\!i}}(z , t)&=&% \hat{\mathcal t}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(z)-\int_0^z\!dz'% \left[\frac{-at}{z - z'}\right]^{1/2}% j_1\left(2\left[-at(z - z')\right]^{1/2}\right)\hat{\mathcal t}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(z')% \nonumber\\% & & -\bar{c}_{13}\epsilon \bar{\mathcal f}_z% \int_0^t\!dt\,j_0\left(2\left[-a(t - t)z\right]^{1/2}\right)% \hat{\xi}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(t)% \label{b.8}%\end{aligned}\ ] ] here @xmath54 is the sample length and @xmath26 is the interaction time . the solution performs an integral transform of the heisenberg operators for the stokes components in the incident light @xmath199 and of initial schrdinger operators for the alignment components of atomic spins @xmath200 . the kernels of the transform are expressed by the cylindrical bessel functions of the zeroth @xmath201 and first @xmath202 order . the original form of atomic and field operators can be recovered with the orthogonal transformations reverse to ( [ b.1 ] ) and ( [ b.2 ] ) . the derived solution preserves the following commutation relations @xmath203&=&% 2i\,\delta(t - t')\bar{\xi}_2% \nonumber\\% \left[\hat{\mathcal t}_{\mathrm{i\!i\!i}}(z , t ) , \hat{\mathcal t}_{\mathrm{i}}(z',t)\right]&=&% -i\bar{c}_{13}\,\delta(z - z')\,\bar{\mathcal f}_{z}% \label{b.9}%\end{aligned}\ ] ] the commutation relation for the stokes components , given by the first line , differs from the original commutation relation in form ( [ 2.4 ] ) . that is direct consequence of our ignoring the retardation effects . the argument of @xmath21-function , which generally is @xmath204 , can be only approximately reproduced in the assumptions we did . the solution also obeys the following important cross - type commutation relations between the field and atomic variables @xmath205=% -2i\bar{c}_{13}\epsilon\,g_{{\mathrm i}\mu}\bar{\xi}_2\bar{\mathcal f}_{z}}% \nonumber\\% & & \times j_0\left(2\left[-a(z - z')(t - t')\right]^{1/2}\right)% \left[\theta(t - t')-\theta(z'-z)\right]% \nonumber\\% \label{b.10}%\end{aligned}\ ] ] where the matrix @xmath206 ( with @xmath207 ) is given by @xmath208 these commutation relations clear indicate that the atomic and field variables always commute before and after interaction . @xmath210^{1/2}% i_1\left(2\left[al(t - t')\right]^{1/2}\right)\hat{\xi}_{\mathrm{i}}^{\mathrm{in}}(t')% \nonumber\\% & & -2\epsilon \bar{\xi}_2\int_0^l\!dz\,% i_0\left(2\left[a(l - z)t\right]^{1/2}\right)% \hat{{\mathcal t}}_{\mathrm{i}}^{\mathrm{in}}(z)% \nonumber\\% \hat{\xi}_{\mathrm{i\!i\!i}}(l , t)&=&% \hat{\xi}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(t)+\int_0^t\!dt'% \left[\frac{al}{t - t'}\right]^{1/2}% i_1\left(2\left[al(t - t')\right]^{1/2}\right)\hat{\xi}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(t')% \nonumber\\% & & + 2\epsilon \bar{\xi}_2\int_0^l\!dz\,% i_0\left(2\left[a(l - z)t\right]^{1/2}\right)% \hat{{\mathcal t}}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(z)% \label{b.12}%\end{aligned}\ ] ] and for the alignment components of the spin subsystem @xmath211^{1/2}% i_1\left(2\left[at(z - z')\right]^{1/2}\right)\hat{\mathcal t}_{\mathrm{i}}^{\mathrm{in}}(z')% \nonumber\\% & & + \bar{c}_{13}\epsilon \bar{\mathcal f}_z% \int_0^t\!dt\,i_0\left(2\left[a(t - t)z\right]^{1/2}\right)% \hat{\xi}_{\mathrm{i}}^{\mathrm{in}}(t)% \nonumber\\% \hat{\mathcal t}_{\mathrm{i\!i\!i}}(z , t)&=&% \hat{\mathcal t}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(z)+\int_0^z\!dz'% \left[\frac{at}{z - z'}\right]^{1/2}% i_1\left(2\left[at(z - z')\right]^{1/2}\right)\hat{\mathcal t}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(z')% \nonumber\\% & & -\bar{c}_{13}\epsilon \bar{\mathcal f}_z% \int_0^t\!dt\,i_0\left(2\left[a(t - t)z\right]^{1/2}\right)% \hat{\xi}_{\mathrm{i\!i\!i}}^{\mathrm{in}}(t)% \label{b.13}%\end{aligned}\ ] ] here the functions @xmath212 and @xmath213 are the modified bessel functions of the zeroth and first order . the solution obeys the commutation relations ( [ b.9 ] ) and the cross - type commutation relations are now given by @xmath205=% -2i\bar{c}_{13}\epsilon\,g_{{\mathrm i}\mu}\bar{\xi}_2\bar{\mathcal f}_{z}}% \nonumber\\% & & \times i_0\left(2\left[a(z - z')(t - t')\right]^{1/2}\right)% \left[\theta(t - t')-\theta(z'-z)\right]% \nonumber\\% \label{b.14}%\end{aligned}\ ] ] k. hammerer , k. mlmer , e.s . polzik , j.i . rev . a**70 * * , 044304 ( 2004 ) ; j. sherson , j. fiurasek , k. mlmer , a. srensen , and e.s . rev . a**74 * * , 011802 ( 2006 ) ; c.a . muschick , k. hammerer , e.s . polzik , j.i . cirac , phys . a**73 * * 062329 ( 2006 ) . eisaman , a. andr , f. massou , m. fleishhauer , a.s . zibrov , m.d . lukin , nature * 438 * , 837 ( 2005 ) ; t. chanilire , d.n . matsukevich , s.d . jenkins , s.y . lan , t.a.b . kennedy , a. kuzmich , nature * 438 * , 833 ( 2005 ) o.s . mishina , d.v . kupriyanov , e.s . polzik , proceedings of the nato advanced research workshop , crete 2005 _ quantum information processing from theory to experiment _ , * 199 * , 346 - 352 ( 2006 ) , quant - ph/0509220 . this definition yields the expansion of the polarization tensor in the projector algebra of pauli matrices with respect to the basis of linear polarizations , see l.d . landau e.m . lifshits _ the classical theory of fields _ ( pergamon press , london , 1951 ) . in literature it is often used an alternative definition with expansion over the pauli matrices with respect to the basis of circular polarizations .

we use semiclassical propagation of singularities to give a general method for gluing together resolvent estimates . as an application we prove estimates for the analytic continuation of the resolvent of a schrdinger operator for certain asymptotically hyperbolic manifolds in the presence of trapping which is sufficiently mild in one of several senses . as a corollary we obtain local exponential decay for the wave propagator and local smoothing for the schrdinger propagator .

1008.3964

in this paper we give a general method for gluing semiclassical resolvent estimates . as an application we obtain the following theorem . [ t : intro ] let @xmath0 be an even asymptotically hyperbolic riemannian manifold . let @xmath1 suppose that the trapped set of @xmath2 , i.e. the set of maximally extended geodesics which are precompact , is either normally hyperbolic in the sense of [ s : wz ] or hyperbolic with negative topological pressure at @xmath3 ( see [ s : nz ] ) . then the cutoff resolvent @xmath4 continues analytically from @xmath5 to @xmath6-i[0,\gamma h]$ ] for every @xmath7 and @xmath8 , where it obeys the bound @xmath9 here @xmath10 and @xmath11 are both determined only by the trapped set . moreover , for @xmath12 $ ] we have for some @xmath13 the following quantitative limiting absorption principle : @xmath14 by an _ even asymptotically hyperbolic manifold _ we mean that @xmath2 is the interior of @xmath15 , a compact manifold with boundary , and @xmath16 here @xmath17 is a boundary defining function and @xmath18 is a family of metrics on @xmath19 , smooth up to @xmath19 , is a 2-cotensor which is a metric on @xmath19 when restricted to @xmath19 : see ( * ? ? ? * proposition 2.1 ) . ] and even in @xmath20 ( see ( * ? ? ? * definition 1.2 ) for a more invariant way to phrase this last condition ) . the assumptions on the trapped set will be discussed in more detail in [ s : trapped ] , but for now we remark that it suffices to take @xmath2 negatively curved with a trapped set consisting of a single closed geodesic . moreover , the compact part of the manifold on which the trapping occurs can be replaced by a domain with several convex obstacles . [ s : apps ] for a stronger result . as already stated our methods work in much greater generality . let @xmath0 be a complete riemannian manifold , @xmath21 a semiclassical schrdinger operator , @xmath22 bounded , @xmath23 . then @xmath24 is essentially self - adjoint , @xmath25 is holomorphic in @xmath26 . moreover , in this set one has uniform estimates on @xmath27 as @xmath28 , namely @xmath29 . on the other hand , as @xmath30 approaches the spectrum , @xmath31 is necessarily not uniformly bounded ( even for a single @xmath32 ) . however , in many settings , e.g. on asymptotically euclidean or hyperbolic spaces with a suitable decay assumption on @xmath33 , the resolvent extends continuously to the spectrum ( perhaps away from some thresholds ) , although only as an operator on weighted @xmath34-spaces . indeed under more restrictive assumptions it continues meromorphically across the continuous spectrum , typically to a riemann surface ramified at thresholds : see e.g. @xcite for a general discussion . it is very useful in this setting to obtain semiclassical resolvent estimates , i.e. estimates as @xmath28 , both at the spectrum of @xmath24 and for the analytic continuation of the resolvent , @xmath31 . by scaling , these imply high energy resolvent estimates for non - semiclassical schrdinger operators , which in turn can be used , for instance , to describe wave propagation , or more precisely the decay of solutions of the wave equation : see [ s : apps ] for examples of such applications in our setting . for this purpose the most relevant estimates are those in a strip near the real axis for the non - semiclassical problem ( which gives exponential decay rates for the wave equation ) , which translates to estimates in an @xmath35 neighborhood of the real axis for semiclassical problems . the best estimates one can expect ( on appropriate weighted spaces ) are @xmath36 ; this corresponds to the fact that this problem is semiclassically of real principal type . however , if there is trapping , i.e. some trajectories of the hamilton flow ( or , in case @xmath37 , geodesics ) do not escape to infinity , the estimates can be significantly worse ( exponentially large : see @xcite ) even at the real axis . nonetheless , if the trapping is mild , e.g. the trapped set is hyperbolic , then one has polynomial , @xmath38 , bounds in certain settings , see the work of nonnenmacher - zworski @xcite , petkov - stoyanov @xcite , and wunsch - zworski @xcite . moreover , on the real axis ( i.e. for @xmath39 with @xmath40 ) @xcite prove @xmath41 bounds , which work of bony - burq - ramond @xcite shows to be optimal when any trapping is present . that @xmath36 bounds hold in strips for nontrapping asymptotically hyperbolic manifolds was proved by the second author in @xcite ( see also @xcite and @xcite ) . this result , along with the accompanying propagation of singularities theorem , is what makes it possible to use our gluing method to prove theorem [ t : intro ] ; see [ s : ahinfinity ] for a discussion of how to . typically , for the settings in which one can prove polynomial bounds in the presence of trapping , one considers particularly convenient models in which one alters the problem away from the trapped set , e.g. by adding a complex absorbing potential . the natural expectation is that if one can prove such bounds in a thus altered setting , one should also have the bounds if one alters the operator in a different non - trapping manner , e.g. by gluing in a euclidean end or another non - trapping infinity . in spite of this widespread belief , no general prior results exist in this direction , though in some special cases this has been proved using partially microlocal techniques , e.g. in work of christianson @xcite on resolvent estimates where the trapping consists of a single hyperbolic orbit , and in the work of the first author @xcite , combining the estimates of nonnenmacher - zworski @xcite with the microlocal non - trapping asymptotically euclidean estimates of the second author and zworski @xcite as well as the more delicate non - microlocal estimates of burq @xcite and cardoso - vodev @xcite . another example is work of the first author @xcite using an adaptation of the method of complex scaling to glue in another class of asymptotically hyperbolic ends to the estimates of sjstrand - zworski @xcite and @xcite . it is important to point out , however , that in the present paper we glue the resolvent estimates _ directly _ , without the need for any information on how they were obtained , so for instance we do not need to construct a global escape function , etc . in addition to the above listed references , bruneau - petkov @xcite give a general method for deducing weighted resolvent estimates from cutoff resolvent estimates , but they require that the operators in the cutoff estimate and the weighted estimate be the same . in this paper we show how one can achieve this gluing in general , in a robust manner . the key point is the following . one can not simply use a partition of unity to combine the trapping model with a new ` infinity ' because the problem is not semiclassically elliptic . thus , semiclassical singularities ( i.e. lack of decay as @xmath28 ) propagate along null bicharacteristics . however , under a convexity assumption on the gluing region , which holds for instance when gluing in asymptotically euclidean or hyperbolic models , following these singularities microlocally allows us to show that an appropriate three - fold iteration of this construction , which takes into account the bicharacteristic flow , gives a parametrix with @xmath42 errors . this in turn allows us to show that the resolvent of the glued operator satisfies a polynomial estimate , and that on asymptotically euclidean and hyperbolic manifolds the order of growth is given by that of the model operator for the trapped region . we state our general assumptions and main result precisely in the next section , and prove the result in [ s : proof ] . in [ s : infinity ] we will show how our assumptions near infinity are satisfied for various asymptotically euclidean and asymptotically hyperbolic manifolds . in [ s : trapped ] we show how our assumptions near the trapped set are satisfied for various types of hyperbolic trapping . in [ s : apps ] we give applications : a more precise version of theorem [ t : intro ] , exponential decay for the wave equation , and local smoothing for the schrdinger propagator . we are very grateful to maciej zworski for his suggestion , which started this project , that resolvent gluing should be understood much better , and for his interest in this project . thanks also to both anonymous referees for their useful comments . before stating the abstract assumptions of our main theorem , we review some terminology from semiclassical analysis . for @xmath43 a symbol supported in a coordinate patch , @xmath44 compactly supported in the patch , @xmath45 on a neighborhood of the projection to @xmath2 of the support of @xmath46 , @xmath47 is a semiclassical quantization given in local coordinates by @xmath48 see , for example , @xcite for more information . we also say that a family of functions @xmath49 on @xmath2 is polynomially bounded if @xmath50 for some @xmath51 . the semiclassical wave front set , @xmath52 , is defined for polynomially bounded @xmath53 as follows : for @xmath54 , @xmath55 iff there exists @xmath56 with @xmath57 such that @xmath58 ( in @xmath34 ) . one can also extend the definition to @xmath59 ( thought of as the cosphere bundle at fiber - infinity in @xmath60 ) by considering @xmath61 , where @xmath62 is the fiber - radial compactification of @xmath60 with @xmath57 for @xmath63 ; then @xmath64 implies @xmath65 ( in @xmath34 ) . let @xmath15 be a compact manifold with boundary , @xmath66 a complete metric on @xmath2 ( the interior of @xmath15 ) , and @xmath67 a semiclassical schrdinger operator on @xmath2 with @xmath68 . we have no further explicit conditions on the potential @xmath33 , although the dynamical assumption and the assumptions on the resolvents below will in practice restrict the potentials the theorem applies to . let @xmath20 be a boundary defining function , and let @xmath69 recall that a bicharacteristic of @xmath24 is an integral curve in @xmath60 of the hamiltonian vector field associated to the hamiltonian function @xmath70 , and that the energy of a bicharacteristic is the level set of @xmath71 in which it lies . suppose that the bicharacteristics @xmath72 of @xmath24 ( by this we always mean bicharacteristics at energy in some fixed range @xmath6 $ ] ) satisfy the convexity assumption @xmath73 in @xmath74 . in particular this rules out the possibility of any trapped trajectory entering @xmath74 . there may be more complicated behavior , including trapping , in @xmath75 . [ rem : constants ] if @xmath20 is a boundary defining function , @xmath76 is a @xmath77 function on @xmath78 with @xmath79 , and @xmath20 satisfies then so does @xmath80 . in particular the specific constants above ( as well as intermediate constants used below ) are chosen only for convenience , and can be replaced by arbitrary constants so long as the ordering of the constants is preserved . let @xmath81 and @xmath82 be model differential operators on @xmath74 and @xmath83 respectively : @xmath84 the spaces @xmath85 on which the operators @xmath86 are globally defined can differ from @xmath2 away from @xmath87 , as the operators will always be multiplied by a smooth cutoff function to the appropriate @xmath87 . assume however that no bicharacteristic of @xmath82 leaves @xmath83 and then returns later , i.e. that @xmath88 note that we do not assume that the @xmath86 are self - adjoint ; this is useful in the applications in [ s : trapped][s : apps ] . however , the condition makes @xmath86 formally self - adjoint on @xmath87 and in particular has real semiclassical symbol on @xmath89 , as a result of which bicharacteristics of @xmath86 are well defined on @xmath89 . note that one difference from the ( in some ways related ) `` black - box '' approach of sjstrand - zworski @xcite is that for us @xmath90 will typically not be a compact manifold . in some applications we may wish to divide @xmath60 , rather than simply @xmath2 , into two overlapping regions with a different model operator on each one . to do this it is enough to take @xmath20 to be a more general function in @xmath91 , rather than a boundary defining function for @xmath92 as we do here . in that case @xmath42 error terms appear in the formulas in , and in several other formulas below , but the construction is essentially the same . for simplicity of exposition we do not pursue this level of generality further below . let @xmath93 , @xmath94 be bounded functions , referred to as _ weights _ ( in typical applications they may be compactly supported , decaying at infinity , or constant ) such that @xmath95 on @xmath96 and @xmath97 on @xmath83 . let @xmath98 have @xmath99 on @xmath74 and @xmath100 otherwise . let @xmath101 denote the weighted resolvent @xmath102 in @xmath5 or its meromorphic continuation where this exists in @xmath103 , and similarly @xmath104 and @xmath105 where those weighted resolvents exist . suppose that @xmath106 and the @xmath107 continue from @xmath5 to @xmath108-i[0,\gamma h ] , \quad 0<h<\tilde h_0,\ ] ] ( note that @xmath109 need not be open , and may in particular consist of @xmath6+i0 $ ] ) , and in that region they obey @xmath110 for some @xmath111 . we call a resolvent satisfying [ def : outgoing ] let @xmath112 and @xmath113 be in the characteristic set of @xmath114 , that is the zero set of @xmath115 , where @xmath116 is the semiclassical principal symbol of @xmath86 . let @xmath117\to t^*x'_j$ ] ( or @xmath118\to t^*x'_j$ ] in case this is not defined for all time ) be the backward @xmath86-bicharacteristic from @xmath119 . we say that the resolvent @xmath120 is _ semiclassically outgoing at @xmath119 _ if @xmath121 implies that @xmath122 we say that the resolvent @xmath120 is _ off - diagonally semiclassically outgoing at @xmath119 _ if holds provided we add @xmath123 to the hypotheses . [ rem : outgoing ] since @xmath53 is compactly supported , the definition involves only the cutoff resolvent . _ in this paper we only need to make the assumption that the resolvents @xmath120 are off - diagonally semiclassically outgoing . however , for brevity , we use the term ` semiclassically outgoing ' in place of ` off - diagonally semiclassically outgoing ' throughout the paper . _ a reason for making the weaker hypothesis of being off - diagonally semiclassically outgoing is that it is sometimes easier to check : typically the schwartz kernel of @xmath120 is simpler away from the diagonal than at the diagonal ( where it may be a semiclassical paired lagrangian distribution ) , and the off - diagonal outgoing property follows easily from the oscillatory nature of the schwartz kernel there ; see the third paragraph of [ s : infinity ] . our microlocal assumption is then that 1 . @xmath124 is semiclassically outgoing at all @xmath125 ( in the characteristic set of @xmath81 ) , 2 . @xmath126 is semiclassically outgoing at all @xmath125 ( in the characteristic set of @xmath82 ) such that @xmath127 is disjoint from @xmath128 , thus disjoint from any trapping in @xmath83 . in fact , for @xmath124 , it is sufficient to have the property in definition [ def : outgoing ] for @xmath129 ( i.e. @xmath53 supported in @xmath130 ) . the main result of the paper is the following general theorem . [ t : main ] under the assumptions of this section , there exists @xmath131 such that for @xmath132 , @xmath31 continues analytically to @xmath109 , where it obeys the bound @xmath133 in particular , when @xmath134 , we find that @xmath135 obeys ( up to constant factor ) the same bound as @xmath105 , the model operator with infinity suppressed . the only way we use convexity is to argue that no bicharacteristics of @xmath24 go from @xmath136 to @xmath137 and back to @xmath138 for some @xmath139 . this is also fulfilled in some settings in which the stronger condition does not hold , for example when @xmath2 has cylindrical or asymptotically cylindrical ends . in particular , some mild concavity is allowed . let @xmath140)$ ] be such that @xmath141 near @xmath142 , and @xmath143 , and let @xmath144 . define a right parametrix for @xmath24 by @xmath145 we then put @xmath146r_0(\lambda ) \chi_0(x ) + [ p,\chi_1(x+1)]r_1(\lambda)\chi_1(x ) { \stackrel{\textrm{def}}=}\operatorname{id}+ a_0 + a_1.\ ] ] this error is large in @xmath32 due to semiclassical propagation of singularities : in general we have only @xmath147 , and @xmath148 , and thus @xmath149 typically does not go to 0 with @xmath32 . ( note that there is no need for a weight on the left of @xmath150 in view of the support of the cutoffs . ) however , using an iteration argument we can replace it by a small error . observe that by disjointness of supports of @xmath151 and @xmath152 , resp . @xmath153 and @xmath154 , we have @xmath155 while lemma [ zigzag : worse ] below implies that @xmath156 this is the step in which we exploit the semiclassical propagation of singularities ( see figure [ f : noreturn ] ) . note that we have @xmath157\rho_0r_0(\lambda ) \rho_0 [ p,\chi_1(x+1)]\rho_1 r_1(\lambda)\rho_1\chi_1(x),\ ] ] that is to say , inserting weights @xmath158 and @xmath159 amounts to multiplying by @xmath160 thanks to the cutoff functions @xmath161 which are present . [ zigzag : worse ] suppose that @xmath162 are compactly supported semiclassical differential operators ( i.e. given in local coordinates by @xmath163 where the sum is over a finite set of multiindices @xmath164 ) with @xmath165 then @xmath166 : the outermost one is @xmath167 and the innermost @xmath168 . the supports of the various cutoffs are indicated ( note that the supports are contained in the interiors of their respective annuli ) . the trajectory @xmath169 is ruled out by the convexity assumption , and this is exploited by lemma [ zigzag : worse ] . the trajectory @xmath170 is possible , and this is the reason a third iteration is needed in the parametrix construction . ] before proving this lemma we show how implies theorem [ t : main ] . we solve away the first error by writing , using @xmath171 similarly we have @xmath172 the last term is already @xmath42 by , but @xmath173 is not yet small . we thus repeat this process for @xmath174 to obtain @xmath175 we now observe that both remaining error terms are of size @xmath42 thanks to . correspondingly , @xmath176 is invertible for sufficiently small @xmath32 , and the inverse is of the form @xmath177 , with @xmath178 . to estimate the resolvent we write out @xmath179 we then find that @xmath180 this completes the proof that lemma [ zigzag : worse ] implies theorem [ t : main ] . note that only @xmath158 , the weight for @xmath181 , and not @xmath159 , appears in the definition of @xmath182 . this is because @xmath126 is already multiplied by a compactly supported cutoff in every place where it appears in our parametrix ( but this is not the case for @xmath124 ) . lemma [ zigzag : worse ] follows from the following two lemmas , for the hypotheses ( 1)-(3 ) of lemma [ zigzag ] are satisfied by @xmath183 as in lemma [ zigzag : worse ] , and ( 4 ) follows from the support properties of @xmath183 and lemma [ lemma : bich - convex ] . [ zigzag ] suppose that @xmath162 are semiclassical differential operators with the properties that 1 . @xmath184 is supported in @xmath83 , 2 . @xmath185 are supported in @xmath130 , 3 . @xmath186 , and @xmath187 , 4 . there is no bicharacteristic of @xmath82 from a point @xmath188 to a point @xmath189 followed by a bicharacteristic of @xmath81 from @xmath190 to a point @xmath191 . then @xmath192 [ lemma : bich - convex ] there is no bicharacteristic of @xmath82 from a point @xmath193 to a point @xmath194 followed by a bicharacteristic of @xmath81 from @xmath190 to a point @xmath195 . we prove this first in the case where the two curves constitute a bicharacteristic of @xmath24 . if there were such a bicharacteristic , say @xmath196\to t^*x$ ] , with @xmath197 , and @xmath198 for some @xmath199 , then the function @xmath200 would attain its minimum in the interior of @xmath201 at some point ( and would be @xmath202 there ) , and the second derivative would be nonnegative there , contradicting our convexity assumption . we now reduce to this case . assume that there are curves , @xmath203\to t^*x'_0 $ ] a bicharacteristic of @xmath81 from @xmath190 to @xmath204 and @xmath205\to t^*x'_1 $ ] a bicharacteristic of @xmath82 from @xmath206 to @xmath190 . now , by the bicharacteristic convexity of @xmath83 in @xmath90 , @xmath207 is completely in @xmath83 ( since its endpoints are there ) , so it is a @xmath24 bicharacteristic . on the other hand , @xmath208 need not be a @xmath24 bicharacteristic since it might intersect @xmath209 . however , taking infimum @xmath210 of times @xmath211 at which @xmath212 , @xmath213}$ ] is a @xmath24 bicharacteristic since it is disjoint from @xmath214 in view of @xmath215 and the intermediate value theorem . thus , @xmath216\to t^*x$ ] given by @xmath207 on @xmath217 $ ] and @xmath208 on @xmath218 $ ] is a @xmath24 bicharacteristic , with @xmath219 , @xmath220 , @xmath221 , completing the reduction to the case in the previous paragraph . first suppose that @xmath222 is polynomially bounded ; we claim that @xmath223 for this , it suffices to show that @xmath224 . note that by the polynomial boundedness assumption on the resolvent , @xmath225 , as well as @xmath226 , are polynomially bounded . so suppose @xmath227 , so in particular @xmath228 and , as @xmath229 is microlocal , @xmath230 . now , if @xmath204 is not in the characteristic set of @xmath81 , then by microlocal ellipticity of @xmath81 , @xmath231 , thus in @xmath232 . this contradicts ( 3 ) . so we may assume that @xmath204 in the characteristic set of @xmath81 ( and hence in particular not in @xmath233 ) . by ( 0-og ) , noting that @xmath234 and @xmath229 have disjoint supports , there is a point @xmath235 on the backward @xmath81-bicharacteristic from @xmath204 . thus @xmath236 and @xmath237 . by ( 1-og ) , noting that @xmath184 and @xmath234 have disjoint supports , either the backward @xmath82 bicharacteristic from @xmath190 intersects @xmath209 , in which case we can take any @xmath206 on it in this region , or there is a point @xmath206 on this backward bicharacteristic in @xmath238 , which is thus in @xmath239 . since this contradicts ( 4 ) , it completes the proof of . to complete the proof of the lemma , we just note that for any @xmath51 , the family of operators @xmath240 dependent on @xmath32 and @xmath30 , is continuous on @xmath34 , and for each @xmath53 , @xmath241 is uniformly bounded in @xmath34 . thus , by the theorem of banach - steinhaus , @xmath242 is uniformly bounded ( in @xmath32 and @xmath30 ) on @xmath34 , completing the proof of the lemma . the application of banach - steinhaus is only needed because we merely made wavefront set assumptions in definition [ def : outgoing ] . in practice , the wave front set statement is proved by means of a uniform estimate , and thus banach - steinhaus is superfluous . lemma [ zigzag : worse ] holds with the same proof if @xmath183 are instead semiclassical pseudodifferential operators with @xmath243 in the cotangent bundle of the corresponding set . note however that in this case slightly more care is needed in defining the @xmath183 since their schwartz kernels may no longer be compactly supported . this could be useful for applications where @xmath24 is not differential , as in @xcite . in this section we describe some examples in which the assumptions on the model at infinity , @xmath81 , are satisfied . recall that the assumptions on @xmath81 are of three kinds : 1 . bicharacteristic convexity of level sets of @xmath20 for @xmath244 , 2 . polynomial bounds for the cutoff resolvent , 3 . semiclassically outgoing resolvent . for simplicity , in this section we consider the case @xmath245 we start with some general remarks . first , in the setting where @xmath246 is diffeomorphic to @xmath247 , has nonpositive sectional curvature and , for fixed @xmath248 the function @xmath249 with @xmath250 , follows from the hessian comparison theorem @xcite . next , the semiclassically outgoing assumption is satisfied for @xmath124 if the restriction of its schwartz kernel to @xmath251 is a semiclassical fourier integral operator with canonical relation @xmath252 corresponding to forward propagation along bicharacteristics , i.e. @xmath253 implies @xmath254 is on the forward bicharacteristic segment from @xmath255 . here @xmath256 is the diagonal in @xmath257 . note that this is where restricting the semiclassical outgoing condition to its off - diagonal version is useful , in that usually the structure of the resolvent at the diagonal is slightly more complicated ( though the condition would still hold ) ; see also remark [ rem : outgoing ] . if @xmath2 is isometric outside of a compact set to euclidean space we may take @xmath258 with the euclidean metric @xmath259 , and @xmath260 the distance function from a point in @xmath247 . thus , the convexity hypotheses holds in view of geodesic convexity of the spheres . moreover , for @xmath261 , @xmath262 , the resolvent continues analytically to @xmath263 as an operator @xmath264 with uniform estimates @xmath265 . finally , @xmath31 is a semiclassical fio associated to the forward flow ; indeed , with @xmath266 the square root on @xmath267 $ ] which is positive for positive @xmath30 , its schwartz kernel is ( see e.g. @xcite ) @xmath268 where @xmath46 is a symbol ( away from the origin ) . the applications in this case have already been treated in @xcite , but for compactly supported cutoff functions . the novelty in the present paper in this setting is that we use exponential weights @xmath269 . more general asymptotically euclidean manifolds , whose metrics have holomorphic coefficients near infinity , could probably also be treated : see @xcite for more details on the needed assumptions and the proof of the analytic continuation of the resolvent , and @xcite for semiclassical estimates and propagation of singularities . the convexity assumption is satisfied for the geodesic flow on a general asymptotically hyperbolic metric . in the following lemma this is proved in a region @xmath270 , but a rescaling of the boundary defining function gives it in the region @xmath271 . the computation is standard , but we include it for the reader s convenience . let @xmath20 be a boundary defining function on @xmath15 , a compact manifold with boundary , and let @xmath66 be a metric on the interior of the form @xmath272 where @xmath18 is a family of metrics on @xmath273 , smooth up to @xmath273 . then for @xmath20 sufficiently small we have @xmath274 along geodesic bicharacteristics . as remarked in the introduction , it is possible to reduce a more general form of the metric @xmath66 to this one . namely , it suffices to assume that @xmath18 is a 2-cotensor which is a metric on @xmath19 when restricted to @xmath19 : see ( * ? ? ? * proposition 2.1 ) . if @xmath275 are coordinates on @xmath2 near @xmath273 such that @xmath276 are coordinates on @xmath273 , and if @xmath277 is dual to @xmath20 and @xmath278 to @xmath276 , then the geodesic hamiltonian is given by @xmath279 where @xmath280 and @xmath281 , and @xmath282 is the bilinear form on @xmath283 induced by @xmath18 . its hamiltonian vector field is @xmath284 we use @xmath285 , @xmath286 and `` @xmath287 '' , where in the last formula the left hand side refers to @xmath288 coordinates , and the right hand side to @xmath289 coordinates . this gives @xmath290{\partial}_\tau + x({\partial}_\mu|\zeta|^2 ) \cdot { \partial}_y - x({\partial}_y|\zeta|^2)\cdot { \partial}_\mu.\end{split}\ ] ] we cancel the @xmath291 terms , write @xmath292 , substitute @xmath293 , and use @xmath294 . now @xmath295 we now observe from this that , along flowlines of @xmath296 , we have @xmath297 and @xmath298 . hence @xmath299 in which case @xmath300 since @xmath301 is positive definite , for sufficiently small @xmath20 this is always negative . if in addition @xmath18 is even in @xmath20 , in the sense that the taylor series at @xmath302 includes only even powers of @xmath20 ( or see ( * * definition 1.2 ) for a more invariantly phrased version of this condition ) , then work of the second author ( * ? ? ? * theorem 4.3 ) , ( * ? ? ? * theorem 5.1 ) implies the polynomial bound and the outgoing condition ( 0-og ) for @xmath303 when the manifold is nontrapping . the outgoing condition which is proved in those theorems , when restricted to data in @xmath304 , is the same as that in condition ( 0-og ) ( for this purpose the weights are irrelevant ) . the resolvent estimate ( * ? ? ? * ( 4.27 ) ) is that with @xmath305 for @xmath306 , for @xmath307 and @xmath308 , @xmath309 we will show that this implies @xmath310 uniformly for @xmath311 $ ] , @xmath312 . this argument is somewhat involved due to the rather different functions spaces appearing in and , as already indicated by the presence of @xmath313 in ; the results of ( * ? ? ? * theorem 4.3 ) , ( * ? ? ? * theorem 5.1 ) are obtained by extending an operator related to the spectral family of the laplacian across @xmath314 to a larger space . here @xmath313 is @xmath315 as a topological manifold with boundary , but with smooth structure given by even ( in @xmath20 ) smooth functions on @xmath315 ; effectively this means that the boundary defining function @xmath20 is replaced by @xmath316 . we first recall the definition of @xmath317 , which is the standard semiclassical ( with @xmath318 playing the role of the semiclassical parameter ) sobolev space on @xmath313 . a straightforward computation gives , see ( * ? ? ? * section 1 ) , @xmath319 where @xmath320 is with respect to any smooth non - degenerate density on the compact manifold @xmath313 ) , while @xmath321 is the metric @xmath34-space . furthermore , in local coordinates @xmath322 , using @xmath323 , for @xmath324 integer , the squared high energy @xmath325 norm of @xmath53 is equivalent to @xmath326 we now convert into an @xmath327-estimate , where @xmath327 are the zero - sobolev spaces of mazzeo and melrose @xcite , i.e. they are the sobolev spaces measuring regularity with respect to @xmath328 , the algebra of differential operators generated by @xmath329 , the lie algebra of @xmath77 vector fields vanishing at the boundary over @xmath330 , in the space @xmath331 . more precisely , we need the semiclassical version @xmath332 of these spaces , in which @xmath318 times @xmath329 is used to generate the differential operators . the square high energy @xmath333 norm of @xmath53 is equivalent to @xmath334 because of the ellipticity of @xmath335 for these spaces ( see @xcite ) , in the precise sense that the standard principal symbol is elliptic even in the semiclassical zero - calculus , this is also equivalent , when @xmath336 is even , to @xmath337 this equivalence identifies the @xmath338 spaces with the usual semiclassical sobolev spaces based on @xmath331 . to make the conversion from into an @xmath332-estimate , we remark that with @xmath339 , @xmath340 and similarly for the high energy spaces , so @xmath341 where the shift of @xmath342 in the exponent is due to the different normalization of the @xmath34-spaces , . thus , taking @xmath343 integer , @xmath344 , @xmath345 , @xmath11 , and simply using @xmath346 , we deduce that @xmath347 notice that there is a loss of @xmath348 in the weight between the two sides . although this simple argument does not give an optimal zero - sobolev space estimate , to minimize losses take @xmath349 , and @xmath350 , so @xmath351 @xmath352 again using the ellipticity of @xmath335 in the zero - calculus ( as in ) , allows one to strengthen the norm on the left hand side to @xmath353 this simply requires a commutator argument or a parametrix with a smoothing ( but not semiclassically trivial ) error . for example , to strengthen the norm to @xmath354 we may write @xmath355{\mathcal{r}}(\sigma)f.\ ] ] multiplying by @xmath356 and taking the @xmath331 norm , we see that the first two terms are both controlled by the estimate , while the last is bounded by @xmath357 this implies that holds with @xmath331 norms replaced by @xmath358 norms , and iterating one can get as far as controlling the @xmath359 norm ( past which point the first term of is no longer controlled ) . to pass from estimates in @xmath359 to estimates in @xmath331 we use a similar procedure . for @xmath360 fixed we have the semiclassical elliptic estimate @xmath361 from which we deduce @xmath362 note that although the resolvent we use is @xmath363 , the shift by @xmath364 is not important when @xmath365 is large and does not interfere with the application of . iterating this we obtain the more standardly phrased weighted estimate @xmath366 which in turn implies . this is not optimal in terms of the weights which could be improved using @xmath313 derivatives to estimate weights in the spirit of ( * ? ? ? * chapter 4 , lemma 5.4 ) , but this would result in a loss in terms of @xmath365 . another approach to obtaining polynomial boundedness of the resolvent and the semiclassical outgoing condition is possible in a special case . let @xmath367 with a metric which is asymptotically hyperbolic in the following stronger sense : @xmath368 where @xmath369 is the hyperbolic metric on @xmath370 and @xmath18 is a smooth symmetric 2-cotensor on @xmath371 , @xmath372 , @xmath373 , @xmath374 supported in @xmath375 , identically @xmath160 near @xmath376 , and @xmath377 sufficiently small . this is the setting considered by melrose , s barreto and the second author @xcite . note that , although we have @xmath378 on a large compact set , the factor @xmath379 does not change @xmath259 near infinity . thus , after possibly scaling @xmath20 , i.e. replacing it by @xmath380 , in the region @xmath381 the cutoff @xmath382 . it is shown in @xcite that the schwartz kernel of @xmath383 is a semiclassical paired lagrangian distribution , which is just a lagrangian distribution away from the diagonal associated to the flow - out of the diagonal by the hamilton vector field of the metric function , hence , as remarked at the beginning of the section , @xmath383 is semiclassically outgoing . this also gives that @xmath383 satisfies the bound in with @xmath384-i[0,\gamma h]$ ] and with @xmath134 for arbitrary @xmath385 with compactly supported cutoffs as a consequence of a semiclassical version of ( * ? ? ? * theorem 3.3 ) . moreover , it is also shown in @xcite that the resolvent satisfies weaker polynomial bounds in weighted spaces , namely @xmath386 , @xmath387 , with @xmath388 . it is highly likely that the better bound @xmath134 could be shown for the weighted spaces in this way as well ; this could be proved by extending the approach of @xcite in a manner that is uniform up to the boundary ( i.e. infinity ) ; this is expected to be relatively straightforward . the same results hold without modification in the case where @xmath389 is a disjoint union of balls with @xmath259 and @xmath81 of the form in each ball . in this section we describe some examples in which the assumptions on the model near the trapped set , @xmath82 , are satisfied . the two main assumptions , polynomial boundedness of the resolvent and the semiclassically outgoing property ( 1-og ) , are the same as in the case of @xmath81 above , with the exception that the latter need only hold at points where the backward bicharacteristic is disjoint from any trapping in @xmath83 . in [ s : cpx ] we prove that the semiclassically outgoing property ( 1-og ) holds for polynomially bounded resolvents when either a complex absorbing barrier is added near infinity ( regardless of the cutoff or weight and regardless of the type of infinite end ) , and in [ s : euc ] we prove it when infinity is euclidean ( with no complex absorption added ) and the resolvent is polynomially bounded and suitably cutoff or weighted . in [ s : wz ] , [ s : nz ] , [ s : ps ] we give examples of assumptions on the trapped set which imply that the resolvent is polynomially bounded . in this subsection we consider model operators of the form @xmath390 where @xmath391 and @xmath392)$ ] has @xmath393 on @xmath83 and @xmath394 off a compact set . suppose that each backward bicharacteristic of @xmath395 at energy @xmath396 $ ] enters either the interior of @xmath397 $ ] in finite time . the strong assumptions on @xmath398 remove the need for any further assumptions on @xmath33 or on the metric . the function @xmath398 in is called a complex absorbing barrier and serves to suppress the effects of infinity . in lemma [ l : w ] we prove the needed semiclassical propagation of singularities in this setting , that is to say that @xmath126 is semiclassically outgoing in the sense of [ s : abstract ] . after this , all that is needed to be in the setting of [ s : abstract ] is the convexity condition and the resolvent estimate . in [ s : wz ] and [ s : nz ] we describe settings in which results of wunsch - zworski @xcite and nonnenmacher - zworski @xcite respectively give the needed bound . for the following lemma we use a positive commutator argument based on an escape function as in @xcite , which is the semiclassical adaptation of the proof of ( * ? ? ? * proposition 3.5.1 ) . the only slight subtlety comes from the interaction of the escape function with the complex absorbing barrier @xmath398 and from the possibly unfavorable sign of @xmath399 , but the positive commutator with the self adjoint part of the operator overcomes these effects . see also ( * ? ? ? * lemma a.2 ) for a similar result . [ l : w ] suppose that @xmath82 is as in . let @xmath400 be preserved by the backward bicharacteristic flow . if @xmath401 has @xmath402 and @xmath403 for all @xmath404 with support in @xmath405 and all @xmath396-i[0,\gamma h]$ ] , then for every @xmath404 with support in @xmath405 we have also @xmath406 the implicit constants in @xmath407 in and are uniform for @xmath396-i[0,\gamma h]$ ] . note that in view of definition [ def : outgoing ] , this lemma implies assumption ( 0-og ) in this proof all norms are @xmath34 norms . in the first step we use ellipticity to reduce to a neighborhood of the energy surface , and then a covering argument to reduce to a neighborhood of a single bicharacteristic segment . in the second step we construct an escape function ( a monotonic function ) along this segment . in the third step we implement the positive commutator method . let @xmath408 * step 1 . * observe first that for any @xmath409 , we can find @xmath410 , a semiclassical elliptic inverse for @xmath82 on the set @xmath411 , such that @xmath412 as long as @xmath413 . since by the semiclassical composition formula the operator @xmath414 is the quantization of a compactly supported symbol with support contained in @xmath415 , plus an error of size @xmath42 ( as an operator @xmath416 ) , we have the lemma for @xmath46 with @xmath413 . it remains to study @xmath46 with @xmath417 . note that this is a precompact set for @xmath418 small , because of the condition that @xmath394 off of a compact set . now fix @xmath419 for which we wish to prove . take @xmath420 with @xmath421 such that @xmath420 is preserved by the backward flow and @xmath422 . for each @xmath423 put @xmath424)\},\ ] ] where @xmath425 is the flow of the hamiltonian vector field of @xmath71 at time @xmath211 , and @xmath139 will be specified later . the supremum is taken over a nonempty set , since each backward bicharacteristic of @xmath71 was assumed to enter either @xmath426 or @xmath427 in finite time , and the second possibility is ruled out by the assumption on @xmath405 . we will prove the lemma for @xmath46 which are supported in a sufficiently small neighborhood @xmath428 of @xmath429 . this gives the full lemma because if @xmath418 small enough these neighborhoods together with @xmath411 cover all of @xmath430 . * step 2 . * to do this we take a tubular neighborhood @xmath431 of @xmath429 , that is a neighborhood of the form @xmath432 where @xmath433 is a hypersurface transversal to the bicharacteristic through @xmath434 , and @xmath435 and @xmath436 are small enough that @xmath437),\ ] ] and also small enough that the map @xmath438 defined by is a diffeomorphism . we now use these ` product coordinates ' to define an escape function as follows . take * @xmath439)$ ] with @xmath440 near @xmath434 , and * @xmath441 with @xmath442 near @xmath443 $ ] and @xmath444 on @xmath445 $ ] . the constant @xmath446 above is the same as the one in the statement of the lemma . put @xmath447 and let @xmath428 be a neighborhood of @xmath429 in which @xmath448 and @xmath449 . take @xmath450 such that @xmath451 if necessary redefining @xmath452 so that @xmath453 is smooth . by taking @xmath454 large , we may ensure that @xmath455 everywhere . note that follows from @xmath456 on @xmath428 . * step 3 . * put @xmath457 , @xmath458 , @xmath459 . now @xmath460 + r + he,\ ] ] where @xmath461 and where the error @xmath462 has @xmath463 . we have @xmath464 u{\rangle}+ { \langle}u , r u { \rangle}+ h { \langle}u , e u { \rangle}\le \frac i h { \langle}u , [ q^*q , p ] u{\rangle}+ h \|e u\| + { \mathcal{o}}(h^\infty),\ ] ] where we used @xmath465 ( which was an assumption ) and @xmath466 ( which follows from step 1 above , since @xmath459 and @xmath467 . next @xmath468 u{\rangle}&= -\frac 2 h \operatorname{im}{\langle}u , q^*q(p_1-\lambda ) u{\rangle}- \frac 2 h \operatorname{re}{\langle}u , q^*qw u{\rangle}- \frac 2 h { \langle}u , q^*q \operatorname{im}\lambda u{\rangle}\\ & \le - \frac 2 h \operatorname{re}{\langle}u , q^*[q , w ] u{\rangle}- \frac 2 h { \langle}u , q^*q \operatorname{im}\lambda u{\rangle}+ { \mathcal{o}}(h^\infty ) \end{split},\ ] ] where we used @xmath469 and @xmath470 ( see ) . we will now show @xmath471 + q \operatorname{im}\lambda ) u{\rangle}\le \frac h 4 \|bu\|^2 + { \mathcal{o}}(h^2 ) \|e ' u\| + { \mathcal{o}}(h^\infty),\ ] ] with @xmath472 with @xmath473 . then we will have @xmath474 after which an iteration argument , for example as in ( * ? ? ? * lemma 2 ) , shows that @xmath475 allowing us to conclude . the iteration argument involves taking a nested sequence of escape functions @xmath476 , with corresponding functions @xmath477 as in such that @xmath478 is contained in the set where @xmath477 is elliptic ( bounded away from @xmath376 ) . this allows us to show that if @xmath479 , then @xmath480 . the estimate is the slight subtlety discussed in the paragraph preceding the statement of the lemma . because @xmath481 has real principal symbol of order @xmath160 , and @xmath482 $ ] has imaginary principal symbol of order @xmath32 , we have @xmath483 u{\rangle}| = { \mathcal{o}}(h^2 ) \|e ' u\| + { \mathcal{o}}(h^\infty),\ ] ] with @xmath484 with @xmath485 . meanwhile @xmath486 with @xmath487 with @xmath488 . for the second inequality we used the sharp grding inequality . indeed , the semiclassical principal symbol of @xmath489 is @xmath490 , and we may apply . the model operator near the trapped set is of the form @xmath491 off of a compact set @xmath492 ( which may contain @xmath83 ) and @xmath493 is isometric to euclidean space there . suppose that each backward bicharacteristic of @xmath494 at energy @xmath396 $ ] which enters @xmath495 also enters either @xmath427 or @xmath496 . in this case , similarly to [ s : aeinfinity ] , the semiclassically outgoing condition , which is only needed in the euclidean region ( i.e. with backward bicharacteristic disjoint from @xmath492 ) , can be proved in several ways . one way is to use an escape function and positive commutator estimate as in lemma [ l : w ] : see ( * ? ? ? * lemma 2 ) for a complete proof in a more general setting , based on the construction and estimates of @xcite . another way , which we only outline here , is to show the ( off - diagonal ) semiclassical fio nature of @xmath497 in this region , with lagrangian given by the flow - out of the diagonal . but this follows from the usual parametrix identity , taking some @xmath498 identically @xmath160 on the compact set , using @xmath499 as the parametrix , with @xmath500 the euclidean resolvent . indeed , first for @xmath501 , @xmath502 , @xmath503 , with @xmath504 and @xmath505 having schwartz kernels with support in the left , resp . right factor in @xmath506 ( e.g. @xmath507\tilde r_0(\lambda)$ ] ) , so @xmath508 this identity thus also holds for the analytic continuation . now , even for the analytic continuation , @xmath509 , @xmath505 and @xmath504 are semiclassical lagrangian distributions away from the diagonal as follows from the explicit formula ( where @xmath399 is @xmath35 ) , and if a point is in the image of the wave front relation of @xmath510 or @xmath505 ( with @xmath152 compactly supported , identically @xmath160 on @xmath506 ) then it is on the forward bicharacteristic emanating from a point in @xmath511 , proving the semiclassically outgoing property of the second and third term of the parametrix identity . we take these conditions from @xcite . let @xmath493 be a manifold which is euclidean outside of a compact set , let @xmath512 , and let @xmath513 with @xmath398 as in and @xmath514 . define the backward / forward trapped sets by @xmath515 where again @xmath516 . the trapped set is @xmath517 we also define @xmath518 assume 1 . there exists @xmath409 such that @xmath519 on @xmath520 . @xmath521 are codimension one smooth manifolds intersecting transversely at @xmath522 . the flow is hyperbolic in the normal directions to @xmath522 in @xmath520 : there exist subbundles @xmath523 of @xmath524 such that @xmath525 where @xmath526 and there exists @xmath527 such that for all @xmath528 @xmath529 here and below , by @xmath530 we mean the differential of @xmath531 as a function of @xmath532 . this is the normal hyperbolicity assumption which we take from @xcite . this type of trapping appears in the setting of a slowly rotating kerr black hole . under these assumptions we have , from ( * ( 1.1 ) ) , @xmath533 for @xmath396-i[0,\gamma h]$ ] , for some @xmath534 . in particular , all the assumptions on @xmath82 and @xmath90 in [ s : abstract ] are satisfied . we take these conditions from @xcite . let @xmath493 be a manifold which is euclidean outside of a compact set , let @xmath512 , and let @xmath535 let @xmath536 denote the set of maximally extended null - bicharacteristics of @xmath82 which are precompact . we assume that @xmath536 is hyperbolic in the sense that for any @xmath537 , the tangent space to @xmath538 ( the energy surface ) at @xmath434 splits into flow , unstable , and stable subspaces ( * ? ? ? * definition 17.4.1 ) : 1 . 2 . @xmath540 . @xmath541 here again @xmath542 . this condition is satisfied in the case where @xmath2 is negatively curved near @xmath536 . the unstable jacobian @xmath543 for the flow at @xmath434 is given by @xmath544 we now define the topological pressure @xmath545 of the flow on the trapped set , following @xcite ( see also ( * ? ? ? * definition 20.2.1 ) ) . we say that a set @xmath546 is @xmath547 separated if , given any @xmath548 , there exists @xmath549 $ ] such that the distance between @xmath550 and @xmath551 is at least @xmath552 . for any @xmath553 define @xmath554 where the supremum is taken over all sets @xmath546 which are @xmath547 separated . the pressure is then defined as @xmath555 the crucial assumption on the dynamics of the bicharacterstic flow on the trapped set is that @xmath556 then from ( * ? ? ? * theorem 3 ) and ( * ? ? ? * ( 1.5 ) ) we have for any @xmath557 and @xmath558 , there exist @xmath559 such that @xmath560 for @xmath396 - i[0,\gamma h]$ ] . in particular , all the assumptions on @xmath82 and @xmath90 in [ s : abstract ] are satisfied . we take these conditions from @xcite . let @xmath561 , where @xmath66 is the euclidean metric and where @xmath562 is a union of disjoint convex bounded open sets with smooth boundary , and let @xmath563 with dirichlet boundary conditions and . assume that the @xmath564 satisfy the no - eclipse condition : namely that for each pair @xmath565 the convex hull of @xmath566 and @xmath567 does not intersect any other @xmath568 . in this setting having negative topological pressure at @xmath3 is equivalent to having negative abscissa of convergence of a certain dynamical zeta function , a condition under which a holomorphic continuation to strip of a polynomially bounded cutoff resolvent was first obtained by ikawa @xcite . to define this , for @xmath72 a primitive periodic reflecting ray with @xmath569 reflections , let @xmath570 be the length of @xmath72 and @xmath571 the associated linear poincar map . let @xmath572 for @xmath573 be the eigenvalues of @xmath571 with @xmath574 . let @xmath575 be the set of primitive periodic rays . set @xmath576 let @xmath577 if @xmath569 is even and @xmath578 if @xmath569 is odd . the dynamical zeta function is given by @xmath579 and the abscissa of convergence is the minimal @xmath580 such that the series is absolutely convergent for @xmath581 . assume that @xmath582 for simplicity , assume in addition that @xmath583 . this assumption can be replaced by another which is weaker and more dynamical but also more complicated : see ( * ? ? ? * theorem 1.3 ) for a better statement . then from ( * ? ? ? * theorem 1.3 ) we have for any @xmath558 @xmath584 for @xmath396-i[0,\gamma h]$ ] , for some @xmath585 and @xmath586 . in particular , all the assumptions on @xmath82 and @xmath90 in [ s : abstract ] are satisfied . we now give an improved version of theorem [ t : intro ] . 1 . suppose for some @xmath589 @xmath24 has a normally hyperbolic trapped set on @xmath590 $ ] in the sense of [ s : wz ] . then there exist @xmath591 such that @xmath592 for @xmath396-i[0,\gamma h]$ ] and @xmath593 . 2 . suppose @xmath24 has a hyperbolic trapped set on @xmath538with @xmath594 as in [ s : nz ] . then for any @xmath595 there exist @xmath596 such that @xmath592 for @xmath396-i[0,\gamma h]$ ] and @xmath593 . 3 . let @xmath597 with @xmath598 has @xmath599 for @xmath454 sufficiently small , @xmath600 for @xmath454 sufficiently large , and @xmath601 for all @xmath454 . let @xmath602 where @xmath603 is a union of disjoint convex open sets all contained in the region where @xmath599 , satisfying the no - eclipse condition , with abscissa of convergence @xmath604 as in [ s : ps ] , and with dirichlet boundary conditions imposed for @xmath605 . then there exist @xmath606 and @xmath586 such that @xmath607 for @xmath396-i[0,\gamma h]$ ] and @xmath593 . note that in case ( 3 ) we assume @xmath79 to rule out geometric trapping and to guarantee . here @xmath608 , perhaps multiplied by a suitable large constant prefactor , playes the role of the boundary defining function @xmath20 . the set @xmath83 encompasses the whole region where @xmath609 , and the set @xmath74 is contained in the region where @xmath600 . the theorem follows immediately from the main theorem , theorem [ t : main ] , together with [ s : ahinfinity ] ( in which we show that the assumptions on @xmath81 are satisfied and derive the weights @xmath610 ) , and [ s : wz ] , resp . [ s : nz ] , resp . [ s : ps ] in the cases ( 1 ) , resp . ( 2 ) , resp . ( 3 ) ( in which we show that the assumptions on @xmath82 are satisfied ) . the same results also hold when @xmath0 has euclidean ends in the sense of [ s : aeinfinity ] ; one merely needs to use the results of [ s : aeinfinity ] instead of those of [ s : ahinfinity ] . in this case the ( mild ) difference with previous authors is that we obtain the analytic continuation and the resolvent estimates for the resolvent with exponential weights ( as in [ s : aeinfinity ] ) rather than with compactly supported cutoff functions . note however that in @xcite bruneau - petkov give a method for passing from cutoff resolvent estimates to weighted resolvent estimates in such a situation . in the special case where @xmath611 and @xmath612 we can obtain a resonant wave expansion as a corollary . indeed , we we have the resolvent estimate @xmath613 where @xmath614 is now @xmath615 for @xmath616 or its meromorphic continuation for @xmath617 . this follows from the substitution @xmath618 on the other hand , work of mazzeo - melrose @xcite and guillarmou @xcite ( see also @xcite ) shows that the weighted resolvent @xmath619 continues meromorphically to @xmath620 . from this the following resonant wave expansion follows . suppose @xmath53 solves @xmath621 for @xmath622 , with support disjoint from the convex obstacles in the case ( 3 ) above ( and with no restriction on the support in cases ( 1 ) and ( 2 ) ) . then @xmath623 the sum is taken over poles of @xmath614 , @xmath624 is the algebraic multiplicity of the pole at @xmath625 , and the @xmath626 are eigenstates or resonant states . the error term obeys the estimate @xmath627 for every @xmath628 and multiindex @xmath164 , uniformly over compact subsets of @xmath2 . note that the sum in is finite , thanks to the resonance free strip established by the estimate . this is a standard consequence of the resolvent estimate and the meromorphic continuation , by taking a fourier transform in time and then performing a contour deformation . see for example @xcite for a similar result , and @xcite for a similar result with an asymptotic extending to infinity in space . we sketch the proof here : see @xcite for more details . when @xmath629 , we can write @xmath630 where @xmath631 . the proof then proceeds by contour deformation from @xmath632 to @xmath633 . the residues at the poles of the resolvent produce the terms of the expansion in , and the resolvent estimate @xmath634 for any @xmath635 , justifies the deformation and controls the @xmath636 norm of the error ( on compact sets , or in suitably weighted spaces ) in terms of the @xmath637 norm of @xmath66 . the estimate can be derived from the @xmath416 estimate following the same procedure as in [ s : ahinfinity ] above . the case where @xmath638 and @xmath639 can be deduced similarly by differentiating the equation in @xmath211 , and then the general case follows by superposition of these two cases . in many settings better resolvent estimates are available in the physical half plane @xmath640 . more specifically , we obtain the following theorem ( see ( * ? ? ? * ( 1.1 ) ) and ( * ? ? ? * ( 1.17 ) ) for the corresponding resolvent estimates for the trapping model operators ) . [ t : logloss ] let @xmath0 be even and asymptotically hyperbolic , let @xmath587 , and let @xmath641 suppose @xmath24 has a normally hyperbolic trapped set in the sense of [ s : wz ] or a hyperbolic trapped set with @xmath594 as in [ s : nz ] . then for any @xmath8 there exist @xmath642 such that @xmath643 for @xmath396+i[0,\infty)$ ] , @xmath593 . in @xcite , bony - burq - ramond prove that for @xmath24 a semiclassical schrdinger operator on @xmath247 , the presence of a single trapped trajectory implies that @xmath644 } \|\chi r(\lambda)\chi\|,\ ] ] provided @xmath8 is @xmath160 on the projection of the trapped set . consequently , in that setting ( and probably in general ) , theorem [ t : logloss ] is optimal . from theorem [ t : logloss ] it follows by a standard @xmath645 argument as in @xcite that the schrdinger propagator exhibits local smoothing with loss : @xmath646 for any @xmath647 . in fact , the main resolvent estimate of @xcite follows from theorem [ t : main ] above , because the model operator near infinity , @xmath81 can be taken to be a nontrapping scattering schrdinger operator , for which the necessary resolvent and propagation estimates were proved in @xcite . moreover , burq - guillarmou - hassell @xcite show that when @xmath648 semiclassical resolvent estimates with logarithmic loss can be used to deduce strichartz estimates with no loss on a scattering manifold ( a manifold with asymptotically euclidean or asymptotically conic ends in a sense which generalizes that of [ s : aeinfinity ] ) , and the same result probably holds on the asymptotically hyperbolic spaces considered here . see also @xcite for more references and a discussion of the history and of recent developments in local smoothing and strichartz estimates . another possible application of the method is to give alternate proofs of cutoff resolvent estimates in the presence of trapping , where the support of the cutoff is disjoint from the trapping . as mentioned in the introduction , estimates of this type were proved by burq @xcite and cardoso and vodev @xcite and take the form @xmath649 for @xmath640 , where @xmath650 vanishes on the convex hull of the trapped set and is either compactly supported or suitably decaying near infinity . indeed a related method based on propagation of singularities was used in @xcite to prove such a result . 00 jean - franois bony , nicolas burq and thierry ramond . minoration de la rsolvante dans le cas captif . [ lower bound on the resolvent for trapped situations ] . _ c. r. math . _ 348(23 - 24):12791282 , 2010 .

we use _ hubble space telescope ( hst ) _ and ground - based imaging to study the multiple populations of 47 tuc , combining high - precision photometry with calculations of synthetic spectra . using filters covering a wide range of wavelengths , our _ hst _ photometry splits the main sequence into two branches , and we find that this duality is repeated in the subgiant and red - giant regions , and on the horizontal branch . we calculate theoretical stellar atmospheres for main sequence stars , assuming different chemical composition mixtures , and we compare their predicted colors through the _ hst _ filters with our observed colors . we find that we can match the complex of observed colors with a pair of populations , one with primeval abundance and another with enhanced nitrogen and a small helium enhancement , but with depleted c and o. we confirm that models of red giant and red horizontal branch stars with that pair of compositions also give colors that fit our observations . we suggest that the different strengths of molecular bands of oh , cn , ch and nh , falling in different photometric bands , are responsible for the color splits of the two populations . near the cluster center , in each portion of the color - magnitude diagram the population with primeval abundances makes up only of the stars , a fraction that increases outwards , approachng equality in the outskirts of the cluster , with a fraction averaged over the whole cluster . thus the second , he / n - enriched population is more concentrated and contributes the majority of the present - day stellar content of the cluster . we present evidence that the color - magnitude diagram of 47 tuc consists of intertwined sequences of the two populations , whose separate identities can be followed continuously from the main sequence up to the red giant branch , and thence to the horizontal branch . a third population is visible only in the subgiant branch , where it includes of the stars .

1109.0900

the presence of multiple stellar populations in globular clusters ( gcs ) has been widely established both by photometric and by spectroscopic studies . more than 40 years ago the red giant branch ( rgb ) of @xmath3 centauri ( ngc 5139 ) was found to have a photometric spread in color ( woolley 1966 ) , associated with a metallicity spread ( freeman & rodgers 1975 , norris & bessel 1975 ) , but the first real challenge to the traditional picture of gcs as simple stellar populations came from the discovery of abundance anomalies among stars within the same gcs ( e.g. , kraft 1979 ) . all galactic gcs studied so far show na - o anti - correlations ( e.g. , carretta et al . 2009a ) , indicative of contamination from products of proton - capture reactions at high temperature ( denisenkov & denisenkova 1989 , langer et al . moreover , the presence of such anti - correlations in unevolved main sequence ( ms ) stars ( e.g. , gratton et al . 2001 ; ramirez & cohen 2002 ) , which have not yet reached sufficiently high temperatures in their interiors , suggests that more than one generation of stars has formed within galactic gcs ( see gratton et al . 2004 for a review ) , possibly in combination with accretion onto low - mass stars of ejecta from either intermediate - mass stars ( dantona et al . 1983 ; renzini 1983 ) , fast - rotating massive stars ( e. g. , decressin et al . 2007 ) , or massive binaries ( de mink et al . 2009 ) . clear evidence of a complex star - formation history in gcs has come from high - precision hubble space telescope ( _ hst _ ) photometry , which showed unequivocally that the phenomenon of multiple sequences in the color - magnitude diagrams ( cmds ) of gcs is not confined to the special case of @xmath3 centauri ( for which see , e.g. , anderson 1997 , bedin et al . split or spread mss have been observed in ngc 2808 , 47 tucanae ( ngc 104 ) , and ngc 6752 ( piotto et al . 2007 , anderson et al . 2009 , milone et al . 2010 ) , while splits in the subgiant branches ( sgbs ) have been detected in ngc 1851 , ngc 6656 ( m22 ) , 47 tuc , and at least five other gcs ( milone et al . 2008 , marino et al . 2009 , anderson et al . 2009 , piotto 2009 ) . the splitting of the red - giant branch ( rgb ) has been observed in all the gcs studied to date with appropriate photometric bands ( e.g. , marino et al . 2008 ; yong et al . 2008 , lee at al . 2009 , lardo et al . the photometric investigations definitively confirm that it is common for gcs to contain multiple stellar populations . more recently it has been possible to connect the photometrically observed multiple sequences with differences of chemical composition among the stars in the same cluster . the aim of such studies has been to understand how successive generations of stars could have formed in a gc , and what could be inferred about the nature of the polluters from their chemical imprint on the second - generation stars . the first notable results were by piotto et al . ( 2005 ) , who showed that the bluer ms in @xmath3 centauri is more metal - rich than the redder one . the only way to reconcile the photometric and spectroscopic results is to assume that the bluer ms is strongly he - enhanced . other examples are given by marino et al . ( 2008 ) , who showed that the presence of two groups of stars with different c , n , na , and o content is at the root of the difference in the @xmath4 color of the rgb stars in ngc 6121 ( m4 ) . na - poor ( cn - weak ) giants define a sequence bluer than the one occupied by the na - rich ( cn - strong ) ones . yong et al . ( 2008 ) found that rgb spreads are present in a large number of gcs , when using the strmgren @xmath5 index , which is a powerful tracer of the n abundance . in general , stellar evolutionary models suggest that observed multiple mss should correspond to stellar populations with different he abundance , the bluer sequences having a higher helium abundance than primordial ( e.g. , norris 2004 ; dantona et al . 2005 ; piotto et al . 2005 , 2007 ; di criscienzo et al . 2010 , 2011 ) . according to this picture , the triple ms in ngc 2808 can also correspond to the three groups of stars with different oxygen and sodium abundances observed among the rgb stars ( carretta et al . 2006 ) , likely to come from three successive episodes of star formation . in fact , hydrogen burning at high temperatures through the cno cycle and subsequent proton captures result in a he enrichment , and at the same time in an enhancement of n , na , and al , and a depletion of c , o , and mg . this scenario has been nicely confirmed for ngc 2808 by bragaglia et al . ( 2010 ) , who measured chemical abundances of one star on the red ms and one on the blue ms , and found that the latter shows an enhancement of n , na , and al , and a depletion in c and mg . to date there are no spectroscopic studies of stars on the middle branch of the ms , however . finally , theoretical models have proposed that the sgb split observed in some gcs could be due to two stellar groups with either an age difference of 12 gyr or a different c+n+o content ( cassisi et al . 2008 , ventura et al . 2009 ) . in support of the chemical - content scenario , a bimodality in the @xmath6-process elements and in the cno has been detected in ngc 1851 and m22 ( yong et al . 2009 , marino et al . 2011a ) . summarizing the observational and theoretical scenarios : * split mss suggest helium enrichment , and also correspond to different groups of stars in the na - o anticorrelation ; * multiple sgbs might be ascribed to different total abundance of c+n+o or to different age ; * photometrically multiple rgbs might be due to differences in c , n , o content , via the different strengths of the corresponding molecular features ; * there appears to be no strong indication of any significant spread in [ fe / h ] ( except for a few clusters : @xmath3 centauri , m22 , terzan 5 , and ngc 2419 ) . multiple stellar populations with different helium abundance also offer an explanation for the complex , extended , and clumpy hb morphology exhibited by some clusters ( e.g. , dantona et al . 2005 , dantona & caloi 2008 , catelan , valcarce , & sweigart 2009 , gratton et al . a direct confirmation of a connection of the hb shape with the chemical content of the hb comes from recent work by marino et al.(2011b ) , who have found that stars on the blue side of the instability strip of the cluster m4 are na - rich and o - poor , whereas stars on the red hb are all na - poor . while the presence of multiple stellar populations in gcs as revealed by multiplicities of the ms , sgb , rgb , or hb has been clearly established in many clusters , efforts to unequivocally connect the various evolutionary stages of each individual stellar generation have met with only modest success so far , because individual sequences appear to merge and even cross each other in some parts of the cmd , strongly depending on the photometric bands used to build the cmd . connecting the various branches from the ms to the hb would greatly help to fully characterize each individual stellar generation , in terms of composition and age . in the present paper we attack this problem by applying high - precision _ hst _ photometry to the globular cluster 47 tucanae ( go-12311 , pi piotto ) . this is one of the gcs in the milky way where multiple stellar populations have recently been detected and studied both photometrically and spectroscopically . from the analysis of a large number of archival _ hst _ images of the inner @xmath7 arcmin , anderson et al . ( 2009 ) found that the sgb is spread in magnitude , with at least two distinct branches : a brighter one with an intrinsic broadening in luminosity , and a second one about 0.05 mag fainter that includes a small fraction of the stars . anderson et al . were also able to study the ms in a less crowded field 6 arcmin from the center and found an intrinsic broadening that increases towards fainter magnitudes . they interpreted the ms spread in terms of a variation of helium abundance of @xmath80.03 . di criscienzo et al . ( 2010 ) suggested that a spread in helium of @xmath8 might be responsible for both the luminosity spread of the bright sgb and the hb morphology , whereas an increase in the overall c+n+o abundance could be responsible for the faint sgb . in substantial agreement with the above estimates of the differences in helium content in 47 tuc , nataf et al . ( 2011 ) have estimated a helium difference of @xmath9 between two sub - populations of this cluster , based on the strength and luminosity of the rgb bump and on the luminosity of the hb . they also noted that the helium - rich population is more centrally concentrated . since the early seventies , spectroscopic investigations have shown that rgb stars in 47 tuc exhibit large star - to - star variations in cn band strength ( e.g. , mcclure & osborn 1974 , bell , dickens , & gustafsson 1975 ) , with two distinct groups of stars showing different cn content ( norris & freeman 1979 , briley 1997 ) ; this dichotomy is also present among ms stars ( cannon 1998 , harbeck et al . a na o anticorrelation has recently been studied by carretta et al . ( 2009a , b ) , with @xmath10% of the stars being na - poor and o - rich , while the remaining @xmath11% are depleted in oxygen and enhanced in sodium . this paper is organized as follows : in section [ data ] we describe the data and the reduction . section [ sec : ms ] reveals a split in the ms , and in section [ ycnoms ] we explore possible theoretical interpretations . sections [ sec : sgb ] , [ sec : rgb ] , and [ sec : hb ] return to the pursuit of multiple sequences along the sgb , the rgb , and the hb respectively , and also explore their interpretation . the spatial distribution of multiple stellar populations is investigated in section [ sec : rd ] , while in section [ sec : con ] we attempt to connect the multiple sequences that we have found along the ms , the sgb , the rgb , and the hb , and to trace the cmd of each of the two stellar generations . a summary and some final discussion follow in section [ sec : discussion ] . for our study of the stellar populations in 47 tuc , we used data sets from two different telescopes . for the crowded central regions of the cluster we used _ images taken with the wide field channel of the advanced camera for surveys ( acs / wfc ) and the uvis channel of wide field camera 3 ( wfc3/uvis ) , while to study the spatial distribution of the populations we made use of _ u _ , _ b _ , _ v _ , and _ i _ ground - based photometry from the data base of 856 original and archival ccd images from stetson ( 2000 ) . among them , 480 images were obtained with the wide - field imager of the eso / mpi 2.2 m telescope , and 200 with the 1.5 m telescope at cerro tololo inter - american observatory , while the remaining 176 images come from various other telescopes . these observations are described in detail in bergbusch & stetson ( 2009 ) . they were reduced following the protocol outlined in some detail by stetson ( 2005 ) and are calibrated on the landolt ( 1992 ) photometric system . table 1 summarizes the characteristics of the _ hst _ images that we used , while fig . [ footprint ] shows their footprints . the acs images , and a number of the wfc3 images , were archival , except that the images of go-12311 ( pi piotto ) were taken expressly for this project , and were crucial to its success . the acs / wfc images were reduced by using the software described in anderson et al . it consists of a package that analyzes all the exposures simultaneously to generate a catalog of stars over the whole field of view . stars are measured in each image independently by using for each filter a spatially variable point - spread - function model from anderson & king ( 2006 ) plus a `` perturbation psf '' that allows for the effects of focus variations . the photometry was put into the vega - mag system following the recipes of bedin et al . ( 2005 ) and using the encircled energy and zero points of sirianni et al . ( 2005 ) . star positions and fluxes in the wfc3 images were measured with software that is mostly based on img2xym_wfi ( anderson et al . 2006 ) ; this will be presented in a separate paper . star positions and fluxes were corrected for pixel area and geometric distortion by using the solution given by bellini & bedin ( 2009 ) and bellini , anderson & bedin ( 2011 ) , and were calibrated as in bedin et al . the work that we present here is based mainly on high - precision photometry , for which our next step was to select a high - quality sample of stars that are relatively isolated and have small photometric and astrometric errors , and are also well fit by the psf . for this we used the quality indices that our photometry software produces , in a procedure that is described in detail by milone et al . ( 2009 , sect . finally we corrected our photometry for some remaining position - dependent errors , due to small inadequacies in our psfs that were quite small but were different for each filter . since all the uses of our photometry would depend on colors , we generated the 36 colors that can be derived from our 9 filters . for each of these colors we drew the main - sequence ridgeline in the corresponding cmd ; then for each star we identified its 50 closest well - measured neighbors , and found their median color offset from the main - sequence ridgeline . since this constituted a good estimate of the systematic color error at the position of the target star , for that pair of filters we corrected the observed color of the star by that amount . [ tabdata ] as already mentioned in section [ introduction ] , the initial evidence that the ms of 47 tuc is not consistent with a single stellar population comes from the recent work by anderson et al . ( 2009 ) , who detected an intrinsic spread in the @xmath12 color ranging from @xmath13 mag near @xmath14 19.0 to @xmath8 mag around @xmath15 . our present data set , however , allowed us to study the main sequence with higher precision than was possible for anderson et al . with the data available at that time . an inspection of the large number of cmds that we obtain from the data set listed in table 1 showed that the multiple populations along the ms are best recognized and separated from photometry that combines @xmath16 with @xmath17 . the left - hand panel of fig . [ ms ] shows the hess diagram of @xmath18 vs. @xmath19 , after the quality selection and photometric corrections described in the previous section . we immediately note a widely spread rgb , a bimodal sgb , and a double ms . to examine these more closely , in the right - hand half of the figure we show a cmd that is zoomed around the upper ms , the sgb , and the start of the rgb . we defer discussion of the sgb and rgb to later sections , and concentrate here on the main - sequence morphology . the cmd in the right - hand panel suggests that the ms of 47 tuc is bimodal , in analogy with the multiple mss observed in @xmath3 centauri , ngc 2808 , and ngc 6752 . in 47 tuc , however , the majority of ms stars populate the blue component ( hereafter msb ) , while a small but significant fraction of the stars lie on a redder ms branch ( msa ) . the two sequences merge close to the turn - off , but on the ms the separation of the two components increases towards fainter magnitudes , from 0.08 mag at @xmath20 to 0.15 mag at @xmath21 as illustrated in more detail in figure [ msnor ] . such large , clear separations allow us to exclude any possibility that the ms split might be due to measuring errors . we note here that the `` verticalizing '' of the ms in fig . [ msnor ] is a process that we will carry out several times in the course of this paper , and we now explain it once and for all : we first designated msb as our target sequence , by means of a hand - drawn first approximation to its ridge line , and we also chose a limited color range around this line . we then put a spline through the median colors in successive short intervals of magnitude , and did an iterated sigma - clipping of outliers ; the result was a fiducial sequence for msb . the verticalization then consisted of subtracting from the color of each star the color of the fiducial sequence at the magnitude of that star . [ msnor ] allows us to estimate the fractions of msa and msb stars by assigning to each star a verticalized color ( left panel of the figure ) , as explained above . the right - hand panels of the figure show the color histograms for five magnitude bins ; in each bin we fitted the histogram with a pair of gaussians , shown in magenta and green . these colors will be used consistently hereafter , to distinguish these two sequences and their post - ms progeny . from the areas under the gaussians we estimate that 82% of the stars belong to msb and 18% to msa ; within the statistical uncertainties of these numbers they have the same values in each of the magnitude intervals . we also note that msa can not be ascribed to a sequence of binaries . the dashed line in fig . [ ms ] is the equal - mass - binary sequence that corresponds to the fiducial sequence of msb . interpreting the stars of msa as binaries would require making the outlandish hypothesis that about a fourth of the ms stars in 47 tuc are in binary systems with mass ratio in the narrow interval 0.70.8 . in addition , assuming that all msa stars are binaries would imply a binary fraction ( @xmath2215% ) , in sharp contrast with recent estimates of a 2% binary fraction by milone et al . ( 2008 , 2011 ) . in view of this , we expect that binaries do not affect the following discussion in any significant way . the double ms is also evident in cmds that use a different combination of magnitude and color , as shown in the @xmath17 vs. @xmath23 hess diagram and cmd of fig . it is important to note that in this color system the less populous ms component is _ bluer _ than the other ms stars . since these two cmds from the filter set f275w , f336w , f435w behave so differently , we construct the two - color diagram that is shown on the left side of figure [ 2c ] . there is a clear separation between the two populations , and in the right panel the stars on opposite sides of the dividing line have been colored green and magenta to identify stars of msa and msb , respectively . with high - accuracy photometric measurements in nine bands , 36 different cmds could be generated , if we were to use all possible combinations of magnitude and color . in the upper panels of figure [ mscriteria ] we show three of the most representative of these cmds , zoomed around the ms region , with each star colored green or magenta , according to its membership in msa or msb as shown in fig . [ 2c ] . in the bottom panels of fig . [ mscriteria ] the same cmds are replicated , and superimposed on the observed cmds are the fiducial ridge lines of msa and msb , derived for each cmd using the method described above . the fiducials for the two mss are then shown again in fig . [ fid275 ] ( @xmath24 vs. @xmath25 ) , fig . [ fid336 ] ( @xmath17 vs. @xmath26 or @xmath27 ) , and fig . [ fid814 ] ( @xmath28 vs. @xmath29 ) , where @xmath30 is indicated on the ordinate of each of row of figures . we note that msa is redder than msb in most of these cmds , but the color ordering of the two fiducial lines is otherwise in some of them : msa stars become bluer than msb stars in the @xmath26 colors ( fig . [ mscriteria ] , lower right panel , and fig . [ fid336 ] ) . surely the bizarre behavior of the two branches of the ms in our various cmds is telling us something about the physical origin of the split . in hopes of clarifying the complicated observational picture , we will focus on a critical subset of the data , then we will compare the observations against synthetic spectra that have been calculated for atmospheres with various chemical compositions . the challenge of choosing a critical subset from our mass of observational results is similar to the one faced by bellini et al . ( 2010 ) in their multicolor study of @xmath3 centauri ; we follow their lead , and look at the separation of the sequences as a function of the color index in which they are observed . figure [ at12 ] shows the separation of msa and msb in each of eight colors ; the figure also compares these observational separations of the sequences with calculated separations , from three theoretical scenarios that we will describe below . on the theoretical side , multiplicity of mss has been attributed to differences in helium content ( e.g. , norris 2004 , dantona et al . 2005 , piotto et al . 2005 , 2007 ) or in light - element abundances ( e.g. , sbordone et al . 2011 ) , so we will explore both of these possibilities , by making three different choices for the abundances of he , c , n , and o. we will calculate synthetic colors for msa and msb stars for each of these options , seeking the composition that best reproduces all the observed color differences between the two mss . all three of the options use the same abundance mixture for msa . for he in msa we choose the primordial he abundance , @xmath31 , and , following cannon et al . ( 1998 ) , we choose [ c / fe ] = 0.06 and [ n / fe ] = 0.20 , which are typical for cn - weak stars . as for [ o / fe ] , we take 0.40 , typical for first - generation stars ( carretta et al . 2009a , b ) . for msb stars we try three different options . in option i we assume that helium is the only cause of the ms split , and adopt @xmath32 . in option ii we keep helium the same in the two populations but instead change the light - element abundances , adopting the values listed in table 2 . finally , in option iii we adjust both helium and the light elements , again as given in table 2 . we chose to characterize the fiducial sequences by measuring the color difference between msa and msb at the reference magnitude of @xmath33 , that is , the magnitude level indicated by the horizontal lines in fig . [ fid814 ] . for an assumed distance modulus @xmath34 and a reddening @xmath35 , this corresponds to an absolute magnitude of @xmath36 . the adopted distance modulus and reddening are those that provide the best fitting of the isochrones to the data in the @xmath37 vs. @xmath12 plane , and are in agreement with those provided by gratton et al . ( 2003 ) and harris ( 1996 , dec . 2010 update ) . to determine the absorption in the f814w band we used the relations given by bedin et al . we assumed [ fe / h ] = @xmath380.75 and [ @xmath39/fe ] = 0.4 , in agreement with the values found by carretta et al . ( 2009 ) . to compare the observations against synthetic photometry , we adopted the isochrones from the teramo group ( pietrinferni et al . 2004 , 2006 ) specifically calculated for the populations listed in table 2 , then determined @xmath40 and @xmath41 for ms stars at @xmath36 . these temperatures and gravities were then used to calculate model atmospheres with the atlas12 code ( kurucz 2005 , castelli 2005 , sbordone 2005 , sbordone et al . 2007 ) , which allows us to use arbitrary chemical compositions . spectral synthesis from 2000 to 10000 was then performed using the synthe code ( kurucz 2005 , sbordone et al . 2007 ) , and the resulting synthetic spectra were integrated over the transmission of each of our nine filters to produce the synthetic magnitudes and colors . we did this separately for an msa star and for an msb star , using for the latter each of the three composition options that table 2 lists for msb . we concentrate here on the eight @xmath42 colors , comparing the synthetic msa @xmath38 msb color differences with the observed ones shown in figure [ at12 ] . the results of our synthetic - spectrum calculations for option i are shown in the figure as blue squares . at wavelengths longer than @xmath43 there is good agreement with the observed color differences between the two mss , but sizable discrepancies appear at shorter wavelengths , in particular in the f336w band , and we conclude that helium alone can not account for the observed ms split . the colors that result from option ii ( same helium but different cno proportions ) are plotted as gray triangles in figure [ at12 ] . again there is a strong discrepancy between the simulated and the observed color differences in the two near - uv bands ( f336w and f390w ) , but now in the opposite direction . turning instead to option iii , with differences in both helium and the cno elements , we see that the red asterisks in figure [ at12 ] are in fair agreement with all of the observed color differences . the agreement is somewhat poorer for the f275w filter , but note that we have not fine - tuned the composition differences between the two mss , and that uv colors can be very sensitive to the adopted mix of he and the cno elements . the effect of chemical composition differences on ms colors can be summarized as follows . the main effect of increased helium is to increase the temperatures , making all colors bluer . the effects of cno are more subtle , since a specific molecule affects some bands but not others . thus nitrogen affects the near - uv f336w and f390w bands via the nh band and cn bands , respectively , whereas oxygen affects the f275w band , via the oh band . this is illustrated in the top panel of figure [ spettri12 ] , showing the synthetic spectra of an msa and an msb star as calculated for option iii ( in green and magenta respectively ) . the difference between the two is shown in the middle panel , while the band - passes of our filters are plotted in the bottom panel . these results indicate that the observed color differences between msa and msb can be understood by assuming that msa corresponds to a first stellar generation , with primordial he , and o - rich / n - poor stars , whereas msb corresponds to a population that is enriched in he and n but depleted in o. this need for differences in both helium and cno to account for all the color differences fits quite well with nucleosynthesis expectations , as helium - enriched stellar regions are also inevitably oxygen - depleted and nitrogen - enriched , and vice versa for example , in the layers subject to the _ second dredge - up _ , at the beginning of the agb phase , or those subject to the so - called envelope - burning process , later during this phase ( renzini & voli 1981 , ventura & dantona 2009 ) . the first photometric evidence for multiple populations in 47 tuc came from the discovery of its bimodal sgb ( anderson et al . 2009 ) . these authors analyzed the large number of acs / wfc archival _ images of the core of 47 tuc , and found that the sgb is split into two distinct components : a brighter sgb showing an intrinsic broadening in the f475w magnitude , and a second one about 0.05 mag fainter , containing a minor fraction ( a few per cent ) of the stars . anderson et al . also examined the sgb in an outer field , @xmath44 arcmin west of the cluster center , finding a similar vertical spread of the sgb . these sgbs have been further investigated by piotto et al . ( 2011 ) , with a multi - band photometric analysis , showing that the magnitude difference between the faint and bright sgbs is almost constant in each of the nine _ hst _ filters they used , and finding a similar behavior in five other galactic gcs that have multiple sgbs . piotto et al . also found that the faint sgb makes up @xmath45 per cent of the stars in the central field of 47 tuc . di criscienzo et al . ( 2010 ) suggest that the spread in luminosity of the bright sgb can be accounted for by a small spread in helium ( @xmath46 ) , whereas they suggest that the minor population that makes up the faint sgb ( fsgb ) would be characterized by a small increase in c+n+o . in our multicolor set of cmds , the sgb region turns out to be unexpectedly complex . figure [ 3sgbs ] shows in the left panel the visible - light cmd of @xmath47 vs. @xmath12 , in which the fsgb component of anderson et al . , marked in red , is clearly visible as a separate sequence . in the right panels are the ultraviolet cmds , with @xmath24 vs. @xmath19 and @xmath24 vs. @xmath48 ; in them the same stars instead fall on top of the numerically dominant sequence , which in turn splits into two separate sequences , bringing the total to three recognizable sgbs . how can this be ? by analogy with what we did for the ms stars , in figure [ 2csgb ] we plot the uv two - color diagram of the sgb stars . here again we see a bimodal distribution , and we therefore repeat the points on the right , and draw a separator line . again by analogy , we color the stars on either side of the line green and magenta , respectively , and attach to the two regions the names sgba and sgbb . in this case , however , we also distinguish with red x s the stars that belonged to the fsgb in the left panel of fig . [ 3sgbs ] . the next step is to bring to bear on the sgb puzzle our full multicolor resources . in figure [ sgbs ] we show a 4 @xmath49 4 array of cmds , with a different magnitude in each row , plotted against a different color in each column . although these cmds are worth looking at one by one , taken together they add up to an _ embarras de richesse_. we therefore note for the reader the characteristics that strike us as systematic and significant : \(1 ) sgba stars share some similarities with msa : in the cmds that use the @xmath24-@xmath17 color they are on average redder than the other sgb stars , but when the @xmath48 color is used instead , the sgba stars are on average bluer than the bulk of sgb stars . no significant color difference between the two sgb groups is evident in the other color combinations . \(2 ) sgba stars are typically brighter than sgbb stars in @xmath17 ( note in particular the third row of panels ) . their being brighter than others in this band explains why they are redder than sgbb and fsgb in the @xmath19 color , and bluer than them in the @xmath48 color . no major systematic magnitude difference between sgba and the two other sgb groups is apparent in any of the other panels . \(3 ) in all of the cmds the fsgb stars follow the same trend as the bulk of the sgbb stars , with the fsgb running parallel to sgbb . the nature of two of the three sgbs is now clear . in the first and second cmds in the bottom row of fig . [ sgbs ] ( @xmath24 vs. @xmath19 and @xmath24 vs. @xmath50 ) the green and magenta sequences interchange their order , exactly as we saw for msa and msb in these same colors ( upper left and upper right cmds in fig . [ mscriteria ] ) . this justifies the name that we gave this pair of sequences , sgba and sgbb , since they appear to be the direct descendants of msa and msb . the sequence that we have marked in red establishes the existence of a third population in 47 tuc , but we see it only here on the sgb ; in other parts of the cmd we find no evidence of this third sequence . it is ironic that the very stars whose existence led us to study multiple populations in 47 tuc should be left as an anomaly that does not fit into the picture that will emerge at the end of the present paper . figure [ rgbshst ] shows the red - giant region of 47 tuc in three different cmds . both the @xmath24 vs. @xmath24-@xmath51 and the @xmath17 vs. @xmath48 diagrams suggest that the rgb is intrinsically broad ; the observed color spread of 0.10.2 mag is much larger than the photometric error , which for these bright rgb stars is less than 0.02 mag in color , and there is even some hint of a double color distribution . however , in the @xmath52 vs. @xmath53 cmd ( rightmost panel ) the rgb is quite narrow . to investigate this behavior further , we again turn to the @xmath54 vs. @xmath48 two - color diagram . in panel _ a _ of figure [ rgbhst2c ] we show a zoomed cmd , in which the rgb stars are indicated in black . b _ shows the two - color diagram of those stars . the distribution of stars in the diagram is clearly bimodal , and in panel _ c _ we arbitrarily draw a line separating two groups of stars , which we name rgba and rgbb , again color - coded green and magenta , respectively . the lower panels of the figure show the steps by which numerical fractions are derived for the rgba and rgbb populations . in accordance with the procedure introduced in sect . 3 , we derived a fiducial line for rgbb in the two - color diagram ; the line is shown in panel _ d_. in panel _ e _ we have subtracted from the color of each star the color of the fiducial line at the magnitude of the star , so as to verticalize the sequences . in panel _ f _ we plot a histogram of the verticalized colors , and fit it with gaussians to represent the rgba and rgbb populations . the result is that rgba represents @xmath55% of the total . ground - based photometry available for a field of view that goes out to 25 arcmin allows us to investigate the radial behavior of the populations . however , the lack of an equivalent to the f275w passband deprives us of our sharpest tool , and we must look for the best population discriminant among the ground - based passbands that are available . fortunately , ground - based @xmath56 , @xmath57 , @xmath58 , and @xmath59 are quite similar to our _ hst _ f336w , f435w , f555w and f814w bands . our procedure , then , was as follows : since our use of the f275w and f336w bands had separated the rgba and rgbb stars , we took those two samples of stars and examined their behavior in the many different diagrams that can be plotted using only the four bands that have ground - based equivalents , in order to see if we could find a surrogate for the two missing bands . as a result , we found that rgb stars separate into two groups when we plot @xmath60 against the combination @xmath17+@xmath61 , as is shown in figure [ rgbhst3ma ] . just to demonstrate the effectiveness of this separation , we use it to separate the stars of rgba and rgbb , instead of using the easier route offered by figure [ rgbhst2c ] . as in similar situations encountered earlier , we draw a fiducial line for rgbb , by putting a spline through the median colors in successive short intervals of magnitude , and doing an iterated sigma - clipping of outliers . we again subtract the color of the fiducial line from the color of each star , to produce the verticalized colors shown in the left panel of figure [ rgbhst3 mb ] . the corresponding two - gaussian fit to the color distributions of the resulting rgba and rgbb stars is shown in the right panel of the figure . finally , in figure [ 3mag ] , which is analogous to figure [ rgbhst2c ] , we use @xmath56 , @xmath57 , and @xmath59 magnitudes to separate rgba and rgbb stars in our entire ground - based field , in exactly the same way as we demonstrated with our _ hst _ photometry using the corresponding f336w , f435w , and f814w passbands . then we calculate the difference between the @xmath62 value of each star and the value of the fiducial line at the magnitude of the star , @xmath63 . this process makes the rgb vertical , and we show that in the left panel of figure [ hallgb ] . the corresponding histograms are plotted in the right panel of the figure , again fitted by two overlapping gaussians . having developed a method of using ground - based filters for separating populations , we now study the spatial distribution of the two rgb components , by dividing our ground - based field of view into three annuli , each containing about the same number of rgb stars , and applying the procedure described above to each group separately . the histograms of the @xmath64 distributions are shown in figure [ rdrgb ] . it is clear from this figure that rgbb is more centrally concentrated than rgba , with the fraction of rgba stars ranging from @xmath65 in the 1.73.5 arcmin bin , to @xmath66 at radial distances from 3.5 to 7.8 arcmin , up to @xmath67 for stars in the 7.825 arcmin bin . this result will be discussed in detail in section [ sec : rd ] , where we also compare the radial distributions of the two rgb components with those of ms and hb components . we note finally that we carefully analyzed the sgb and the ms in the @xmath57 vs. @xmath62 diagram , the same as we have just done for the rgb , but for those other regions we found no evidence for multiple sequences , presumably because of the larger errors in the ground - based photometry of those fainter stars . since the red giants are the brightest stars in 47 tuc , their spectroscopy has a long history . in the early seventies , large star - to - star cyanogen variations showed that the cluster is not chemically homogeneous ( e.g. , mcclure & osborn 1974 ; bell , dickens , & gustafsson 1975 ; hesser , hartwick , & mcclure 1977 ; norris 1978 ; hesser 1978 ) , and in a paper that set a standard of spectroscopic accuracy at that time , norris & freeman ( 1979 ) measured cn in 142 rgb stars of 47 tuc and found a bimodal distribution , with cn - strong stars more centrally concentrated . our figure [ norrisrgb ] is adapted from their figure 1 , and here we plot their values of the ddo c(4142 ) index ( which is sensitive to the blue cn bands ) against our @xmath58 magnitude . the dashed line separates the cn - strong and cn - weak stars , which for added emphasis we have indicated by magenta circles and green triangles , respectively . we have 80 rgb stars in common with norris & freeman , and in figure [ cn3rgb ] we have marked their cn - strong and cn - weak stars with distinctive symbols as defined above , in a repeat of our figure [ 3mag ] . note that most of the cn - strong stars belong to rgbb , and nearly all the cn - weak stars lie on rgba . it is obviously very tempting to associate the two groups of cn - weak and cn - strong stars with the two rgbs . more accurate measurements would be needed , however , to establish whether the few exceptions to this rule are real , or are due to errors in either the photometric or the spectroscopic measures . it is now well established that a chemical signature of multiple stellar populations in gcs is offered by the na - o anticorrelation that has been noticed in virtually every cluster observed so far . the na - o anticorrelation in rgb stars of 47 tuc has recently been studied by carretta et al . ( 2009a ) , and their data are shown in figure [ nao ] , where we have made an arbitrary division into na - poor / o - rich stars ( green triangles ) and na - rich / o - poor stars ( magenta circles ) . _ vs. @xmath68 cmd shown in figure [ naocmd ] gives photometric evidence for a spread in color that extends from the base of the rgb to its tip . we then identify stars from carretta et al . ( 2009a ) in our ground - based cmds of 47 tuc , and the two groups of stars defined in figure [ nao ] are found to segregate on the rgb as illustrated in figure [ naocmd ] , with na - rich / o - poor stars being systematically bluer in @xmath69 compared to the na - poor / o - rich stars , while mixing with them in the _ b _ vs. @xmath70 cmd , in analogy with the results of marino et al . ( 2008 ) for m4 . finally , in figure [ nao3 m ] we plot the na - poor / o - rich and na - rich / o - poor stars in the _ b _ vs. @xmath62 diagram and in the _ b _ vs. @xmath64 diagram , to investigate the abundances of na and o in the two rgbs . we find that rgba is populated mainly by na - poor / o - rich stars , while all the na - rich / o - poor stars belong to rgbb . we now turn to the horizontal branch of 47 tuc . figure [ hbhst ] shows the _ hst _ data in the same plots as were used for the other sequences . in order to point out the stars that we are studying here , we show in panel _ a _ a long - wavelength cmd , with the hb emphasized in black . in panel _ b _ we show the ultraviolet two - color diagram of these stars , with the two hb sequences hba and hbb identified as usual in panel _ c_. in panel _ d _ we show the same plot again , but with a fiducial line through the hbb locus . we verticalize the sequences in the usual way in panel _ e_. panel _ f _ shows the corresponding histogram , with the fit by two gaussians that are colored appropriately . perhaps quite unsurprisingly at this stage , the result that hba makes up @xmath55% of the hb stars is very similar to what we found for the rgb . to study the radial distribution of the two hb populations , we again use ground - based photometry . in panel _ a _ of figure [ hbgb ] we show the @xmath57 vs. @xmath70 cmd of all the stars that passed the selection criteria described in section [ data ] , and the selected hb stars are marked in black . b _ shows a zoom of the hb , in the same color system , and in panel _ c _ we plot the same stars in the @xmath57 vs. @xmath62 diagram , where the distribution shows some bimodality . the red dashed line is the fiducial line of the more populous hb component , and by analogy with the previous figure we refer to the lower - left stars as hba and the upper - right stars as hbb . the rectified @xmath57 vs. @xmath62 plot is shown in panel _ d _ , and the corresponding histogram is shown in panel _ it has been suspected for a long time that the hb of 47 tuc might contain multiple populations . norris & freeman ( 1982 ) measured the strengths of cn and ch bands in the spectra of 14 hb stars and concluded that their results were `` consistent with a dichotomy as found for the giants '' , which agrees with the result that we have just presented . they also noted that cn - weak stars are on average @xmath71 0.04 mag brighter in @xmath58 than cn - strong stars which is similar to what we see in panel _ e _ of figure [ hbhst ] and panel @xmath72 of figure [ hbgb ] . 1a from norris & freeman ( 1982 ) is reproduced here as fig . [ cnnf82 ] . we plotted their s(3839 ) index measurements for 14 hb stars as a function of the _ ( b - v ) _ color and drew the line that shows the dependence of s(3839 ) on effective temperature , as discussed by norris & freeman ( 1982 ) . the two groups of cn - weak and cn - strong stars defined by norris & freeman are plotted as green triangles and magenta circles , respectively . our ground - based catalog has in common nine out these fourteen stars . these stars are marked with black circles . among them three are cn - strong and six cn - weak stars . in fig . [ hbcn ] we show the position in the _ b _ vs. _ b - i _ cmd ( right panel ) and in the _ b _ vs. @xmath64 diagram the nine stars in common ( left panel ) . cn - strong stars are on average brighter than cn - weak ones in the _ b _ band . the right panel shows that cn - strong and cn - weak stars have @xmath64 values consistent to that of the bulk of the hbb and hba groups , respectively . an exception to this rule is given by a cn - weak star which has a _ b _ magnitude and @xmath64 value close to that of hbb stars . it is not clear if this anomalous position in the latter diagram is an intrinsic property of this star or is due to either photometric or spectroscopic errors . the connection between the hb morphology in gcs and the groups of stars with different abundances of light elements ( na , o , c , n ) is not a peculiarity of 47 tuc but has been also observed in the gc ngc 6121 ( m4 ) . as already mentioned in sect . [ introduction ] , this cluster hosts two groups of stars with different na and o content ( marino et al . 2008 ) , as well as a bimodal hb . marino et al . ( 2010 ) have recently measured oxygen and sodium for stars in the blue and the red hb segments of m4 , and found that the red hb is made by stars with low na and high o content , while blue hb stars are all na - enhanced and o - depleted ( possibly he - rich ) . these results on 47 tuc and ngc 6121 provide direct evidence that the hb morphology of these gcs is strictly related to the multiple stellar generations they host , and suggest that the multiple sequences discovered in the cmds of many gcs may be connected with the hb morphology . in order to better understand the origin of the multicolor distribution of sgb and hb stars , we present here an analysis similar to that described in section [ ycnoms ] for ms stars , using a number of synthetic spectra over the wavelength range of interest ( i. e. , @xmath73 ) . they were computed using the kurucz ( 1993 ) model atmospheres ( with the overshooting option switched off ) , and line lists from the kurucz cd - roms . although these lists may be incomplete , especially in the uv , and the models adopt the 1-dimensional approximation , the resulting model atmospheres are still useful for the present purposes , which consist of identifying the major spectroscopic features that can affect the photometry of stars in 47 tuc . in the rest of this discussion , we limit ourselves to consideration of differential effects between stars that should have very similar spectra unless their chemical compositions differ : a subgiant star with @xmath74 k and @xmath75 , and a red hb star with @xmath76 k and @xmath77 . one set of synthetic spectra was computed with the typical abundance pattern observed in metal - poor stars in the field , [ c / fe ] = [ n / fe ] = 0.0 , [ o / fe ] = 0.4 ; spectra computed with these parameters are called `` n - poor '' . they should mimic the composition of the stars that carretta et al.(2009a ) call primordial - generation . we computed another pair of spectra for the same atmospheric parameters , now with [ c / fe ] = @xmath380.2 , [ n / fe ] = + 1.3 , and [ o / fe ] = @xmath380.1 , but with the other element abundances the same as in the `` n - poor '' stars . we call this second group `` n - rich '' ; they should mimic the spectra of the stars that carretta et al . ( 2009a ) call second - generation . for both the `` n - poor '' and `` n - rich '' stars we assumed [ fe / h ] = @xmath380.75 and [ @xmath39/fe ] = 0.4 . the spectra were then integrated over the transmission of _ hst _ filters f275w , f336w , f390w , and f435w , to derive the fluxes expected in those bands . figure [ f : band ] compares a pair of synthetic spectra ( those corresponding to subgiant stars ) , and shows the transmissions of the filters . this figure shows that the differences between the `` n - poor '' and `` n - rich '' spectra are essentially due to different strengths of the molecular bands : the oh band ( in the wavelength range 26003200 ) , is stronger in `` n - poor '' stars and falls within the f275w band ; the nh band at @xmath71 3400 is stronger in the `` n - rich '' spectra and falls within the f336w band . the cn violet bands ( stronger in n - rich spectra ) at 3883 and 4216 , and the ch g - band ( stronger in n - poor spectra ) falls in the f390w band . the last two molecular bands are also within the f435w pass - band . as a consequence , the flux predicted for both the f336w and the f390w bands is smaller for the `` n - rich '' spectra than for the `` n - poor '' ones . the difference is greater for the f336w band , where it can be as much as 0.1 mag . the opposite holds for the f275w band . abundance variations of c , n , o elements do not appreciably affect the stellar flux for passbands at longer wavelengths . we can now compare our theoretical values of three uv colors with the observed ones : * @xmath19 : this color index is predicted to be larger ( redder ) for n - poor and smaller ( bluer ) for n - rich stars ; this is the combined effect of less oh absorption in the f275w band and more nh absorption in the f336w band in n - rich compared to n - poor stars . if n varies while he and mg do not , the predicted color difference for the abundances given above is 0.17 mag for the subgiant , and 0.19 mag for the red hb star . * @xmath48 : this index is larger ( redder ) for the n - rich and smaller ( bluer ) for the n - poor stars , again a result of stronger nh absorption in the f336w band . if there is again no difference in he and mg , the predicted color difference is 0.09 mag for the subgiant and 0.10 mag for the red hb star . * @xmath78 : this index is larger ( redder ) for the n - rich and smaller ( bluer ) for the n - poor star , again due to stronger nh and especially cn absorption in the f390w band , and less ch absorption in the f435w band . if there is no difference in he and mg , the predicted color difference is 0.05 mag for the subgiant and 0.07 mag for the red hb star . these predicted differences are indeed similar to those observed in the two - color diagram in panels _ b _ and _ c _ of fig . [ hbhst ] . it is of course a complication that additional differences are expected if the abundances of he and/or mg are also different . the f275w band includes the very strong resonance doublet of mg ii , so that the impact of mg is appreciable ( up to @xmath71 0.05 mag ) . the main effect of a helium difference is a change in the effective temperatures of the stars : n - rich stars that are also he - rich are expected to be warmer and bluer . even a small temperature difference has a quite dramatic effect on the uv bands : a difference of 100 k ( corresponding to a change of @xmath71 0.07 in @xmath79 ) makes @xmath19 bluer by a further @xmath71 0.2 mag , and more than offsets the difference in @xmath80 color between n - rich and n - poor stars . such a large difference is clearly excluded by the photometry of 47 tuc , which on the whole agrees quite well with only a minimal variation in helium ( @xmath81 0.015 ) , as suggested in sect . [ ycnoms ] in our attempt to explain the multicolor observations of the double ms . in the previous sections , we examined the radial gradients of the populations one sequence at a time . here , we put all the information together to develop a comprehensive picture of the cluster . since from an abundance perspective , most studies have focused on cn , we will frame the discussion in terms of that molecule . the spatial distribution of stellar populations with different cn in 47 tuc has been widely studied and debated in the literature , and little doubt remains concerning the presence of a significant radial gradient . on the basis of cn measurements of 142 rgb stars norris & freeman ( 1979 ) found that the cn - strong population is more centrally concentrated . but the same data were further analyzed by hartwick & mcclure ( 1980 ) , who found no evidence of differences in radial distribution of stars with different cn strength . norris & smith ( 1981 ) found cn - strong stars in the majority in the inner @xmath71 3 arcmin , a nearly equal fraction of cn - strong and cn - weak stars between @xmath71 3 and @xmath71 15 arcmin , and a predominance of cn - weak stars at larger radial distances . langer et al . ( 1989 ) concluded that because of the small sample size the results of norris & freeman ( 1979 ) should be considered inconclusive . finally , briley ( 1997 ) studied the radial distribution of the cn - strong and cn - weak populations on the basis of @xmath71 300 rgb stars with radial distances larger than 4 arcmin . he found that the relative numbers of cn - rich and cn - poor stars are roughly constant within @xmath71 13 arcmin of the cluster center , while the distribution exterior to 13 arcmin is clearly biased toward cn - poor stars , thus confirming the radial gradient first detected by norris & freeman ( 1979 ) . an obvious advantage of photometric measurements is that we can study more stars and therefore get better statistics . in the following we take advantage of the large size of our photometric catalogs to analyze the radial distributions of the multiple stellar sequences along the rgb and the hb . in section [ sec : rgb ] we used both _ hst _ and ground - based photometry to estimate the rgb population ratio in four radial intervals , while in section [ sec : hb ] we determined the fraction of stars in the two hb segments at three radial distances . the results for the radial distributions of hb and rgb stars are summarized in figure [ rdall ] , where we have plotted the fraction of hbb with respect to the total hb stars ( green triangles ) and the fraction of rgbb stars with respect to the total number of rgb stars ( red dots ) . for completeness we also show , ( as a blue square ) the fraction of msb stars with respect to the total number of ms stars . ( the ms split could be measured only at the center , where we have deep hst images ; the ground - based images are too shallow and crowded to allow the ms populations to be discerned ) . we find that in the central field the fraction of msb , rgbb , and hbb stars with respect to the total number of ms , rgb , or hb stars , respectively , is about 8082% for each . for the rgb and hb this fraction falls to about 60% in the outer parts of the cluster . thus the rgbb and hbb populations , which likely represent a second generation , appear to be more centrally concentrated . an integration of the @xmath82 population ratio , adopting a king model appropriate for 47 tuc , reveals that globally the _ first generation _ ( msa , sgba , rgba , hba ) accounts for @xmath1 of the present - day stellar content of the cluster , while the _ second generation _ ( msb , sgbb , rgbb , hbb ) accounts for the @xmath83 majority share of the cluster , in agreement with the fraction of first and second generation stars measured by carretta et al . ( 2009a ) on the basis of the na and o content . one final note : it should be remembered that the proportions that we have quoted refer only to the sum of the components that we have called @xmath84 and @xmath85 , which together make up only 92% of the total , the other 8% being the third population , which we glimpse only as the faint component of the sgb . in the other regions of the cmd the third population presumably makes up some small fraction of the parts that we call @xmath84 and @xmath85 . such a global predominance of the second generation sets strong constraints on scenarios for the formation of multiple populations in gcs , and actually for the formation of gcs _ tout court_. the fact that the second generation is much more centrally concentrated has also been observed in other gcs ( such as @xmath3 centauri , sollima et al . 2007 , bellini et al . 2009 ) and suggests that much of the first generation ( whatever it was ) might have been tidally stripped from the progenitor of 47 tuc . this is consistent with a similar suggestion for @xmath3 cen by bekki & norris ( 2006 ) . we note that hydrodynamic plus @xmath86-body simulations of the formation of multiple stellar populations in gcs predict a larger concentration of second generation stars in the cluster central regions ( dercole et al . 2008 , decressin et al . so far we have analyzed each evolutionary phase separately from the others , using the @xmath87 vs. @xmath80 two - color diagram to separate multiple stellar populations along each of the evolutionary phases : ms , sgb , rgb , and hb . the various cmds show that in our central _ hst _ field the stellar distribution is bimodal for all such evolutionary phases , with about 20% of the stars of each phase falling in the lower left part of the two - color diagram , and the remaining 80% populating the upper right part of the diagram . this behavior is summarized in fig . [ allsel ] , where we have used the cmd in the left panel to select five regions labeled r1r5 , respectively , marking the ms , the turn - off region , the sgb , the rgb , and the hb . in the right panels are plotted the two - color diagrams for stars in each of these cmd regions , with the black dash - dot lines drawn so as to separate the two groups of stars that are colored green and magenta , just as in all previous sections . the behavior of the two groups is strikingly similar from the ms all the way to the hb . each of these stages shows a similar 80/20 ratio of star numbers in the two groups , in the central field , and a similarly decreasing radial trend of this ratio . the most straightforward interpretation is that we are seeing two populations that wind their way , along nearly parallel paths , through the various successive stages of stellar evolution . this continuity is pictured in figure [ cmdpops ] , whose two cmds emphasize the crucial role that the f275w and f336w filters play in seeing such simplicity in what would otherwise have been a perplexing mlange of details . finally , the main properties of the two populations are summarized in table [ popolazioni ] . we have analyzed a large set of _ hst _ and ground - based images of the galactic globular cluster ngc 104 ( 47 tuc ) in nine photometric bands , finding multiple sequences throughout the various cmds , from the main sequence all the way to the horizontal branch . exploiting this wealth of _ hst _ data to investigate the behavior of the multiple populations as seen in several different combinations of magnitudes and colors , we found that among the rainbow of possible cmds , those involving the f275w and f336w filters are particularly effective in separating components of otherwise entangled cluster populations . taking a cue from _ hst _ , we were able to construct a color system based on the @xmath56 band that exhibited a similarly effective separation of the populations in ground - based data . we found that the distribution of stars along the ms , sgb , rgb , and hb was in every case bimodal in the @xmath87 vs. @xmath48 two - color diagram , and we finally put together the groups of stars that we had separated in this way , so as to draw a continuous connection between their successive evolutionary phases . near the cluster center all evolutionary phases split into two near - parallel sequences , with the richer one making up about 80% of the cluster stars , and the poorer one the remaining 20% . wide - field ground - based photometry allowed us to identify and separate the two sub - populations at larger radial distances from the cluster center , with the result that the majority population is more centrally concentrated , but its relative fraction decreases outward and approaches 50/50 in the outskirts of the cluster . globally , the majority population accounts for @xmath83 of the whole population of 47 tuc , most of the remainder consisting of the minority population . radial gradients in the stellar populations of this cluster have been known for a long time , with a cn - strong population more centrally concentrated than the cn - weak one . this suggests that the numerically dominant population is cn - strong . along these same lines , we used cn band strengths and na and o abundances that are available from the literature for some rgb and hb stars of both populations , to investigate their chemical content . it appears that the more populous rgbb and hbb sequences consist of cn - strong / na - rich / o - poor stars , while the bulk of the cn - weak / na - poor / o - rich stars belong to the numerically poorer rgba and hba . on the theoretical side , we calculated synthetic spectra of main - sequence stars with different chemical compositions , derived the corresponding colors for our filter set , and compared them with the observed colors in the two distinct populations . the colors of the minority population are well reproduced by stars with primordial helium abundance and an oxygen - rich / nitrogen - poor composition that is typical of halo stars of metallicity similar to that of 47 tuc . on the other hand , the colors of the majority population stars are well reproduced by a composition in which nitrogen is greatly enhanced , along with a slightly increased helium , while carbon and oxygen are depleted . synthetic spectra for rgb and hb stars confirm this result . the most straightforward interpretation of these differences is that the minority population is the remnant of the first stellar generation , which formed out of the interstellar medium of its time , and shared its chemical composition . the chemical composition of the majority population , by contrast , carries the signatures of cno and proton - capture processing at high temperatures , such as depletion of oxygen in favor of nitrogen and sodium , accompanied by helium enhancement . therefore the majority population should be regarded as the second stellar generation of 47 tuc , which formed out of material that had been partly processed through stars of the first generation . since both populations share the same iron abundance , one can exclude from the enrichment history of the second - generation material any significant contribution by massive stars , exploding as core - collapse supernovae . this leaves intermediate - mass stars of the first generation as the obvious candidates for having processed the material that is now incorporated into the second generation , a view that is indeed widely entertained in the literature . furthermore , it is certainly striking that the second generation is today more populous than the first . this sets strong constraints on the nature of the progenitor of the present cluster 47 tuc , and on the initial mass of the first generation , which must have been one to two orders of magnitude more massive than the portion that is still bound to the cluster . finally , we must again remind the reader that besides the two populations that we have called first and second , we have found unmistakable evidence for the presence of a third population , including some @xmath2 of the stars , which we can distinguish it only on the sgb , but it appears to have the abundance distribution of the `` b '' population . this remains as a strident reminder that the stellar populations in 47 tuc are more complex than the two - generation picture that we have just sketched . , s.l . , and a.p . acknowledge partial support by prin miur 20075tp5k9 and prin inaf " formation and early evolution of massive star clusters . , s.c . , and a.r . acknowledge partial support by asi under the program asi - inaf i/016/07/0 . j. a. and i. r. k. acknowledge support from stsci grant go-12311 .

for an absolutely continuous ( integer - valued ) r.v . of the pearson ( ord ) family , we show that , under natural moment conditions , a stein - type covariance identity of order holds ( cf . [ goldstein and reinert , _ j. theoret . probab . _ * 18 * ( 2005 ) 237260 ] ) . this identity is closely related to the corresponding sequence of orthogonal polynomials , obtained by a rodrigues - type formula , and provides convenient expressions for the fourier coefficients of an arbitrary function . application of the covariance identity yields some novel expressions for the corresponding lower variance bounds for a function of the r.v . , expressions that seem to be known only in particular cases ( for the normal , see [ houdr and kagan , _ j. theoret . probab . _ * 8 * ( 1995 ) 2330 ] ; see also [ houdr and prez - abreu , _ ann . probab . _ * 23 * ( 1995 ) 400419 ] for corresponding results related to the wiener and poisson processes ) . some applications are also given .

1007.3662

for an r.v . @xmath0 with density @xmath2 , mean @xmath3 and finite variance @xmath4 , goldstein and reinert @xcite showed the identity ( see also @xcite ) @xmath5,\ ] ] which holds for any absolutely continuous function @xmath6 with a.s . derivative @xmath7 such that the right - hand side is finite . in ( [ eq1.1 ] ) , @xmath8 is defined to be the r.v . with density @xmath9 , @xmath10 . identity ( [ eq1.1 ] ) extends the well - known stein identity for the standard normal @xcite ; a discrete version of ( [ eq1.1 ] ) can be found in , for example , @xcite , where the derivative has been replaced by the forward difference of @xmath11 . in particular , identities of the form ( [ eq1.1 ] ) have many applications to variance bounds and characterizations @xcite , and to approximation procedures @xcite . several extensions and applications can be found in @xcite . in @xcite , the ( continuous ) pearson family is parametrized by the fact that there exists a quadratic @xmath12 such that @xmath13 typically , the usual definition of a pearson r.v . is related to the differential equation @xmath14 , with @xmath15 being a polynomial of degree at most 2 . in fact , the set - up of ( [ eq1.2 ] ) ( including , e.g. , the standard uniform distribution with @xmath16 ) will be the framework of the present work and will hereafter be called `` the pearson family of continuous distributions '' . it is easily seen that under ( [ eq1.2 ] ) , the support of @xmath0 , @xmath17 , must be an interval , say @xmath18 with @xmath19 , and @xmath20 remains strictly positive for @xmath21 . clearly , under ( [ eq1.2 ] ) , the covariance identity ( [ eq1.1 ] ) can be rewritten as @xmath22={\mathbb{e}}[q(x)g'(x)].\ ] ] it is known that , under appropriate moment conditions , the functions @xmath23,\qquad x\in(r , s),\ k=0,1,\ldots , m\ ] ] ( where @xmath24 can be finite or infinite ) are orthogonal polynomials with respect to the density @xmath2 so that the quadratic @xmath20 in ( [ eq1.2 ] ) generates a sequence of orthogonal polynomials by the rodrigues - type formula ( [ eq1.4 ] ) . in fact , this approach is related to the sturm liouville theory , @xcite , section 5.2 ; see also @xcite . in the present paper , we provide an extended stein - type identity of order @xmath1 for the pearson family . this identity takes the form @xmath25= { \mathbb{e}}\bigl [ q^k(x)g^{(k)}(x)\bigr],\ ] ] where @xmath26 is the @xmath1th derivative of @xmath11 ( since @xmath27 , ( [ eq1.5 ] ) for @xmath28 reduces to ( [ eq1.3 ] ) ) . identity ( [ eq1.5 ] ) provides a convenient formula for the @xmath1th fourier coefficient of @xmath11 , corresponding to the orthogonal polynomial @xmath29 in ( [ eq1.4 ] ) . for its proof , we make use of a novel `` rodrigues inversion '' formula that may be of some interest in itself . an identity similar to ( [ eq1.5 ] ) holds for the discrete pearson ( ord ) family . application of ( [ eq1.5 ] ) and its discrete analog yields the corresponding lower variance bounds , obtained in section [ sec4 ] . the lower bound for the @xmath30 distribution , namely , @xmath31\ ] ] ( cf . @xcite ) and the corresponding one for the normal@xmath32 distribution @xcite , @xmath33,\ ] ] are particular examples ( examples [ ex4.1 ] and [ ex4.5 ] ) of theorems [ th4.1 ] and [ th4.2 ] , respectively . both ( [ eq1.6 ] ) and ( [ eq1.7 ] ) are particular cases of the finite form of bessel s inequality and , under completeness , they can be extended to the corresponding parseval identity . in section [ sec5 ] , we show that this can be done for a fairly large family of r.v.s , including , of course , the normal , the poisson and , in general , all the r.v.s of the pearson system which have finite moments of any order . for instance , when @xmath0 is @xmath34 , inequality ( [ eq1.7 ] ) ( and identity ( [ eq1.5 ] ) ) can be strengthened to the covariance identity @xmath35= \sum_{k=1}^{\infty } \frac{(\sigma^2)^{k}}{k ! } { \mathbb{e}}\bigl[g_1^{(k)}(x)\bigr ] { \mathbb{e}}\bigl[g_2^{(k)}(x)\bigr],\ ] ] provided that for @xmath36 , @xmath37 , @xmath38 , @xmath39 and that @xmath40 ^ 2<\infty$ ] . similar identities hold for poisson , negative binomial , beta and gamma distributions . these kinds of variance / covariance expressions may sometimes be useful in inference problems see , e.g. , the applications [ ap5.1 ] and [ ap5.2 ] at the end of the paper . in order to simplify notation , we assume that @xmath0 is a non - negative integer - valued r.v . with mean @xmath41 . we also assume that there exists a quadratic @xmath42 such that @xmath43 where @xmath44 is the probability function of @xmath0 . relation ( [ eq2.1 ] ) describes the discrete pearson system ( ord family ) @xcite . let @xmath45 be the forward difference operator defined by @xmath46=g(x+1)-g(x)$ ] and @xmath47=\delta[\delta^{k-1}[g(x)]]$ ] ( @xmath48\equiv g$ ] , @xmath49 ) . we also set @xmath50}(x)=q(x)q(x+1)\cdots q(x+k-1)$ ] ( with @xmath51}\equiv1 $ ] , @xmath52}\equiv q$ ] ) . we first show some useful lemmas . [ le2.1 ] if @xmath53 for @xmath54 and @xmath55 \delta^{k - j}[g(x)]|&<&\infty \qquad\mbox{for } j=0,1,\ldots , k , \\ \label{eq2.3}\lim_{x\to\infty } \delta^j [ h(x - j ) ] \delta^{k - j-1}[g(x)]&=&0\qquad \mbox{for } j=0,1,\ldots , k-1,\end{aligned}\ ] ] then @xmath56g(x ) \nonumber \\[-8pt ] \\[-8pt ] \nonumber & & \quad= \sum_{x=0}^{\infty } h(x)\delta^k[g(x)].\end{aligned}\ ] ] we have @xmath57 & = & \lim_{n\to\infty } \sum_{x=0}^n h(x ) \bigl(\delta^{k-1}[g(x+1)]-\delta ^{k-1}[g(x)]\bigr ) \\ & = & \lim_{n\to\infty } \biggl [ h(n+1)\delta^{k-1}[g(n+1)]-\sum_{x=0}^{n+1 } \delta[h(x-1 ) ] \delta^{k-1}[g(x ) ] \biggr ] \\ & = & \lim_{n\to\infty } h(n+1)\delta^{k-1}[g(n+1)]-\sum_{x=0}^{\infty } \delta[h(x-1 ) ] \delta^{k-1}[g(x ) ] \\ & = & -\sum_{x=0}^{\infty } \delta[h(x-1 ) ] \delta^{k-1}[g(x)].\end{aligned}\ ] ] by the same calculation , it follows that @xmath58 \delta^{k - j}[g(x)]=(-1)^{j+1 } \sum_{x=0}^{\infty } \delta^{j+1 } [ h(x - j-1 ) ] \delta^{k - j-1}[g(x)]\ ] ] for any @xmath59 . [ le2.2 ] for each @xmath60 , there exist polynomials @xmath61 , @xmath62 , such that the degree of each @xmath63 is at most @xmath64 and @xmath65}(x - n)p(x - n)\bigr ] = q^{[n - i]}(x - n+i)p(x - n+i)q_{i , n}(x),\qquad i=0,1,\ldots , n.\hspace*{26pt}\end{aligned}\ ] ] moreover , the leading coefficient ( i.e. , the coefficient of @xmath66 ) of @xmath67 is given by @xmath68 , where an empty product should be treated as @xmath69 . for @xmath70 , the assertion is obvious and @xmath71 . for @xmath72 , the assertion follows from the assumption ( [ eq2.1 ] ) with @xmath73 , @xmath74 . for the case @xmath75 , the assertion will be proven using ( finite ) induction on @xmath64 . indeed , for @xmath76 , ( [ eq2.5 ] ) holds with @xmath77 . assuming that the assertion holds for some @xmath78 and setting @xmath79}(x - n)p(x - n)$ ] , it follows that @xmath80 \\ & = & \delta[h_{n - i}(x ) q_{i , n}(x ) ] \\ & = & \delta\bigl [ q(x - n+i)p(x - n+i)\bigl(q^{[n - i-1]}(x - n+i+1)q_{i , n}(x)\bigr)\bigr ] \\ & = & q(x - n+i+1)p(x - n+i+1 ) \delta\bigl[q^{[n - i-1]}(x - n+i+1)q_{i , n}(x)\bigr ] \\ & & { } + \bigl(\mu-(x - n+i+1)\bigr)p(x - n+i+1)q^{[n - i-1]}(x - n+i+1)q_{i , n}(x),\end{aligned}\ ] ] where the obvious relation @xmath81=(\mu-(x+1))p(x+1)$ ] , equivalent to ( [ eq2.1 ] ) , has been used with @xmath82 in place of @xmath83 . moreover , @xmath84}(x - n+i+1)q_{i , n}(x)\bigr]\\ & & \quad= q^{[n - i-1]}(x - n+i+2)\delta[q_{i , n}(x ) ] + q_{i , n}(x)\delta\bigl[q^{[n - i-1]}(x - n+i+1)\bigr]\end{aligned}\ ] ] and @xmath85}(x - n+i+1)\bigr ] = q^{[n - i-2]}(x - n+i+2 ) \bigl(q(x)-q(x - n+i+1)\bigr)\ ] ] ( where , for @xmath86 , the right - hand side of the above should be treated as @xmath87 ) . observing that @xmath88}(x - n+i+2)\bigl(q(x)-q(x - n+i+1)\bigr ) \\ & & \quad = q^{[n - i-1]}(x - n+i+1)\bigl(q(x)-q(x - n+i+1)\bigr)\end{aligned}\ ] ] ( which is also true in the case where @xmath86 ) , the above calculations show that ( [ eq2.5 ] ) holds with @xmath89,\ ] ] where @xmath90 is a linear polynomial or a constant . from the above recurrence , it follows immediately that @xmath91 , @xmath92 ; this , combined with the fact that @xmath93 , yields the desired result . it is easy to see that under ( [ eq2.1 ] ) , the support of @xmath0 , @xmath94 , is a finite or infinite integer interval . this integer interval will be denoted by @xmath95 . here , the term `` integer interval '' means that `` if @xmath96 and @xmath97 are integers belonging to @xmath95 , then all integers between @xmath96 and @xmath97 also belong to @xmath95 '' . [ le2.3 ] for each @xmath98 define the functions @xmath99 , @xmath100 , by the rodrigues - type formula @xmath101}(x - k)p(x - k)\bigr ] \nonumber \\[-8pt ] \\[-8pt ] \nonumber & = & \frac{1}{p(x)}\sum_{j=0}^k ( -1)^{k - j}{k\choose j}q^{[k]}(x - j)p(x - j).\end{aligned}\ ] ] we then have the following : each @xmath29 is a polynomial of degree at most @xmath1 , with @xmath102 . let @xmath11 be an arbitrary function defined on @xmath95 and , if @xmath95 is infinite , assume in addition that the functions @xmath11 and @xmath103}(x)p(x)$ ] satisfy the requirements and of lemma [ le2.1 ] . then , @xmath104}(x)|\delta^{k}[g(x)]|\bigr]<\infty\ ] ] and the following identity holds : @xmath105={\mathbb{e}}\bigl[q^{[k]}(x)\delta^{k}[g(x)]\bigr].\ ] ] part ( a ) follows from lemma [ le2.2 ] since @xmath106 . for part ( b ) , assume first that @xmath107 , that is , the support @xmath95 is either @xmath108 or @xmath109 for a positive natural number @xmath110 . in the unbounded case , ( [ eq2.7 ] ) follows by an application of lemma [ le2.1 ] to the functions @xmath111}(x)p(x)$ ] and @xmath11 since they satisfy the conditions ( [ eq2.2 ] ) and ( [ eq2.3 ] ) . for the bounded case , it follows by ( [ eq2.1 ] ) that @xmath112 is a zero of @xmath20 so that @xmath113 , where , necessarily , @xmath114 . thus , in the case where @xmath115 , @xmath50}(x)=0 $ ] for all @xmath100 so that the right - hand side of ( [ eq2.7 ] ) vanishes . on the other hand , the left - hand side of ( [ eq2.7 ] ) ( with @xmath29 given by ( [ eq2.6 ] ) ) is @xmath116}(x - j)p(x - j)$ ] and it is easy to verify that for all @xmath117 and all @xmath118 , the quantity @xmath50}(x - j)p(x - j)$ ] vanishes . thus , when @xmath119 , ( [ eq2.7 ] ) holds in the trivial sense @xmath120 . for @xmath121 , the left - hand side of ( [ eq2.7 ] ) equals @xmath122}(x ) p(x)g(x+j)\\ & & \quad= \sum_{j=0}^k(-1)^{k - j } \pmatrix{k\cr j } \sum_{x=0}^{n - j } q^{[k]}(x)p(x)g(x+j ) \\ & & \quad= \sum_{x=0}^{n } q^{[k]}(x ) p(x)\sum_{j=0}^{\min\{k , n - x\}}(-1)^{k - j } \pmatrix{k\cr j}g(x+j ) \\ & & \quad= \sum_{x=0}^{n - k } q^{[k]}(x ) p(x ) \sum_{j=0}^k ( -1)^{k - j } \pmatrix{k\cr j}g(x+j ) , \ ] ] where we have made use of the facts that @xmath123 for @xmath54 and @xmath50}(x)=0 $ ] for @xmath124 . thus , ( [ eq2.7 ] ) is proved under the assumption @xmath107 . for the general case , where the support of @xmath0 is either @xmath125 or @xmath126 , it suffices to apply the same arguments to the r.v . @xmath127 . [ th2.1 ] suppose that @xmath0 satisfies and has @xmath128 finite moments for some @xmath129 . the polynomials @xmath29 , @xmath130 , defined by , then satisfy the orthogonality condition @xmath131=\delta_{k , m } k!{\mathbb{e}}\bigl[q^{[k]}(x)\bigr]\prod_{j = k-1}^{2k-2}(1-j\delta),\qquad k , m=0,1,\ldots , n , \ ] ] where @xmath132 is kronecker s delta . let @xmath133 . first , observe that @xmath50}$ ] is a polynomial of degree at most @xmath134 and @xmath135 a polynomial of degree at most @xmath136 . note that the desired result will be deduced if we can apply ( [ eq2.7 ] ) to the function @xmath137 ; this can be trivially applied when @xmath95 is finite . in the remaining case where @xmath95 is infinite , we have to verify conditions ( [ eq2.2 ] ) and ( [ eq2.3 ] ) of lemma [ le2.1 ] for the functions @xmath138}p$ ] and @xmath137 , or , more generally , for any polynomial @xmath11 of degree less than or equal to @xmath1 . since the case @xmath139 is obvious ( @xmath140 ) , we assume that @xmath141 . observe that ( [ eq2.2 ] ) is satisfied in this case because of the assumption @xmath142 . indeed , from lemma [ le2.2 ] , @xmath143}(x - j)p(x - j)]=p(x ) q^{[k - j]}(x)q_{j , k}(x - j+k)$ ] , where @xmath144 is of degree at most @xmath145 and @xmath146}$ ] is of degree at most @xmath147 so that their product is a polynomial of degree at most @xmath148 , while @xmath149 $ ] is a polynomial of degree at most @xmath145 . on the other hand , since @xmath150 is eventually decreasing and @xmath151 , we have , as @xmath152 , that @xmath153 hence , @xmath154 . for @xmath155 , we have , from lemma [ le2.2 ] , that @xmath156 \delta^{k - j-1}[g(x)]=q(x)p(x)r(x ) , \ ] ] where @xmath157}(x+1)q_{j , k}(x - j+k)\delta^{k - j-1}[g(x ) ] \ ] ] is a polynomial of degree at most @xmath158 . this , combined with the above limit , verifies ( [ eq2.3 ] ) and a final application of ( [ eq2.7 ] ) completes the proof . the covariance identity ( [ eq2.7 ] ) enables the calculation of the fourier coefficients of any function @xmath11 in terms of its differences @xmath159 $ ] , provided that conditions ( [ eq2.2 ] ) and ( [ eq2.3 ] ) are fulfilled for @xmath11 and @xmath111}(x)p(x)$ ] . since this identity is important for the applications , we state and prove a more general result that relaxes these conditions . the proof of this result contains a novel inversion formula for the polynomials obtained by the rodrigues - type formula ( [ eq2.6 ] ) . [ th2.2 ] assume that @xmath0 satisfies and has @xmath134 finite moments , and suppose that for some function @xmath11 defined on @xmath95 , @xmath160}(x ) |\delta^{k}[g(x)]|\bigr]<\infty . \ ] ] then , @xmath161 and the covariance identity holds . moreover , provided that @xmath162 and @xmath0 has @xmath158 finite moments , the following `` rodrigues inversion '' formula holds : @xmath163}(x)p(x ) = \frac{1}{(k-1)!}\sum_{y = x+1}^{\infty } ( y - x-1)_{k-1}p_k(y ) p(y),\qquad x=0 , 1,\ldots , \ ] ] where , for @xmath10 , @xmath164 , @xmath165 and @xmath166 . relation ( [ eq2.7 ] ) is obvious when @xmath167 . also , the assertion follows trivially by lemma [ le2.3](b ) if @xmath95 is finite . assume , now , that @xmath162 and that @xmath95 is unbounded , fix @xmath168 and consider the function @xmath169 it is easily seen that the pair of functions @xmath170 and @xmath111}(x)p(x)$ ] satisfy ( [ eq2.2 ] ) and ( [ eq2.3 ] ) and also that @xmath171=i(x = s)$ ] . thus , by ( [ eq2.7 ] ) , we get ( [ eq2.9 ] ) ( cf . formulae ( [ eq3.6 ] ) and ( [ eq3.7 ] ) , below , for the continuous case ) , where @xmath29 is defined by ( [ eq2.6 ] ) and the series converges from lemma [ le2.3](a ) and the fact that @xmath0 has @xmath158 finite moments ( note that @xmath158 finite moments suffice for this inversion formula ) . since @xmath29 is a polynomial and the left - hand side of ( [ eq2.9 ] ) is strictly positive for all large enough @xmath83 , it follows that @xmath172 for all large enough @xmath173 . using ( [ eq2.9 ] ) , the assumption on @xmath159 $ ] is equivalent to the fact that @xmath174| \sum_{y = x+1}^{\infty } ( y - x-1)_{k-1}p_k(y ) p(y)<\infty \ ] ] and arguments similar to those used in the proof of theorem [ th3.1](b ) below show that @xmath174| \sum_{y = x+1}^{\infty } ( y - x-1)_{k-1}|p_k(y)| p(y)<\infty . \ ] ] therefore , we can interchange the order of summation , obtaining @xmath175}(x)\delta^{k}[g(x)]\bigr ] & = & \frac{1}{(k-1 ) ! } \sum_{x=0}^\infty \delta^{k}[g(x ) ] \sum_{y = x+1}^{\infty } ( y - x-1)_{k-1}p_k(y ) p(y ) \\ & = & \frac{1}{(k-1)!}\sum_{y=1}^\infty p_k(y ) p(y ) \sum_{x=0}^{y-1 } ( y - x-1)_{k-1 } \delta^{k}[g(x ) ] \\ & = & { \mathbb{e}}[p_k(x ) g(x ) ] , \ ] ] where @xmath176,\qquad x=0,1,\ldots , \ ] ] and where an empty sum should be treated as @xmath87 . taking forward differences , it follows that @xmath177=\delta^{k}[g(x)]$ ] so that @xmath178 , where @xmath179 is a polynomial of degree at most @xmath180 , and the desired result follows from the orthogonality of @xmath29 to polynomials of degree lower than @xmath1 . this completes the proof . the orthogonality of polynomials ( [ eq1.4 ] ) has been shown , for example , in @xcite ; see also @xcite . for our purposes , we review some details . the induction formula , as in ( [ eq2.5 ] ) , here takes the form @xmath181=q^{k - i}(x)f(x)q_{i , k}(x),\qquad i=0,1,\ldots , k , \ ] ] with @xmath182 , where @xmath183 is a polynomial of degree at most @xmath64 and the explicit recurrence for @xmath183 is @xmath184 this immediately implies that @xmath185 , as in the discrete case ; see also @xcite . the covariance identity , as in ( [ eq2.7 ] ) , here takes the form ( after repeated integration by parts ; cf . @xcite ) @xmath186 = { \mathbb{e}}\bigl[q^k(x)g^{(k)}(x)\bigr ] , \ ] ] provided that the expectations are finite and @xmath187 \\[-8pt ] \nonumber & & \quad=\lim_{x\to s-}q^{i+1}(x)f(x)q_{k - i-1,k}(x ) g^{(i)}(x)=0,\qquad i=0,1,\ldots , k-1 \ ] ] ( here , @xmath188 ; that the support is an interval follows from ( [ eq1.2 ] ) ) , where @xmath183 are the polynomials defined by the recurrence above . obviously , an alternative condition , sufficient for ( [ eq3.3 ] ) ( and , hence , also for ( [ eq3.2 ] ) ) , is @xmath189 \\[-8pt ] \nonumber & & \quad i=0,1,\ldots , k-1 , j=0,1,\ldots , k - i-1 . \ ] ] assuming that @xmath0 has @xmath134 finite moments , @xmath162 , it is seen that ( [ eq3.4 ] ) , and hence ( [ eq3.2 ] ) , is fulfilled for any polynomial @xmath11 of degree at most @xmath1 . for example , for the upper limit , we have @xmath190 so that ( [ eq3.4 ] ) trivially holds if @xmath191 ; also , if @xmath192 , then ( [ eq3.4 ] ) follows from @xmath193 as @xmath194 , which can be shown by observing that @xmath195 is eventually decreasing , positive and , by the assumption of finite @xmath134th moment , satisfies @xmath196 . therefore , the explicit orthogonality relation is @xmath197=\delta_{k , m } k!{\mathbb{e}}[q^{k}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta),\qquad k , m=0,1,\ldots , n , \ ] ] where @xmath132 is kronecker s delta , provided that @xmath0 has @xmath128 finite moments . the proof follows by a trivial application of ( [ eq3.2 ] ) to @xmath198 , for @xmath133 ( cf . . it should be noted , however , that the condition ( [ eq3.4 ] ) or ( [ eq3.3 ] ) imposes some unnecessary restrictions on @xmath11 . in fact , the covariance identity ( [ eq3.2 ] ) ( which enables a general form of the fourier coefficients of @xmath11 to be constructed in terms of its derivatives ) holds , in our case , in its full generality ; the proof requires the novel inversion formula ( [ eq3.6 ] ) or ( [ eq3.7 ] ) below , which may be of some interest in itself . [ th3.1 ] assume that @xmath0 satisfies and consider the polynomial @xmath99 defined by , where @xmath188 . if @xmath0 has @xmath158 finite moments , then the following `` rodrigues inversion '' formula holds : @xmath199 if @xmath0 has @xmath134 finite moments and @xmath200 , then @xmath201 and the covariance identity holds . \(a ) let @xmath202 , @xmath203 be the left - hand side and right - hand side , respectively , of ( [ eq3.6 ] ) . it is easy to see that the integral @xmath203 is finite ( this requires only @xmath158 finite moments ) . moreover , expanding @xmath204 in the integrand of @xmath203 according to newton s formula , it follows that @xmath205 , @xmath206 , and , thus , by the definition ( [ eq1.4 ] ) , @xmath207 vanishes identically in @xmath18 . therefore , @xmath208 is a polynomial of degree at most @xmath180 . by ( [ eq3.1 ] ) , @xmath209 for all @xmath210 because @xmath211 as @xmath212 and , for the case @xmath213 , @xmath214 as @xmath215 since the @xmath216th moment is finite , @xmath217 is of degree at most @xmath218 and @xmath195 is increasing and positive in a neighborhood of @xmath219 . similarly , using the fact that @xmath29 is a polynomial of degree at most @xmath1 and observing that @xmath220 for all @xmath210 , @xmath221 ( again , @xmath158 finite moments suffice for this conclusion ) , it follows that @xmath222 for all @xmath210 . this proves that @xmath208 vanishes identically in @xmath18 and ( [ eq3.6 ] ) follows . finally , ( [ eq3.7 ] ) follows from ( [ eq3.6 ] ) and ( [ eq3.2 ] ) with @xmath223 ( the validity of ( [ eq3.2 ] ) for polynomials @xmath11 of degree at most @xmath180 can be shown directly , using repeated integration by parts , as above ) . \(b ) suppose that @xmath162 ; otherwise , since @xmath140 , we have nothing to show . since @xmath224={\mathbb{e}}[p_k(x)p_0(x)]=0 $ ] from ( [ eq3.5 ] ) , either @xmath99 vanishes identically for @xmath21 ( in which case , ( [ eq3.2 ] ) trivially holds ) or , otherwise , it must change sign at least once in @xmath18 . assume that @xmath29 is not identically zero and consider the change - sign points of @xmath29 in @xmath18 , say , @xmath225 ( of course , @xmath226 because @xmath29 is a polynomial of degree at most @xmath1 ) . fix a point @xmath227 in the finite interval @xmath228 $ ] and write , with the help of ( [ eq3.6 ] ) and ( [ eq3.7 ] ) , @xmath229\nonumber\\ & & \quad= \frac{(-1)^k}{(k-1)!}\int_r^{\rho } \bigl|g^{(k)}(x)\bigr| \int_{r}^x ( x - y)^{k-1}p_k(y ) f(y)\,\mathrm{d}y \,\mathrm{d}x \\ & & \qquad { } + \frac{1}{(k-1)!}\int_{\rho}^s \bigl|g^{(k)}(x)\bigr| \int_{x}^s ( y - x)^{k-1}p_k(y ) f(y)\,\mathrm{d}y \,\mathrm{d}x.\nonumber \ ] ] because of the assumption on @xmath11 , both integrals on the right - hand side of ( [ eq3.8 ] ) are finite . we wish to show that we can change the order of integration in both integrals in the right - hand side of ( [ eq3.8 ] ) . this will follow from fubini s theorem if it can be shown that @xmath230 and similarly for the other integral in ( [ eq3.8 ] ) . since @xmath231 for all @xmath21 ( see ( [ eq1.2 ] ) ) , it follows that @xmath232 for all @xmath21 as well . thus , from ( [ eq3.7 ] ) , we get @xmath233 on the other hand , @xmath99 does not change sign in the interval @xmath234 and , hence , @xmath235 for all @xmath236 , showing that @xmath237 \\[-8pt ] \nonumber & & \quad = \int_{\rho_m}^s \bigl|g^{(k)}(x)\bigr| \int_{x}^s ( y - x)^{k-1 } p_k(y ) f(y)\,\mathrm{d}y \,\mathrm{d}x . \ ] ] by the above considerations , it follows that @xmath238 } h(x):=\inf_{x\in[\rho,\rho_m ] } \int_{x}^s ( y - x)^{k-1 } p_k(y ) f(y)\,\mathrm{d}y > 0 , \\ \label{eq3.13 } s(\rho)&:=&\sup_{x\in[\rho,\rho_m ] } s(x ) : = \sup_{x\in[\rho,\rho_m]}\int_x^{\rho_m } ( y - x)^{k-1 } \\ \label{eq3.14}d(\rho)&:=&\sup_{x\in[\rho,\rho_m]}{d}(x):=\sup_{x\in[\rho,\rho_m ] } \int_{\rho_m}^s ( y - x)^{k-1 } p_k(y ) f(y)\,\mathrm{d}y < \infty\end{aligned}\ ] ] because the three positive functions @xmath239 , @xmath240 and @xmath241 defined above are obviously continuous and @xmath83 lies in the compact interval @xmath242 $ ] ( note that @xmath243 by ( [ eq3.10 ] ) ) . now , from the inequalities @xmath244 and @xmath245 , @xmath246 $ ] , we conclude that @xmath247 combining ( [ eq3.11 ] ) and ( [ eq3.15 ] ) , we see that there exists a finite constant @xmath248 ( take , for example , @xmath249 ) such that @xmath250 and , thus , by ( [ eq3.9 ] ) , we can indeed interchange the order of integration in the second integral in the right - hand side of ( [ eq3.8 ] ) . similar arguments apply to the first integral . by the above arguments and by interchanging the order of integration in both integrals in the right - hand side of ( [ eq3.8 ] ) ( with @xmath26 in place of @xmath251 ) , we obtain @xmath252 = { \mathbb{e}}[p_k(x ) g(x ) ] , \ ] ] where @xmath253 differentiating ( [ eq3.17 ] ) or ( [ eq3.18 ] ) @xmath1 times , it is easily seen that @xmath254 so that @xmath255 is a polynomial of degree at most @xmath180 and the desired result follows by ( [ eq3.16 ] ) and the orthogonality of @xmath29 to polynomials of degree lower than @xmath1 . this completes the proof of the theorem . a simple application of theorem [ th2.2 ] leads to the following lower variance bound . [ th4.1 ] fix @xmath256 and assume that @xmath0 satisfies and has @xmath128 finite moments . then , for any function @xmath11 satisfying @xmath257}(x)|\delta^{k}[g(x)]| \bigr]<\infty \qquad\mbox{for } k=0,1,\ldots , n , \ ] ] the bound @xmath258}(x)\delta ^{k}[g(x)]]}{k!{\mathbb{e}}[q^{[k]}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta ) } \ ] ] holds with equality if and only if @xmath11 is a polynomial of degree at most @xmath259 . assume that @xmath260<\infty$ ] ( otherwise , we have nothing to show ) . by theorem [ th2.1 ] , the polynomials @xmath261 } ; k=0,1,\ldots,\min\{n , n\}\}$ ] form an orthonormal basis of all polynomials with degree up to @xmath259 , where @xmath262 is the cardinality of @xmath95 . observing that the @xmath1th fourier coefficient for @xmath11 is , by ( [ eq2.7 ] ) , ( [ eq2.8 ] ) and theorem [ th2.2 ] , @xmath263}{{\mathbb{e}}^{1/2}[p_k^2(x ) ] } = \frac{{\mathbb{e}}[q^{[k]}(x)\delta^{k}[g(x)]]}{(k!{\mathbb{e}}[q^{[k]}(x ) ] \prod_{j = k-1}^{2k-2}(1-j\delta))^{1/2}},\qquad k\leq \min\{n , n\ } , \ ] ] the desired result follows by an application of the finite form of bessel s inequality . it is worth mentioning here the similarity of the lower variance bound ( [ eq4.2 ] ) with the poincar - type ( upper / lower ) bound for the discrete pearson family , obtained recently in @xcite , namely @xmath264}(x ) ( \delta^{k}[g(x)])^2\bigr]\biggr)\geq0 . \ ] ] the following examples can be verified immediately . [ ex4.1]if @xmath0 is @xmath30 , then @xmath265 so that @xmath266 and ( [ eq1.6 ] ) follows from ( [ eq4.2 ] ) ( see also @xcite ) . moreover , the equality in ( [ eq1.6 ] ) holds if and only if @xmath11 is a polynomial of degree at most @xmath259 . [ ex4.2 ] for the @xmath267 distribution , @xmath268 so that @xmath266 and @xmath269}(x)]=(n)_k p^k(1-p)^k$ ] . thus , ( [ eq4.2 ] ) yields the bound @xmath270\bigr ] \ ] ] with equality only for polynomials of degree at most @xmath259 . note that there is equality if @xmath271 . [ ex4.3 ] for the negative @xmath272 with @xmath273 , @xmath274 ( [ eq2.1 ] ) is satisfied with @xmath275 . thus , @xmath266 , @xmath269}(x)]=(1-p)^k [ r]_k /p^{2k}$ ] and ( [ eq4.2 ] ) produces the bound @xmath276_k } { \mathbb{e}}^2\bigl [ [ r+x]_k \delta^{k}[g(x)]\bigr ] \ ] ] ( in the above formulae and elsewhere in the paper , @xmath277_n = x(x+1)\cdots(x+n-1)$ ] if @xmath129 and @xmath277_0=1 $ ] ) . [ ex4.4 ] for the discrete uniform@xmath278 , ( [ eq2.1 ] ) is satisfied with @xmath279 ; thus , @xmath280 and ( [ eq4.2 ] ) entails the bound ( which is an identity if @xmath281 ) @xmath282_k(n - x)_k \delta^{k}[g(x)]\bigr].\ ] ] the lower variance bound for the continuous pearson system is stated in the following theorem ; its proof , being an immediate consequence of ( [ eq3.2 ] ) , ( [ eq3.5 ] ) , theorem [ th3.1 ] and a straightforward application of the finite form of bessel s inequality ( cf . the proof of theorem [ th4.1 ] above ) , is omitted . [ th4.2 ] assume that @xmath0 satisfies and has finite moment of order @xmath128 for some fixed @xmath129 . then , for any function @xmath11 satisfying @xmath283<\infty$ ] , @xmath130 , we have the inequality @xmath284}{k!{\mathbb{e}}[q^{k}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta ) } \ ] ] with equality if and only if @xmath11 is a polynomial of degree at most @xmath259 . some examples now follow . [ ex4.5 ] if @xmath0 is @xmath34 , then @xmath285 and ( [ eq4.4 ] ) yields the houdr kagan variance bound ( [ eq1.7 ] ) ( see @xcite ) , under the weaker assumptions @xmath286 , @xmath130 ; equality holds if and only if @xmath11 is a polynomial of degree at most @xmath259 . [ ex4.6 ] if @xmath0 is @xmath287 with density @xmath288 , @xmath289 , then @xmath290 , @xmath291=[a]_k/\lambda^{2k}$ ] and ( 4.4 ) entails the bound @xmath292_k } { \mathbb{e}}^2\bigl [ x^k g^{(k)}(x)\bigr ] \ ] ] with equality only for polynomials of degree at most @xmath259 . [ ex4.7 ] for the standard uniform density , @xmath16 , @xmath280 , @xmath293=(k!)^2/(2^k ( 2k+1)!)$ ] and we get the bound @xmath294 \ ] ] with equality only for polynomials of degree at most @xmath259 . similar inequalities hold for all beta densities . of course , in the above three examples , the corresponding orthogonal polynomials ( [ eq1.4 ] ) are the well - known hermite , laguerre and jacobi ( legendre ) , respectively , so that one can alternatively obtain the results using explicit expressions for the variance and generating functions for the polynomials ( see , e.g. , @xcite ) ; this is also the case for the discrete polynomials corresponding to examples [ ex4.1][ex4.4 ] ( namely , charlier , krawtchouck , meixner and hahn , respectively ) . however , as will become clear from the following example , the considerations given here do not only simplify and unify the calculations , but also go beyond the classical polynomials . [ ex4.8 ] if @xmath0 follows the @xmath295 distribution ( student s @xmath296 with @xmath110 degrees of freedom ) with density @xmath297 then it is well known that @xmath0 has only @xmath298 finite integral moments . however , for @xmath299 , @xmath0 satisfies ( [ eq1.2 ] ) and its quadratic @xmath300 has @xmath301 . thus , ( [ eq4.4 ] ) applies for sufficiently large @xmath110 ( see also @xcite for a poincar - type bound corresponding to ( [ eq4.3 ] ) ) . to this end , it suffices to calculate @xmath302= \biggl(\frac{n}{n-1}\biggr)^k \prod_{j=1}^{k}\biggl(1+\frac{1}{n-2j}\biggr),\qquad k\leq ( n-1)/2 , \ ] ] and @xmath303 . theorem [ th4.2 ] yields the ( non - classic ) bound @xmath304}{k!n^k ( n - k)_k\prod_{j=1}^{k}(1 + 1/(n-2j))},\qquad n\leq\frac{n-1}{2 } , \ ] ] with equality only for polynomials of degree at most @xmath259 . it seems that it would be difficult to work with the explicit forms of the corresponding orthogonal polynomials , obtained by ( [ eq1.4 ] ) . note that a similar bound can be easily obtained for the fisher snedecor @xmath305 distribution and that schoutens @xcite has obtained the corresponding stein s equation , useful in approximating the @xmath295-distribution . the main application of the present article ( section [ sec4 ] ) presents a convenient procedure for approximating / bounding the variance of @xmath306 when @xmath11 is smooth enough and when @xmath0 is `` nice '' enough ( pearson ) . although the procedure is based on the corresponding orthonormal polynomials , @xmath307 $ ] , the main point is that we do not need explicit forms for @xmath29 . all we need are the fourier coefficients of @xmath11 , @xmath308 $ ] , but , due to the identities ( [ eq3.2 ] ) and ( [ eq2.7 ] ) , the fourier coefficients can be simply expressed in terms of the quadratic @xmath217 and the derivatives / differences of @xmath11 when @xmath0 belongs to the pearson / ord family , that is , when ( [ eq1.2 ] ) or ( [ eq2.1 ] ) is satisfied . a natural question thus arises : _ is it true that the @xmath259th partial sums , given by the right - hand sides of and , converge to the variance of @xmath306 as @xmath309 _ ? more generally , assume that we want to calculate @xmath310 , when @xmath0 is an r.v . with finite moments of any order . let @xmath3 be the probability measure of @xmath0 , that is , @xmath3 satisfies @xmath311=\pr(x\leq x)$ ] , @xmath10 , and assume that the support of @xmath3 , @xmath312 , is not concentrated on a finite number of points ( otherwise , the following considerations become trivial ) . it is well known that there exists an orthonormal polynomial system ( ops ) @xmath313 , which can be obtained by an application of the gram schmidt orthonormalization process to the real system @xmath314 . each ( real ) polynomial @xmath315 is of degree ( exactly ) @xmath1 and satisfies the orthonormality condition @xmath316=\int_{{\mathbb{r } } } \phi_i(x)\phi_j(x ) \,\mathrm{d}\mu(x)=\delta_{ij } $ ] . the members of @xmath317 are uniquely determined by the moments of @xmath0 ( of course , any element of @xmath317 can be multiplied by @xmath318 ) . for any function @xmath319 ( i.e. , with finite variance ) , we first calculate the fourier coefficients @xmath320=\int_{{\mathbb{r}}}\phi_k(x)g(x)\,\mathrm{d}\mu(x),\qquad k=0,1,2,\ldots , \ ] ] and then use the well - known bessel inequality , @xmath321 to obtain the desirable lower variance bound . clearly , theorems [ th4.1 ] and [ th4.2 ] just provide convenient forms of the @xmath259th partial sum in ( [ eq5.2 ] ) for some particularly interesting cases ( pearson / ord system ) . it is well known that the sum in the right - hand side of ( [ eq5.2 ] ) is equal to the variance ( for all @xmath319 ) if and only if the ops @xmath317 is complete in @xmath322 ( and , thus , it is an orthonormal basis of @xmath322 ) . this means that the set of real polynomials , @xmath323 $ ] , is dense in @xmath322 . if this is the case , then bessel s inequality , ( [ eq5.2 ] ) , is strengthened to parseval s identity , @xmath324 the following theorem summarizes and unifies the above observations . [ th5.1 ] assume that @xmath0 has probability measure @xmath3 and finite moments of any order . if @xmath0 satisfies and if @xmath325 , then can be written as @xmath326 } { k!{\mathbb{e}}[q^{k}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta)},\ ] ] provided that the polynomials are dense in @xmath322 and that @xmath327<\infty,\qquad k=0,1,2,\ldots.\ ] ] similarly , if @xmath0 satisfies , if @xmath3 is not concentrated on a finite integer interval and if @xmath319 , then yields the identity @xmath328}(x)\delta ^k[g(x)]]}{k!{\mathbb{e}}[q^{[k]}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta)},\ ] ] provided that the polynomials are dense in @xmath322 and that @xmath329}(x)|\delta^k[g(x)]|\bigr]<\infty,\qquad k=0,1,2,\ldots.\ ] ] it should be noted that ( [ eq5.6 ] ) is always true ( and reduces to a finite sum ) if @xmath0 belongs to the discrete pearson ( ord ) family and the support of @xmath3 is finite . in this case , the sum adds zero terms whenever @xmath269}(x)]=0 $ ] . on the other hand , it is well known ( due to m. riesz ) that the polynomials are dense in @xmath322 whenever @xmath3 is determined by its moments ; see @xcite or @xcite , page 45 . an even simpler sufficient condition is when @xmath3 has a finite moment generating function at a neighborhood of zero , that is , when there exists @xmath330 such that @xmath331 a proof can be found in , for example , @xcite . since condition ( [ eq5.8 ] ) can evidently be checked for all pearson distributions , we include an alternative proof in the . taking into account the above , we have the following covariance representation . [ th5.2 ] assume that @xmath0 has a finite moment generating function at a neighborhood of zero . if @xmath0 satisfies and if @xmath332 , @xmath36 , then @xmath333= \sum_{k=1}^{\infty } \frac{{\mathbb{e}}[q^{k}(x)g_1^{(k)}(x ) ] \cdot{\mathbb{e}}[q^{k}(x)g_2^{(k)}(x ) ] } { k!{\mathbb{e}}[q^{k}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta ) } , \ ] ] provided that for @xmath36 , @xmath334<\infty,\qquad k=0,1,2,\ldots . \ ] ] similarly , if @xmath0 satisfies , then @xmath335= \sum_{k=1}^{\infty } \frac{{\mathbb{e}}[q^{[k]}(x)\delta ^k[g_1(x ) ] ] \cdot{\mathbb{e}}[q^{[k]}(x)\delta^k[g_2(x)]]}{k!{\mathbb{e}}[q^{[k]}(x)]\prod_{j = k-1}^{2k-2}(1-j\delta ) } , \ ] ] where each term with @xmath269}(x)]=0 $ ] should be treated as zero , provided that for @xmath36 , @xmath336}(x)|\delta^k[g_i(x)]|\bigr]<\infty,\qquad k=0,1,2,\ldots . \ ] ] let @xmath337 $ ] and @xmath338 $ ] be the fourier coefficients of @xmath339 and @xmath340 . it is then a standard inner product property in the hilbert space that @xmath341=\sum_{k=1}^{\infty } \alpha_k\beta_k$ ] . substituting , for example , @xmath342=\frac{{\mathbb{e}}[p_k(x)g_1(x)]}{{\mathbb{e}}^{1/2}[p_k^2(x ) ] } = \frac{{\mathbb{e}}[q^{[k]}(x)\delta^k[g_1(x ) ] ] } { ( k!{\mathbb{e}}[q^{[k]}(x ) \prod_{j = k-1}^{2k-2}(1-j\delta)])^{1/2}}\ ] ] ( and similarly for the continuous case and for @xmath343 ) , we obtain ( [ eq5.9 ] ) and ( [ eq5.11 ] ) . [ ap5.1 ] assume that @xmath344 is a random sample from @xmath345 , , with probability function @xmath346 and let @xmath347 be the complete sufficient statistic . the uniformly minimum variance unbiased estimator , umvue , of @xmath348 is then given by ( see @xcite ) @xmath349 since no simple form exists for the variance of @xmath350 , the inequalities ( [ eq4.3 ] ) have been used in @xcite in order to prove asymptotic efficiency . however , @xmath0 is negative @xmath351 and @xmath352=\frac{(-1)^{k-1 } ( k-1)!}{[\nu+x]_k},\qquad k=1,2,\ldots , \ ] ] so that one finds from ( [ eq5.6 ] ) the exact expression ( cf . example [ ex4.3 ] ) @xmath353 observe that the first term in the series , @xmath354 , is the cramr rao lower bound . the above series expansions are in accordance with the corresponding results on bhattacharyya bounds given in @xcite ; these results are also based on orthogonality and completeness properties of bhattacharyya functions , obtained by seth @xcite . similar series expansions for the variance can be found in @xcite . next , we present a similar application for the exponential distribution . [ ap5.2 ] assume that @xmath344 is a random sample from @xmath365 , @xmath366 , with density @xmath367 , @xmath289 , and let @xmath347 be the complete sufficient statistic . we wish to obtain the umvue of @xmath368 and its variance . setting @xmath369 , we find that @xmath370 where @xmath371 is euler s constant . therefore , @xmath372 is unbiased and it follows that the umvue of @xmath368 is of the form @xmath373=-\log(x)-\gamma+\sum _ { j=1}^{\nu-1 } \frac{1}{j}. \ ] ] since @xmath0 follows a @xmath374 distribution and @xmath375 , we obtain from ( [ eq5.4 ] ) ( cf . example [ ex4.6 ] ) the formula @xmath376 which is quite similar to the formula for @xmath377 in application [ ap5.1 ] . ( once again , the first term in the series , @xmath378 , is the cramr rao bound . ) moreover , the series @xmath379 can be simplified in a closed form . indeed , observing that @xmath380 and taking into account the identity @xmath381_{\nu}}-\frac{1}{[k+1]_{\nu}}\biggr ) , \qquad \nu , k=1,2,\ldots , \ ] ] we have @xmath382_{\nu}}-\frac{1}{[k+1]_{\nu } } \biggr)=\frac{1}{\nu^2 } \ ] ] so that @xmath383 finally , using the last expression and the fact that @xmath384 , we get the inequalities @xmath385 , that is , @xmath386 which shows that @xmath387 is asymptotically efficient . _ a completeness proof under ( _ . it is well known that , under ( [ eq5.8 ] ) , @xmath0 has finite moments of any order so that the ops exists ( and is unique ) . from the general theory of hilbert spaces , it is known that @xmath317 is a basis ( i.e. , it is complete ) if and only if it is total , that is , if and only if there does not exist a non - zero function @xmath319 such that @xmath11 is orthogonal to each @xmath315 . therefore , it suffices to show that if @xmath319 and if @xmath388=0\qquad \mbox{for all } k=0,1,\ldots , \ ] ] then @xmath389 . since each @xmath315 is a polynomial with non - zero leading coefficient , ( [ eqa.1 ] ) is equivalent to @xmath390=\int_{{\mathbb{r}}}x^n g(x ) \,\mathrm{d}\mu(x)=0 , \qquad n=0,1,\ldots , \ ] ] and it thus suffices to prove that if ( [ eqa.2 ] ) holds , then @xmath391 . since , by assumption , the functions @xmath392 and @xmath393 ( @xmath394 ) belong to @xmath322 , it follows from the cauchy schwarz inequality that @xmath395<\infty$ ] for @xmath394 and , thus , that @xmath396\leq{\mathbb{e}}[\mathrm{e}^{-tx}|g(x)|]+{\mathbb{e}}[\mathrm{e}^{tx}|g(x)|]<\infty,\qquad |t|<t_0/2 . \ ] ] from beppo levi s theorem and ( [ eqa.3 ] ) it follows that @xmath397<\infty , \qquad |t|<t_0/2 , \ ] ] and , therefore , from fubini s theorem and ( [ eqa.2 ] ) , we get @xmath398= { \mathbb{e}}\biggl [ \sum_{n=0}^{\infty } \frac{t^n x^n}{n ! } g(x)\biggr]= \sum_{n=0}^{\infty } \frac{t^n}{n ! } { \mathbb{e}}[{x^n}g(x)]=0 , \qquad |t|<t_0/2 . \ ] ] write @xmath399 , @xmath400 and observe that from ( [ eqa.2 ] ) with @xmath70 , @xmath401 . it follows that @xmath402 , say , where @xmath403 . clearly , @xmath404 because both @xmath405 and @xmath406 are dominated by @xmath407 and , by assumption , @xmath408 . we assume now that @xmath409 and we shall obtain a contradiction . under @xmath409 , we can define two borel probability measures , @xmath410 and @xmath411 , as follows : @xmath412 by definition , both @xmath413 and @xmath411 are absolutely continuous with respect to @xmath3 , with radon nikodym derivatives @xmath414 since , by ( [ eqa.4 ] ) , @xmath415 it follows that the moment generating functions of @xmath413 and @xmath411 are finite ( for @xmath394 ) and identical because , from ( [ eqa.5 ] ) , we have that , for any @xmath416 , @xmath417 thus , @xmath418 and choosing @xmath419 , we are led to the contradiction @xmath420 . this work was partially supported by the university of athens research fund under grant no . we would like to thank professor roger w. johnson for providing us with a copy of his paper . thanks are also due to an anonymous referee for the careful reading of the manuscript and for some helpful comments . n. papadatos is devoting this work to his six years old little daughter , dionyssia , who is suffering from a serious health disease . cacoullos , t. , papadatos , n. and papathanasiou , v. ( 2001 ) . an application of a density transform and the local limit theorem . i primenen . _ * 46 * 803810 ; translation in _ theory probab . * 46 * ( 2003 ) 699707 . cacoullos , t. , papathanasiou , v. and utev , s.a . variational inequalities with examples and an application to the central limit theorem . * 22 * 16071618 . chen , l.h.y . poisson approximation for dependent trials . probab . _ * 3 * 534545 . papadatos , n. and papathanasiou , v. ( 2001 ) . unified variance bounds and a stein - type identity . in _ probability and statistical models with applications _ ( c.a . charalambides , m.v . koutras and n. balakrishnan , eds . ) 87100 . new york : chapman & hall / crc . stein , c.m . a bound for the error in the normal approximation to the distribution of a sum of dependent random variables . in _ proc . sixth berkeley symp . * 2 * 583602 . berkeley , ca : univ . california press .

we use ideas from algebraic geometry and dynamical systems to explain some ways that control points influence the shape of a bzier curve or patch . in particular , we establish a generalization of birch s theorem and use it to deduce sufficient conditions on the control points for a patch to be injective . we also explain a way that the control points influence the shape via degenerations to regular control polytopes . the natural objects of this investigation are irrational patches , which are a generalization of krasauskas s toric patches , and include bzier and tensor product patches as important special cases .

0812.1275

the control points and weights of a bzier curve , bzier patch , or tensor - product patch govern many aspects of the curve or surface . for example , they provide an intuitive means to control its shape . through de castlejau s algorithm , they enable the computation of the curve or surface patch . finer aspects of the patch , particularly continuity and smoothness at the boundary of two patches are determined by the control points and weights . global properties , such as the location of a patch in space due to the convex hull property , also depend upon the control points . when the control points are in a particular convex position , then the patch is convex @xcite . we apply methods from algebraic geometry , specifically toric geometry , to explain how some further global properties of a patch are governed by the control points . we first investigate the self - intersection , or injectivity of a patch . we give a simple and easy - to - verify condition on a set of control points which implies that the resulting patch has no self - intersection , for any choice of weights . for 3-dimensional patches as used for solid modeling , injectivity is equivalent to the patch properly parameterizing the given solid . this uses craciun and feinberg s injectivity theorem @xcite from the theory of chemical reaction networks , which may be seen as a generalization of birch s theorem from algebraic statistics . a second global property that we investigate is how the shape of the patch is related to the shape of a . this is a piecewise linear triangulated surface whose vertices are the control points . it is if the underlying triangulation comes from a regular triangulation of the domain polytope of the patch . we show that regular control polytopes are the limits of patches as the weights undergo a toric deformation corresponding to the underlying regular triangulation , and that non - regular control polytopes can never be such a limit . this gives a precise meaning to the notion that the shape of the control net governs the shape of the patch . this line of inquiry is pursued in terms of krasauskas s toric patches @xcite , as it relies upon the structure of toric varieties from algebraic geometry . the correct level of generality is however that of , which are analytic subvarieties of the simplex ( realized as a compactified positive orthant ) that are parameterized by monomials , where is a vector of positive numbers and the exponent vector has real - number coordinates . ( this is a usual toric variety when @xmath0 has integer coordinates . ) while irrational patches may seem exotic for modeling , they occur naturally in statistics as discrete exponential families @xcite and their blending functions may be computed using iterative proportional fitting ( ipf ) @xcite , a popular numerical algorithm from statistics . furthermore , these blending functions have linear precision . for toric patches , this was observed in @xcite and developed in @xcite , and the analysis there carries over to irrational patches . while we work in this generality , our primary intent ( and the main application ) is to shed light on properties of bzier curves , surfaces , and 3-dimensional patches . we recall the standard definition of a mapping via control points and blending functions , and then the definitions of toric bzier patches in sect . [ cgs - s : defs ] . there , we also illustrate some of our results on examples of bzier curves . in sect . [ cgs - s : irrational ] , we introduce irrational toric patches , recalling the geometric formulation of a toric patch and the use of iterative proportional fitting to compute these patches , explaining how these notions from @xcite for toric patches extend to irrational patches . the next two sections contain our main results . we study injectivity of patches in sect . [ cgs - s : inj ] , and discuss degenerations to control polytopes in sect . [ cgs - s : degeneration ] . appendices a and b contain technical proofs of some theorems . we interpret the standard definition of a mapping via control points and blending functions ( see for example @xcite ) in a general form convenient for our discussion . all functions here are smooth ( @xmath1 ) where defined and real - valued . let be the set of strictly positive real numbers and the set of non - negative real numbers . we will use the following typographic conventions throughout . vector constants ( control points , indexing exponents , and standard basis vectors ) will be typeset in * boldface * , while vector variables will be typeset in standard math italics . let @xmath2 be a finite set of points that affinely span @xmath3 , which we shall use as geometrically meaningful indices . a control point scheme for parametric patches , or , is a collection @xmath4 of non - negative functions , called . the common domain of the blending functions is the convex hull @xmath5 of @xmath2 , which we call the . we also assume that the blending functions do not vanish simultaneously at any point of @xmath5 , so that there are no basepoints . lists @xmath6 of and positive @xmath7 together give a map @xmath8 defined by @xmath9 the denominator in is positive on @xmath5 and so the map @xmath10 is well - defined . we will refer to both @xmath11 and @xmath12 as the blending functions of a patch . this generality of separating the weights from the blending functions will be used in sect . [ cgs - s : degeneration ] when we investigate the effect of systematically varying the weights of a patch while keeping the control points and blending functions constant . the control points and weights affect the shape of the patch which is the image of the map @xmath10 . for example , the asserts that the image @xmath13 of the patch lies in the convex hull of the control points . to see this , note that if we set @xmath14 then @xmath15 and @xmath16 . then formula becomes @xmath17 so that @xmath18 is a convex combination of the control points and therefore lies in their convex hull . in fact , if there is a point @xmath19 at which no blending function vanishes , then _ any _ point in the interior of the convex hull of the control points is the image @xmath18 of some patch for some choice of weights . in this way , the convex hull property is the strongest general statement that can be made about the location of a patch . another well - known manifestation of control points is the relation of a bzier curve to its control polygon . fix a positive integer @xmath20 and let @xmath21 so that @xmath5 is the unit interval . the blending functions of a bzier curve are the bernstein polynomials , @xmath22 the of a bzier curve with control points @xmath23 is the union of the line segments @xmath24 between consecutive control points . figure [ cgs - f : one ] displays two quintic plane bzier curves with their control polygons ( solid lines ) . @xmath25 the convex hulls of the control points are indicated by the dashed lines . the first curve has no points of self - intersection , while the second curve has one point of self - intersection . while this self - intersection may be removed by varying the weights attached to the control points , by theorem [ cgs - th : inj ] it is impossible to find weights so that a curve with the first set of control points has a point of self - intersection . we will also show that the control polygon may be approximated by a bzier curve . we state a simplified version of theorem [ cgs - th : bez_curve_approx ] from sect . [ cgs - s : degeneration ] . * theorem . * _ given control points in @xmath26 for a bzier curve and some number @xmath27 , there is a choice of weights so that the image @xmath28 $ ] of the bzier curve lies within a distance @xmath29 of the control polygon . _ in figure [ cgs - f : degenerating ] , we display one of the quintic curves from figure [ cgs - f : one ] , but with weights on @xmath30@xmath31 of @xmath32 and @xmath33 , respectively . the control polygon for the second curve is omitted , as it would obscure the curve . the first curve lies within a distance @xmath34 of the control polygon and the second within a distance @xmath35 , if the control polygon has height 1 . @xmath36 krasauskas @xcite introduced toric patches as a generalization of the classical bzier and tensor product patches . these are based upon toric varieties from algebraic geometry and their shape may be any polytope with integer vertices . the articles @xcite provide an introduction to toric varieties for geometric modeling . a polytope @xmath5 is defined by its @xmath37 here , @xmath5 has @xmath38 facets ( faces of maximal dimension ) and for each @xmath39 , @xmath40 is the linear function defining the @xmath41th facet , where @xmath42 is the ( inward oriented ) primitive vector normal to the facet and @xmath43 . for example , if our polytope is the triangle with vertices @xmath44 , @xmath45 , and @xmath46 , @xmath47 then we have @xmath48 , @xmath49 , and @xmath50 . here , @xmath51 is the unit triangle @xmath52 with vertices @xmath44 , @xmath53 , and @xmath54 scaled by a factor of @xmath20 . let @xmath55 be any subset of the integer points of @xmath5 which includes its vertices . for every @xmath56 , krasauskas defined the @xmath57 which is non - negative on @xmath5 , and the collection of all @xmath58 has no common zeroes on @xmath5 . these are blending functions for the of shape @xmath2 . if we choose weights @xmath59 and multiply the formula by @xmath60 , we obtain blending functions for the toric patch of shape @xmath61 . [ cgs - ex : beztri ] when @xmath5 is a scaled triangle or a product of such triangles , gives the blending functions of the bzier patch or bzier simploid @xcite with the corresponding shape . to see this for the scaled triangle @xmath51 , note that given an integer point @xmath62 , and weight the multinomial coefficient @xmath63 , then the corresponding blending function is @xmath64 this is almost the bivariate bernstein polynomial , which is obtained by substituting @xmath65 and @xmath66 for @xmath67 and @xmath68 , respectively , and dividing by @xmath69 . ( this has the effect of changing the domain from @xmath51 to the unit triangle @xmath52 . ) krasauskas s definition of toric bzier functions still makes sense if we relax the requirement that the points @xmath70 have integer coordinates . this leads to the notion of an irrational patch ( as its blending functions are no longer rational functions ) , which provides the level of generality appropriate for our investigation . let @xmath70 be a finite collection of points and set @xmath71 to be the convex hull of @xmath2 , which we assume is a full - dimensional polytope . we may also realize @xmath5 as an intersection of half - spaces through its facet inequalities , @xmath72 here , @xmath5 has @xmath38 facets with the @xmath41th facet supported by the affine hyperplane @xmath73 where @xmath74 with @xmath75 and @xmath76 _ an _ inward pointing normal vector to the @xmath41th facet of @xmath5 . there is no canonical choice for these data ; multiplying a pair @xmath77 by a positive scalar gives another pair defining the same half - space . following krasauskas , we provisionally define @xmath78 by the same formula as , @xmath79 these are blending functions for the of shape @xmath2 . while these functions do depend upon the choice of data @xmath77 for the facet inequalities defining @xmath5 , we will see that the image @xmath13 of such a patch given by weights and control points is independent of these choices . we follow sect . 2.2 of @xcite , but drop the requirement that our objects are algebraic . let @xmath70 be a finite subset indexing a collection of blending functions @xmath80 , where @xmath5 is the convex hull of @xmath2 . let @xmath81 be a real vector space with basis @xmath82 . set @xmath83 to be the points with non - negative coordinates and let @xmath84 be those points with strictly positive coordinates . for @xmath85 , set @xmath86 . the , @xmath87{figures / bsimplex.eps}}^{\mathcal a}\subset{{\mathbb r}}^{\mathcal a}_\geq$ ] , is the set @xmath88{figures / bsimplex.eps}}^{\mathcal a}\ : = \ \ { z\ \in\ { { \mathbb r}}^{\mathcal a}_{\geq } : { \textstyle \sum z}=1\}\enspace.\ ] ] we introduce homogeneous coordinates for @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] . if @xmath89 , then we set @xmath90\ : = \ \frac{1}{\sum z}(z_{\bf a } \mid { \bf a}\in{\mathcal a})\ \in\ { \includegraphics[height=7pt]{figures / simplex.eps}}^{\mathcal a}\enspace.\ ] ] the blending functions @xmath80 give a @xmath1 map , @xmath91{figures / simplex.eps}}^{\mathcal a}\\ x&\longmapsto&[\beta_{\bf a}(x ) \mid { \bf a}\in{\mathcal a}]\ , . \end{aligned}\ ] ] the reason for this definition is that a mapping @xmath92 given by the blending functions @xmath93 , weights @xmath94 , and control points @xmath95 factors through the map @xmath96{figures / simplex.eps}}^{\mathcal a}$ ] . to see this , first note that the weights @xmath97 act on @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] : if @xmath98\in { \includegraphics[height=7pt]{figures / simplex.eps}}^{\mathcal a}$ ] , then @xmath99\enspace.\ ] ] the control points @xmath95 define the map @xmath100{figures / simplex.eps}}^{\mathcal a}\to{{\mathbb r}}^n$ ] via @xmath101 then the mapping @xmath10 is simply the composition @xmath102{figures / simplex.eps}}^{\mathcal a}\ \xrightarrow{\;w.\;}\ { \includegraphics[height=7pt]{figures / simplex.eps}}^{\mathcal a}\ \xrightarrow{\;\pi_{\mathcal b}\;}\ { { \mathbb r}}^n\enspace.\ ] ] in this way , we see that the image @xmath103{figures / simplex.eps}}^{\mathcal a}$ ] of @xmath5 under the map @xmath93 determines the shape of the patch @xmath13 . internal structures of the patch , such as the mapping of texture , are determined by how @xmath93 maps @xmath5 to @xmath104 . for example , precomposing @xmath93 with any homeomorphism of @xmath5 gives blending functions with the same image in @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] , but with a different internal structure . for an irrational toric patch of shape @xmath2 with blending functions , the image @xmath103{figures / simplex.eps}}^{\mathcal a}$ ] is independent of the choice of normal vectors . for this , we first define the map @xmath105{figures / simplex.eps}}^{\mathcal a}$ ] by @xmath106\enspace.\ ] ] let be the closure of the image of the map @xmath107 . when @xmath108 , this is the positive part @xcite of the toric variety parameterized by the monomials of @xmath2 . when @xmath2 is not integral , we call @xmath109 the parameterized by monomials in @xmath2 . in appendix [ cgs - a : b ] we prove the following theorem . [ cgs - th : x_cala ] suppose that @xmath70 is a finite set of points with convex hull @xmath5 . let @xmath110 be a collection of irrational toric bzier functions for @xmath2 . then @xmath111 , the closure of the image of @xmath107 . we prove this by showing that the restriction of the map @xmath93 to the interior @xmath112 of @xmath5 factors through the map @xmath107 . by theorem [ cgs - th : x_cala ] , the image of the irrational toric blending functions for @xmath2 depends upon @xmath2 and not upon the choice of toric blending functions for @xmath2 . thus the shape of the corresponding patch @xmath13 depends only upon @xmath2 , the weights @xmath94 , and the control points @xmath95 . however , the actual parameterization of @xmath109 by @xmath5 , and hence of @xmath13 does depend upon the choice of toric blending functions for @xmath2 . to ensure that the patch @xmath13 has shape reflecting that of @xmath5 , we require that the map @xmath113 be injective . this also guarantees that the patch @xmath13 is typically an immersion . in the context of irrational toric patches , this injectivity is guaranteed by birch s theorem from algebraic statistics . for a standard reference , see @xcite . when @xmath108 , birch s theorem follows from general results on the moment map in symplectic geometry ( * ? ? ? * and notes to chapter 4 , p. 140 ) . [ cgs - th : birch ] suppose @xmath70 is finite and let @xmath93 be a collection of toric bzier functions . if we choose control points @xmath95 to be the corresponding points of @xmath2 , @xmath114 , then the composition @xmath115 is a homeomorphism onto @xmath5 . by birch s theorem and theorem [ cgs - th : x_cala ] , any two sets @xmath116 of toric bzier functions of shape @xmath2 differ only by a homeomorphism @xmath117 of the polytope @xmath5 , so that @xmath118 . in fact @xmath119 restricts to a homeomorphism on all faces of @xmath5 . as we are concerned with the shape of a patch and not its internal structure , we follow krasauskas lead and make the following definition . a(n irrational ) toric patch of shape @xmath2 is any set of blending functions @xmath120 such that the map @xmath121 is a homeomorphism . the projection map @xmath100{figures / simplex.eps}}^{\mathcal a}\to{{\mathbb r}}^d$ ] appearing in birch s theorem induced by the choice @xmath122 of control points is called the and written . restricting the tautological projection to @xmath109 gives the @xmath123 . the components of its inverse @xmath124 provide a preferred set of blending functions for the patch . when @xmath108 , these were studied in @xcite , where they were shown to have linear precision , and that they may be computed by , a numerical algorithm from statistics @xcite . these same arguments apply to irrational toric patches the preferred blending functions have linear precision and are computed by ipf . any patch has unique blending functions with linear precision ( * ? ? ? * theorem 1.11 ) . while the classification of toric patches for which these preferred blending functions are rational functions remains open in general , it has been settled for surface patches ( @xmath125 ) @xcite , and this places very strong restrictions on higher - dimensional patches . in algebraic statistics , @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] is identified with the probability simplex parameterizing probability distributions on data indexed by @xmath2 . the image @xmath126 of @xmath127 under the map @xmath128 and translation by @xmath94 is known as a @xcite . it is more common to call this a , as the logarithms of the coordinates of @xmath128 are linear functions in the logarithms of the coordinates of @xmath127 , or a as the coordinates of @xmath128 are exponentials in the logarithms of the coordinates of @xmath127 . the tautological map appears in statistics as follows . given ( observed ) normalized data @xmath129{figures / simplex.eps}}^{\mathcal a}$ ] , the problem of asks for a probability distribution in the toric model , @xmath130 , with the same sufficient statistics as @xmath131 , @xmath132 . by birch s theorem , the point @xmath130 is unique and hence @xmath133 thus inverting the tautological projection is necessary for maximum likelihood estimation . darroch and ratcliff @xcite introduced the numerical algorithm of iterative proportional fitting , also known as generalized iterative scaling , for computing the inverse @xmath134 of the tautological projection . we now describe their algorithm . observe first that the toric patch @xmath135 does not change if we translate all elements of @xmath2 by a fixed vector @xmath136 , ( @xmath137 ) , so we may assume that @xmath2 lies in the positive orthant @xmath138 . scaling the exponent vectors in @xmath2 by a fixed positive scalar @xmath139 also does not change @xmath135 as @xmath140 is a homeomorphism of @xmath141 which extends to a homeomorphism of @xmath138 . thus we may assume that @xmath2 lies in the standard simplex @xmath142 in @xmath3 , @xmath143 lastly , we lift this to the probability simplex @xmath87{figures / simplex.eps}}^{d+1 } \subset{{\mathbb r}}^{d+1}_{\geq}$ ] , @xmath88{figures / bsimplex.eps}}^{d+1}\ : = \ \ { y\in { { \mathbb r}}^{d+1}_\geq\mid { \textstyle \sum y}=1\}\enspace,\ ] ] by @xmath144{figures / simplex.eps}}^{d+1}\enspace.\ ] ] since for @xmath139 and @xmath145 , @xmath146 we see that replacing @xmath2 by this homogeneous version also does not change @xmath147 . we describe the algorithm of , which is theorem 1 in @xcite . [ cgs - p : ipf ] suppose that @xmath148{figures / simplex.eps}}^{d+1}$ ] has convex hull @xmath5 and @xmath149{figures / simplex.eps}}^{\mathcal a}$ ] . set @xmath150 . then the sequence of points @xmath151{figures / simplex.eps}}^{\mathcal a}\ ] ] whose @xmath152-coordinates are defined by @xmath153 and , for @xmath154 , @xmath155 converges to the unique point @xmath130 such that @xmath156 . we remark that if @xmath2 is not homogenized then to compute @xmath157 for @xmath158 , we first put @xmath2 into homogeneous form using an affine map @xmath159 , and then use iterative proportional fitting to compute @xmath160 . we also call this modification of the algorithm of proposition [ cgs - p : ipf ] iterative proportional fitting . thus iterative proportional fitting computes the inverse image of the tautological projection . birch s theorem ( theorem [ cgs - th : birch ] ) states that for one particular choice of control points , namely @xmath161 and all weights @xmath97 , the mapping @xmath10 of a toric patch of shape @xmath61 is a homeomorphism onto its image . from this , we can infer that for _ most _ choices of control points and weights , this mapping is at least an immersion . to study dynamical systems arising from chemical reaction networks , craciun and feinberg @xcite prove an injectivity theorem for certain maps , which we adapt to generalize birch s theorem . this will give conditions on control points @xmath162 which guarantee that for _ any _ choice @xmath94 of weights , the resulting mapping @xmath10 of a toric patch of shape @xmath61 is a homeomorphism onto its image . this result has several consequences concerning the injectivity of toric patches . let us first give the craciun - feinberg theorem . let @xmath163 be a finite set of points which affinely spans @xmath26 . for @xmath164 and @xmath165 , consider the map @xmath166 defined by @xmath167 [ cgs - th : cf ] the map @xmath168 is injective for every @xmath164 if and only if the determinant of the jacobian matrix , @xmath169 does not vanish for any @xmath170 and any @xmath164 . we give a proof in appendix [ cgs - a : a ] . the condition of theorem [ cgs - th : cf ] that the jacobian @xmath171 does not vanish for any @xmath170 is reminiscent of the jacobian conjecture @xcite , which is that the jacobian of a polynomial map @xmath172 does not vanish if and only if the map @xmath173 is an isomorphism . since we are restricted to @xmath170 , it is closer to the real jacobian conjecture , which is however false @xcite , and therefore not necessarily relevant . the condition of theorem [ cgs - th : cf ] is conveniently restated in terms of @xmath174 and @xmath175 . for a list @xmath176 , which we write as @xmath177}{n}}}$ ] , let @xmath178 be the determinant of the matrix whose columns are the vectors @xmath179 , and define @xmath180 similarly . in appendix [ cgs - a : a ] , we deduce the following corollary . [ cgs - c : matroid ] the map @xmath168 @xmath181 is injective for all @xmath164 if and only if @xmath182 for every @xmath183}{n}$ ] and at least one product @xmath184 is non - zero . this leads to a generalization of birch s theorem . an ordered list @xmath185 of affinely independent points in @xmath3 determines an orientation of @xmath3simply consider the basis @xmath186 let @xmath2 and @xmath187 be finite sets of points in @xmath3 . suppose that @xmath188 is an affinely independent subset of @xmath2 . if the corresponding subset @xmath189 of @xmath95 is also affinely independent , then each subset determines an orientation , and the two orientations are either the same or they are opposite . we say that @xmath2 and @xmath95 are if either every such pair of orientations is the same , or if every such pair of orientations is opposite . we further need that there is at least one affinely independent subset of @xmath2 such that the corresponding subset of @xmath95 is also affinely independent . observe that compatibility is preserved by invertible affine transformations acting separately on @xmath2 and @xmath95 . in fig . [ cgs - f : compatible ] shows three sets of labeled points . the first and second sets are compatible , but neither is compatible with the third . @xmath190 we give our generalization of birch s theorem . suppose that @xmath71 is the convex hull of @xmath2 and @xmath191 are toric bzier functions for @xmath2 . for any @xmath97 , let @xmath192 be the toric patch of shape @xmath61 given by the control points @xmath162 : @xmath193 [ cgs - th : new_birch ] the map @xmath194 is injective for all @xmath97 if and only if @xmath2 and @xmath95 are compatible . as any set @xmath2 is compatible with itself , this implies birch s theorem ( theorem [ cgs - th : birch ] ) . [ cgs - ex : cubic_move ] let @xmath195 be the convex hull of @xmath196 . set @xmath197 and let @xmath97 be the weights of a cubic bzier patch ( example [ cgs - ex : beztri ] with @xmath198 ) . we consider choices @xmath199 of control points that are compatible with @xmath2 . for convenience , we will require that @xmath200 when @xmath152 is a vertex and that if @xmath152 lies on an edge of @xmath201 , then so does @xmath202 . for these edge control points , compatibility imposes the restriction that they appear along the edge in the same order as the corresponding exponents from @xmath2 . the placement of the center control point is however constrained . we show two compatible choices of @xmath95 in fig . [ cgs - f : compatible_control ] . on the left is the situation of birch s theorem , in which @xmath200 , and on the right we have moved the edge control points . the region in which we are free to move the center point is shaded in each picture . @xmath203 let @xmath204 be coordinates for @xmath205 and consider the map @xmath206 defined by @xmath207 we claim that @xmath194 is injective if and only if @xmath208 is injective . since @xmath194 is the composition @xmath209 , with the first map an isomorphism , @xmath194 is injective if and only if the composition of the last two maps is injective . since @xmath109 is compact , this will be injective if and only if its restriction to the interior @xmath210 of @xmath109 is injective . precomposing with the monomial parametrization of @xmath210 , we see that @xmath194 is injective if and only if the map @xmath211 defined by @xmath212 is injective . since @xmath213 , these values lie on a ray through the origin , and we invite the reader to check that this ray meets the hyperplane with first coordinate 1 at the point @xmath214 . thus @xmath208 is injective if and only if @xmath215 is injective , which is equivalent to the map @xmath194 being injective . we deduce the theorem by showing that @xmath208 is injective . this follows from corollary [ cgs - c : matroid ] as the condition that @xmath2 and @xmath95 are compatible is equivalent to @xmath216 for all @xmath217}{d+1}$ ] , where @xmath218 , @xmath219 and we have @xmath220 . we now describe two applications of theorem [ cgs - th : new_birch ] to modeling . in solid modeling , we represent a 3-dimensional solid by covering it with 3-dimensional patches , for example using bzier toric patches as finite elements . besides the obvious @xmath221 or higher continuity along the boundary as required , such bzier finite elements should at least provide a one - to - one parametrization of their image , i.e. they should be injective . by theorem [ cgs - th : new_birch ] , we may guarantee injectivity by requiring that the control points @xmath95 be compatible with the exponents @xmath2 . moreover , if these sets are incompatible , then there is some choice of weights for which the patch is not injective . typically , the exponents @xmath2 and the control points do not lie in the same space ; surfaces ( @xmath222 ) are modeled in 3-space ( @xmath223 ) , or curves ( @xmath224 ) in 2 or 3space ( @xmath225 ) . nevertheless , theorem [ cgs - th : new_birch ] gives conditions that imply injectivity of patches . let @xmath226 be a point disjoint from a hyperplane , @xmath227 . the projection @xmath228 with center @xmath229 is the map which associates a point @xmath230 to the intersection of the line @xmath231 with @xmath227 . we use a broken arrow as the projection is not defined on the plane through @xmath229 parallel to @xmath227 . identifying @xmath227 with @xmath26 gives a projection map @xmath232 . a coordinate projection @xmath233 is a projection with center at infinity . more generally , a @xmath234 is a sequence of such projections from points . [ cgs - th : inj ] let @xmath70 , @xmath97 , and @xmath235 be the exponents , weights , and control points of a toric patch and let @xmath5 be the convex hull of @xmath2 . if there is a projection @xmath236 such that @xmath2 is compatible with the image @xmath237 of @xmath95 , then the mapping @xmath92 given by the toric blending functions associated to @xmath2 , the weights @xmath94 , and control points @xmath95 is injective . by theorem [ cgs - th : new_birch ] , the composition @xmath238 is injective , from which it follows that @xmath10 must have been injective . for the curve on the left in fig . [ cgs - f : one ] ( which is reproduced below ) , the vertical projection @xmath239 maps the control points @xmath240 to points on the line in the same order as the exponents @xmath241 , which implies that the curve has no self - intersections . the convex hull property asserts that the image , @xmath13 , of a toric bzier patch of shape @xmath61 given by control points @xmath243 and weights @xmath97 lies in the convex hull of the control points . when @xmath13 is a curve , the control points may be joined sequentially to form the , which is a piecewise linear representation of the curve . when @xmath13 is however a surface patch , there are many ways to interpolate the control points by triangles or other polygons to obtain a piecewise linear surface , called a , that represents the patch . the shape of this control polytope affects the shape of the patch . for example , when the control points have the form @xmath244 for @xmath245 a convex function , then the patch is convex @xcite . also , leroy @xcite uses a particular control polytope for the graph of a function to obtain certificates of positivity for polynomials . among all control polytopes for a given set of control points , we identify the class of regular control polytopes , which come from regular triangulations of the exponents @xmath2 . these regular control polytopes are related to the shape of the patch in the following precise manner : there is a choice of weights so that a toric bzier patch is arbitrarily close to a given control polytope if and only if that polytope is regular . it is instructive to begin with bzier curves . a bzier curve of degree @xmath20 in @xmath26 with weights @xmath94 is the composition , @xmath246\ \xrightarrow{\;\beta\;}\ { \includegraphics[height=7pt]{figures / simplex.eps}}^{m+1}\ \xrightarrow{\;w.\;}\ { \includegraphics[height=7pt]{figures / simplex.eps}}^{m+1}\ \xrightarrow{\;\pi_{\mathcal b}\;}\ { { \mathbb r}}^n\enspace,\ ] ] where @xmath247 with @xmath248 for @xmath249 $ ] . then the map @xmath93 is given by @xmath250 , for @xmath251 . here , @xmath252 with @xmath253 are the coordinates for @xmath87{figures / simplex.eps}}^{m+1}$ ] . the image @xmath254\subset{\includegraphics[height=7pt]{figures / simplex.eps}}^{m+1}$ ] is defined by the binomials @xmath255 to see this , suppose that @xmath256 satisfies . setting @xmath257 , then we may solve these equations to obtain @xmath258 . if @xmath259 are weights , then @xmath260 $ ] is defined in @xmath87{figures / simplex.eps}}^{m+1}$ ] by @xmath261 suppose that we choose weights @xmath262 . dividing by @xmath263 , becomes @xmath264 since @xmath265 , we may assume that @xmath266 . setting @xmath267 , we see that @xmath268 if we consider the limit of these binomials as @xmath269 , we obtain @xmath270 these define the polygonal path in @xmath87{figures / simplex.eps}}^{m+1}$ ] whose @xmath41th segment is the edge @xmath271\enspace,\ ] ] and whose projection to @xmath26 is the of the bzier curve , which is the collection of line segments @xmath272 , @xmath273 , @xmath274 , @xmath275 . we illustrate this when @xmath198 in fig . [ cgs - f : cubic_bezier ] , which shows three different bzier curves having the same control points , but different weights @xmath276 for @xmath277 . in algebraic geometry , altering the weights in this manner is called a . the bzier cubics are displayed together with the cubics @xmath260 $ ] lying in the 3-simplex , @xmath278 , which is drawn in @xmath279 . @xmath280{figures / cubic.1.eps } } \put(52,0){$t=1 $ } \put(0,105){$\pi_{\mathcal b}$ } \put(16,135){\vector(0,-1){60 } } \end{picture } \begin{picture}(133,187 ) \put(-1,10){\includegraphics[height=185pt]{figures / cubic.3.eps } } \put(52,0){$t=3 $ } \put(0,105){$\pi_{\mathcal b}$ } \put(16,135){\vector(0,-1){60 } } \end{picture } \begin{picture}(133,187 ) \put(-1,10){\includegraphics[height=185pt]{figures / cubic.9.eps } } \put(52,0){$t=9 $ } \put(0,105){$\pi_{\mathcal b}$ } \put(16,135){\vector(0,-1){60 } } \end{picture}\ ] ] in these pictures , the projection @xmath281 is simply the vertical projection forgetting the third coordinate . the progression indicated in fig . [ cgs - f : cubic_bezier ] , where the bzier curve approaches the control polygon as the parameter @xmath282 increases , is a general phenomenon . let @xmath283 be the usual euclidean distance in @xmath26 . [ cgs - th : bez_curve_approx ] suppose that @xmath284\to{{\mathbb r}}^n$ ] is a bzier curve of degree @xmath20 with control points @xmath95 and weights @xmath285 . set @xmath286 . for any @xmath27 , if we have @xmath287 , then the distance between the control polygon and any point of the bzier curve @xmath288 $ ] is less than @xmath29 . let @xmath289\subset{\includegraphics[height=7pt]{figures / simplex.eps}}^{m+1}$ ] , where the weights are @xmath285 with @xmath287 . suppose that @xmath290 are integers in @xmath291 $ ] . then there exist integers @xmath292 with @xmath266 and @xmath293 . by , we have @xmath294 since the binomial coefficients are log - concave , we have @xmath295 using @xmath296 and @xmath297 , we see that @xmath298 in particular , if @xmath299 , then at most one of @xmath300 or @xmath301 exceeds @xmath302 . we conclude that at most two , necessarily consecutive , coordinates of @xmath303 may exceed @xmath302 . suppose that @xmath41 is an index such that @xmath304 if @xmath305 and let @xmath306 , a point along the @xmath41th segment of the control polygon . since @xmath307 we have @xmath308 , and we see that @xmath309 this proves the theorem as @xmath289 $ ] is an arbitrary point of the curve @xmath28 $ ] . in dimensions @xmath310 and higher , the analog of theorem [ cgs - th : bez_curve_approx ] requires the notion of a regular triangulation from geometric combinatorics . let @xmath70 be a finite set of points and consider a lifting function @xmath311 . let be the convex hull of the lifted points @xmath312 we assume that @xmath313 is full - dimensional in that @xmath205 is its affine span . the of @xmath313 are those facets whose outward pointing normal vector has positive last coordinate . any face of an upper facet is an . we illustrate this below when @xmath314 , where the displayed arrows are outward pointing normal vectors to upper facets . @xmath315{figures / upper_facets.eps } } \put(-32,40){$\lambda({\mathcal a})$ } \put(40,45){$p_\lambda$ } \put(-20,0){${\mathcal a}$ } \end{picture}\ ] ] projecting these upper facets to @xmath3 yields a of the convex hull of @xmath2 , which is the image of @xmath313 . for our purposes , we will need to assume that the lifting function is generic in that all upper facets of @xmath313 are simplices . in this case , we obtain a of @xmath2 . this consists of a collection @xmath316 of subsets of @xmath2 , where each subset @xmath317 consists of @xmath318 elements and spans a @xmath319-dimensional simplex . we regard all subsets of the facets @xmath317 as faces of the triangulation . these simplices form a subdivision in that they cover the convex hull of @xmath2 and any two with a non - empty intersection meet along a common face . the subdivision induced by the lifting function in fig . [ cgs - f : upper_face ] consists of three intervals resulting from removal of the middle point of @xmath2 , which does not participate in the subdivision as it is not lifted high enough . a set may have many regular triangulations , and not every point needs to participate in a given triangulation . figure [ cgs - f : regular_triangulations ] shows the edges in four regular triangulations of @xmath320 . @xmath321{figures / cubicreg.1.eps}\qquad \includegraphics[height=90pt]{figures / cubicreg.2.eps}\qquad \includegraphics[height=90pt]{figures / cubicreg.3.eps}\qquad \includegraphics[height=90pt]{figures / cubicreg.4.eps}\ ] ] not every triangulation is regular . we may assume that a lifting function @xmath245 for the triangulation of @xmath322 in fig . [ cgs - f : irrregular_triangulation ] takes a constant value at the three interior points . the clockwise neighbor of any vertex of the big triangle must be lifted lower than that vertex . ( consider the figure they form with the parallel edge of the interior triangle . ) since the edge of the big triangle is lifted to a convex path , this is impossible , except in some m.c . escher woodcuts . @xmath323{figures / quartic_not.eps}\ ] ] let @xmath324 be a collection of control points indexed by a finite set of exponents @xmath325 with @xmath326 . given a regular triangulation @xmath327 of @xmath2 we define the control polytope as follows . for each @xmath319-simplex @xmath317 in @xmath328 , the corresponding points of @xmath95 span a ( possibly degenerate ) simplex @xmath329 the union of these simplices in @xmath26 forms the that is induced by the regular triangulation @xmath328 of @xmath2 . this is a simplicial complex in @xmath26 with vertices in @xmath95 that has the same combinatorial type as the triangulation @xmath328 of @xmath2 . if the coordinate points @xmath330 of @xmath81 are our control points ( these are the vertices of @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] ) , then the regular control polytope is just the geometric realization of the simplicial complex @xmath328 , which is a subcomplex of the simplex @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] . in general , @xmath331 is the image of this geometric realization @xmath332{figures / simplex.eps}}^{\mathcal a}$ ] under the projection @xmath281 . [ cgs - ex : control_polytope ] let @xmath197 , the exponents for a cubic bzier triangle . cgs - f : control_polytopes ] shows the three control polytopes corresponding to the last three regular triangulations of figure [ cgs - f : regular_triangulations ] , all with the same control points . @xmath333{figures / controlpolytope.1.eps}\quad \includegraphics[height=95pt]{figures / controlpolytope.2.eps}\quad \includegraphics[height=95pt]{figures / controlpolytope.3.eps}\ ] ] the reason that we introduce regular control polytopes is that they may be approximated by toric bzier patches . [ cgs - th : bez_approx ] let @xmath70 , @xmath97 , and @xmath235 be exponents , weights , and control points for a toric bzier patch . suppose that @xmath328 is a regular triangulation of @xmath2 induced by a lifting function @xmath311 . for each @xmath334 , let @xmath335 be the toric bzier patch of shape @xmath2 with control points @xmath95 and weights @xmath336 . then , for any @xmath27 there exists a @xmath337 such that if @xmath338 , the image @xmath339 lies within @xmath29 of the control polytope @xmath331 . we prove theorem [ cgs - th : bez_approx ] in appendix [ cgs - a : b ] . figure [ cgs - f : control_polytope_degeneration ] illustrates theorem [ cgs - th : bez_approx ] for a cubic bzier triangle with the control points of example [ cgs - ex : control_polytope ] . the patch on the left is the cubic bzier triangle with the weights of example [ cgs - ex : beztri ] . the second and third patches are its deformations corresponding to the lifting function inducing the leftmost control polytope of fig . cgs - f : control_polytopes ] . the values of @xmath282 are @xmath340 , @xmath341 , and @xmath342 , as we move from left to right . @xmath333{figures / a.eps}\quad \includegraphics[height=95pt]{figures / b.eps}\quad \includegraphics[height=95pt]{figures / c.eps}\ ] ] an absolutely unpractical consequence of theorem [ cgs - th : bez_approx ] is a universality result : any surface which admits a triangulation that forms a regular control polytope may be approximated by a single bzier patch . as with theorem [ cgs - th : bez_curve_approx ] , the main idea behind the proof of theorem [ cgs - th : bez_approx ] ( which is given in appendix [ cgs - a : b ] ) is that for @xmath282 large enough , the translated patch @xmath343{figures / simplex.eps}}^{\mathcal a}$ ] can be made arbitrarily close to the geometric realization @xmath332{figures / simplex.eps}}^{\mathcal a}$ ] of the regular triangulation @xmath328 . the result follows by projecting this into @xmath26 using @xmath281 . in appendix [ cgs - a : b ] we also prove a weak converse to theorem [ cgs - th : bez_approx ] . namely if @xmath344 is sufficiently close to the geometric realization @xmath345 of a triangulation @xmath328 , then @xmath328 is in fact the regular triangulation of @xmath2 induced by the lifting function @xmath346 . [ cgs - th : conv_bez_approx ] let @xmath70 be a finite set of exponents . suppose that @xmath347{figures / simplex.eps}}^{\mathcal a}$ ] is the geometric realization of a triangulation @xmath328 of @xmath2 and there is a weight @xmath94 such that the distance between @xmath135 and @xmath345 is less than @xmath348 . then @xmath328 is the regular triangulation induced by the lifting function @xmath346 . * theorem [ cgs - th : cf ] . * ( craciun - feinberg ) _ the map @xmath168 is injective for every @xmath164 if and only if the determinant of the jacobian matrix , @xmath169 does not vanish for any @xmath170 and any @xmath164 . _ we show the equivalence of the two statements , transforming one into the other . first , suppose there is a @xmath164 so that @xmath168 is not injective . then there exist @xmath349 so that @xmath350 . then we have @xmath351 where @xmath352 is defined by @xmath353 , and @xmath354 has @xmath41th coordinate @xmath355 . in particular , @xmath356 , where @xmath357 . define @xmath358 by @xmath359 so that @xmath360 , where @xmath361 is the euclidean dot product , and we now have @xmath362 define the univariate function by @xmath363 for @xmath364 and set @xmath365 . then @xmath366 is an increasing continuous bijection between @xmath367 and @xmath141 . define @xmath368 by @xmath369 . then @xmath370 , and becomes @xmath371 we claim that @xmath372 lies in the kernel of the jacobian matrix of @xmath373 evaluated at the point @xmath374 . indeed , let @xmath375 be the unit vector in the @xmath376th direction . then @xmath377 where @xmath378 . since @xmath379 and @xmath380 , we see that @xmath381 so that @xmath372 lies in the kernel of the jacobian matrix of @xmath373 evaluated at @xmath374 , which implies that the jacobian determinant of @xmath373 vanishes at @xmath374 . the theorem follows as these arguments are reversible . * corollary [ cgs - c : matroid ] . * _ the map @xmath168 @xmath181 is injective for all @xmath164 if and only if @xmath182 for every @xmath183}{n}$ ] and at least one product @xmath184 is non - zero . _ observe first that the jacobian matrix @xmath382 factors as the product of matrices @xmath383 , where @xmath384 is the diagonal matrix with entries @xmath385 , @xmath386 is the diagonal matrix with entries @xmath387 and @xmath174 and @xmath175 are the matrices whose columns are the vectors @xmath388 and @xmath389 , respectively . if we apply the binet - cauchy theorem to this factorization , we see that @xmath390}{n } } \prod_{i\in i } k_i x^{{\bf y}_i } \cdot y_i\cdot z_i\enspace,\ ] ] where @xmath391 . suppose that @xmath182 for every @xmath183}{n}$ ] , and at least one product @xmath184 is non - zero . then all terms in the sum have the same sign and not all are zero , and so the jacobian does not vanish for any @xmath170 and @xmath164 . thus @xmath168 is injective for all @xmath164 , by theorem [ cgs - th : cf ] . suppose that there are two subsets @xmath183}{n}$ ] such that @xmath392 and @xmath393 . for @xmath139 and @xmath394}{n}$ ] , define @xmath395 by @xmath396 if we fix @xmath145 , then the expansion , implies that @xmath397 has the same sign as @xmath398 when @xmath399 , at least when @xmath400 . we conclude that there is some @xmath401 such that @xmath402 and some @xmath401 such that @xmath403 , and therefore some @xmath401 such that @xmath404 . this implies the corollary . * theorem [ cgs - th : x_cala ] . * _ suppose that @xmath70 is a finite set of points with convex hull @xmath5 . let @xmath110 be a collection of irrational toric bzier functions for @xmath2 . then @xmath111 , the closure of the image of @xmath107 . _ let @xmath405 be the interior of @xmath5 , which we assume has @xmath38 facets and is given by the facet inequalities @xmath406 . define two maps @xmath407 and @xmath408{figures / simplex.eps}}^{\mathcal a}$ ] by @xmath409\ , . \end{aligned}\ ] ] then the map @xmath410{figures / simplex.eps}}^{\mathcal a}$ ] ( whose image is dense in @xmath109 ) is the composition of the maps @xmath227 and @xmath159 . @xmath411{figures / simplex.eps}}^{\mathcal a}\enspace.\ ] ] let us recall the definiton of the map @xmath412{figures / simplex.eps}}^{\mathcal a}$ ] , @xmath413\enspace.\ ] ] the theorem follows once we show that the map @xmath159 factors through the map @xmath107 . for this , define a new map @xmath414 by @xmath415 then we claim that @xmath416{figures / simplex.eps}}^{\mathcal a}\enspace.\ ] ] to see this , we compute the component of @xmath417 corresponding to @xmath56 , @xmath418 thus @xmath419 , as maps to @xmath84 . the common factor @xmath420 does not affect the image in @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] , which shows that @xmath421 and proves the theorem . a consequence of this proof of theorem [ cgs - th : x_cala ] is the derivation of equations which define the points of @xmath109 . this derivation is similar to , but easier than , the development of toric ideals in ( * ? ? ? * ch . 4 ) , as we have monomials with arbitrary real - number exponents . suppose that we have a linear relation among the points of @xmath2 , @xmath422 for some @xmath423 . then the analytic binomial @xmath424 vanishes on @xmath425 , considered as a point in @xmath84 . this follows from the easy calculation @xmath426 even after clearing denominators , the common zero set of the binomials is not exactly the image @xmath425 in the simplex @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] , as the point @xmath427{figures / simplex.eps}}^{\mathcal a}$ ] is where the ray @xmath428 meets the simplex @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] . if we require that the binomial is homogeneous in that @xmath429 , then it vanishes at every point of this ray and therefore on the image of @xmath107 in @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] . since the coordinates are positive numbers , we may further assume that is an affine relation in that @xmath430 these necessary conditions are also sufficient . one way to see the sufficiency is to pick an affinely independent subset @xmath434 of @xmath2 that affinely spans @xmath3 and use the formula @xmath435 for @xmath436 to solve for @xmath437 . then the point @xmath438{figures / simplex.eps}}^{\mathcal a}$ ] satisfies this collection of binomials if and only if @xmath439 . this is also evident if we take logarithms of the coordinates . these arguments only work for points @xmath303 in the interior of @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] . for points of @xmath109 with some coordinates zero , we use the recursive nature of polytopes and the toric bzier functions . namely , if we restrict the collection @xmath440 of toric bzier functions to a face @xmath441 of the convex hull of @xmath2 , then those whose index @xmath152 does not lie in @xmath441 vanish , while those indexed by points of @xmath2 lying in @xmath441 specialize to toric bzier functions for @xmath441 . * theorem [ cgs - th : bez_approx ] . * _ let @xmath70 , @xmath97 , and @xmath235 be exponents , weights , and control points for a toric bzier patch . suppose that @xmath328 is a regular triangulation of @xmath2 induced by a lifting function @xmath311 . for each @xmath334 , let @xmath335 be the toric bzier patch of shape @xmath2 with control points @xmath95 and weights @xmath336 . then , for any @xmath27 there exists a @xmath337 such that if @xmath338 , the image @xmath339 lies within @xmath29 of the control polytope @xmath331 . _ the lifting function @xmath311 inducing the triangulation @xmath328 also induces an action of @xmath141 on @xmath87{figures / simplex.eps}}^{\mathcal a}$ ] where @xmath139 acts on a point @xmath438{figures / simplex.eps}}^{\mathcal a}$ ] by scaling its coordinates , @xmath442 . then @xmath339 is the image of @xmath443 under the projection @xmath100{figures / simplex.eps}}^{\mathcal a}\to{{\mathbb r}}^n$ ] . it suffices to show that @xmath443 can be made arbtrarily close to @xmath332{figures / simplex.eps}}^{\mathcal a}$ ] , if we choose @xmath282 large enough . we single out some equations from proposition [ cgs - prop : eqs ] . suppose that @xmath444 are points that do not lie in a common simplex of @xmath328 . that is , the segment @xmath445 is not an edge in the triangulation @xmath328 , and therefore it meets the interior of some face @xmath441 of @xmath328 so that there is a point common to the interiors of @xmath441 and of @xmath445 . ( if @xmath446 , so that @xmath152 does not participate in the triangulation @xmath328 , then this point is just @xmath152 . ) this gives the equality of convex combinations @xmath447 where all coefficients are positive and @xmath448 . thus @xmath449 holds on @xmath109 . the corresponding equation on @xmath443 is @xmath450 since @xmath445 is not in the triangulation , points in the interior of the lifted segment @xmath451 lie below points of upper faces of the polytope @xmath313 . we apply this observation to the point . its height in the lifted segment is @xmath452 , while its height in the lift of the face @xmath441 is @xmath453 , and so @xmath454 this implies that the exponent of @xmath282 in is negative . since the other terms on the right hand side are bounded , we see that the left hand side , and in fact the simple product @xmath455 , may be made as small as we please by requiring that @xmath282 be sufficiently large . suppose that @xmath2 consists of @xmath456 elements . repeating the previous argument for the ( finitely many ) pairs of points @xmath457 which are not both in any simplex of @xmath328 , we see that for any @xmath27 , there is a @xmath337 such that if @xmath338 and @xmath458 , then @xmath459 whenever @xmath457 do not lie in a common simplex of @xmath328 . in particular , at most one of @xmath460 or @xmath461 can exceed @xmath462 . let @xmath458 . then there is some facet @xmath441 of @xmath328 such that if @xmath463 , then @xmath464 . suppose that @xmath465 and set @xmath466 which is a point of the facet @xmath467 of the geometric realization @xmath332{figures / simplex.eps}}^{\mathcal a}$ ] . then @xmath468 as @xmath469 . we say that two subsets @xmath470 and @xmath174 of euclidean space are within a distance @xmath29 if for every point @xmath67 of @xmath470 there is some point @xmath68 of @xmath174 within a distance @xmath29 of @xmath67 , and vice - versa . * theorem [ cgs - th : conv_bez_approx ] . * _ let @xmath70 be a finite set of exponents . suppose that @xmath347{figures / simplex.eps}}^{\mathcal a}$ ] is the geometric realization of a triangulation @xmath328 of @xmath2 and there is a weight @xmath94 such that the distance between @xmath135 and @xmath345 is less than @xmath348 . then @xmath328 is the regular triangulation induced by the lifting function @xmath346 . _ to show that @xmath328 is the regular triangulation induced by the lifting function @xmath245 whose value at @xmath152 is @xmath471 , we must show that if a segment @xmath445 between two points of @xmath2 does not lie in the triangulation @xmath328 , then its lift by @xmath245 lies below the lift of some face @xmath441 of @xmath328 . set @xmath472 . for each face @xmath441 of @xmath328 , let @xmath473{figures / simplex.eps}}^{\mathcal a}$ ] be the barycenter of @xmath441 , @xmath474 where @xmath475 is the number of points of @xmath2 in @xmath441 , which is at most @xmath318 . if @xmath303 is a point of @xmath135 within a distance @xmath29 of @xmath476 , so that @xmath477 , then in particular no component of the vector @xmath478 has absolute value exceeding @xmath29 . thus we have the dichotomy @xmath479 now suppose that the segment @xmath445 does not lie in the triangulation @xmath328 . then there is a face @xmath441 of the triangulation whose interior meets the interior of this segment . that is , there is an equality of convex combinations and a corresponding equation that holds for points @xmath458 , @xmath480 suppose that @xmath303 is a point of @xmath135 that lies within a distance of @xmath481 of the barycenter @xmath482 of the face @xmath441 . then , by the estimates , we have @xmath483 canceling the common factor of @xmath481 and taking logarithms , we obtain @xmath484 which implies that the point common to the segment @xmath445 and the face @xmath441 of @xmath328 is lifted higher in the face @xmath441 than in the segment @xmath445 , and so the lift of the segment @xmath445 by @xmath245 lies below the lift of the face @xmath441 . as this is true for all segments , we see that @xmath328 is the triangulation induced by the lifting function @xmath245 . lawrence d. brown , _ fundamentals of statistical exponential families with applications in statistical decision theory _ , institute of mathematical statistics lecture notes monograph series , 9 , institute of mathematical statistics , hayward , ca , 1986 . kstutis kariauskas and rimvydas krasauskas , _ comparison of different multisided patches using algebraic geometry _ , curve and surface design : saint - malo 1999 ( p .- j . laurent , p. sablonniere , and l.l . schumaker , eds . ) , vanderbilt university press , nashville , 2000 , pp . 163172 . frank sottile , _ toric ideals , real toric varieties , and the moment map _ , topics in algebraic geometry and geometric modeling , contemp . math . 334 , amer . soc . , providence , ri , 2003 , pp .

we present a systematic method for reducing an arbitrary onelooppoint massless feynman integral with generic 4dimensional momenta to a set comprised of eight fundamental scalar integrals : six box integrals in , a triangle integral in , and a general twopoint integral in space time dimensions . all the divergences present in the original integral are contained in the general twopoint integral and associated coefficients . the problem of vanishing of the kinematic determinants has been solved in an elegant and transparent manner . being derived with no restrictions regarding the external momenta , the method is completely general and applicable for arbitrary kinematics . in particular , it applies to the integrals in which the set of external momenta contains subsets comprised of two or more collinear momenta , which are unavoidable when calculating oneloop contributions to the hardscattering amplitude for exclusive hadronic processes at large momentum transfer in pqcd . the iterative structure makes it easy to implement the formalism in an algebraic computer program . irbth/03 * * reduction method for dimensionally regulated one - loop n - point feynman integrals * + g. duplani ' c and b. nii ' c + _ theoretical physics division , rudjer bokovi institute , p.o . box 180 , hr-10002 zagreb , croatia _ + *

hep-ph0303184

scattering processes have played a crucial role in establishing the fundamental interactions of nature . they represent the most important source of information on short - distance physics . with increasing energy , multiparticle events are becoming more and more dominant . thus , in testing various aspects of qcd , the high - energy scattering processes , both exclusive and inclusive , in which the total number of particles ( partons ) in the initial and final states is @xmath5 , have recently become increasingly important . owing to the well - known fact that the lo predictions in perturbative qcd ( pqcd ) do not have much predictive power , the inclusion of higher - order corrections is essential for many reasons . in general , higher - order corrections have a stabilizing effect , reducing the dependence of the lo predictions on the renormalization and factorization scales and the renormalization scheme . therefore , to achieve a complete confrontation between theoretical predictions and experimental data , it is very important to know the size of radiative corrections to the lo predictions . obtaining radiative corrections requires the evaluation of one - loop integrals arising from the feynman diagram approach . with the increasing complexity of the process under consideration , the calculation of radiative corrections becomes more and more tedious . therefore , it is extremely useful to have an algorithmic procedure for these calculations , which is computerizable and leads to results which can be easily and safely evaluated numerically . the case of feynman integrals with massless internal lines is of special interest , because one often deals with either really massless particles ( gluons ) or particles whose masses can be neglected in high@xmath0energy processes ( quarks ) . owing to the fact that these integrals contain ir divergences ( both soft and collinear ) , they need to be evaluated in an arbitrary number of space - time dimensions . as it is well known , in calculating feynman diagrams mainly three difficulties arise : tensor decomposition of integrals , reduction of scalar integrals to several basic scalar integrals and the evaluation of a set of basic scalar integrals . considerable progress has recently been made in developing efficient approaches for calculating one@xmath0loop feynman integrals with a large number ( @xmath5 ) of external lines @xcite . various approaches have been proposed for reducing the dimensionally regulated ( @xmath5)@xmath0point tensor integrals to a linear combination of @xmath1 and lower@xmath0point scalar integrals multiplied by tensor structures made from the metric tensor @xmath6 and external momenta @xcite . it has also been shown that the general @xmath7point scalar one@xmath0loop integral can recursively be represented as a linear combination of @xmath8@xmath0point integrals provided the external momenta are kept in four dimensions @xcite . consequently , all scalar integrals occurring in the computation of an arbitrary one@xmath0loop @xmath9@xmath0point integral can be reduced to a sum over a set of basic scalar box @xmath10 integrals with rational coefficients depending on the external momenta and the dimensionality of space@xmath0time . despite , the considerable progress , the developed methods still can not be applied to all cases of practical interest . the problem is related to vanishing of various relevant kinematic determinants . as far as the calculation of one - loop @xmath11-point massless integrals is concerned , the most complete and systematic method is presented in @xcite . it does not , however , apply to all cases of practical interest . namely , being obtained for the non - exceptional external momenta it can not be , for example , applied to the integrals in which the set of external momenta contains subsets comprised of two or three collinear on - shell momenta . the integrals of this type arise when performing the leading - twist nlo analysis of hadronic exclusive processes at large@xmath0momentum transfer in pqcd . with no restrictions regarding the external kinematics , in this paper we formulate an efficient , systematic and completely general method for reducing an arbitrary one@xmath0loop @xmath1point massless integral to a set of basic integrals . although the method is presented for massless case , the generalization on massive case is straightforward . the main difference between the massive and massless cases manifests itself in the basic set of integrals , which in former case is far more complex . among the one@xmath0loop faynman integrals there exist both massive and massless integrals for which the existing reduction methods break down . the massless integrals belonging to this category are of more practical interest at the moment , so in this paper we concentrate on massless case . the paper is organized as follows . section 2 is devoted to introducing notation and to some preliminary considerations . in sec . 3 , for the sake of completenes , we briefly review a tensor decomposition method for @xmath1point tensor integrals which was originally obtained in ref . @xcite . in sec . 4 we present a procedure for reducing one@xmath0loop @xmath12@xmath0point massless scalar integrals with generic @xmath13@xmath0dimensional external momenta to a fundamental set of integrals . since the method is closely related to the one given in @xcite , similarities and differences between the two are pointed out . being derived with no restrictions to the external momenta , the method is completely general and applicable for arbitrary kinematics . section 5 contains considerations regarding the fundamental set of integrals which is comprised of eight integrals . section 6 is devoted to some concluding remarks . in the appendix a we give explicit expressions for the relevant basic massless box integrals in @xmath2 space@xmath0time dimensions . these integrals constitute a subset of the fundamental set of scalar integrals . as an illustration of the tensor decomposition and scalar reduction methods , in the appendix b we evaluate an one@xmath0loop 6@xmath0point feynman diagram with massless internal lines , contributing to the nlo hard@xmath0scattering amplitude for @xmath14 exclusive reaction at large momentum transfer in pqcd . in order to obtain one@xmath0loop radiative coorections to physical processes in massless gauge theory , the integrals of the following type are required : @xmath15 this is a rank @xmath16 tensor one@xmath0loop @xmath12@xmath0point feynman integral with massless internal lines in @xmath4@xmath0dimensional space@xmath0time , where @xmath17 , @xmath18 are the external momenta , @xmath19 is the loop momentum , and @xmath20 is the usual dimensional regularization scale . the feynman diagram with @xmath12 external lines , which corresponds to the above integral , is shown in fig . [ fig1 ] . _ one - loop n - point diagram_,width=377 ] for the momentum assignments as shown , i.e. with all external momenta taken to be incoming , the massless propagators have the form @xmath21 where the momenta @xmath22 are given by @xmath23 for @xmath24 from 1 to @xmath12 , and @xmath25 . the quantity @xmath26(@xmath27 ) represents an infinitesimal imaginary part , it ensures causality and after the integration determines the correct sign of the imaginary part of the logarithms and dilogarithms . it is customary to choose the loop momentum in a such a way that one of the momenta @xmath22 vanishes . however , for general considerations , such a choice is not convenient , since by doing so , one loses the useful symmetry of the integral with respect to the indices @xmath28 . the corresponding scalar integral is @xmath29 if @xmath30 , the integral ( [ f1 ] ) is uv divergent . in addition to uv divergence , the integral can contain ir divergence . there are two types of ir divergence : collinear and soft . a feynman diagram with massless particles contains a soft singularity if it contains an internal gluon line attached to two external quark lines which are on mass@xmath0shell . on the other hand , a diagram contains a collinear singularity if it contains an internal gluon line attached to an external quark line which is on mass @xmath0shell . therefore , a diagram containing a soft singularity at the same time contains two collinear singularities , i.e. soft and collinear singularities overlap . when evaluating feynman diagrams , one ought to regularize all divergences . making use of the dimensional regularization method , one can simultaneously regularize uv and ir divergences , which makes the dimensional regularization method optimal for the case of massless field theories . the tensor integral ( [ f1 ] ) is , as it is seen , invariant under the permutations of the propagators @xmath31 , and is symmetric with respect to the lorentz indices @xmath32 . lorentz covariance allows the decomposition of the tensor integral ( [ f1 ] ) in the form of a linear decomposition consisting of the momenta @xmath17 and the metric tensor @xmath33 . various approaches have been proposed for reducing the dimensionally regulated @xmath1point tensor integrals to a linear combination of @xmath1 and lower@xmath0point scalar integrals multiplied by tensor structures made from the metric tensor @xmath6 and external momenta . in this section we briefly review the derivation of the tensor reduction formula originally obtained in ref . @xcite . for the purpose of the following discussion , let us consider the tensor integral @xmath34 and the corresponding scalar integral @xmath35 the above integrals represent generalizations of the integrals ( [ f1 ] ) and ( [ f3 ] ) , in that they contain arbitrary powers @xmath36 of the propagators in the integrand , where @xmath37 is the shorthand notation for ( @xmath38 ) . also , for notational simplicity , the external momenta are omitted from the argument of the integral . the feynman parameter representation of the tensor integral @xmath39 , given in ( [ f4 ] ) , which is valid for arbitrary values of @xmath12 , @xmath16 , @xmath22 and @xmath40 , for the values of @xmath4 for which the remaining integral is finite , and the @xmath41function does not diverge is given by @xmath42^k [ r_1]^{j_1}\cdots [ r_{n}]^{j_{n}}\right\}_{\mu _ 1\cdots \mu_p}}{\nonumber}\\ & & \times\,\frac{\gamma\left ( \sum\nolimits_i \nu_i - d/2-k\right)}{2^k \,\left[\prod\nolimits_i \gamma ( \nu_i)\right]}(-1)^{\sigma_i \nu_i+p - k}\int_0 ^ 1\ ! \left ( \prod\nolimits_i { \rm d}y_i y_i^{\nu_i+j_i-1}\right){\nonumber}\\ & & \times \delta\left ( \sum\nolimits_{i=1}^{n}y_i-1\right ) \left [ \,-\sum_{i , j=1\atop i < j}^n y_i y_j \left ( r_i - r_j\right)^2 -{{\rm i}\epsilon}\ , \right]^{k+d/2-\sigma_i \nu_i},\label{f12}\end{aligned}\ ] ] where @xmath43^k [ r_1]^{j_1}\cdots [ r_{n}]^{j_{n}}\}_{\mu _ 1\cdots \mu_p}$ ] represents a symmetric ( with respect to @xmath44 ) combination of tensors , each term of which is composed of @xmath45 metric tensors and @xmath46 external momenta @xmath22 . thus , for example , @xmath47 as for the integral representation of the corresponding scalar integral ( [ f5 ] ) the result is of the form @xmath48^{d/2-\sigma_{i=1}^{n } \nu_i}\hspace{-1.5cm}.\label{f13}\end{aligned}\ ] ] now , on the basis of ( [ f13 ] ) , ( [ f12 ] ) can be written in the form @xmath49^k [ r_1]^{j_1}\cdots [ r_{n}]^{j_{n}}\right\}_{\mu _ 1\cdots \mu_p } } { \nonumber}\\ & & { } \times \frac { ( 4 \pi \mu^2)^{p - k}}{(-2)^{k } } \left[\prod_{i=1}^n \frac{\gamma ( \nu_i+j_i ) } { \gamma(\nu_i)}\right ] i^n_0(d+2(p - k);\ { \nu_i+j_i\ } ) . \label{f15}\end{aligned}\ ] ] this is the desired decomposition , of the dimensionally regulated @xmath1point rank @xmath16 tensor integral . it is originally obtained in ref . @xcite . based on ( [ f15 ] ) , any dimensionally regulated @xmath1point tensor integral can be expressed as a linear combination of @xmath1point scalar integrals multiplied by tensor structures made from the metric tensor @xmath6 and external momenta . therefore , with the decomposition ( [ f15 ] ) , the problem of calculating the tensor integrals has been reduced to the calculation of the general scalar integrals . it should be pointed out that among the tensor reduction methods presented in the literature one can find methods , e.g. @xcite which for @xmath5 completely avoid the terms proportional to the metric tensor @xmath6 . compared with the method expressed by ( [ f15 ] ) , these reduction procedures lead to decomposition containing a smaller number of terms . the methods of this type are based on the assumption that for @xmath5 , one can find four linearly independent 4-vectors forming a basis of the 4-dimensional minkowski space , in terms of which the metric tensor can then be expressed . this assumption is usually not realized when analyzing the exclusive processes at large - momentum - transfer ( hard scattering picture ) in pqcd . thus , for example , in order to obtain the next - to - leading order corrections to the hard - scattering amplitude for the proton - compton scattering , one has to evaluate one - loop @xmath50 diagrams . the set of external momenta contains two subsets comprised of three collinear momenta ( representing the proton ) . the kinematics of the process is thus limited to the 3-dimensional subspace . if this is the case , the best way of doing the tensor decomposition is the one based on formula ( [ f15 ] ) , regardless of the fact that for large @xmath12 the number of terms obtained can be very large . as is well known , the direct evaluation of the general scalar integral ( [ f5 ] ) ( i.e. ( [ f13 ] ) ) represents a non - trivial problem . however , with the help of the recursion relations , the problem can be significantly simplified in the sense that the calculation of the original scalar integral can be reduced to the calculation of a certain number of simpler fundamental ( basic ) intgerals . recursion relations for scalar integrals have been known for some time @xcite . however , as it turns out , the existing set of relations that can be found in the literature is not sufficient to perform the reduction procedure completely , i.e. for all one - loop integrals appearing in practice . the problem is related to vanishing of various kinematic determinants , it is manifest for the cases corresponding to @xmath51 , and it is especially acute when evaluating one - loop feynman integrals appearing in the nlo analysis of large momentum transfer exclusive processes in pqcd . as is well known , these processes are generally described in terms of feynman diagrams containing a large number of external massless lines . thus , for example , for nucleon compton scattering is @xmath50 . a large number of external lines implies a large number of diagrams to be considered , as well as a very large number of terms generated when performing the tensor decomposition using ( [ f15 ] ) . in view of the above , to treat the feynman integrals ( diagrams ) with a large number of external lines the use of computers is unavoidable . this requires that the scalar reduction procedure be generally applicable . it is therefore absolutely clear that any ambiguity or uncertainty present in the scalar recursion relations constitutes a serious problem . the method presented below makes it possible to perform the reduction completely regardless of the kinematics of the process considered and the complexity of the structure of the contributing diagrams . for the reason of completeness and clearness of presentation and with the aim of comparison with the already existing results , we now briefly present a few main steps of the derivation of recursion relations . it should be pointed out that the derivation essentially represents a variation of the derivation originally given in @xcite . recursion relations for scalar integrals are obtained with the help of the integration - by - parts method @xcite . owing to translational invariance , the dimensionally regulated integrals satisfy the following identity : @xmath52 where @xmath53 are arbitary constants , while @xmath31 are the propagators given by ( [ f2 ] ) . the identity ( [ f16 ] ) is a variation of the identity used in @xcite , where it was assumed that @xmath54 . performing the differentiation , expressing scalar products in the numerator in terms of propagators @xmath31 , choosing @xmath55 , ( which we assume in the following ) and taking into account the scalar integral ( [ f5 ] ) , the identity ( [ f16 ] ) leads to the relation @xmath56 z_i\right ) \nu_j i_0^n(d;\{\nu_k+\delta_{k j}\ } ) } { \nonumber}\\ & = & \!\!\!\sum_{i , j=1}^{n}z_i \nu_j i_0^n(d;\{\nu_k+\delta_{k j}-\delta_{k i}\ } ) - ( d-\sum_{j=1}^{n}\nu_j ) z_0 i_0^n(d;\{\nu_k\ } ) , \label{f20}\end{aligned}\ ] ] where @xmath57 is the kronecker delta symbol . in arriving at ( [ f20 ] ) , it has been understood that @xmath58 the relation ( [ f20 ] ) represents the starting point for the derivation of the recursion relations for scalar integrals . we have obtained the fundamental set of recursion relations by choosing the arbitrary constants @xmath59 so as to satisfy the following system of linear equations : @xmath60 where @xmath61 is an arbitrary constant . introducing the notation @xmath62 , the system ( [ f25 ] ) may be written in matrix notation as @xmath63 it should be pointed out that the expression of the type ( [ f20 ] ) and the system of the type ( [ f27 ] ) , for the case of massive propagators ( @xmath64 ) , see ref . @xcite , can simply be obtained from the relation given above by making a change @xmath65 . consequently , considerations performed for the massive case @xcite apply to the massless case , and vice versa . it should be mentioned that , in the existing literature , the constant @xmath61 used to be chosen as a real number different from zero . however , it is precisely this fact that , at the end , leads to the breakdown of the existing scalar reduction methods . namely , for some kinematics ( e.g. collinear on - shell external lines ) the system ( [ f27 ] ) has no solution for @xmath66 . however , if the possibility @xmath67 is allowed , the system ( [ f27 ] ) will have a solution regardless of kinematics . this makes it possible to obtain additional reduction relations and formulate methods applicable to arbitrary number of external lines and to arbitrary kinematics . if ( [ f25 ] ) is taken into account , and after using the relation @xcite @xmath68 which can be easily proved from the representation ( [ f13 ] ) , the relation ( [ f20 ] ) reduces to @xmath69 where @xmath59 are given by the solution of the system ( [ f25 ] ) , and the infinitesimal part proportional to @xmath26 has been omitted this is a generalized form of the recursion relation which connects the scalar integrals in a different number of dimensions @xcite . the use of the relation ( [ f31 ] ) in practical calculations depends on the form of the solution of the system of equations ( [ f25 ] ) . for general considerations , it is advantageous to write the system ( [ f25 ] ) in the following way : @xmath70 in writing ( [ f32 ] ) , we have taken into account the fact that @xmath55 . in this way , the only free parameter is @xmath71 and by choosing it in a convenient way , one can always find the solution of the above system and , consequently , be able to use the recursion relations ( [ f31 ] ) . in the literature , for example in ref . @xcite , the recursion relations are obtained by inserting the general solution of the system ( [ f25 ] ) , i.e. the system ( [ f32 ] ) , into the relation ( [ f31 ] ) . the recursion relations thus obtained are of limited practical use if the matrices of the mentioned systems are very singular . this happens when there are either two or more collinear external lines or , in general , for @xmath51 . when this is the case , the analysis of the general coefficient of the recursion becomes very complicated and in many cases unmanageable . there are cases when all coefficients vanish . as stated in @xcite , for @xmath72 , owing to the drastic reduction of the recurrence relations these cases need a separate investigation . in addition , the above - mentioned problems with @xmath73 appear . to avoid these problems , a different approach to recursion relations can be taken . it is based on the fact , that finding any solution of the systems of equations mentioned above , makes it possible to perform the reduction . being forced to use computers , it is very convenient and important that the reduction procedure be organized in a such a way that the recursion relations are classified and used depending on the form of the solutions of the above systems . if this is done , the increased singularity of the kinematic determinants turns out to be working in our favour by making it easy to find a solution of the systems of linear equations relevant to the reduction . in the following we frequently refer to two determinants , for which we introduce the notations : for the determinant of the system ( [ f27 ] ) we introduce det@xmath74 , while for the determinant of the system ( [ f32 ] ) we use det@xmath75 . depending on whether the kinematic determinants det@xmath74 and det@xmath75 are equal to zero or not , we distinguish four different types of recursion relations following from ( [ f31 ] ) . before proceeding to consider various cases , note that in the case when det@xmath76 , it holds @xmath77 it should be mentioned that for some of the recursion relations presented below one can find similar expressions in the literature . for reasons of clearness , connections of the relations given below with those existing in the literature are commented upon after the analysis of all possible cases has been considered . let us now discuss all possible cases separately . the most convenient choice in this case is @xmath80 . it follows from ( [ f33 ] ) that @xmath66 , so that the recursion relation ( [ f31 ] ) can be written in the following form : @xmath81 .\label{f34}\end{aligned}\ ] ] as it is seen , this recursion relation connects the scalar integral in @xmath4 dimensions with the scalar integrals in @xmath82 dimensions and can be used to reduce the dimensionality of the scalar integral . since det@xmath83 , some more recursion relations can be directly derived from ( [ f20 ] ) . by directly choosing the constants @xmath59 in ( [ f20 ] ) in such a way that @xmath84 , for @xmath85 , we arrive at a system of @xmath12 equations which is always valid : @xmath86 in the system ( [ f35 ] ) we have again disregarded the non - essential infinitesimal term proportional to @xmath26 . the matrix of the system ( [ f35 ] ) is the same as the matrix of the system ( [ f25 ] ) , whose determinant is different from zero , so that the system ( [ f35 ] ) can be solved with respect to @xmath87 , @xmath88 . the solutions represent the recursion relations which can be used to reduce the powers of the propagators in the scalar integrals . making use of these relations and the relation ( [ f34 ] ) , each scalar integral @xmath89 belonging to the type for which det@xmath90 , det@xmath83 can be represented as a linear combination of integrals @xmath91 and integrals with the number of propagators which is less than @xmath12 . for the dimension @xmath92 , one usually chooses @xmath93 , where @xmath94 is the infinitesimal parameter regulating the divergences . even in the case when one starts with @xmath95 , one can make use of the recursion ( [ f34 ] ) to change from the dimension @xmath4 to the dimension @xmath92 . the most convenient choice in this case is @xmath80 . unlike in the preceding case , it follows from ( [ f33 ] ) that @xmath67 , so that the recursion relation ( [ f31 ] ) can be written as @xmath101 .\label{f39}\ ] ] it follows from ( [ f39 ] ) that it is possible to represent each integral of this type as a linear combination of scalar integrals with the number of propagators being less than @xmath12 . this possibility arises only if the first row of the matrix of the system ( [ f32 ] ) is a linear combination of the remaining rows . then , the system ( [ f32 ] ) has a solution only for the choice @xmath103 . with this choice , the remaining system of equations reduces to the system ( [ f25 ] ) , where the constant @xmath61 can be chosen at will . after the parameter @xmath61 is chosen , the constants @xmath59 are uniquely determined . thus the recursion relation ( [ f31 ] ) with the choice @xmath104 leads to @xmath105 consequently , as in the preceding case , the scalar integrals of the type considered can be represented as a linear combination of scalar integrals with a smaller number of propagators . unlike in the preceding cases , in this case two different recursion relations arise . to derive them , we proceed by subtracting the last , @xmath106@xmath0th , equation of the system ( [ f32 ] ) from the second , third , ... and @xmath12@xmath0th equation , respectively . as a result , we arrive at the following system of equations : @xmath107 as it is seen , the first @xmath12 equations of the above system form a system of equations in which the constant @xmath61 does not appear , and which can be used to determine the constants @xmath59 , @xmath108 . the fact that det@xmath109 implies that the determinant of this system vanishes . therefore , for the system in question to be consistent ( for the solution to exist ) , the choice @xmath103 has to be made . consequently , the solution of the system , @xmath110 , will contain at least one free parameter . inserting this solution into the last , @xmath106@xmath0th , equation of the system ( [ f41 ] ) , we obtain @xmath111 now , by arbitrarily choosing the parameter @xmath61 , one of the free parameters on the left - hand side can be fixed . sometimes , ( for instance , when there are collinear external lines ) the left - hand side of eq . ( [ f42 ] ) vanishes explicitly , although the solution for @xmath59 contains free parameters . in this case the choice @xmath67 has to be made . therefore , corresponding to the case when det@xmath75=det@xmath112 , one of the following two recursion relations holds : @xmath113 obtained from ( [ f31 ] ) by setting @xmath103 and @xmath104 , or @xmath114 obtained from ( [ f31 ] ) by setting @xmath103 and @xmath67 . in the case ( [ f43 ] ) , it is clear that the integral with @xmath12 external lines can be represented in terms of the integrals with @xmath115 external lines . what happens , however , in the case ( [ f44 ] ) ? with no loss of generality , we can take that @xmath116 . the relation ( [ f44 ] ) can then be written in the form @xmath117 we can see that , in this case too , the integral with @xmath12 external lines can be represented in terms of the integrals with @xmath115 external lines . in this reduction , @xmath118 remains conserved . based on the above considerations , it is clear that in all the above cases with the exception of that when det@xmath90 , det@xmath119 , the integrals with @xmath12 external lines can be represented in terms of the integrals with smaller number of external lines . consequently , then , there exists a fundemantal set of integrals in terms of which all integrals can be represented as a linear combination . before moving on to determine a fundamental set of integrals , let us briefly comment on the recursion relations for scalar integrals that can be found in the literature . as we see below , det@xmath75 is proportional to the gram determinant . all recursion relations for which the gram determinant does not vanish are well known . thus the relation of the type ( [ f34 ] ) can be found in refs . @xcite , while the solutions of the systems ( [ f35 ] ) and ( [ f38 ] ) correspond to the recursion relations ( 28 ) and ( 30 ) , respectively , given in ref . even though case ii also belongs to the class of cases for which the gram determinant is different from zero , the system ( [ f27 ] ) has no solution for @xmath66 . this is a reason why the problem with using recursion relations appear in all approaches in which it is required that @xmath66 . this can be seen from the discussion in @xcite ( the method is based on the choice @xmath104 ) where the authors state that the reduction can not be done for @xmath120 with on - shell external lines , and for @xmath121 when one of the mandelstam variables @xmath122 or @xmath123 vanishes . such cases , however , are unavoidable when obtaining leading twist nlo pqcd predictions for exclusive processes at large momentum transfer . on the other hand , in the approach of ref . @xcite , where all coefficients of the recursion are given in terms of det@xmath75 and the minors of the matrix @xmath124 , the relation of the type ( [ f39 ] ) can be obtained ( eq . ( 35 ) in ref . @xcite ) . cases iii and iv , for which the gram determinant vanishes are of special interest . one of the most discussed cases in the literature , belonging to case iii , is @xmath125 . the recursion relations of the type ( [ f40 ] ) can be found in ref . @xcite . as for case iv , it is especially interesting owing to the fact that it includes all cases for @xmath72 . in this case , the systems ( [ f27 ] ) and ( [ f32 ] ) have no unique solution . case iv causes a lot of trouble for approaches in which the recursion coefficients are given in terms of det@xmath75 and the minors of the matrix @xmath124 , for example in ref . the problem consists in the fact that all determinants vanish , making it impossible to formulate the recursion relation , so these cases need a separate investigation . on the other hand , the method of ref . @xcite , based on using pseudo - inverse matrices , can be used to construct the most general solution of the system ( [ f27 ] ) for the case of the vanishing gram determinant . even though the authors of ref . @xcite claim that using their approach one can always perform the reduction of the @xmath12-point function ( @xmath126 ) , that does not seem to be the case . namely , the method in @xcite is based on the choice @xmath104 , and as it has been shown above , in some cases belonging to case iv the system ( [ f25 ] ) has no solution for @xmath66 , implying that the reduction can not be performed . the impossibility of performing the reduction manifests itself such that @xmath127 ( see ( 15 ) and ( 19 ) in @xcite ) , a consequence of which is that the recursion coefficients become divergent . the situation of this kind arises regularly when dealing with integrals containing collinear external lines , i.e. for exceptional kinematics . the method of ref . @xcite has been obtained for non - exceptional kinematics . in view of what has been said above , most of the problems with existing reduction methods appear when dealing with the integrals with a large number of external lines . in all considerations in the literature that happens for @xmath51 . it is very important to point out that , this is valid for the case when external momenta span the 4-dimensional minkowski space . if the dimensionality of space span by external momenta is smaller , the problems start appearing for smaller @xmath12 . even though one can find the statements that such cases are at the moment of minor physical interest , we disagree . namely , as stated earlier , the analysis of exclusive processes in pqcd , even for simple processes , requires evaluation of the diagrams with @xmath128 . since these diagrams contain collinear external lines , the kinematics is limited to ( @xmath129)-dimensional subspace . thus , for example , for nucleon compton scattering the integrals with @xmath50 external lines contribute and the kinematics is limited to the 3-dimensional subspace . a consequence of this is that the problem with using existing reduction methods will start appearing at the level of one - loop @xmath130 diagrams . the reduction method presented in this paper is formulated with an eye on exclusive processes in pqcd . the main point of the method is that the reduction is defined in terms of the solution of the linear systems given by ( [ f27 ] ) and ( [ f32 ] ) . a consequence of this is that the method is quite general , very flexible , practical and easily transfered to the computer program . to perform reduction , one only needs to find solution of the above systems which can always be done . a very pleasing feature of this reduction is that the increased singularity of the kinematic determinants facilitates reduction , since finding a solution of the relevant linear systems becomes easy . we now turn to determine the fundamental set of integrals . to this end , let us first evaluate the determinant of the system ( [ f32 ] ) , det@xmath75 , and determine the conditions under which this kinematic determinant vanishes . by subtracting the last column from the second , third , ... and @xmath1th column , respectively , and then the last row from the second , third , ... and @xmath1th row , respectively , we find that the determinant det@xmath75 is given by the following expression : @xmath131 , \qquad i , j=1,\ldots , n-1 . \label{f46}\ ] ] as it can be seen , det@xmath75 is proportional to the gram determinant . denote by @xmath132 the dimension of the vector space spanned by the vectors @xmath133 , ( @xmath134 ) . owing to the linear dependence of these vectors , the determinant vanishes when @xmath135 . as in practice , we deal with the 4@xmath0dimensional minkowski space , the maximum value for @xmath132 equals 4 . an immediate consequence of this is that all integrals with @xmath136 can be reduced to the integrals with @xmath137 . in view of what has been said above , all one - loop integrals are expressible in terms of the integrals @xmath138 , @xmath139 , @xmath140 , belonging to case i , and the general two@xmath0point integrals @xmath141 , which are simple enough to be evaluated analytically . next , by substituting @xmath142 , @xmath130 and @xmath143 into the recursion relation ( [ f31 ] ) , one finds @xmath144 owing to the fact that the integral @xmath145 is ir finite@xcite , the relation ( [ f47 ] ) implies that the @xmath130 scalar integral , @xmath140 , can be expressed as a linear combination of the @xmath121 scalar integrals , @xmath139 , plus a term linear in @xmath94 . in massless scalar theories , the term linear in @xmath94 can simply be omitted , with a consequence that the @xmath130 integrals can be reduced to the @xmath121 integrals . on the other hand , when calculating in renormalizable gauge theories ( like qcd ) , the situation is not so simple , owing to the fact that the rank @xmath146 tensor integrals are required . in the process of the tensor decomposition and then reduction of scalar integrals all way down to the fundamental set of integrals , there appears a term of the form @xmath147 , which implies that one would need to know an analytical expression for the integral @xmath140 , to order @xmath94 . going back to the expression ( [ f47 ] ) , we notice that all such terms can be written as a linear combination of the box ( 4@xmath0point ) integrals in @xmath93 dimensions and 5@xmath0point integrals in @xmath148 dimensions . therefore , at this point , the problem has been reduced to calculating the integral @xmath145 , which is ir finite , and need to be calculated to order @xmath149 . it is an empirical fact @xcite that in final expressions for physical quantities all terms containing the integral @xmath145 always combine so that this integral ends up being multiplied by the coefficients @xmath150 , and as such , can be omitted in one@xmath0loop calculations . a few theoretical proofs of this fact can be found in literature @xcite , but , to the best of our knowledge , the proof for the case of exceptional kinematics is still missing . that being the case , in concrete calculations , ( to be sure and to have all the steps of the calculation under control ) , it is absolutely necessary to keep track of all the terms containing the integral @xmath145 , add them up and check whether the factor multiplying it is of order @xmath151 . even though the experience gained in numerous calculations shows that this is so , a situation in which the integral @xmath145 would appear in the final result for a physical quantity accompained by a factor @xmath152 would not , from practical point of view , present any problem . namely , being ir finite , although extremely complicated to be evaluated analyticaly , the integral @xmath145 can always , if necessary , be evaluated numerically . based on the above considerations we may conclude that all one@xmath0loop integrals occurring when evaluating physical processes in massless field theories can be expressed in terms of the integrals @xmath153 these integrals , therefore , constitute a minimal set of fundamental integrals . in view of the above discussion , we conclude that the set of fundamental integrals is comprised of integrals with two , three and four external lines . integrals with two external lines can be calculated analytically in arbitrary number of dimensions and with arbitrary powers of the propagators . they do not constitute a problem . as far as the integrals with three and four external lines are concerned , depending on how many kinematic variables vanish , we distinguish several different cases . we now show that in the case @xmath120 we have only one fundamental integral , while in the case corresponding to @xmath121 there are six integrals . for this purpose , we make use of the vanishing of the kinematic determinants det@xmath74 and det@xmath75 . according to ( [ f5 ] ) , the general massless scalar two@xmath0point integral in @xmath4 space@xmath0time dimensions is of the form @xmath155 the closed form expression for the above integral , valid for arbitrary @xmath156 , and arbitrary propagator powers @xmath157 and @xmath158 , is given by @xmath159 where @xmath160 it is easily seen that in the formalism of the dimensional regularization the above integral vanishes for @xmath161 . the massless scalar one@xmath0loop triangle integral in @xmath162 dimensions is given by @xmath163 making use of the representation ( [ f13 ] ) , and introducing the external masses @xmath164 @xmath165 , the integral ( [ f51 ] ) can be written in the form @xmath166 it is evident that the above integral is invariant under permutations of external massess @xmath167 . depending on the number of the external massless lines , and using the above mentioned symmetry , there are three relevant special cases of the above integral . we denote them by @xmath168 the integrals @xmath169 and @xmath170 are ir divergent and need to be evaluated with @xmath171 , while the integral @xmath172 is finite and can be calculated with @xmath173 . now , it is easily found that the determinants of the systems of equations ( [ f27 ] ) and ( [ f32 ] ) are , for @xmath120 , given by @xmath174 as is seen from ( [ f56 ] ) , if at least one of the external lines is on mass - shell , the determinant det@xmath175 vanishes . consequently , using the recursion relations ( case ii or iv ) the integrals @xmath169 and @xmath170 can be reduced to the integrals with two external lines . therefore , we conclude that among the scalar integrals with three external lines the integral @xmath172 is the only fundamental one . the result for this integral is well known @xcite . in @xcite it is expressed in terms of the dimensionless quantities of the form @xmath176 \label{f58}\ ] ] and , being proportional to @xmath177 , appears to have a pole at @xmath178 . it appears that the final expression @xcite is not well defined when @xmath178 . on the basis of eqs.([f57 ] ) and ( [ f58 ] ) , one finds that @xmath179 this equation implies that when @xmath180 , instead of examining the limit of the general expression in @xcite , one can utilize the reduction relations ( [ f40 ] ) ( corresponding to @xmath181 and @xmath182 ) to reduce the ir finite integral with three external lines , @xmath172 , to the integrals with two external lines . the massless scalar one@xmath0loop box integral in @xmath162 space@xmath0time dimensions is given by @xmath183 making use of ( [ f13 ] ) , introducing the external `` masses '' @xmath184 @xmath185 , and the mandelstam variables @xmath186 and @xmath187 , the integral ( [ f59 ] ) becomes @xmath188 introducing the following set of ordered pairs @xmath189 one can easily see that the integral ( [ f60 ] ) is invariant under the permutations of ordered pairs , as well as under the simultaneous exchange of places of elements in any two pairs . the determinants of the coefficient matrices of the systems of equations ( [ f27 ] ) and ( [ f32 ] ) , corresponding to the above integral , are @xmath190~.\label{f63}\end{aligned}\ ] ] by looking at the expression for @xmath191 given in ( [ f62 ] ) it follows that all box integrals @xmath192 that are characterized by the fact that in each of the ordered pairs in ( [ f61 ] ) at least one kinematic variable vanishes , are reducible . therefore , for a box integral to be irreducible , it is necessary that both kinematic variables in at least one of the ordered pairs should be different from zero . owing to the symmetries valid for the box integrals it is always possible to choose that pair to be @xmath193 . taking into account symmetries , and the number of external massless lines , there are six potentially irreducible special cases of the integral ( [ f60 ] ) . adopting the notation of ref . @xcite , we denote them by @xmath194 with all kinematic variables appearing above being different from zero . the results for these integrals are well known @xcite . the integrals ( [ f65]-[f69 ] ) are ir divergent , and as such need to be evaluated with @xmath171 , while the integral ( [ f64 ] ) is finite and can be calculated in @xmath3 . the results for these integrals , obtained in @xcite for arbitrary values of the relevant kinematic variables , and presented in a simple and compact form , have the following structure : @xmath195 + { \cal o}({\varepsilon } ) , \nonumber \\ & & { } k\in \{0m,1m,2me,2mh,3m,4m\}\ , . \label{f70}\end{aligned}\ ] ] the ir divergences ( both soft and collinear ) of the integrals are contained in the first term within the square brackets , while the second term is finite . the function @xmath196 is represented by a sum of powerlike terms , it depends on @xmath94 and is finite in the @xmath197 limit . as for the function @xmath198 , it is given in terms of dilogarithm functions and constants . in the above , @xmath199 is the determinant corresponding to the integral @xmath200 given in ( [ f64]-[f69 ] ) . for the purpose of numerical integration , it is very useful to have the exact limit of the integral @xmath200 when @xmath201 . this limit can be determined in an elegant manner by noticing that for @xmath202 the reduction relations corresponding to cases ii and iv apply , making it possible to represent the box integral @xmath200 as a linear combination of the triangle integrals . this result can be made use of to combine box and triangle integrals ( or pieces of these integrals ) with the aim of obtaining numerical stability of the integrand @xcite . the integrals ( [ f64]-[f69 ] ) are irreducible only if the corresponding kinematic determinant @xmath199 does not vanish . with the help of the tensor decomposition and the scalar reduction procedures , any dimensionally regulated one@xmath0loop @xmath1point feynman integral can be represented as a linear combination of the integrals : @xmath203 multiplied by tensor structures made from the external momenta and the metric tensor . the integrals in ( [ f71 ] ) constitute a fundamental set of integrals . an alternative and more convenient set of fundamental integrals is obtained by noticing that all the relevant box integrals are finite in @xmath2 . on the basis of eq . ( [ f31 ] ) , all ir divergent box integrals can be expressed as linear combinations of triangle integrals in @xmath162 dimension and a box integral in @xmath204 dimension . next , using the same equation , all triangle integrals can be decomposed into a finite triangle integral and two@xmath0point integrals . in the final expression thus obtained all divergences , ir as well as uv , are contained in the general two - point integrals and associated coefficients . therefore , an alternative fundamental set of integrals is comprised of @xmath205 where @xmath206 denotes box integrals in @xmath2 dimensions , explicit expressions for which are given in the appendix a. a characteristic feature of this fundamental set of integrals , which makes it particularly interesting , is that the integral @xmath207 is the only divergent one , while the rest of integrals are finite . in this work we have considered one@xmath0loop scalar and tensor feynman integrals with an arbitrary number of external lines which are relevant for construction of multi@xmath0parton one@xmath0loop amplitudes in massless field theories . main result of this paper is a scalar reduction approach by which an arbitrary @xmath1point scalar one@xmath0loop integral can be reqursively represented as a linear combination of eight basic scalar integrals with rational coefficients depending on the external momenta and the dimensionality of space@xmath0time , provided the external momenta are kept in four dimensions . the problem of vanishing of the kinematic determinants , which is a reflection of very complex singularity structure of these integrals , has been solved in an elegant and transparent manner . namely , the approach has been taken according to which instead of solving the general system of linear equations given in ( [ f25 ] ) , and then finding the limit , which sometimes does nt exists , of the obtained solution corresponding to a given singular kinematic situation , we first obtain and then solve the system of equations appropriate to the situation being considered . our method has been derived without any restrictions regarding the external momenta . as such , it is completely general and applicable for arbitrary kinematics . in particular , it applies to the integrals in which the set of external momenta contains subsets comprised of two or more collinear momenta . this kind of integrals are encountered when performing leading@xmath0twist nlo pqcd analysis of the hadronic exclusive processes at large@xmath0momentum@xmath0transfer . trough the tensor decomposition and scalar reduction presented , any massless one - loop feynman integral with generic 4-dimensional momenta can be expressed as a linear combination of a fundamental set of scalar integrals : six box integrals in @xmath2 , a triangle integral in @xmath3 , and a general two@xmath0point integral . all the divergences present in the original integral are contained in the general two - point integral and associated coefficients . in conclusion , the computation of ir divergent one@xmath0loop integrals for arbitrary number of external lines can be mastered with the reduction formulas presented above . the iterative structure makes it easy to implement the formalism in algebraic computer program . with this work all the conceptual problems concerning the construction of multi@xmath0parton one@xmath0loop amplitudes are thus solved . we thank t.binoth and g.heinrich for useful discussions and helpful comments . this work was supported by the ministry of science and technology of the republic of croatia under contract no . 00980102 . in addition to the explicit calculation , the irreducible box integrals in @xmath2 dimensions can be obtained using the existing analytical expressions for the irreducible box integrals in @xmath162 dimensions and the reduction formula ( [ f31 ] ) . to this end , we substitute @xmath204 , @xmath121 , @xmath143 and @xmath104 into the relation ( [ f31 ] ) and find @xmath208 note that the ir divergences in @xmath162 box integrals are exactly cancelled by the divergences of the triangle integrals . the expressions for the relevant basic massless scalar box integrals in @xmath2 space@xmath0time dimensions are listed below : @xmath209 + \mbox{li}_{2}\big[\ , 1-(t+{\rm i}\epsilon)\ , f^{3m}\ , \big ] \nonumber \\ & & { } -\mbox{li}_{2}\big[\ , 1-(m_{2}^{2}+ { \rm i}\epsilon)\ , f^{3m}\ , \big ] - \mbox{li}_{2}\big[\ , 1-(m_{4}^{2}+{\rm i}\epsilon)\ , f^{3m}\ , \big ] \nonumber \\ & & { } -\frac{1}{2 } \left ( t - m_2 ^ 2-m_3 ^ 2 + 2 m_2 ^ 2 m_3 ^ 2 \frac{t - m_4 ^ 2}{s t - m_2 ^ 2 m_4 ^ 2}\right ) { \cal i}_3(m_2 ^ 2,m_3 ^ 2,t)\nonumber \\ & & { } -\frac{1}{2 } \left ( s - m_3 ^ 2-m_4 ^ 2 + 2 m_3 ^ 2 m_4 ^ 2 \frac{s - m_2 ^ 2}{s t - m_2 ^ 2 m_4 ^ 2}\right){\cal i}_3(m_3 ^ 2,m_4 ^ 2,s ) \bigg\ } .\label{f74}\end{aligned}\ ] ] @xmath210 \nonumber \\ & & { } + \mbox{li}_{2}\big[\ , 1-(t+{\rm i}\epsilon)\ , f^{2mh}\ , \big ] -\mbox{li}_{2}\big[\ , 1-(m_{4}^{2}+{\rm i}\epsilon)\ , f^{2mh}\ , \big ] \nonumber \\ & & { } -\frac{1}{2 } \left ( s - m_3 ^ 2-m_4 ^ 2 + 2 \frac{m_3 ^ 2 m_4 ^ 2}{t } \right){\cal i}_3(m_3 ^ 2,m_4 ^ 2,s ) \bigg\ } .\label{f75}\end{aligned}\ ] ] @xmath211 + \mbox{li}_{2}\big[\ , 1-(t+{\rm i}\epsilon)\ , f^{2me}\ , \big ] \nonumber \\ & & { } -\mbox{li}_{2}\big[\ , 1-(m_{2}^{2}+{\rm i}\epsilon)\ , f^{2me}\ , \big ] -\mbox{li}_{2}\big[\ , 1-(m_{4}^{2}+{\rm i}\epsilon)\ , f^{2me}\ , \big]\ , \bigg\}. \label{f76}\end{aligned}\ ] ] @xmath212 + \mbox{li}_{2}\big[\ , 1-(t+{\rm i}\epsilon)\ , f^{1m}\ , \big ] \nonumber \\ & & { } -\mbox{li}_{2}\big[\ , 1-(m_{4}^{2}+{\rm i}\epsilon)\ , f^{1m}\ , \big ] -\frac{{\pi}^2}{6}\ , \bigg\}. \label{f77}\end{aligned}\ ] ] @xmath213 + \mbox{li}_{2}\big[\ , 1-(t+{\rm i}\epsilon)\ , f^{0m}\ , \big ] -\frac{{\pi}^2}{3}\ , \bigg\ } , \label{f78}\end{aligned}\ ] ] where @xmath214 and @xmath215 the functions appearing above are given by @xmath216 as an illustration of the tensor decomposition and scalar reduction methods , we evaluate an one - loop 6-point feynman diagram shown in fig . [ fig2 ] . note that , due to the kinematics which is bounded to three dimensional minkowski subspace there are no four linearly independent four - vectors . consequently , this diagram is of complexity of the 7-point one - loop diagram with four dimensional external kinematics . we choose this particular diagram because of the compactness of intermediate and final expressions . in the @xmath218 centre - of - mass frame , the 4-momenta of the incoming and outgoing particles are @xmath219 while the polarization states of the photons are @xmath220 where @xmath221 is the total centre - of - mass energy of the @xmath218 system ( or the invariant mass of the @xmath222 pair ) . with the momenta @xmath17 @xmath226 @xmath227 the quantities @xmath228 and @xmath229 ( @xmath230 and @xmath231 ) are the fractions of the momentum @xmath232 ( @xmath233 ) shared between the quark and the antiquark in the @xmath234 ( @xmath235 ) . the integral @xmath237 is composed of one - loop 6-point tensor integrals of rank 0 , 1 , 2 , 3 and 4 . performing the tensor decomposition and evaluating the trace , we obtain the integral @xmath237 in the form @xmath238}.\label{td}\end{aligned}\ ] ] next , performing the scalar reduction using the method described in the paper , we arrive at the following expression for the integral written in terms of the basic integrals : @xmath239 { \nonumber}\\ & & { } + \frac{1}{s}\left [ \frac{1}{({\overline{x}}-x)\,y}\left ( \frac{2\,{\overline{x}}}{{\varepsilon}\,({\overline{x}}-x)}+2-\frac{{\overline{x}}}{x } \right)\,i_2(4 + 2\,{\varepsilon};-s\,{\overline{x}}/2)\right.{\nonumber}\\ & & { } + \frac{1}{(x-{\overline{x}})\,{\overline{y}}}\left ( \frac{2\,x}{{\varepsilon}\,(x-{\overline{x}})}+2-\frac{x}{{\overline{x } } } \right)\,i_2(4 + 2\,{\varepsilon};-s\,x/2){\nonumber}\\ & & { } + \frac{1}{({\overline{y}}-y)\,x}\left ( \frac{2\,{\overline{y}}}{{\varepsilon}\,({\overline{y}}-y)}+2-\frac{{\overline{y}}}{y } \right)\,i_2(4 + 2\,{\varepsilon};-s\,{\overline{y}}/2){\nonumber}\\ & & { } + \frac{1}{(y-{\overline{y}})\,{\overline{x}}}\left ( \frac{2\,y}{{\varepsilon}\,(y-{\overline{y}})}+2-\frac{y}{{\overline{y } } } \right)\,i_2(4 + 2\,{\varepsilon};-s\,y/2){\nonumber}\\ & & { } + \left ( \frac{(1-x\,{\overline{y}}-3\,y\,{\overline{x}})(1-y\,{\overline{x}}-3\,x\,{\overline{y}})}{x\,{\overline{x}}\,y\,{\overline{y}}\ , ( x-{\overline{x}})(y-{\overline{y}})}+\frac{2 ( { \overline{x}}\,{\overline{y}}+x\,y)(8\,x\,{\overline{x}}\,y\,{\overline{y}}-x\,{\overline{x}}-y\,{\overline{y } } ) } { { \varepsilon}\,x\,{\overline{x}}\,y\,{\overline{y}}\ , ( x-{\overline{x}})^2(y-{\overline{y}})^2}\right){\nonumber}\\ & & { } \left.\left.\times\,i_2(4 + 2\,{\varepsilon};-s\,({\overline{x}}\,{\overline{y}}+x\,y)/2)+\frac{({\overline{x}}\,{\overline{y}}+x\,y)}{x\,{\overline{x}}\,y\,{\overline{y}}}\,i_2(4 + 2\,{\varepsilon};-s/2)\right ] \right \ } .\label{fin}\end{aligned}\ ] ] here , @xmath240 is the two - point scalar integral in @xmath241 with @xmath242 , while @xmath243 and @xmath244 are box scalar integrals in @xmath245 . analytic expressions for these integrals are given in the appendix a. expanding eq . ( [ fin ] ) up to order @xmath246 , we finally get @xmath247 \right \ } , \label{fin1}\end{aligned}\ ] ] where @xmath248 . 99 g. passarino and m. veltman , nucl . phys . b * 160 * ( 1979 ) 151 ; g. j. van oldenborgh and j. a. m. vermaseren , z. phys . c * 46 * ( 1990 ) 425 ; w. l. van neerven and j. a. m. vermaseren , phys . b * 137 * ( 1984 ) 241 . t. binoth , j. p. guillet and g. heinrich , nucl . b * 572 * ( 2000 ) 361 [ arxiv : hep - ph/9911342 ] ; g. heinrich and t. binoth , nucl . * 89 * ( 2000 ) 246 [ arxiv : hep - ph/0005324 ] . t. binoth , j. p. guillet , g. heinrich and c. schubert , nucl . b * 615 * ( 2001 ) 385 [ arxiv : hep - ph/0106243 ] . j. m. campbell , e. w. glover and d. j. miller , nucl . b * 498 * ( 1997 ) 397 [ arxiv : hep - ph/9612413 ] . a. denner and s. dittmaier , nucl . b * 658 * ( 2003 ) 175 [ arxiv : hep - ph/0212259 ] . g. duplani ' c and b. nii ' c , eur . j. c * 20 * ( 2001 ) 357 [ arxiv : hep - ph/0006249 ] . g. duplani ' c and b. nii ' c , eur . j. c * 24 * ( 2002 ) 385 [ arxiv : hep - ph/0201306 ] . a. i. davydychev , j. phys . a * 25 * ( 1992 ) 5587 . k. g. chetyrkin and f. v. tkachov , nucl . b * 192 * ( 1981 ) 159 ; f. v. tkachov , phys . b * 100 * ( 1981 ) 65 . f. jegerlehner and o. tarasov , nucl . proc . suppl . * 116 * ( 2003 ) 83 [ arxiv : hep - ph/0212004 ] . a. t. suzuki , e. s. santos and a. g. schmidt , eur . j. c * 26 * ( 2002 ) 125 [ arxiv : hep - th/0205158 ] . c. anastasiou , e. w. glover and c. oleari , nucl . b * 565 * ( 2000 ) 445 [ arxiv : hep - ph/9907523 ] . a. t. suzuki , e. s. santos and a. g. schmidt , arxiv : hep - ph/0210083 .

based on classical density functional theory ( dft ) , we investigate the demixing phase transition of a two - dimensional , binary heisenberg fluid mixture . the particles in the mixture are modeled as gaussian soft spheres , where one component is characterized by an additional classical spin - spin interaction of heisenberg type . within the dft we treat the particle interactions using a mean - field approximation . for certain magnetic coupling strengths we calculate phase diagrams in the density - concentration plane . for sufficiently large coupling strengths and densities , we find a demixing phase transition driven by the ferromagnetic interactions of the magnetic species . we also provide a microscopic description ( i.e. , density profiles ) of the resulting non - magnetic / magnetic fluid - fluid interface . finally , we investigate the phase separation using dynamical density functional theory ( ddft ) , considering both nucleation processes and spinodal demixing .

1109.6756

the theoretical description of the phase separation of fluid mixtures is a long - standing problem with importance in many areas of soft matter physics such as , e.g. , the stability of molecular and colloidal solutions@xcite , the interactions between nanoparticles and macromolecules including novel phenomena such as casimir forces @xcite , as well as interfacial and confinement effects occurring in the presence of surfaces . indeed , even simple " mixtures consisting of spherical particles can display non - trivial phase behaviour including triple points and critical end points not present for one - component systems @xcite . correspondingly , even more complex behaviour is observed for particles with internal degrees of freedom such as magnetic particles @xcite and mixtures involving shape - anisotropic particles such as colloidal rods @xcite . besides purely numerical approaches such as monte - carlo or molecular ( brownian ) dynamics computer simulations , classical density functional theory ( dft ) has proved to be a very accurate tool for describing both the homogeneous phase behaviour of mixtures and , at least for simple models , also the inhomogeneous structure occurring at interfaces . the key quantity in dft is the one - particle density , which is obtained through minimization of a grand canonical free energy functional corresponding to the microscopic hamiltonian of the system @xcite . in addition to yielding the equilibrium phase diagram and microscopic fluid structure , dft techniques have been successfully used to calculate nucleation barriers for state points in the metastable region of the phase diagrams @xcite . based on the success of these approaches , it seems very tempting to use dft techniques also to tackle the _ non - equilibrium dynamics _ of the phase separation , including the growth of nuclei , the actual nucleation pathway and coarsening processes during spinodal decomposition . indeed , one motivation for studying the phase separation dynamics accompanying demixing transitions is their important role in the context of pattern formation and self - organization @xcite . traditionally , phase separation dynamics is studied using mesoscopic models involving equations of motion for coarse - grained order parameters @xcite . the advantage of addressing these topics using dft is that the latter allows one to establish the link between the macroscopic behaviour of the system to the microscopic hamiltonian , which is naturally incorporated via the excess helmholtz free energy functional . in the last few years , the first steps in these directions have already been made on the basis of the so - called dynamical density functional theory@xcite ( ddft ) , which consists of a generalized continuity equation for the one - body density distribution of a many - particle systems of overdamped ( brownian ) colloidal particles . recent applications of the ddft to phase separation kinetics include spinodal decomposition in spherical fluids @xcite and heterogeneous nucleation at solid surfaces @xcite . however , most of these studies have been devoted to simple fluids with no internal degrees of freedom . in the present work , we use both static and dynamic dft to explore the phase separation of one of the simplest examples of a mixtures with internal degrees of freedom , that is , a binary fluid of spherical particles where one species carries a classical , ( 3d ) heisenberg spin . heisenberg fluids @xcite are basic models for continuum systems exhibiting ferromagnetic order , particularly for the description of ferromagnetism in undercooled liquid metal alloys . mixtures of such systems and , in particular , mixtures of magnetic and nonmagnetic particles are promising candidates for the controlled fabrication of patterns on the micron scale @xcite . moreover , an obvious attractive feature of these systems is that the phase separation and thus , the occurrence of patterns can be tuned by external magnetic fields . from the theoretical side , the equilibrium properties of one - component heisenberg fluids @xcite , as well as other spin fluids with two - dimensional ( xy ) and ising spins and mixtures thereof have been extensively studied by mc simulations , integral equation methods , and ( mean field ) density functional theories ( see ref . and references therein ) . however , this microscopic level of description for the dynamic behaviour is essentially unexplored . as a starting point to fill this gap we consider here a heisenberg mixture in two spatial dimensions in the absence of an external field . the restriction to a 2d situation is actually close to many experiments ( see , e.g. , ref . ) and has the advantage that the calculated structure can be easily visualized . within the vast parameter space characterizing our model , we focus on a situation where the system demixes into a non - magnetic and a ferromagnetic phase . for this situation , we first use conventional ( static ) dft to calculate a complete phase diagram ( involving a first - order transition and a tricritical point ) , as well as the inhomogeneous fluid density and magnetization profiles characterizing the liquid interface . based on this information we then consider the phase separation dynamics , focussing on the nucleation of non - magnetic bubbles within the ferromagnetic liquid phase . for this problem , we compare the results of three different approaches , namely classical nucleation theory ( cnt ) , which is based on macroscopic concepts , an approach based on equilibrium dft , and finally ddft . we demonstrate that both dft approaches yield consistent results for the nucleation barriers , but predict different pathways due to the fact that the ddft conserves the densities ( contrary to dft ) . moreover , the dft nucleation barriers differ from the cnt predictions when the size of the critical nucleus becomes small . we also present evidence that the ddft can describe the coarsening process during spinodal decomposition . the remainder of this paper is organized as followed . in sec . ii we formulate the model hamiltonian for the binary system . the equilibrium theory of the demixing transition is presented in sec . iii , which includes a calculation of the phase diagram and the interfacial structure . in sec . iv we turn to discuss the dynamics of the demixing transition . finally , we summarize the results in sec . v. + + the fluid system that we investigate is a binary mixture composed of two species . one species ( @xmath0 ) is composed of spherical particles which interact via purely isotropic and repulsive forces . the other species ( @xmath1 ) consists of magnetic particles . in addition to the repulsive interaction induced by the particle cores , these particles carry magnetic moments . the interaction part of the hamiltonian may therefore be decomposed into a core part and a contribution from the spin - spin interaction@xmath2 where @xmath3 and @xmath4 in our model , the particles are confined to a ( two - dimensional ) plane , so that the position of particle @xmath5 , denoted @xmath6 , but the magnetic moment is represented by a three - dimensional normalized classical spin @xmath7 whose orientation is described by the euler angles @xmath8 . for the magnetic interaction , we choose the heisenberg model , @xmath9 where @xmath10 determines the range of the spin - spin interaction . we further assume that @xmath10 can be described by yukawa s potential , that is , @xmath11 for interparticle distances @xmath12 the interaction between two magnetic particles is assumed to be small as compared to the repulsion from the core potentials ( see below ) and we therefore set the yukawa potential in our model to zero in this region . the sign of the coupling constant @xmath13 in eq . ( [ eq.4 ] ) ( where @xmath14 is boltzmann s constant and @xmath15 is the temperature ) determines which type of spin ordering is preferred . as we show below in sec . [ subsect.demixing ] , the choice @xmath16 yields a spontaneous ferromagnetic ordering of the magnetic component of the mixture at temperatures @xmath15 below a curie temperature @xmath17 . on the other hand , the choice @xmath18 favors antiferromagnetic ordering . to model the repulsion between the particles , we choose a gaussian with height @xmath19 and width @xmath20 . the resulting gaussian core " model ( gcm ) , first studied by stillinger @xcite , is given by @xmath21 the gcm is often used as an approximation for the effective interactions between the centre of mass of two `` soft '' particles , such as polymers and star - polymers@xcite or dendrimers@xcite . the dimensionless quantity @xmath22 determines the strength of the repulsion , and the range parameter @xmath20 roughly corresponds to the radius of gyration of the ` particles ' . for the magnetic interaction , we choose a positive coupling constant @xmath16 . hence , in the ferromagnetic phase the magnetic contribution to the pair potential acts effectively as an attractive tail to the repulsive ( soft ) core . the central quantity in density functional theory is the singlet ( one body ) density distribution @xmath23 . following other studies of molecular magnetic fluids @xcite we assume that the singlet density can be factorized into a translational ( number density ) part , @xmath24 , and an orientational distribution function , @xmath25 , that is , @xmath26 the orientational distribution is normalized , so that , @xmath27 which yields @xmath28 for the particles without orientational degrees of freedom ( species @xmath0 ) . the equilibrium fluid singlet density distribution is that which minimizes the grand free energy functional @xcite @xmath29=\mathcal{f}[\{\rho_\alpha\}]-\sum_\alpha\negthickspace\int\negthickspace d\omega\negthickspace\int\negthickspace d\mathbf{r } \big[\mu_\alpha - v_\text{ext}(\mathbf{r},\omega)\big]\rho_\alpha(\mathbf{r},\omega),\label{eq.7}\end{aligned}\ ] ] where @xmath30 is the helmholtz free energy functional and @xmath31 denotes the chemical potential for species @xmath32 . note also that @xmath33 denotes a two - dimensional spatial integral . for a given external potential @xmath34 and interaction potentials , the functional in eq . ( [ eq.7 ] ) has a minimum at the equilibrium density @xmath35 . the functional @xmath36 $ ] is then identical to the grand canonical potential @xmath37 ( cf . ref . ) . in the present study we set the external potential @xmath38 . the helmholtz free energy functional can be split up into two contributions : @xmath39=&\mathcal{f}_\text{id}[\{\rho_\alpha\}]+\mathcal{f}_\text{ex}[\{\rho_\alpha\}]\label{eq.8},\end{aligned}\ ] ] where @xmath40=\sum\limits_\alpha\int d\mathbf{r}\int d\omega \rho_\alpha(\mathbf{r},\omega)[\ln(\lambda_\alpha^2\rho_\alpha(\mathbf{r},\omega))-1]$ ] is the ideal gas contribution and @xmath41 $ ] is the excess part . using eqs . ( [ eq.7a ] ) and ( [ eq.7b ] ) , the ideal gas contribution becomes @xmath42=&k_bt\negthickspace\int\negthickspace d\mathbf{r}\rho_a(\mathbf{r})\big[\ln(\lambda_a^2\rho_a(\mathbf{r } ) ) -\ln 4\pi-1\big]\\ & + k_bt\negthickspace\int\negthickspace d\mathbf{r}\rho_b(\mathbf{r})\big[\ln(\lambda_b^2\rho_b(\mathbf{r } ) ) -1\big]\nonumber\\ & + k_bt\negthickspace\int\negthickspace d\mathbf{r}\rho_b(\mathbf{r})\negthickspace\int\negthickspace d\omega h_b(\mathbf{r},\omega ) \ln\big[h_b(\mathbf{r},\omega)\big]\label{eq.9},\end{aligned}\ ] ] where @xmath43 denotes the thermal de broglie wavelength of species @xmath32 . the particle interactions enter into the excess part of the free energy functional , which can be written as@xcite @xmath44=&\frac12\sum_{\alpha,\beta}\int\limits_0 ^ 1\negthickspace d\lambda\negthickspace\int\negthickspace d\mathbf{r}\negthickspace \int\negthickspace d\mathbf{r'}\negthickspace\int\negthickspace d\omega\negthickspace\int\negthickspace d\omega ' \nonumber\\ & \times \rho^{(2)}_{\alpha\beta}(\mathbf{r},\mathbf{r'},\omega,\omega';\lambda)v^{\alpha\beta}(|\mathbf{r}-\mathbf{r'}|,\omega,\omega')\label{eq.10}.\end{aligned}\ ] ] equation ( [ eq.10 ] ) is exact for systems with pair interactions . the function @xmath45 is the two - body density distribution function which is determined by the pair interactions @xmath46 , and @xmath47 is a charging " parameter@xcite . in general , the function @xmath48 is not known exactly . here we employ a mean - field ( mf ) approximation by setting @xmath49 , i.e. that the pair correlation function is set to one . previous studies@xcite have shown that this simple mf approximation for the gcm yields reliable results for the fluid structure and thermodynamics , particularly at higher densities . within the mf approximation , the excess part of the helmholtz free energy may be written as the following sum @xmath50 where the contribution due to the interactions between the non - magnetic particles is @xmath51=\dfrac{1}{2}\int\negthickspace d\mathbf{r } \negthickspace\int\negthickspace d\mathbf{r}'\rho_a(\mathbf{r})\rho_a(\mathbf{r}')v_{\text{core}}(|\mathbf{r}-\mathbf{r}'|)\label{eq.12},\end{aligned}\]][eq.10a ] and for the magnetic particles @xmath52=\dfrac{1}{2}\int\negthickspace d\omega\negthickspace\int\negthickspace d\omega ' h_b(\mathbf{r},\omega)h_b(\mathbf{r}',\omega')\int\negthickspace d\mathbf{r } \negthickspace\int\negthickspace d\mathbf{r}'\nonumber\\ & \thickspace\thickspace\thickspace\thickspace\times\rho_b(\mathbf{r})\rho_b(\mathbf{r}')\big[v_{\text{core}}(|\mathbf{r}-\mathbf{r}'|)+j(|\mathbf{r}-\mathbf{r}'|)\mathbf{s}\cdot\mathbf{s'}\big].\label{eq.13}\end{aligned}\ ] ] the contributions @xmath53 and @xmath54 to the excess free energy functional are equal because of the symmetry of the pair potentials between species @xmath0 and @xmath1 ( cf . ( [ eq.1 ] ) ) , that is , @xmath55&=\dfrac{1}{2}\int\negthickspace d\mathbf{r } \negthickspace\int\negthickspace d\mathbf{r}'\rho_a(\mathbf{r})\rho_b(\mathbf{r}')v_{\text{core}}(|\mathbf{r}-\mathbf{r}'|)\nonumber\\&=\mathcal{f}_\text{ex}^{ba}[\left\{\rho_\alpha\right\}]\end{aligned}\ ] ] the equilibrium densities @xmath56 are found by minimizing the grand free energy functional given in eq . ( [ eq.7 ] ) : @xmath57}{\partial \rho_\alpha(\mathbf{r})}\right|_{\rho_\alpha^{(0)}(\mathbf{r})}=0\nonumber,\\ & \left . \dfrac{\partial \omega[\rho_\alpha , h]}{\partial h_\alpha(\mathbf{r},\omega)}\right|_{h_\alpha^{(0)}(\mathbf{r},\omega)}=0.\label{eq.14a}\end{aligned}\ ] ] equation ( [ eq.14a ] ) yields an implicit equation for @xmath58 , @xmath59 where the effective field @xmath60 thus , the orientational distribution is determined solely by the scalar product between the spin and the effective field @xmath61 . this is an exact result within the mf approximation , which has been previously applied also to three - dimensional heisenberg fluids@xcite as well as in other contexts such as in liquid crystal theory.@xcite in the low - temperature ferromagnetic state , the orientational order is uniaxial with respect to a director @xmath62 . thus , the angular distribution reduces to @xmath63 , where @xmath64 . the integration over orientation in eq . ( [ eq.15 ] ) then becomes@xcite @xmath65 where @xmath66 is the langevin function . the latter also defines the local magnetization @xmath67 . inserting eq . ( [ eq.14 ] ) into eq . ( [ eq.13 ] ) , the heisenberg contribution to the free energy functional can be written as @xmath68=&\dfrac{1}{2 } \int\negthickspace d\mathbf{r}\negthickspace \int\negthickspace d\mathbf{r}'\rho_b(\mathbf{r})\rho_b(\mathbf{r}')\nonumber\\ & \thickspace\thickspace\thickspace\thickspace\times\big[v_{\text{core}}(|\mathbf{r}-\mathbf{r}'|)+l(\mathbf{r})l(\mathbf{r}')j(|\mathbf{r}-\mathbf{r}'|)\big]\label{eq.16}.\end{aligned}\ ] ] in this section , we employ the density functional approach introduced above to investigate the phase behavior of the bulk binary mixture . we restrict ourselves to fluid phases . the state of the system can be characterized by the strength of the repulsion @xmath69 , the magnetic coupling parameter @xmath70 , the reduced total number density @xmath71 in the system , and the concentration @xmath72 of the magnetic component . the densities of the individual component can then be written as @xmath73 and @xmath74 . assuming that the system is homogeneous , the helmholtz free energy per particle @xmath75 follows from eqs . ( [ eq.8])-([eq.10 ] ) as @xmath76+\nonumber\\ & \dfrac12\rho\big[(1-x)^2\hat{v}^{aa}(0)+2x(1-x)\hat{v}^{ab}(0)+x^2 \hat{v}^{bb}(0)\big]\label{eq.18}.\end{aligned}\ ] ] in eq . ( [ eq.18 ] ) , the coupling matrix elements @xmath77 and @xmath78 denote the integrated strength of the repulsive core potential , or equivalently the @xmath79 limit of the fourier transform of eq . ( [ eq.2 ] ) , that is , @xmath80 the element @xmath81 involves , in addition , an integral over the magnetic interaction , i.e. @xmath82 in the following , we investigate the possibility of fluid - fluid phase separation . the thermodynamic stability conditions for a binary mixture are given by @xcite @xmath83 where @xmath84 is the volume per particle . the first stability condition expresses that the compressibility must be positive , the second ensures stability against spontaneous demixing at constant volume , and the last inequality is the condition for stability at constant pressure . as shown in previous studies@xcite , it is more convenient to use these stability conditions in a constant - pressure ensemble . to this end , we perform a legendre transform of the helmholtz free energy per particle , yielding the gibbs free energy per particle @xmath85 for two phases @xmath86 and @xmath87 to coexist in equilibrium , the chemical potentials of each species @xmath88 have to be equal , and the same holds for the pressure and the temperature . in other words , one has @xmath89 , @xmath90 and @xmath91 , where @xmath92 are the densities of the components in the two phases . these equilibrium conditions lead to a common - tangent construction on the gibbs free energy , @xmath93 where @xmath94 is the ( bulk ) pressure at coexistence . the spinodal is given by the inflection points of @xmath95 , that is @xmath96 figure [ fig.1 ] shows the phase diagram for the ( 2d ) bulk binary mixture whose free energy is given by eq . ( [ eq.18 ] ) . we consider a fixed repulsion strength @xmath97 and two different magnetic coupling parameters , @xmath98 and @xmath99 . for both parameter sets we find a first - order demixing phase transition appearing at densities above a critical density @xmath100 . moreover , the demixing is coupled to a transition from a paramagnetic phase ( @xmath101 ) , which is rich in @xmath0-particles , to a ferromagnetic phase ( @xmath102 ) rich in @xmath1-particles . the onset of magnetic order is determined by the curie line ( blue line in fig . 1 ) , which is obtained by making a taylor expansion of the langevin function @xmath103 [ see eq . ( [ eq.14 ] ) and below ] around @xmath101 combined with the expression for the effective field @xmath1 [ see eq . ( [ eq.15 ] ) ] ( note that we consider spatially homogeneous systems here ) . the resulting critical concentration @xmath104 as function of the total density is given by @xmath105 inspecting the position of the curie line in the phase diagrams in fig . [ fig.1 ] , we see that the system is entirely disordered , regardless of the concentration , for values of the total density @xmath106 ( @xmath107 ) for @xmath108 ( @xmath109 ) . increasing the density from these values towards the critical density , the transition from the paramagnetic into the ferromagnetic phase is of second order , until the curie line meets with the demixing coexistence curve . this merging occurs directly at the demixing critical density ( and critical concentration ) , corresponding to a _ tricritical _ point . at densities @xmath110 , the curie line then coincides with the low - concentration branch of the demixing spinodal . this reflects the fact that it is the spin - spin interaction [ see eq . ( [ eq.1 ] ) ] which drives the phase separation . indeed , as can be seen from eq . ( [ eq.16 ] ) , the spin - spin interaction reduces the free energy of the system whenever the magnetization is non - zero . the demixing spinodal has been calculated using eq . ( [ eq.27 ] ) . inside the spinodal the mixture is thermodynamically unstable , as indicated by the red areas in fig . [ fig.1 ] . the black areas in fig . [ fig.1 ] indicate the metastable regions between the spinodal and the coexistence curve . the corresponding coexisting densities @xmath111 , @xmath112 and concentrations @xmath113 , @xmath114 are calculated using eq . ( [ eq.26 ] ) . recall that coexisting phases are at equal pressure ; we display a number of isobars in the density - concentration plane , which are indicated by the yellow lines in fig . [ fig.1 ] . for each value of the interaction parameter @xmath115 we also display , for one particular exemplary pressure value , a pair of coexisting state points ( see circles ) . from this one may observe that the first - order phase transition is indeed mainly a demixing transition in the sense that the total density change on crossing the transition is small . finally , a comparison of figs . 1a ) and b ) reveals that as @xmath115 is reduced the critical point shifts to larger values of @xmath116 . this is because an increase of the number density of the magnetic component supports the ferromagnetic phase transition . in sec . [ subsect.demixing ] we showed that the magnetic mixture displays a first - order demixing phase transition for a broad range of densities @xmath116 and concentrations @xmath72 . in this section , we focus on the structure of the fluid - fluid interface between the two demixed phases for densities above @xmath100 . the grand canonical free energy @xmath37 for the non - uniform binary mixture is given by eq . ( [ eq.7 ] ) . setting the external potential @xmath117 yields a well - defined ( one - dimensional ) interface between one region enriched with particles from the non - magnetic species and a second region enriched with magnetic particles . setting the functional derivative of eq . ( [ eq.7 ] ) to zero yields the euler - lagrange equations for the chemical potential of species @xmath0 @xmath118+\int\negthickspace d\mathbf{r}'\big[\rho_a(z')+\rho_b(z')\big]v_{\text{core}}(|\mathbf{r}-\mathbf{r'}|)\label{eq.mua},\end{aligned}\ ] ] and species @xmath1 , respectively , @xmath119+b(z)\coth\big[b(z)\big]-1\nonumber\\ & + \ln\left[\frac{b(z')}{\sinh\left[b(z')\right]}\right]\bigg\}\nonumber\\ & + \int\negthickspace d\mathbf{r'}\big[\rho_a(z')+\rho_b(z')\big]\beta v_{\text{core}}(|\mathbf{r}-\mathbf{r'}|)\nonumber\\ & + l\big[b(z)\big]\int\negthickspace d\mathbf{r ' } l\big[b(z')\big]\rho_b(z')\beta j(|\mathbf{r}-\mathbf{r'}|)\label{eq.mub},\end{aligned}\ ] ] where @xmath120 . ) at the interface for different values of @xmath121 . the coupling parameters are @xmath97 and @xmath108.,title="fig:",width=302 ] + ) at the interface for different values of @xmath121 . the coupling parameters are @xmath97 and @xmath108.,title="fig:",width=302 ] using the ( bulk ) coexisting densities @xmath122 , @xmath123 found from the calculation of the binodal in section [ subsect.demixing ] , the chemical potentials @xmath124 and @xmath125 can be eliminated from eqs . ( [ eq.mua ] ) and ( [ eq.mub ] ) . this leads to the equations for the one - body density profiles across the free interface . specifically , we obtain for the non - magnetic component @xmath126,\label{eq.32}\end{aligned}\ ] ] and for the magnetic component @xmath127}+b^\text{bulk}l^\text{bulk}-b(z')l(z')\nonumber\\ & \left .- l\big[b(z)\big]\int\negthickspace d\mathbf{r'}l\big[b(z')\big]\big(\rho_b^{\text{bulk}}-\rho_2(z')\big)\beta j(|\mathbf{r}-\mathbf{r'}|)\right]\label{eq.33}.\end{aligned}\ ] ] equations ( [ eq.32 ] ) and ( [ eq.33 ] ) can be solved self - consistently . the results for the density profiles and the magnetization are shown in fig . [ fig.2 ] for a fixed magnetic coupling parameter @xmath108 ( see fig . 1a ) for the corresponding phase diagram ) . the decay of the density profiles into the two bulk phases in fig . [ fig.2 ] is monotonic . in previous studies of binary mixtures of soft particles @xcite and also of colloid - polymer mixtures @xcite it was found that non - monotonic oscillatory decay of the density profiles can occur for the free interface between coexisting state points that are sufficiently far removed in the phase diagram from the critical point . we expect a similar scenario for the present system as that observed in the systems studied in refs . , i.e. we expect to observe oscillatory behavior in the density profiles also for the present system , but for higher values of the total bulk density , i.e. further away from the critical point . note that the freezing transition@xcite of the pure gcm fluid in three dimensions occurs only at much lower temperatures ( i.e. much larger values of @xmath128 ) than we consider here . approaching the critical density from above , the interface softens , resulting in a reduced pressure parallel to the interface . physically , this softening is reflected by a decreasing line tension ( in 2d ) . to calculate the line tension we assume that the dividing surface is a line that is orientated along the @xmath72-direction . the pressure tensor @xmath129 is a ( @xmath130)-matrix characterized by one component parallel to the interface ( @xmath131 ) and one component normal to the interface ( @xmath132 ) , that is @xmath133 with the interface being located at @xmath134 . the coupling parameters are @xmath97 and @xmath108.,width=302 ] here , @xmath135 and @xmath136 are normalized unit vectors in the @xmath72-direction and @xmath137-direction , respectively . in equilibrium , the normal pressure @xmath138 is constant and equals the bulk pressure @xmath94 at coexistence . furthermore , the tangential component @xmath131 only depends on @xmath137 . the interfacial ( line ) tension @xmath139 is then defined as the excess force resulting from the dividing interface@xcite , that is , @xmath140 to actually calculate @xmath141 , we use the relation@xcite @xmath142 , where @xmath143 is defined as the grand canonical free energy per unit length calculated in the absence of an external potential . numerical results for the function @xmath144 [ i.e. , the integrand in eq . ( [ eq.surftens ] ) ] at various total densities @xmath71 are shown in fig . we only find non - vanishing values of the function @xmath144 near the interface . this reflects the simple fact that the interfacial tension stems from the density inhomogeneities at the interface ( see fig . 2 ) . from fig . 2 we see that the density profiles become smoother as the critical point is approached . this behavior is mirrored by the function @xmath144 ( see fig . 3 ) , resulting in a vanishing interfacial tension at the critical point . the behavior of @xmath139 as a function of the pressure difference @xmath145 is displayed in fig . [ fig.4 ] . between the demixed phases as function of the pressure difference relative to the critical point . the coupling parameters are @xmath108 , @xmath97.,width=302 ] we now turn to discuss the dynamics of phase separation in the present system . to this end , we employ the ddft approach,@xcite in which the time evolution of the one - particle densities are governed by a generalized continuity equation . the latter may be derived by integrating the smoluchowski equation , that is , the fokker - planck equation for a system of ( colloidal ) particles with _ overdamped _ stochastic equations of motion ( i.e. the inertial terms in the microscopic langevin equations of motion are neglected ) . the key approximation of the ddft approach is that the non - equilibrium two - body density distribution functions at time @xmath146 are set equal to those of an _ equilibrium _ system with the same one - body density profile@xcite . as a consequence , the currents entering the ddft equations are determined by ( functional derivatives of ) the _ equilibrium _ helmholtz free energy functional . as in most ddft applications so far , we neglect here the effect of hydrodynamic ( solvent - induced ) interactions between the particles . for the present system , where one species ( the @xmath1-particles ) has internal degrees of freedom , one should employ the ddft equations for anisotropic particles recently proposed in ref . , which shows that the equations of motion for the position- and angle - dependent densities involve both the usual translational currents @xmath147 , and also ` rotational current ' terms resulting from application of the angular momentum operator to the thermodynamic driving force @xmath148 . however , the numerical solution of the resulting set of equations of motions for the demixing problems that we consider here ( see below ) involve simultaneously determining not only the two - dimensional ( isotropic ) number density profiles of the @xmath0- and @xmath1-particles , respectively , but also of the orientational distribution function @xmath149 . the angle - dependence of the latter induces additional dimensions making the numerical calculations rather involved . in the present study , we somewhat simplify the problem by assuming that the magnetic degrees of freedom are _ at each moment in time _ in equilibrium with the density profiles . physically , this assumption implies that the relaxation time of the magnetic moments is much shorter than that of the translational degrees of freedom . this assumption implies that the functional derivative @xmath150 at _ all times _ @xmath146 , i.e. , there is no driving torque . under these conditions , the ddft equations for the present system reduce to a coupled set of equations for the number density profiles of the @xmath0- and @xmath1- particles @xmath151}{\delta \rho_\alpha(\mathbf{r},t)}\right],\label{eq.34}\end{aligned}\ ] ] combined with the self - consistency relation @xmath152 where the time - dependent effective field @xmath153 is given by eq . and where @xmath30 is the mf helmholtz free energy functional developed in sec . [ section : thedft ] . the mobility coefficients @xmath154 in eq . ( [ eq.34 ] ) are related to the diffusion constants via @xmath155 , where @xmath156 or @xmath1 . in what follows we assume that these are equal : @xmath157 . when the external potential @xmath158 , the uniform density distributions , @xmath159 , always correspond to a stationary solution of the ddft eq . , since in this case the functional derivatives @xmath160 are constants . however , if one considers applying small harmonic perturbations to the uniform densities @xmath161 , where @xmath162 , with wave number @xmath163 , and where @xmath164 , then one finds that inside the spinodal region ( the red region in fig . [ fig.1 ] ) density fluctuations with certain wavenumbers @xmath165 grow with time@xcite i.e. within the spinodal the fluid is linearly unstable . outside the spinodal , the system is linearly stable , i.e. the amplitude of any small amplitude density fluctuations decrease over time . in the region between the binodal and the spinodal ( the black regions in fig . [ fig.1 ] ) , the fluid is linearly stable , but is not absolutely stable : if the amplitude of a given density perturbation is large enough , then the amplitude of this density fluctuation will grow over time . this is due to the non - linear terms in eq . . for densities outside of the binodal , the uniform fluid is absolutely stable and all density modulations diminish in amplitude over time . the above description of the system is therefore qualitatively very similar to the results from cahn - hilliard theory@xcite . thus , there are two phase separation mechanisms : firstly , spinodal demixing , which is triggered by the presence of small amplitude density modulations , which in reality are always present due to thermal fluctuations . this is the dominant mechanism inside the spinodal . secondly , for state points in the region between the spinodal and the binodals , where the uniform fluid is linearly stable , phase separation must proceed via the nucleation of density fluctuations with sufficiently large amplitude . we present results pertaining to these two mechanisms below . to study the phase separation dynamics in the spinodal region , we set the time @xmath166 density profiles to be @xmath167 , where @xmath168 is a small amplitude random white noise field , which is equivalent to adding many harmonic density perturbations , with randomly chosen amplitude , phase and wavenumbers @xmath165 . the density and magnetization profiles are then evolved forward in time using eqs . and . note that we only add noise to the initial @xmath166 density profiles and do not add noise at any other subsequent time . this corresponds to taking a uniform system and then rapidly quenching it into the unstable region of the phase diagram , by decreasing the temperature . inside the spinodal , the density perturbations with wave numbers @xmath169 grow over time@xcite . the density modulations with wavenumber @xmath170 , where @xmath171 , grows fastest in amplitude over time , leading to density profiles having density modulations with a typical length scale @xmath172 , at short times after the quench . as is illustrated by the results displayed in fig . [ fig.5 ] , the domains of demixed fluid then coarsen over time . since in the present system this demixing is between a magnetic and a non - magnetic phase , we also observe a similar pattern in the local magnetization ; see the bottom row in fig . [ fig.5 ] . the results in this figure are for a fluid with total density @xmath173 , concentration @xmath174 and with @xmath97 and @xmath108 . the phase diagram for this system is displayed in fig . [ fig.1 ] . for a given total density @xmath116 within the spinodal region , on varying the concentration @xmath72 one may observe bicontinuous labyrinthine patterns , such as those displayed in fig . [ fig.5 ] , or alternatively one observes phase separated morphologies consisting of ` islands ' of the minority phase surrounded by a ` sea ' of the majority phase . a more detailed discussion of spinodal phase separation and the resulting structures that we observe in the present system will be published elsewhere . before using ddft to investigate the dynamics of nucleation , it is worth recalling the main results from classical nucleation theory ( cnt ) and also approaches to nucleation using equilibrium dft . nucleation is normally considered to be the relevant phase separation mechanism within the _ metastable _ regions of the phase diagram , i.e. , within the regions between the binodals and the spinodal . in these metastable regions , one may consider forming ( circular , in 2d ) clusters of the new ( globally stable ) phase with radius @xmath175 , surrounded by the metastable bulk phase . one finds that the free energy as a function of @xmath175 initially increases , reaches a maximum at @xmath176 , the critical radius , and then decreases for @xmath177 . it is assumed that clusters of a given radius are randomly formed in the system by thermal fluctuations . clusters with radius @xmath178 then typically shrink , since this lowers the system free energy . on the other hand , clusters with @xmath177 must grow without limit ( in an infinite size system ) , since doing this also reduces the system free energy , thereby initiating the transformation into the new ( stable ) phase . thus , the excess free energy corresponding to the ` critical cluster ' , i.e. the cluster with radius @xmath179 , is important , because this is the free energy barrier which must be surmounted for the phase separation to occur . cnt treats the nucleation process on the simplest possible level . the key assumption of cnt is that _ any _ cluster ( regardless of its actual size ) can be regarded as a macroscopic object with a homogeneous density ( and thus , pressure ) inside and outside its surface . moreover , interfacial curvature effects are typically neglected . as a consequence , the free energy related to creation of a nucleus can be written as a sum of two terms : a negative contribution stemming from the difference of the ( bulk ) pressures inside and outside the nucleus , and a positive contribution related to the increase of surface free energy . the latter is determined by the interfacial tension @xmath139 of a planar interface ( we use the interfacial tension calculated in sec . [ sec : interface_structure ] above ) . applying this concept to the two - dimensional system at hand , the grand potential for the creation of an @xmath0-rich phase nucleus in the surrounding sea " of @xmath1-rich phase is @xmath180 where @xmath181 is the total system area , @xmath182 is the grand potential of the uniform @xmath1-rich phase and @xmath183 is the difference between the bulk pressure @xmath184 and the pressure of the @xmath0-rich phase that is being nucleated . the nucleation barrier is then given by the maximum of @xmath185 which follows from eq . ( [ eq.35 ] ) as @xmath186 the corresponding critical radius is @xmath187 for @xmath177 the grand potential decreases , indicating the absence of a sustaining force against the growth of a drop of the new phase . obtained from classical nucleation theory as a function of the concentration @xmath72 for fixed total density @xmath173 . the inset shows the excess grand potential of the nucleus as a function of the radius @xmath175 for fixed concentration @xmath188 . the coupling parameters are @xmath97 and @xmath108.,width=302 ] of the nucleus obtained from the dft approach as function of the excess number @xmath189 for different concentrations . the parameters are @xmath173 , @xmath97 and @xmath108.,width=302 ] in the present calculations we focus on nucleation on the right - hand side of the phase diagram , i.e. , the formation of non - magnetic clusters predominantly containing @xmath0-particles , out of the ferromagnetic , @xmath1-dominated liquid . we consider the nucleus and its environment ( i.e. , the undersaturated magnetic liquid ) at the same pair of chemical potentials @xmath124 , @xmath125 . because we are working on the right side of the phase diagram , these chemical potentials are typically smaller than those corresponding to phase coexistence . the quantity @xmath183 appearing in eqs . ( [ eq.35 ] ) , ( [ eq.barrier ] ) and ( [ eq.36 ] ) is then defined as the difference between the pressure corresponding to the actual ( magnetic ) state on the right side , and the pressure of the corresponding nonmagnetic state on the left hand side of the phase diagram . these states with equal chemical potentials ( as well as the associated pressures ) are found from the bulk free energy given in eq . , which yields the desired quantities through the relations @xmath190 in fig . [ fig.6 ] we plot cnt results for the free energy barrier height @xmath191 for a range of concentrations @xmath72 of the magnetic species within the metastable regime . the chosen path corresponds to states at constant total density @xmath173 . the inset shows the grand potential [ see eq . ( [ eq.35 ] ) ] as a function of @xmath175 for one particular state point . as expected from the structure of eq . ( [ eq.barrier ] ) , the nucleation barrier according to cnt becomes infinitely large at the coexistence line , where @xmath192 . for the total density @xmath173 , the binodal is at the concentration @xmath193 . decreasing then the concentration towards the spinodal value , the barrier height decreases . note , however , that the barrier directly at the spinodal is not exactly zero , as one would expect at the limit of metastability . this deficiency is a well - known artefact of cnt . indeed , given that cnt is a macroscopic theory , it is not surprising that its predictions become unreliable when the critical clusters become so small that they contain only a few particles . under such conditions , a microscopic theory such as dft is clearly more appropriate . since the important work of oxtoby and evans@xcite there has been much work done using dft to obtain a more reliable ( microscopic ) estimate for the free energy barrier @xmath194 see e.g. refs . @xcite for examples of recent work . here , we use the oxtoby - evans approach to study the nucleation of isotropic clusters of the phase rich in @xmath0-particles , and compare the resulting free energy barrier with the corresponding cnt results discussed in sec . [ sec : cnt ] . the key idea in all equilibrium dft based approaches to nucleation is that the density profile characterizing the critical nucleus corresponds to a _ saddle point _ of the grand canonical free energy@xcite . assuming a symmetric droplet in the center of the system , the goal is thus to find the density profiles @xmath195 for which @xmath196}{\delta \rho_\alpha({\bf r},\omega)}=\frac{\delta \mathcal{f}[\{\rho_\alpha\}]}{\delta \rho_\alpha({\bf r},\omega)}-\mu_\alpha=0,\ ] ] with the boundary conditions @xmath197 where @xmath198 . as demonstrated by oxtoby and evans@xcite , these density profiles may be found by iterating the euler - lagrange equations , beginning with a simple approximation for the initial profiles . the latter are characterized by spherical symmetry and a sharp ( step - wise ) change of the number density at a radius @xmath175 . the idea then is that if the guessed radius is too small ( large ) , the droplet will shrink ( grow ) during the iteration procedure until the profiles eventually approach the density values corresponding to the globally stable ( unstable ) phase . however , to identify the critical droplet one must iterate the euler - lagrange equations a limited number of times ( for each initial guess @xmath175 ) . this procedure allows for an estimate of the grand potential @xmath37 as function of @xmath175 . the critical droplet then follows as the position of the maximum . inspired by these ideas we have performed additional calculations , not with the ddft ( which will be discussed below ) , but with a simple , relaxational algorithm which is equivalent to the oxtoby - evans method . importantly , this algorithm keeps the chemical potentials fixed ( just as in the original oxtoby - evans work@xcite ) , while the number densities themselves are not conserved . using this algorithm we investigated the evolution of several initial profiles of the form @xmath199 the initial values for the densities inside ( @xmath200 ) and outside ( @xmath201 ) the nucleus , are set to the bulk densities determined by the chosen value of the chemical potentials . the magnetization for these state points is obtained self - consistently from eq . . for each initial guess , we performed about 300 iterations of the density profiles and the corresponding orientational profile , keeping the chemical potentials fixed . in this way we obtained an estimate of the excess free energy @xmath202 for each given cluster size . as it turns out , the actual density profiles characterizing the clusters are rather smooth , such that the definition of a radius becomes ambiguous . as an alternative ` reaction coordinate ' , we thus consider the quantity @xmath203 which counts the number of @xmath0-particles in the cluster ( recall that we are considering the nucleation of isotropic @xmath0-dominated clusters ) . numerical results for the functions @xmath204 at various chemical potentials within the metastable regime are plotted in fig . [ fig.7 ] . more precisely , the chemical potentials considered are associated to concentrations ( of @xmath1-particles ) between the spinodal and the binodal along a path with fixed total number density @xmath173 ( cnt results along this path are presented in fig . [ fig.6 ] ) . as may be seen from fig . [ fig.7 ] , all the curves reveal a clear maximum and thus , a clearly identifiable nucleation barrier , the height and position of which increase upon increasing the associated ( bulk ) concentration @xmath72 . this finding is fully consistent with our expectation that the nucleation barrier and the size of the critical droplet are smallest close to the spinodal and then increase monotonically upon approaching coexistence . from the positions of the maxima in @xmath204 we also see that typical critical droplets contain between a few ten and a few hundred of particles , consistent with results of other dft studies@xcite . our data for the height of the nucleation barrier as function of the concentration are summarized in fig . [ fig.8 ] , where we have included the corresponding macroscopic ( cnt ) results from fig . [ fig.6 ] . the main difference between the two approaches is that the microscopic dft calculation yields , contrary to cnt , a _ vanishing _ nucleation barrier at the spinodal , as one should expect on physical grounds . on the other hand , approaching the binodal the two curves merge , reflecting the increasingly macroscopic character of the critical cluster . obtained from classical nucleation theory ( black curve ) and from the dft approach ( red curve ) as function of the concentration @xmath72 . the dashed curve denotes the state point of coexistence . the parameters are @xmath173 , @xmath97 and @xmath108.,width=283 ] of the nucleus as function of the excess number @xmath189 . the black curve is obtained from the dft approach and the red curve corresponds to ddft calculations for initial configurations with radius @xmath205 ( a ) , @xmath206 ( b ) , @xmath207 ( c ) , @xmath208 ( d ) , @xmath209 ( e ) , @xmath210 ( f ) . the parameters are @xmath173 , @xmath188 , @xmath97 and @xmath108.,width=283 ] whilst it is clear that the above and other equilibrium dft based approaches are able to calculate the density profiles corresponding to the critical droplet , it is not clear whether the other density profiles ( i.e. those not corresponding to the critical droplet ) have any physical significance . of particular interest are the density profiles corresponding to the most likely pathway ( mlp ) that go up to and then descend from the critical droplet state@xcite . these mlp profiles should correspond to what one would observe experimentally for nucleation in the system . recently , lutsko@xcite argued that one should be able to determine the mlp using ddft by initiating the system at the critical droplet density profiles . since these profiles correspond to a saddle point , and in reality one is never able to initiate the system _ exactly _ at the saddle point , the density profiles evolve under the ddft away from this point . there are two directions the system may move : firstly , towards a state corresponding to the drop disappearing , with the excess density being spread uniformly throughout the system and secondly , the drop may grow . the performance of ddft in this context is not yet fully understood ( see ref . ) , contrary to with the static dft approaches mentioned above . one issue in this context is the fact that , within the ddft , the number densities are _ conserved _ quantities [ as reflected by the appearance of a divergence in front of the free energy derivative in eq . ( [ eq.34 ] ) ] . this is in contrast to traditional ( static ) dft approaches towards nucleation , where the fixed quantity is the chemical potential(s ) . in view of this subtle point , and given the rather plausible results from the dft approach described so far , it is an important question whether a different algorithm , and particularly the conserved dynamics implied by the ddft [ see eqs.([eq.34])-([ddftangle ] ) ] , yields consistent results . we recall that the ddft is constructed such that the density evolves towards a profile which minimizes the free energy ( one can prove that the free energy always decreases or remains constant under the time evolution of the ddft , unless the system is externally driven@xcite ) . therefore , one would expect that the critical profile found in the approach discussed above , also ` behaves ' as a saddle point within ddft calculations . to confirm this , we have performed a number of ddft calculations with initial density profiles stemming from the dft calculations described above ( after 300 iterations ) . some results of these calculations are illustrated in fig . [ fig.9 ] . note that for the ddft results we have used a slightly different definition of @xmath189 . the ddft calculations are performed on a finite size square area of length @xmath211 , with periodic boundary conditions . we start with the drop located at the center and we define @xmath212 where @xmath213 is the value of the density at the corners of the ( square , periodic ) system . during the initial stages of the evolution @xmath214 , and so the value of @xmath189 remains constant due to the conserved dynamics . however , at later times @xmath215 changes , and so @xmath189 changes with time . this is because either excess density from the center of the system diffuses out to the corner as the drop disappears , or because as the drop grows it removes particles of species @xmath0 from the surrounding fluid and so @xmath215 decreases . in fig . [ fig.9 ] the curves labelled by a , b , c correspond to calculations where the radius characterizing the initial profile is smaller than that characterizing the critical droplet predicted by the previous dft approach . on the other hand , the curves labelled by d , e , f have been started from supercritical " clusters . in all cases , the ddft algorithm evolves in the direction predicted by the previous free - energy approach . that is , when starting from a subcritical or supercritical profile , respectively , the droplet vanishes or grows without restriction . we also see from fig . [ fig.9 ] that the actual values of @xmath216 on the way " towards the final state are strongly different from those predicted by the previous approach . the interpretation of this issue clearly needs further investigation ( see also the discussion in ref . ) . furthermore , in the limit @xmath217 the ddft curves a , b , c approach different values , which depend on the initial profile . this is a consequence of the fact that the initial profiles a - c correspond to different space - averaged densities . this stems from the fact that the ddft conserves the total densities , and so the final state emerging for subcritical clusters does not necessarily have chemical potentials equal to those chosen in the previous dft calculations . we conclude that , at least , the ddft predictions for the critical nucleus are consistent with the traditional dft theory . typical density profiles illustrating the nucleation dynamics according to ddft on a microscopic ( space - resolved ) level are shown in fig . [ fig.10 ] . besides the shrinking / growth process , we remark in particular the diffuse character of the density profiles , which directly reflects the difficulty in associating a fixed radius to the instantaneous droplets . in this paper we have investigated the demixing phase transition of a binary mixture of magnetic and non - magnetic soft - core particles from both a static and a dynamic point of view . to this end we have employed classical equilibrium dft as well as the recently developed ddft@xcite , a time - dependent extension of dft involving a generalized continuity equation for the density . our analysis of the equilibrium phase behavior of the system shows that the magnetic ( heisenberg ) interaction within one species is capable of inducing macroscopic phase separation . more precisely , for sufficiently large ferromagnetic coupling strength we find a combined phase transition where the system both demixes and develops global ferromagnetic ordering in one of the phases . depending on the total density , the transition may either be second or first order ( in terms of the magnetization and composition , respectively ) , with the two regimes being separated by a tricritical point . thus , the general topology of the phase diagram ( in the fluid regime ) , particularly the appearance of a curie line and a tricritical point , coincides with that of 3d heisenberg and xy - fluids@xcite . based on dft we have also calculated the structure of the fluid - fluid interface in the first - order regime , i.e. we calculate the density and magnetization profiles , as well as the resulting interfacial ( line ) tension . as expected , the latter vanishes upon approaching the tricritical point from the high - density side . we note that the present dft results are based on a mean - field ( mf ) approximation for the excess free energy contribution from both the soft - core repulsion ( where mf theory has already proven to be very accurate@xcite ) and from the heisenberg interactions . thus , the quantitative reliability of our results remain to be checked against those of more refined free energy functionals and/or computer simulations . however , judging from previous theoretical studies of 3d heisenberg fluids@xcite we would expect the mf results to be quite good . the second part of the paper has been devoted to the dynamics of the ( first - order ) phase separation , concentrating mainly on the nucleation of non - magnetic clusters within the metastable ferromagnetic phase . we have analyzed and compared results from three different approaches to nucleation . the simplest one is cnt , a macroscopic approach , involving the bulk pressure and line tension . the resulting nucleation barriers show familiar behavior ( as compared to other model systems@xcite ) and predicts that the barrier height increases as one approaches to the binodal . in this regime cnt is expected to give a good approximation for @xmath194 . however , cnt incorrectly predicts that the barrier is finite at the spinodal . as a second step we have investigated the nucleation using the microscopic dft approach originally proposed by oxtoby and evans@xcite . in this approach one calculates the free energy as function of an appropriate reaction coordinate " , which we set to the excess number of particles in the nucleus . the results for the nucleation barrier are very similar to those from cnt when the fluid state point is near to the binodal . however , in contrast to cnt , the dft predicts that the nucleation barriers vanishes upon approaching the spinodal , as one should expect on physical grounds . finally , we have touched the issue of the nucleation pathway by comparing the results of the microscopic dft with those from ddft . since both of these approaches are based on the same free energy functional , they are both probing the same underlying free energy landscape . the choice of reaction coordinate to some extent determines what regions of this landscape are accessed and what path to and from the saddle point are predicted . in addition , the ` dynamics ' inherent in each approach also determines this path . the ddft has a real ( physical ) dynamics , which conserves the total number of particles in the system , whilst in contrast the oxtoby evans dft approach , which has a finite number of iterations starting from an initial guess , leads to having an effective ( unphysical ) dynamics , which does not conserve the number of particles . thus , the ddft is essentially a canonical theory , whereas the dft approach is more grand canonical in character . as we have shown , these subtleties are irrelevant for the actual height of the nucleation barrier , but yield marked differences when we consider the evolution of profiles away from that of the critical nucleus towards equilibrium . to our knowledge , these two dft approaches have not been directly compared before . what is clear is that more investigations must be done to understand in detail the role of ddft for the nucleation pathway , and more importantly which approach is the relevant one for determining the true mlp . there are several directions which we believe require further investigation . first , as a direct extension of the present work , it would be interesting to consider in more detail the spinodal decomposition occurring in the unstable range of the phase diagram . our results in the present study already indicate that the ddft is capable of describing the coarsening process characterizing this regime in a qualitative way ( see sec . however , this issue clearly calls for a more systematic investigation including a ( linear ) stability analysis@xcite and also an investigation of dynamic correlations such as the time - dependent structure factor . this quantity may be obtained using recent extensions of the ddft to determine dynamical two - particle correlation functions@xcite . another ( maybe more technical ) open point concerns the effect of the actual formulation of the ddft . in the present study we have restricted ourselves to a simplified version where the spin degrees of freedom relax instantaneously ; therefore , we were essentially dealing with the ddft equations for isotropic particles , and the magnetization entered only via the selfconsistency relation for the ( local ) magnetization . however , to further proceed one should also explore results from the full ddft for anisotropic particles , which involves an equation for the rotational dynamics on top of that of the translational dynamics@xcite . a particular interesting question in this context concerns the role of different time scales of rotational and translational motion for the demixing dynamics . finally , it should be interesting to extend the present study towards 2d ( magnetic ) systems exposed to _ patterned _ surfaces and , optionally , additional external magnetic fields . in fact , a number of recent experiments@xcite have addressed the question of phase behavior and transport of magnetic particles on complex substrates@xcite . all these question can be , in principle , tackled by the present ( d)dft approach . work in these directions is in progress . aja gratefully acknowledges rcuk for support . kl and shlk acknowledge financial support via the collaborative research center ( sfb ) 910 control of self - organizing nonlinear systems : theoretical methods and concepts of application " and the research training group ( grk ) 1558 nonequilibrium collective dynamics in condensed matter and biological systems " . 54ifxundefined [ 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 _ _ , ( , ) \doibase http://dx.doi.org/10.1038/nature06443 [ * * , ( ) ] link:\doibase 10.1098/rsta.1980.0266 [ * * , ( ) ] link:\doibase 10.1103/physreve.58.2201 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.255703 [ * * , ( ) ] \doibase doi:10.1063/1.2178356 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.208302 [ * * , ( ) ] link:\doibase 10.1080/00018737900101365 [ * * , ( ) ] link:\doibase 10.1063/1.455285 [ * * , ( ) ] link:\doibase 10.1063/1.467183 [ * * , ( ) ] @noop * * , ( ) @noop ( ) link:\doibase 10.1103/revmodphys.65.851 [ * * , ( ) ] http://books.google.com/books?id=21ibv50evdmc[__ ] ( , ) link:\doibase 10.1103/revmodphys.49.435 [ * * , ( ) ] link:\doibase 10.1063/1.478705 [ * * , ( ) ] http://stacks.iop.org/0953-8984/12/i=8a/a=356 [ * * , ( ) ] \doibase doi:10.1063/1.1778374 [ * * , ( ) ] @noop * * , ( ) http://stacks.iop.org/0953-8984/21/i=46/a=464101 [ * * , ( ) ] link:\doibase 10.1039/c0sm01219j [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.108302 [ * * , ( ) ] link:\doibase 10.1103/physreve.58.3426 [ * * , ( ) ] link:\doibase 10.1103/physreve.49.5169 [ * * , ( ) ] link:\doibase 10.1103/physreve.58.3478 [ * * , ( ) ] link:\doibase 10.1103/physreve.52.1915 [ * * , ( ) ] link:\doibase 10.1103/physreve.55.7242 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.94.045701 [ * * , ( ) ] link:\doibase 10.1063/1.432891 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.2522 [ * * , ( ) ] \doibase doi : 10.1016/s0370 - 1573(00)00141 - 1 [ * * , ( ) ] link:\doibase 10.1021/ma001346x [ * * , ( ) ] link:\doibase 10.1063/1.1486209 [ * * , ( ) ] link:\doibase 10.1063/1.2172596 [ * * , ( ) ] link:\doibase 10.1103/physreve.62.7961 [ * * , ( ) ] link:\doibase 10.1103/physreva.38.2022 [ * * , ( ) ] link:\doibase 10.1103/physreve.64.041501 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/14/46/311 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) http://linkinghub.elsevier.com/retrieve/pii/s0370157300001411 [ * * , ( ) ] \doibase doi : 10.1016/0378 - 4371(93)90344 - 4 [ * * , ( ) ] http://www.amazon.com/exec/obidos/redirect?tag=citeulike07-20&path=asin/0486425444[__ ] ( , ) link:\doibase 10.1103/physreve.76.021403 [ * * , ( ) ] @noop * * , ( ) \doibase doi:10.1063/1.1730447 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/83/46007 [ * * , ( ) ] http://scitation.aip.org/getabs/servlet/getabsservlet?prog=normal&id=jcpsa6000129000024244501000001&idtype=cvips&gifs=yes [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physreve.68.061510 [ * * , ( ) ] link:\doibase 10.1103/physreve.75.040501 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.148304 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.99.038303 [ * * , ( ) ] link:\doibase 10.1021/jp1118585 [ * * , ( ) ]

x - ray observations of synchrotron rims in supernova remnant ( snr ) shocks show evidence of efficient electron acceleration and strong magnetic field amplification ( a factor of between the upstream and downstream medium ) . this amplification may be due to plasma instabilities driven by shock - accelerated particles , or cosmic rays ( crs ) , as they propagate ahead of the shocks . one candidate process is the cosmic ray current - driven ( crcd ) instability ( bell 2004 ) , caused by the electric current of unmagnetized " crs ( i.e. , crs whose larmor radii are much larger than the length scale of the crcd modes ) propagating parallel to the upstream magnetic field . particle - in - cell ( pic ) simulations have shown that the back - reaction of the amplified field on crs would limit the amplification factor of this instability to less than in galactic snrs ( not including the additional field compression at the shock ) . in this paper , we study the possibility of further amplification driven near shocks by magnetized " crs , whose larmor radii are smaller than the length scale of the field that was previously amplified by the crcd instability . we find that additional amplification can occur due to a new instability , driven by the cr current perpendicular to the field , which we term the _ perpendicular current - driven instability _ ( pcdi ) . we derive the growth rate of this instability , and , using pic simulations , study its non - linear evolution . we show that the maximum amplification of pcdi is determined by the disruption of cr current , which happens when cr larmor radii in the amplified field become comparable to the length scale of the instability . we find that , in regions close to the shock , pcdi grows on scales smaller than the scales of the crcd instability , and , therefore , it results in larger amplification of the field ( amplification factor up to ) . one possible observational signature of pcdi is the characteristic dependence of the amplified field on the shock velocity , , which contrasts with the one corresponding to the crcd instability acting alone , . our results strengthen the idea of crs driving a significant part of the magnetic field amplification observed in snr shocks .

0912.4990

cosmic rays ( crs ) up to @xmath5ev are most likely accelerated in galactic supernova remnant ( snr ) shocks via the diffusive shock acceleration ( dsa ) mechanism @xcite . this mechanism is believed to be highly non - linear , due to the expected coupling between the evolution of the thermal plasma , the crs , and the magnetic turbulence in the shock vicinity . indeed , there is now observational evidence suggesting that electrons are being accelerated in the non - relativistic shocks of snrs , and that significant magnetic turbulence is being produced as part of the process ( see , e.g. * ? ? ? * ; * ? ? ? the evidence comes from x - ray observations of snrs that show the existence of thin , non - thermal rims , which are interpreted as synchrotron emission by tev electrons accelerated at the shocks . the rapid variability and thinness of the rims ( which depend on the synchrotron cooling time of the electrons ) have allowed to estimate the strength of the field , suggesting downstream amplitudes @xmath0 times larger than typically expected in the ism of the galaxy . this implies a significant amplification even accounting for the compression of the field at the shock , which would contribute a factor of @xmath6 to the growth of the field . the origin of this strong amplification is an open question . on the one hand , using mhd shock simulations , @xcite showed that the presence of density and magnetic fluctuations in the upstream medium of non - relativistic shocks may introduce vorticity into the flow and produce field amplification at the shock itself ( see , also , * ? ? ? this mechanism would produce a magnetic enhancement larger than expected from the simple shock compression of the upstream field . on the other hand , it has been proposed that the magnetic field may be amplified by plasma instabilities driven by the crs themselves , as they propagate through the upstream medium of shocks @xcite . this idea is interesting because , besides helping to explain the field amplification , the implied magnetic turbulence would increase the cr confinement to the shock vicinity , enhancing the efficiency of dsa . due to this , understanding the extent to which magnetic fields are amplified by crs , as well as characterizing the corresponding magnetic turbulence , is essential for any realistic model of dsa . the cosmic ray current - driven ( crcd ) instability proposed by @xcite has gathered considerable attention recently . the crcd instability consists of circularly polarized alfvn - type waves , driven by the electric current of crs propagating along the magnetic field lines . it requires the cr larmor radii , @xmath7 , to be much larger than the wavelength of the fastest growing mode , @xmath8 , where @xmath9 , @xmath10 , and @xmath11 are the magnitude of the initial magnetic field , the cr current , and the speed of light , respectively . this feature makes this instability different from the previously proposed resonant instability , in which the waves grow at wavelengths comparable to @xmath7 @xcite . the growth rate of the crcd instability is given by @xmath12 @xcite , where @xmath13 is the mass density of the background plasma . for the typical conditions of cr - modified snr shocks ( where the energy densities of crs , @xmath14 , downstream thermal plasma , @xmath15 , and magnetic field,@xmath16 , satisfy @xmath17 ) , this rate is expected to be significantly larger than that of the resonant instability . the non - linear evolution of the crcd instability has been studied using both mhd @xcite and particle - in - cell ( pic ) @xcite simulations . both kinds of studies showed that the crcd instability has the potential of growing to very non - linear amplitudes as long as the cr current is kept constant , i.e. , if the back - reaction on the cr trajectories is neglected . in particular , @xcite found that , at constant cr current , the field growth stops when the alfvn velocity of the background plasma , @xmath18 , gets close to the cr drift velocity , @xmath19 . however , pic studies show that if the cr back - reaction is considered , the field can also saturate due to the trapping of the crs in the amplified field . when @xmath7 becomes comparable to the dominant length scale of the magnetic turbulence , the crs get strongly deflected , which significantly reduces their current and quenches the magnetic growth . this is the most likely saturation mechanism in the case of snrs , which implies a _ maximum _ amplification factor of @xmath1 in the upstream medium of shocks , and confirms the crcd instability as a viable mechanism for magnetic turbulence generation in snrs @xcite . when the field compression at the shock is considered , this amplification may increase by an extra factor of @xmath6 . however , since the upstream amplification factor of @xmath20 constitutes an upper limit , the crcd instability alone would not be enough to account for the factor of @xmath0 inferred from the x - ray observations . in this paper we study the possibility of magnetic amplification beyond the saturation of the crcd instability . the situation we explore is one where crs propagate through a plasma where previous crcd magnetic turbulence has already been produced on scales larger than the typical cr larmor radius , @xmath7 . thus , our analysis would be applicable to the lower energy crs , which are more confined to the shock vicinity . this is motivated by the fact that the crcd instability is probably driven first by the highest energy crs , in regions far upstream from the shock . this instability saturates when the larmor radius of these high energy particles is about the length scale of the pre - amplified magnetic fluctuations , @xmath21 . thus , as the shock approaches , a large fraction of the cr energy will be carried by magnetized " crs , i.e. , crs with @xmath7 smaller than @xmath21 . then , near the shock , crs will find regions of pre - amplified , quasi - transverse field , which can be considered homogeneous on scales @xmath22 . if @xmath23 , then crs will not easily diffuse through the regions of pre - amplified field . instead , they will protrude into these regions only by a distance of @xmath22 . this situation will produce a cr current , @xmath24 , perpendicular to the initial , pre - amplified field , @xmath25 , due to the coherent deflection experienced by crs in the homogeneous " ( on scales of @xmath22 ) magnetic field . we propose that , under these conditions , an extra magnetic amplification is possible due to a new instability : the _ perpendicular current - driven instability _ ( pcdi ) , which consists of purely growing , compressional waves that arise when @xmath24 is perpendicular to @xmath25 . as we show below , the pcdi would produce larger magnetic amplifications compared to the crcd instability acting alone . in addition , this instability can amplify magnetic fluctuations on scales comparable to the larmor radii of the lowest energy crs , which would improve their diffusion and increase the efficiency of their acceleration at the shock . if this did not happen , the large - scale , transverse fields generated by crcd instability would hamper the low energy cr diffusion , which would probably shut off the acceleration of the crs that cause the crcd amplification in the first place . in [ sec : physics ] , we explain the physics of the pcdi . first , we show how the penetration of crs into the regions of pre - amplified field can give rise to a current , @xmath24 , perpendicular to the pre - amplified field , @xmath26 , and why we expect that situation to happen in the upstream medium of snr shocks . then , we show how the presence of a @xmath24 perpendicular to @xmath26 can produce the pcdi , and derive its dispersion relation . in [ sec : simulations ] , we show the results of our study of the pcdi using pic simulations . first , we model the non - linear evolution of the instability assuming a constant cr current , i.e. , ignoring the back - reaction on the crs , and provide a simple analytical model for the non - linear behavior of the pcdi modes . second , we show the results of a series of simulations that model the non - linear evolution of the pcdi including the full cr dynamics . the main focus is to show the role of the cr back - reaction on the final saturation of the instability . in [ sec : discussion ] , we apply our results to the case of snr shocks . finally , in [ sec : conclusions ] we summarize our results and present our conclusions . the perpendicular current - driven instability ( pcdi ) is caused by a cr current , @xmath24 , perpendicular to the ambient magnetic field , @xmath25 . in [ sec : anticorrelation ] , we show why we expect this @xmath27 to actually occur in the precursor of snr shocks . then , in [ sec : dispersion ] we explain the physics of the pcdi growth , and derive its dispersion relation . , in the upstream medium of snr shocks . the * upper panel * shows the downstream and upstream regions of a generic shock . regions of pre - amplified field in the upstream are represented by grey , doughnut - like shapes . black dots represent magnetized " crs , i.e. , crs whose larmor radii are smaller than the size of the regions of pre - amplified field . the * lower panel * shows a zoomed and rotated ( notice the orientation of coordinate axes in both panels ) image of the small , solid - line rectangle shown in the upper panel . this image is divided into region a " ( with low initial magnetic field ) , and region b " ( with the pre - amplified magnetic field @xmath25 ) . the crs , coming from region a " , can penetrate into region b " by a distance @xmath28 , which defines the interface into which crs can protrude . as indicated in the figure , the coherent deflection of the crs in this region produces a net mean velocity along @xmath29 , of magnitude @xmath30 . finally , the small , dashed - line rectangle depicted in the upper panel represents the type of regions that we study in [ sec : crback ] by using two - dimensional pic simulations.,width=325 ] in this section , we show that a significant cr mean velocity perpendicular to an initial field , @xmath25 , can arise in the upstream medium of snr shocks , if there is a significant anti - correlation between the cr number density , @xmath31 , and the strength of the magnetic field . if crs propagate through a smooth medium , as may be the case shortly after the shock formation , there is no reason to think that a significant anti - correlation between @xmath31 and @xmath32 can happen . however , if some magnetic turbulence has already been amplified , then this anti - correlation can exist . for instance , numerical mhd and pic simulations of the crcd instability ( see fig . 4 of @xcite or figs . 10 and 12 of @xcite ) show how this instability produces turbulence characterized by typical plasma density contrast of @xmath1 , with frozen - in magnetic field concentrated in the high density regions . as discussed in @xcite , this kind of turbulence can be produced far upstream from the shock by high - energy crs escaping " at @xmath30 ( see also * ? ? ? this implies that lower energy crs , which are more confined to the shock , will encounter the upstream medium with previously amplified magnetic turbulence . this is the situation depicted in the upper panel of figure [ fig : figure2 ] , which shows the regions in the upstream of a shock containing a pre - amplified magnetic field , @xmath33 ( diagrammatically represented by doughnut - like shapes in the figure ; note that , in reality , these doughnuts " are not necessarily in one plane ) surrounded by crs ( indicated by black dots ) . since the saturation of the crcd instability happens when the larmor radius of the highest energy crs is close to the length scale of the pre - amplified field , @xmath21 , then most crs near the shock are expected to have larmor radii , @xmath7 , that satisfy @xmath34 . we will refer to crs satisfying this condition as magnetized " crs . in this scenario , crs will not be able to diffuse through the regions of pre - amplified field . instead , they will only be able to penetrate into these regions by a distance of @xmath35 . the lower panel in figure [ fig : figure2 ] depicts this situation . it shows a zoomed image of the interface between a region of low and high upstream magnetic field ( rectangular box in the upper panel).this plot is divided into two regions . region a " ( left ) , which contains the crs and where there is almost no initial magnetic field , and region b " ( right ) , where there is a pre - amplified magnetic field , @xmath25 , which has a length scale , @xmath21 , larger than the cr larmor radii , @xmath7 . since @xmath34 , crs protruding into region _ b _ ( by a distance @xmath35 ) will not experience random scatterings due to fluctuations in the field . instead , they will just complete a fraction ( typically @xmath36 ) of a larmor gyration before being scattered back into region _ a_. the fact that crs complete only a fraction of a larmor cycle in region _ _ implies that , while in this region , they will tend to move parallel to @xmath37 ( i.e. , along @xmath29 in figure [ fig : figure2 ] ) . thus , if crs have positive electric charge , they will produce a current , @xmath24 perpendicular to the pre - amplified field . ] . since crs are relativistic , their mean velocity perpendicular to @xmath25 will be a fraction ( of the order of @xmath36 ) of the speed of light , larger than the typical drift velocities along the field that drive the crcd instability . this suggests that it is possible to find upstream regions near snr shocks where @xmath24 can be quasi - perpendicular to @xmath25 . the formation of the perpendicular mean cr velocity at @xmath30 is confirmed in [ sec : crback ] using two - dimensional pic simulations . the presence of the cr current , @xmath24 , perpendicular to the initial field , @xmath25 , induces a return current , @xmath38 , almost equal to @xmath39 in the background , @xmath38 is still almost equal to @xmath39 . in this case the compensating current is due to the polarization drift of particles , @xmath40 , where @xmath41 and @xmath42 are the particle mass and charge , and @xmath43 is the electric field . it is possible to show from the ampres s law that , if @xmath44 , the return current produced by this drift ( acting mainly on the ions ) almost completely compensates the cr current ( see also lyutikov 2009 ) . ] . because of this , a constant force , @xmath45 , will push the plasma homogeneously . thus , if plasma density fluctuations , @xmath46 , exist , this @xmath45 force will accelerate the low density regions more than the ones with higher density . as we will see below , this differential plasma acceleration in the direction perpendicular to @xmath25 can amplify the transverse component of the magnetic field , @xmath47 , and enhance the density fluctuations in the plasma , producing the exponential growth of the pcdi modes . we note that this instability starts from a non - equilibrium state , where the whole plasma is being accelerated in the direction of @xmath45 . we derive the dispersion relation of the pcdi in the mhd limit . we assume the existence of an externally imposed cr current , @xmath27 . the way this cr current is produced and the back - reaction of the amplified field , @xmath48 ( @xmath49 ) , on the cr trajectories is not modeled in this derivation . the current carried by the background plasma is @xmath50 , which includes the return current induced in the background . so the mhd equations will only be modified by an extra term in the force per unit volume on the plasma : @xmath51 @xmath52 @xmath53 where @xmath54 is the plasma fluid velocity and @xmath55 is its pressure . although the crs can also include accelerated electrons , which would reduce the cr net charge , in general crs may carry a charge density , @xmath56 , that must be compensated by the background plasma . this extra plasma charge should introduce an electrostatic term in equation ( [ eq : euler2 ] ) due to plasma charge density fluctuations . it is possible to show that this term is second order in perturbation theory , so it can be neglected in our linear analysis . we refer the reader to appendix [ app : b ] for further details . if each physical quantity @xmath57 is separated into an initial , homogeneous component , @xmath58 , plus a small perturbation , @xmath59 , ( so @xmath60 , @xmath61 , and @xmath62 ) , then the mhd momentum equation will have a zeroth order part : @xmath63 which represents homogeneous acceleration of the plasma . the first order quantities read : @xmath64 @xmath65 @xmath66 with @xmath67 being the sound speed in the plasma and where we have chosen a reference frame where @xmath68 . the physical meaning of the different force terms in equation ( [ eq : prob2 ] ) can be understood with the help of figure [ fig : figure1 ] . let perturbations @xmath47 and @xmath46 be such that the magnetic field lines and plasma density are represented by the solid and dotted lines of figure [ fig : figure1 ] , respectively . then , since the force @xmath69 is the same everywhere , regions of low density will accelerate more than regions of high density , making the overdense regions lag behind with respect to the rest of the plasma . this is the origin of the force @xmath70 in equation ( [ eq : prob2 ] ) . as shown by the solid black arrows in figure [ fig : figure1 ] , this force tends to stretch the lines of magnetic field and , therefore , produces the amplification of @xmath47 . this amplification can happen exponentially , because the growth of @xmath47 increases the force @xmath71 , which , as depicted in figure [ fig : figure1 ] , enhances the density contrast even more . in a sense , the pcdi is similar to parker instability , since the magnetic growth implies the existence of regions of the plasma that accelerate more than others in the direction perpendicular to @xmath25 . however , whereas in the case of parker instability the amplification is driven by a homogeneous gravitational _ acceleration _ , the amplification of pcdi is caused by a homogeneous _ force _ , given by @xmath72 . performing the fourier transform of eqs . ( [ eq : cont2 ] ) , ( [ eq : prob2 ] ) , and ( [ eq : law2 ] ) , and assuming that the fourier modes are stationary , with @xmath73 , it is straightforward to obtain the dispersion relation : @xmath74 where @xmath75 is the growth rate corresponding to a given @xmath76 . this dispersion relation can also be obtained by taking the limit @xmath77 of the more general mhd result shown in eq . ( 4 ) of @xcite . a derivation using the multifluid approach can be found in appendix ( [ sec : appendixa ] ) . from equation ( [ eq : dispersion ] ) we obtain that the fastest growing mode has the wave number , @xmath78 and the growth rate , @xmath79 where @xmath80(@xmath81 ) is the initial alfvn velocity in the plasma . thus , if the background plasma is initially cold ( @xmath82 ) , then @xmath83 and @xmath84 . however , if we assume that @xmath67 is close to the typical turbulent velocity in the plasma , then it is possible to show from eqns . [ eq : cont2]-[eq : law2 ] that the sound speed due to pcdi turbulence would be @xmath85 . thus , the thermal effects would make @xmath86 and would keep the same growth rate @xmath84 as long as @xmath87 . as @xmath88 , the expressions for the wave number of fastest growth , @xmath89 , and its corresponding growth rate , @xmath90 , look the same as the ones for crcd instability : @xmath91 and @xmath92 @xcite . however , the cr currents that drive the pcdi and crcd instabilities are the components of @xmath24 perpendicular and parallel to @xmath26 , respectively . thus , if @xmath24 is quasi - perpendicular to @xmath25 ( as may be the case in regions close to the shock ) , the pcdi would grow faster than the crcd instability . is applied perpendicular to the plane of the figure and also perpendicular to the initial field , @xmath25 , the background plasma will develop a return current , @xmath93 pointing out of the plane of the figure . this will produce a homogeneous force @xmath72 accelerating the plasma , as shown in the diagram . the pcdi begins from this non - equilibrium state . let us consider the magnetic field lines and the plasma density are represented by the solid and dotted lines of the figure , respectively . then , since the force @xmath72 is the same everywhere , regions of low density will accelerate more than regions of high density , making the overdense regions lag behind with respect to the rest of the plasma . this is the origin of the force @xmath70 of eq . ( [ eq : prob2 ] ) ( shown by the solid , black arrows ) , which tends to stretch the lines of magnetic field and amplify @xmath47 . the other forces from the same equation : @xmath71 , @xmath94 , and @xmath95 , are represented by the solid grey arrow , the dotted grey arrow , and the dotted black arrow , respectively . for unstable modes , this amplification happens because the growth of @xmath47 increases the force @xmath71 , which enhances the density fluctuations . thus , the growth of @xmath47 increases the force that produces it , resulting in an exponential growth of @xmath47 and @xmath46.,width=325 ] in this section we use two - dimensional pic simulations to study the pcdi . first , in [ sec : constcurr ] , we explore the case of a constant , externally imposed cr current , @xmath24 , perpendicular to the ambient field , @xmath25 . then , in [ sec : crback ] , we study the full dynamical evolution of the crs , allowing a non - constant @xmath24 . we pay special attention to the way a current @xmath24 perpendicular to @xmath25 can set in , and to the effect of the cr back - reaction on the saturation of the instability . in each section we specify the numerical set up of the corresponding simulations , listing their common features in [ sec : common ] . we use the electromagnetic pic code tristan - mp ( buneman 1993 ; spitkovsky 2005 ) . our simulations use periodic , two - dimensional boxes whose normal vector is chosen to point along @xmath29 . the boxes contain a charge- and current - neutral plasma , consisting of cold ions and electrons ( with initial thermal velocities @xmath96 ) , which constitute the background plasma , and a population of relativistic particles or crs that may either be composed of ion - crs ( positive crs ) only , or a mixture of both ion- and electron - crs . the boxes also contain the initial magnetic field , @xmath25 , in the plane of the simulation , pointing along the @xmath97 axis . particles are spread within the box in such a way that the local charge density is zero everywhere . since typically in a snr the cr density @xmath31 is much smaller than the background ion density , @xmath98 , we increase cr statistics by modifying the charge of the plasma and cr macroparticles ( for instance , by making @xmath99 , where @xmath100 and @xmath101 are the charge of the ion and ion - cr macroparticles ) . the mass of the macroparticles is modified accordingly in order to keep their charge to mass ratios unchanged . in this section we focus on the case of a constant @xmath24 pointing along @xmath29 ( out of the plane of the simulation ) . thus , the back - reaction of the field on the cr trajectories is explicitly suppressed in these simulations . we show the results of four runs with different @xmath102 ratios and cr charge densities , so that we can test the dispersion relation ( eq . [ eq : dispersion ] ) and investigate the effect of having different cr net charge . in two of our simulations the crs have no net charge ( they are made of equal amount of ion- and electron - crs ) , while in the other two runs crs are only positively charged . in the case of zero cr net charge , the cr current is produced by two counter - streaming cr beams of equal density but opposite charge . for both the zero and non - zero cr charge cases we test @xmath103 and 0.04 , where @xmath31 is the total density of crs ( which in the case of zero cr charge includes the positive and negative crs ) , and @xmath104 ( where @xmath105 represents the number of macroparticle per cell for species @xmath106 " ) . in the non - zero case , @xmath107 . then , in order to have local charge neutrality we need @xmath108 , and @xmath109 ( this way there is a small excess of electron charge in the background plasma , which compensates the charge of the crs ) . ] . in both types of simulations , particles are randomly distributed in the box , but placed such that , initially , there is local charge neutrality . also , the initial local current is made zero by giving electrons a small initial velocity so that they carry a compensating current @xmath39 . the other physical and numerical parameters of the simulations are : @xmath110 , mass ratio @xmath111 , cr current velocity @xmath112 , speed of light @xmath113 , the skin depth of electrons @xmath114 , and the number of macroparticles per cell per species is 16 . here , @xmath115 is the grid cell size and @xmath116 is the time step . first , we tested the dominant length scale for the pcdi amplification . figure [ fig : magden ] shows the space distribution of the field and background plasma density for the run with @xmath117 and neutral cr beam at three different times ( @xmath118 , and 35 , where @xmath90 is the theoretical growth rate of the simulation ) . when @xmath119 ( @xmath120 , in panel @xmath121 ) , the dominant wavelength is smaller than @xmath122 ( which is chosen to be the wavelength of the fastest growing mode when @xmath123 ) . when @xmath124 ( @xmath125 , in panel @xmath126 ) , the length scale of the fluctuations becomes very close to @xmath122 , confirming our analytical result . as the field becomes non - linear , it adopts the shape of loops , which tend to merge and grow in size . we see that in the very non - linear regime ( @xmath127 ) the dominant scale of the loops grows by a factor comparable to the magnitude of the amplification , which is consistent with our predictions . our simulations also confirm the analytic growth rate , given in eq . ( [ eq : gammamax ] ) . figure [ fig : magenergy ] shows the evolution of the energy in the three magnetic field components in our simulations . the left and right panels represent the cases with @xmath128 and @xmath129 , respectively . solid and dotted lines show the cases with zero and finite cr charge . the colors black , green , and red represent the magnetic energy of the @xmath130 , @xmath131 , and @xmath132 field components . the time is measured in units of @xmath133 , where @xmath90 is the pcdi theoretical growth rate for @xmath134 . in the linear regime , the pic results are consistent with our theoretical estimates for the pcdi growth rates . also , we see that , as long as @xmath135 , the growth rate is independent of the net cr charge . this situation changes in the non - linear regime . while in the zero cr charge cases the magnetic field keeps growing with about the same growth rate as in the linear regime , the non - zero cr charge simulations saturate at @xmath136 . the non - linear behavior of pcdi can be understood as follows . as the amplified field becomes non - linear , it adopts the shape of magnetic loops that surround background plasma holes on scales of @xmath137 ( as shown in figure [ fig : magden ] ) . the evolution of magnetic loops in the presence of a constant , perpendicular cr current was already considered by @xcite , whose argument we revise here . if @xmath24 is constant , the background plasma will create a return current , @xmath93 . thus , this return current will produce a force @xmath138 that tends to expand the loops . in the mhd regime , as a loop expands , the magnetic flux freezes in the fluid , keeping the ratio @xmath139 constant ( where @xmath140 is the radius of the loop ) . thus , if no extra forces are considered , we obtain that the acceleration of the loop is @xmath141 , which is solved by @xmath142 , where @xmath122 is equal to @xmath143 in the limit @xmath144 . thus , since flux freezing keeps @xmath139 constant , the magnetic field should also grow exponentially at approximately the same rate as @xmath145 . , and 35 ( represented by panels a ) , b ) , and c ) , respectively ) , for the run with @xmath117 and neutral cr beam . the initial magnetic field , @xmath25 , and the cr current , @xmath24 , point along @xmath97 and @xmath29 , respectively . the distance is normalized in terms of the theoretical @xmath122 at @xmath123 . we see that in the linear regime ( panel a ) the size of the fluctuations , @xmath146 , is smaller than @xmath122 , and that it evolves into @xmath147 as the amplification becomes non - linear ( panels b ) and c ) . when @xmath148 ( panel c ) , the pcdi modes acquire a loop - like shape.,width=321 ] , @xmath131 , and @xmath132 axes , respectively ) for the four simulations in [ sec : constcurr ] . the left and right panels show the cases with @xmath128 and 0.04 , respectively . the solid lines show the situations when the background plasma ( and the crs ) have a zero net charge , while the dotted lines represent the cases when the background plasma has a negative net charge compensated by the positive crs . the time is measured in units of @xmath133 , where @xmath90 is the theoretical growth rate of the instability for the case @xmath134.,width=321 ] however , we see that , when crs are made of only positively charged particles , this non - linear amplification gets suppressed . in that case , the expansion of the negatively charged loops would produce charge separation in the plasma . due to this , a monopolar electric field will form in the loops , which would suppress their expansion . although this force plays an important role in the cases analyzed in this section , in cases where @xmath24 is caused by crs protruding into regions of pre - amplified field ( like in the lower plot of figure [ fig : figure2 ] ) , they would not stop the growth . the reason has to do with the particular way the charges are distributed within each loop . we will analyze in detail this point in appendix [ app : b ] . in summary , in this section we have used two - dimensional pic simulations to show that , if the cr current is kept constant , the pcdi grows at the wavelengths and rates predicted by the linear theory . in the non - linear regime , the exponential growth can continue with about the same rate as in the linear regime , and with a length scale proportional to @xmath149 . however , if crs are mainly positively charged particles , this growth may be affected by electrostatic fields due to charge separation in the plasma ( see appendix [ app : b ] ) . in this section , we include the full dynamics of crs our the study . in [ sec : protru ] , we show how crs protruding into regions of pre - amplified field can produce a @xmath27 , and drive the growth of pcdi modes . then , in [ sec : saturation ] we study the effect of the cr back - reaction on the saturation of the instability . in our simulations we want to reproduce the situation pictured in figure [ fig : figure2 ] , where regions with a pre - amplified field ( represented by grey doughnut - like " shapes ) are surrounded by a population of magnetized crs ( i.e. , crs with larmor radii smaller than the length scales of these regions ) . in particular , we want to model small sections of these regions , like the one marked with a dotted - line box in the upper panel of figure [ fig : figure2 ] . the idea is to study the growth of the pcdi due to crs protruding into the regions of pre - amplified field region . since we use a monoenergetic population of crs , these particles are expected to penetrate by a distance comparable to two times their larmor radius , @xmath7 , which defines a region or interface where the field grows . the other parts of the plasma , which are not reached by these crs , should experience compression due to the acceleration of the interface . as crs turn in the field of the magnetized region , the acquire the mean velocity in the @xmath132 direction ( out of the box ) , and the associated perpendicular current drives the rippling of the field characteristic of the pcdi . the numerical setup we use is similar to the one of the constant cr current case described in [ sec : common ] and [ sec : constcurr ] . the only difference is that , in this case , crs are positively charged and monoenergetic particles , and are isotropically injected within a narrow strip along the @xmath97 axis . this strip is located in the left part of the simulation box and its width is @xmath150th of the width of the box . thus , whereas the strip represents the region with high cr density and low initial magnetic field , the rest of the simulation box would represent the inner part of the region of pre - amplified field . extra electrons are injected along with the crs so that the net injected charge is zero . the injection criterion is such that the mean density of crs in the strip , @xmath151 , is kept constant , where @xmath152 is the physical density of ions in the strip at @xmath153 . ] . at @xmath154 , where @xmath90 is the growth rate of the pcdi calculated using @xmath151 and @xmath155 as the number density and mean velocity of crs . the distance is shown in units of @xmath156 , which is the larmor radius of the crs in the initial field @xmath25 ( @xmath157 ) . the simulation box is periodic . a ) magnitude of the generated field ; b ) density of different species averaged over @xmath131 as a function of the @xmath130 coordinate . the red and green lines represent the background ions and electrons , whose densities are normalized by @xmath152 , and the black and blue lines represent the densities of crs and injected electrons ( to compensate cr charge ) , which are normalized in terms of @xmath151 ; c ) mean velocities of crs along @xmath158 ( black line ) and @xmath29 ( blue line ) , normalized by @xmath11 ; d ) magnetic field amplification as a function of @xmath130 , normalized in terms of @xmath9 . the red and blue lines show the maximum and average amplifications , respectively . black line shows the field magnitude corresponding to local energy equipartition with the crs . dashed lines in panels b ) , c ) and d ) show the results of a simulation analogous to @xmath159 , but with the @xmath131 dimension of the box significantly reduced ( quasi-1d ) . this simulation shows only the effect of compressional amplification.,width=302 ] , but at @xmath160.,width=302 ] , but at @xmath161.,width=302 ] we ran a series of simulations that share the following parameters : @xmath111 , @xmath162 ( where @xmath152 is the initial background ion density in the injection strip ) , @xmath163 , @xmath113 , @xmath164 , and use 8 macroparticle per cell per species . the difference between the runs is given by their initial alfvn velocity , @xmath165 , which ranges from @xmath166 to 1/40 . the evolution of the relevant physical quantities for a simulation with @xmath167 , which we will call @xmath159 " , is depicted in figures [ fig : protru32 ] , [ fig : protru40 ] , and [ fig : protru44 ] . these figures correspond to @xmath168 and 44 , where @xmath90 is the growth rate of the pcdi considering @xmath169 and @xmath170 . in panels @xmath171 of these figures , we depict the space structure of the generated field . as crs penetrate into the regions of zero initial cr density , we see the formation of magnetic loops , similar to the ones shown in the case of constant @xmath24 ( [ sec : constcurr ] ) . in panels @xmath172 we show the density of the different species as a function of the @xmath130 coordinate . it shows how crs can penetrate into the magnetized region by a distance @xmath173 , while the injected electrons ( necessary to compensate the cr charge ) stay close to the injection strip is caused by our particular choice of a monoenergetic cr energy distribution . in reality , crs should have a wide distribution of energies , so they should be able to penetrate by different distances , depending on their energy . we will get back to this point at the end of [ sec : crback ] . ] . the plots @xmath174 represent the cr mean velocities along @xmath158 and @xmath29 ( pointing out of the simulation plane ) . we can see how the mean cr velocity is dominated by its @xmath132 component ( in the three figures of the order of @xmath30 ) , due to their deflection in the @xmath175 field , which confirms the model presented in the lower plot of figure [ fig : figure2 ] . finally , the plots @xmath176 show the mean and maximum magnetic amplification as a function of @xmath130 . we can see that the typical wavelengths of the loops are consistent with the analytical theory , which predicts the dominant wavelength to be such that @xmath177 , where we have assumed that the mean cr velocity along @xmath132 is @xmath30 . applying this formula to the cases of figures [ fig : protru32 ] , [ fig : protru40 ] , and [ fig : protru44 ] , we see that the dominant length scales of the fluctuations are about @xmath178 times smaller than what the theory predicts . also , as anticipated at the end of [ sec : constcurr ] , the charge separation due to loop expansion does not stop the non - linear growth of the pcdi in this case . the reason is related to the way the background plasma compensates the charge of crs , as they penetrate into regions of low initial cr density . we refer the reader to appendix [ app : b ] for details on this point . ) , magnetic field strength ( @xmath126 ) , and the @xmath130 , @xmath131 , and @xmath132 components of @xmath48 ( plots @xmath11 , @xmath179 , and @xmath42 ) are depicted for a region of simulation @xmath159 at @xmath180 . the arrows in plot @xmath172 represent the field projectin on the plane @xmath181 . the density and the field are normalized in terms of @xmath182 and @xmath9 , respectively.,width=340 ] a clearer view of the spatial structure of the pcdi can be seen from figure [ fig : protruden ] , which shows the components of the field and the density fluctuations for a selected region of simulation @xmath159 at @xmath180 . the observed features of the pcdi turbulence are essentially the same as for the case of the externally imposed current : the non - linear field adopts a loop - like structure with very strong density contrasts ( @xmath183 ) . as was discussed before , the pcdi happens while the background plasma is being pushed by the @xmath184 force . this pushing can also amplify the field due to a simple field compression . in order to differentiate the growth due to the compression from the one due to the pcdi modes , the dashed lines in figs . [ fig : protru32 ] , [ fig : protru40 ] , and [ fig : protru44 ] show the results of an analogous simulation where the @xmath131 dimension of the box has been significantly reduced . this simulation behaves as a one - dimensional run , where the pcdi modes ( whose @xmath185 ) can not grow . we see that the magnetic amplification corresponding to this one - dimensional case is significantly smaller than the one due to the growth of the pcdi . our simulations show that , similarly to the case of the crcd instability , the field saturation happens when the larmor radii of the crs , @xmath7 , become comparable to the size of the generated magnetic fluctuations . in our fiducial simulation @xmath159 , the magnetic field reaches saturation at @xmath160 ( figure [ fig : protru40 ] ) , where it is amplified by a factor of @xmath1 . this implies that the larmor radii of the crs,@xmath7 , should be @xmath186th of its initial value , @xmath187 . this coincides with the typical length scale of the magnetic fluctuations observed in figure [ fig : protru40 ] , where the distance is measured in units of @xmath187 . this saturation criterion can be understood in the same way as for the case of the crcd instability . since the magnetic growth is driven by the cr current ( in this case perpendicular to @xmath33 ) , the growth will stop when this current is significantly reduced , which happens when cr trajectories are strongly perturbed by the generated field . this requires the length scale of the fluctuations to be comparable to @xmath7 . we check this in figure [ fig : protru44 ] , where the @xmath132 component of the cr mean velocity ( in regions where @xmath188 ) has been reduced to almost half of its value in figure [ fig : protru40 ] . by considering the expected evolution of the size of the magnetic fluctuations ( @xmath189 ) and assuming that the cr mean velocity perpendicular to the field is @xmath30 , it is possible to estimate the maximum field reachable by the pcdi . if we make @xmath190 , where @xmath191 and @xmath192 are the mass and charge of the crs , its possible to show that saturation will happen at @xmath193 . this implies that the field at saturation should be a factor of @xmath194 below energy equipartition with the crs . we can test this criterion by looking at panels d ) of figs . [ fig : protru40 ] and [ fig : protru44 ] . we see that the maximum amplification is about @xmath195 times smaller than the one corresponding to local energy equipartition between crs and magnetic field . the mean magnetic amplification is an extra factor of @xmath195 smaller , due to the existence of regions of very low density and magnetic field amplitude in the pcdi turbulence . we can also check this condition by seeing the saturation values of the field for simulations with the same cr energy density as in simulation @xmath159 , but with different values of @xmath9 . ) as a function of @xmath130 at the moment of saturation for simulations analogue to @xmath159 , but with @xmath196 and 1/40 , respectively . the red and blue lines show the maximum and average amplification , respectively . and the black line shows the magnetic amplitude of a field at local energy equipartition with the crs.,width=302 ] the left and right panels in figure [ fig : prosatura ] show the mean and maximum magnetic amplification as a function of @xmath130 at saturation for simulations with @xmath196 and 1/40 , respectively . again , the observed saturation is consistent with the mean magnetic field being @xmath197 times smaller than the one corresponding to energy equipartition with the crs . the main uncertainty in this saturation criterion is given by the particular cr energy distribution we use , which corresponds to monoenergetic crs . this implies that crs are able to penetrate into the simulation box by a distance shorter than @xmath173 . this defines a region or interface " where pcdi modes grow , with the rest of the plasma only being compressed due to the acceleration of the plasma in the interface . the beginning of the compression of the central part of the region of pre - amplified field can be seen in figure [ fig : protru44 ] , where the density of the background plasma shows prominent spikes at @xmath198 and 3 . however , in reality , crs should have a wide distribution of energies . so , in principle , the magnetized crs ( the ones with @xmath199 ) should be able to fill most of the volume of pre - amplified field . also , since the number density of the crs is expected to be a strongly decreasing function of their energy , low energy crs ( with small @xmath7 and , therefore , filling a small fraction of the volume of the pre - amplified field region ) will produce pcdi modes on length and time scales significantly shorter than the ones of the highest energy crs ( that still satisfy the magnetization condition , @xmath199 ) . thus , in a realistic setup , the growth of pcdi field should start at the boundary of the regions of pre - amplified field and then propagate towards their interior due to the action of higher energy particles . by choosing monoenergetic crs , we ignore this effect , and concentrate on capturing the basic physics of field amplification by the pcdi instability , including the growth of pcdi modes due to the formation of a @xmath24 perpendicular to @xmath26 , and the effect of the cr back - reaction on the final saturation of the instability . a more detailed study including the effect of a broad cr energy distribution will be presented elsewhere . we have found that the saturation of the pcdi is determined by the cr back - reaction , and that it would imply magnetic fields @xmath194 times below energy equipartition with the crs . we can now apply this result to the expected conditions in the upstream medium of snr shocks . if we assume that all cr energy is carried by a monoenergetic cr population , we can estimate the maximum amplification factor given by @xmath200 where @xmath201 is the ratio between the energies of the crs , @xmath14 , and the one of the downstream thermal plasma , @xmath15 . the main uncertainty in this estimate is given by the assumption that crs are monoenergetic , so they will only amplify the field in an interface of width @xmath35 within the regions of pre - amplified magnetic field ( see lower panel in figure [ fig : figure2 ] ) . in reality , crs will have a wide distribution of energies , so they are expected to enter most of the volume where the field has been previously amplified . since the density of crs decreases with their energy , the growth of pcdi modes should happen faster in the outer part of the magnetized regions , and gradually propagate towards the inner parts due to the action of the higher energy crs . the exact details of how this process unfolds , and how it would change the estimated amplification of eq . ( [ eq : estimate ] ) , requires further study , taking into account more realistic cr energy distribution . the maximum amplification [ eq : estimate ] is about 4 times larger than for the case of the crcd instability . since the saturation criteria of the two mechanisms are analogous , then the factor of @xmath6 is explained by the different mean cr velocities used for calculating the corresponding @xmath24 ( perpendicular and parallel to @xmath25 for the pcdi and crcd instability , respectively ) . while for the crcd instability the mean velocity of the crs can be considered close to @xmath202 , in the case of the pcdi it is @xmath30 . thus , the maximum amplification should differ by a factor @xmath203 ( @xmath204 for @xmath205km / s ) . if the estimate given in eq . ( [ eq : estimate ] ) were calculated considering the crs that are already diffusing with a speed of about @xmath202 , then both instabilities would saturate at about the same level . it is important to make sure that the pcdi would have time to grow before being advected into the shock . the growth of the pcdi field is expected to happen at a distance from the shock that is _ at least _ the distance that crs protrude into the pre - amplified field , @xmath173 ( this would correspond to the field loop that is the closest to the shock , in figure [ fig : figure2 ] ) . then , we can define an advection time , @xmath206 , that would be at least @xmath207 . in addition to that , the typical time scale of growth is given by the inverse of the growth rate of the instability , @xmath208 . therefore , the product @xmath209 ( assuming crs moving on average at @xmath30 ) shows that , as long as the growth of the field starts at a distance a few times larger than the typical cr larmor radius , the advection time would not be a restriction to the growth of the pcdi . the pcdi would also contribute to the efficiency of the acceleration of particles at the shock . indeed , the fact that the field saturation happens at @xmath210 means that the lower energy crs would be able to generate magnetic fluctuations at length scales comparable to their larmor radii , which would increase their diffusion and , therefore , the efficiency of their shock acceleration . in this work we proposed a new plasma instability , the perpendicular current - driven instability ( pcdi ) , as a candidate to amplify magnetic fields due to crs diffusing in front of snr shocks . the pcdi consists of purely growing , compressional waves produced by the cr current , @xmath24 , perpendicular to the initial magnetic field , @xmath25 . the time and length scale of growth of the pcdi are similar to the ones of the crcd instability , i.e. , @xmath211 and @xmath212 . however , whereas the crcd instability is driven by a current _ parallel _ to @xmath25 , the pdci is produced by a current _ perpendicular _ to the initial field . thus , the fastest growing instability will be determined by whether @xmath24 is quasi - perpendicular or quasi - parallel to the @xmath25 . we show that , in the upstream medium of snr shocks , the required perpendicular current can be due to crs with larmor radii , @xmath7 , smaller than the length scale of a previously amplified magnetic turbulence , @xmath213 . this scenario is motivated by earlier pic studies showing that , far from the shock , the crcd instability can produce non - linear magnetic amplification , driven by the highest energy crs . this amplification , although limited to a factor of less than @xmath1 for the expected energy carried by crs , would be characterized by non - linear fluctuations in the plasma density and magnetic field strength @xcite . the typical length scale of these fluctuations would be close to the larmor radius of the highest energy crs ( which is the saturation criterion that implies a maximum amplification factor of @xmath1 ) , producing a situation where most crs would have a @xmath7 significantly smaller than the typical size of the pre - amplified fluctuations . we show that , under these conditions , the current perpendicular to @xmath25 can be produced by lower energy crs protruding into the regions of pre - amplified field by a distance @xmath173 , which would give rise to a mean velocity of magnitude @xmath30 perpendicular to the field . due to this , in regions close to the shock , @xmath24 would be quasi - perpendicular to the field , making the pcdi grow faster and on length scales smaller than the scales of the crcd instability . we used the two - dimensional pic simulations to study the evolution of the pcdi . we showed that the instability exists and confirmed its theoretical growth rate . we also studied its possible saturation mechanisms , and found that the strong deflection of crs in the amplified field ultimately stops the magnetic growth , which happens when the the larmor radius of the crs , @xmath7 , is close to the size of the magnetic fluctuations , @xmath122 . this saturation mechanism is qualitatively the same as the one found for the crcd instability @xcite . however , since the perpendicular cr current is due to a cr mean velocity of magnitude @xmath30 , then the maximum magnetic field amplification would be @xmath194 times below energy equipartition with the crs . this would imply a maximum amplification factor in the upstream medium of the snr shocks of @xmath2 . the maximum downstream amplification could increase to @xmath214 , including the compression at the shock . the pcdi would also contribute to the efficiency of the acceleration of low energy particles at the shock , by providing magnetic fluctuation on scales comparable to their larmor radii , which allows them to diffuse . given the geometry of the pcdi ( see figure [ fig : figure1 ] ) , where the fluctuations in the fields and particle properties happen in a plane perpendicular to @xmath24 , we believe that our two - dimensional simulations ( with @xmath24 pointing out of the plane of the simulation ) are already capturing the essence of the instability . however , future three - dimensional simulations will be important to investigate possible extra effects that may appear by including the additional dimension . one interesting signature of the pcdi is that it predicts the amplified field , @xmath48 , that satisfies @xmath3 , assuming a weak dependence of @xmath215 ( @xmath216 ) on @xmath202 . this relationship contrasts with the one obtained for the crcd instability acting alone , @xmath4 @xcite , and constitutes a possible observational test to shed light on the nature of the magnetic amplification in snr shocks ( see observational results of * ? ? ? although we have focused on the case of non - relativistic shocks in snrs , the pcdi may also be relevant for the case of upstream magnetic amplification in the relativistic shocks of jets and gamma ray bursts . in that case , as seen from the reference frame of the upstream medium , the crs will only experience a small deflection before being advected into the downstream medium of the shock . thus , crs will provide an electric current parallel to the shock normal that may also amplify pcdi modes . the amplification of magnetic field due to the expansion of magnetic loops ( as in the non - linear regime of the pcdi ) in the upstream medium of grb shocks was already discussed by @xcite . they suggest that the growth of the field would saturate when neighboring loops collide . our simulations show that these collisions do not stop the growth . instead , they make neighboring loops merge and grow in size . we also found that under certain conditions , electrostatic forces due to the charge separation involved in the expansion of the loops may contribute to the quenching of the nonlinear growth . this possibility requires a dedicated study , which will be presented elsewhere . in conclusion , we have shown that the pcdi constitutes a viable amplification mechanism that , in combination with the crcd instability , strengthens the idea of crs being responsible for the significant fraction of magnetic amplification inferred from snr shock observations . axford , w. i. , leer , e. , & skadron , g. 1977 , 15th int . cosmic ray conf . , 11 , 132 ballet , j. 2006 , adv . in space res . , 37 , 1902 bell , a. r. 1978 , , 182 , 147 bell , a. r. 2004 , , 353 , 550 bell , a. r. 2005 , , 358 , 181 blandford , r. d. , & ostriker , j. p. 1978 , , 221 , l29 buneman , o. 1993 , `` computer space plasma physics '' , terra scientific , tokyo , 67 giacalone , j. & jokipii , j.r . 2007 , , 663 , 41 krymsky , g. f. 1977 , sov . dokl . , 23 , 327 kulsrud , r. , & pearce , w. p. 1969 , , 156 , 445 lyutikov , m. 2009 , arxiv:0912.1784 milosavljevic , m. , & nakar , e. 2006 , , 651 , 979 niemiec , j. , pohl , m. , stroman , t. , & nishikawa , k. 2008 , , 684 , 1189 riquelme , m. a. , & spitkovsky , a. 2009 , , 694 , 626 sironi , l. , & goodman , j. 2007 , , 671 , 1858 spitkovsky , a. 2005 , aip conf . proc , 801 , 345 , astro - ph/0603211 stroman , t. , pohl , m. , niemiec , j. 2009 , , 706 , 38 uchiyama , y. , aharonian , f. a. , tanaka , t. , takahashi , t. , & maeda , t. 2007 , nature , 449 vlk , h. j. , berezhko , e. g. , & ksenofontov , l. t. 2005 , , 433 , 229 zirakashvili , v. n. , ptuskin , v. s. , & volk , h. j. 2008 , , 678 , 255 we calculate the dispersion relation corresponding to the growth of magnetic field fluctuations , @xmath217 , due to crs propagating at velocity @xmath218 perpendicular to the initial magnetic field , @xmath25 . we focus on the case @xmath219 and @xmath220 , where @xmath221 is the wave vector of the fourier modes of the fluctuations . ions , electrons , and cr are modeled as three separate fluids with densities , @xmath98 , @xmath222 , and @xmath31 , such that @xmath223 . electrons are approximated as massless particles , whereas ions and crs have finite masses and sound speeds @xmath224 and @xmath225 , respectively . the essential part of the calculation consists of determining the perturbations to the plasma currents , @xmath226 , expected in the presence of the amplified field . to first order , the current perturbation contributed by species @xmath106 " is given by @xmath227 , where @xmath228 and @xmath229 represent the perturbations to the density and velocity of species @xmath106 " . in the case of the crs , these quantities can be calculated from the perturbed part of the euler equation , @xmath230 where @xmath231 is the mass of the ions ( assumed equal to the mass of the crs ) , @xmath232 is the lorentz factor of the crs , @xmath42 is the charge of the ions ( also , equal to the one of the crs ) , and @xmath233 corresponds the the electric field fluctuations . for the kind of fluctuations we are interested in , the relevant currents will be the ones perpendicular to @xmath25 . if we assume that @xmath228 , @xmath229 , @xmath233 , and @xmath47 are stationary waves that grow exponentially at a rate @xmath75 , then @xmath234 where the subscript @xmath235 stands for perpendicular " , @xmath236 is the cyclotron frequency of the crs , and @xmath97 is the axis parallel to @xmath25 . the cr density can be obtained from the @xmath131 component of equation ( [ eq : euler ] ) . if we neglect the term proprotional to @xmath233 , we obtain @xmath237 this approximation is justified if we look at the @xmath131 component of the euler equation for electrons , @xmath238 since they are assumed to have zero mass , then @xmath239 . the unperturbed electron velocity @xmath240 is given mainly by the @xmath241 drift of the background plasma , which we make initially zero by choosing the appropriate frame of reference . in the case of ions and electrons , the velocity fluctuations are calculated from the fluctuations of their drift velocities . these drifts are the @xmath241 drift , which we call @xmath242 , and an extra drift due to the time variation of the fields , @xmath243 ( see appendix a of * ? ? ? thus , it is possible to show that @xmath244 and @xmath245 where @xmath246 affects equally the ions and the electrons , while @xmath247 only affects the ions ( given that electrons are massless ) . the density fluctuations of the ions can be obtained directly from the @xmath131 component of their euler equation : @xmath248 it can be shown that in the regime in which @xmath249 , the time variation of the zeroth order component of the electric field is given by @xmath250 , where @xmath80 is the initial alfvn velocity of the background plasma . thus , the homogeneous part of @xmath251 can be expressed in terms of the time variations of the electric field to obtain , @xmath252 since we are choosing a frame where @xmath253 is initially zero , the term @xmath254 can also be neglected . thus , from the poisson equation we obtain that @xmath255 . this way we have all the density and velocity fluctuations needed to calculate @xmath226 . if we are interested in the case where the cr back - reaction can be neglected , the value of @xmath226 can be calculated by taking the limit @xmath256 . in that case we obtain that @xmath257 thus , combining the ampere s and faraday s laws with eq . ( [ eq : curr ] ) we recover eq . ( [ eq : dispersion ] ) . at @xmath180 . the electric field is normalized in terms of @xmath9.,width=340 ] if crs are made of only positively charged particles , then the background plasma is expected to carry a negative , compensating charge that nearly cancels it . in [ sec : constcurr ] we saw that this background charge may suppress the non - linear growth of the pcdi . in regions where the pcdi produces a decrease in the plasma density , a net positive charge will grow . in the linear regime , this net charge growth is not important . it creates an electric field proportional to the product of the charge fluctuations in the background plasma , which is approximately equal to @xmath258 , times the characteristic length scale of the fluctuations , which is proportional to @xmath259 . thus , the force on the plasma due to this effect is of second order in perturbation theory , and can be neglected . in the non - linear regime , however , this electrostatic force can be relevant . as the pcdi magnetic loops expands , the electric field in the loops will be given by @xmath260 , where @xmath261 points in the radial direction . thus , the loops will feel an electrostatic force @xmath262 , that will push them back to their center . recalling that , when the amplified field becomes non - linear , @xmath263 , then this force will be of the order of @xmath264 if the crs move at an average velocity close to @xmath11 . this implies that the force driving the expansion ( @xmath265 ) would be roughly compensated by this electrostatic force , which would quench the non - linear growth of the pcdi . this situation , however , would not happen when the pcdi is driven by crs protruding into regions of pre - amplified field , as suggested by the simulations presented in [ sec : crback ] . the reason has to do with the way the compensating charge is distributed in the loops . as crs penetrate into the pre - amplified field regions , their electric current induces a return current in the plasma ( @xmath266 ) , which is what compensates the charge . however , if this happens while the loops are being formed , this return current will tend to separate charges within the loops , producing a dipolar charge distribution in them . this can be seen from the shape of the electric field , @xmath43 , shown in figure [ fig : protruelec ] , which depicts the same simulation region as figure [ fig : protruden ] . in the magnetic loops locations , @xmath43 does not have the monopolar shape needed to stop the loop expansion . instead , we see that , whereas the @xmath131 component has a monopolar shape , the @xmath130 component looks like a dipolar field , indicating the presence of charge separation within the loops . in the situation depicted in figure [ fig : protruelec ] , electrons accumulate to the right of the loops . thus , while the negative charge of the plasma concentrates mainly in regions where @xmath267 , the places where @xmath268 will be more neutral . thus , there will be parts of the loops where the electrostatic forces are less important than in the cases where the charge in the loops is distributed homogeneously ( like in the simulations presented in [ sec : constcurr ] ) , and their evolution will be dominated by the @xmath269 force that expands the loops . this way , the pcdi can grow to non - linear amplitudes , even when the crs may be mainly composed of positively charged particles .

the structure , formation , and evolution of the milky way bulge is a matter of debate . important diagnostics for discriminating between models of bulge formation and evolution include-abundance trends with metallicity , and spatial abundance and metallicity gradients . due to the severe optical extinction in the inner bulge region , only a few detailed investigations have been performed of this region . here we aim at investigating the inner 2 degrees of the bulge ( projected galactocentric distance of approximately 300 pc ) , rarely investigated before , by observing the [/fe ] element trends versus metallicity , and by trying to derive the metallicity gradient in the region . [/fe ] and metallicities have been determined by spectral synthesis of 2 spectra of 28 m - giants in the bulge , lying along the southern minor axis at , , and . these were observed with the crires spectrometer at the _ very large telescope , vlt _ at high spectral resolution . low - resolution k - band spectra , observed with the isaac spectrometer at the _ vlt _ , are used to determine the effective temperature of the stars . we present the first connection between the galactic center and the bulge using similar stars , high spectral resolution , and analysis techniques . the [/fe ] trends in all our three fields show a large similarity among each other and with trends further out in the bulge . all point to a rapid star - formation episode in the bulge . we find that there is a lack of an [/fe ] gradient in the bulge all the way into the centre , suggesting a homogeneous bulge when it comes to the enrichment process and star - formation history . we find a large range of metallicities from}<+0.3 $ ] , with a lower dispersion in the galactic center :}<+0.3 $ ] . the derived metallicities of the stars in the three fields get , in the mean , progressively higher the closer to the galactic plane they lie . we could interpret this as a continuation of the metallicity gradient established further out in the bulge , but due to the low number of stars and possible selection effects , more data of the same sort as presented here is necessary to conclude on the inner metallicity gradient from our data alone . our results firmly argues for the center being in the context of the bulge rather than very distinct .

1510.02622

the current knowledge and main ideas of the structure , formation , evolution of the milky way bulge are under debate and in rapid development , mainly due to the recent large amount of new observations of kinematics , photometry , and abundances of bulge stars , as well as sophisticated modelling and interpretation . the milky way bulge or bulge / bar can nowadays be seen upon as the inner structure of the bar @xcite and is certainly a complex environment . it might be a composite system with a mixture of different populations , similarly to many bulges at low redshift @xcite . for recent reviews of the milky way bulge , see @xcite , @xcite , and @xcite . originally , it was thought that the milky way bulge was a classical bulge formed by an initial collapse , with an early , rapid , and intense star formation , in an inside - out formation scenario ( e.g. * ? ? ? * ) , or through accretion of substructures and/or large violent merger events in the @xmath8cdm scenario ( e.g. * ? ? ? * ; * ? ? ? however , the observational evidence for a classical bulge has weakened and the classical contribution seems to be small @xcite , see however the discussions in @xcite and @xcite . at present , a range of observations points toward a complex situation which may be consistent with multiple components / populations or gradients within populations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? on the one hand , kinematic surveys , e.g. @xcite and @xcite are strongly consistent with the bar / x - structure being the predominant population by mass ( see also * ? ? ? * ) . however , there are strong correlations observed between abundances and kinematics , which pose challenges to any unified picture ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? according to the chemical evolution models of @xcite , the modelled spheroidal component formed rapidly on a short timescale via an intense burst of star formation in a classical gravitational gas collapse , and the component created by bar evolution formed later on a longer time scale . note , however , that much of the metal - rich component has elevated alphas ( * ? ? ? * ; * ? ? ? * see also this work ) . the nature of these components are not certain ( see e.g. * ? ? ? * ) and associating them with specific formation scenarios should be done with caution @xcite . important diagnostics for discriminating between models of bulge formation and evolution are _ ( i ) _ @xmath0-abundance trends with metallicity , and _ ( ii ) _ spatial abundance - gradients in the bulge : _ ( i ) _ abundance trends can set constraints on the formation time - scales of the different components @xcite . the dynamics and abundance trends of bulge stars from the metal - poor component ( even though there are differences , see e.g. * ? ? ? * ) point to similarities especially with the ( local ) thick disk ( e.g. * ? ? ? * ; * ? ? ? @xcite show that from a chemical point of view , an inside - out formation , modelled with a initial collapse , be it a due to an early star burst in a disk , a classical bulge , or a thick disk , fits the chemical data . large samples with chemistry and dynamics are needed in order to constrain the formation in detail . _ ( ii ) _ abundance gradients are important outcomes of formation models . there is observational evidence that there is a metallicity gradient from @xmath9 to @xmath10 latitude ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but the origin of this is debated . it could either reflect varying relative proportions of the different components @xcite or it could reflect an intrinsic metallicity gradient produced by the bar due to an original disk gradient @xcite . within @xmath11 , only a few investigations of the metallicity distribution based on stellar spectra at medium- and high resolutions exist . the galactic centre ( gc ) and the region defined by the inner few hundred parsecs have escaped detailed study of stellar abundances and abundance gradients from giants , due to the extreme optical obscuration , resulting in only a handful studies in the literature . @xcite provide evidence for a break in the gradient , being absent with a narrow metallicity distribution at [ fe / h]@xmath12 between @xmath13 and @xmath14 to the south of the galactic centre . these studies are based on @xmath15 spectroscopy of 50 giants observed in the h band . recently , @xcite found a double - peaked distribution with peaks at slightly sub - solar and super - solar metallicities at a northern latitude of @xmath16 , based on medium - resolution spectra of 112 stars in the field . @xcite and @xcite obtained k - band spectra of the supergiant population in the galactic center , finding [ fe / h]@xmath17 , with a narrow dispersion . recently , @xcite also find a metallicity peak in the galactic centre at [ fe / h]@xmath18 , but with a slightly larger spread , also from k - band spectra . all these galactic centre studies are based on approximately 10 stars each , so caution should be exercised when drawing conclusions . recently , @xcite determined metallicities of 83 giants within the inner 1 pc of the galactic center , based on low - resolution ifu spectra ( @xmath19 ) , and found an surprisingly large dispersion ( @xmath20[fe / h]@xmath21 ) . high - resolution spectra are needed to validate these results . in this paper , we will discuss these issues for the inner bulge , within a projected distance of 300 pc ( @xmath22 ) from the center of the galaxy . to overcome the extreme optical extinction in these regions , we observe all our stars at @xmath23@xmath2 . the extinction in the k band is a factor of 10 lower that in the v band @xcite . the large crowding of stars can be dealt with thanks to adaptive optics . to ensure a detailed abundance analysis , we record our spectra with high spectral resolution . we will be able to link the galactic center and the bulge by analysing similar stars and using the same analysis techniques including the same spectral region and high spectral resolution , which to our knowledge has not really been done before . the luminosity and temperature ranges of our stars in the galactic center and the other two fields is similar as well . as presented in @xcite , we also argue that this population is at least 1 gyr old , with no substantial membership from the youngest 10 myr stars . l c c c c c c c c c c r + gc1 & 17:45:35.43 & -28:57:19.28 & 14.74 & 11.90 & 10.48 & 9.40 & 3000 & 3668 & 1.37 & @xmath240.15 & 6.6 + gc20 & 17:45:34.95 & -28:55:20.17 & 14.47 & 11.87 & 10.21 & 9.37 & 3600 & 3683 & 0.76 & @xmath240.14 & 8.1 + gc22 & 17:45:42.41 & -28:55:52.99 & 13.41 & 11.54 & 9.15 & 9.04 & 3600 & 3618 & 0.70 & @xmath240.04 & 6.0 + gc25 & 17:45:36.34 & -28:54:50.41 & 14.35 & 11.60 & 10.09 & 9.10 & 2400 & 3340 & 0.77 & @xmath250.20 & 7.6 + gc27 & 17:45:36.72 & -28:54:52.37 & 14.31 & 11.64 & 10.05 & 9.14 & 3600 & 3404 & 1.16 & @xmath240.23 & 8.1 + gc28 & 17:45:38.13 & -28:54:58.32 & 14.08 & 11.67 & 9.82 & 9.17 & 3000 & 3773 & 1.24 & @xmath250.04 & 5.7 + gc29 & 17:45:43.12 & -28:55:37.10 & 14.39 & 11.59 & 10.13 & 9.09 & 3600 & 3420 & 0.74 & @xmath240.12 & 8.0 + gc37 & 17:45:35.94 & -28:58:01.43 & 13.77 & 11.50 & 9.51 & 9.00 & 3600 & 3754 & 0.93 & @xmath250.08 & 7.7 + gc44 & 17:45:35.95 & -28:57:41.52 & 13.78 & 11.74 & 9.52 & 9.24 & 3600 & 3465 & 0.83 & @xmath240.18 & 6.9 + + bm1 - 06 & 17:49:33.42 & -29:27:28.75 & 11.84 & 11.04 & 10.58 & 10.26 & 1800 & 3814 & 1.18 & 0.29 & 5.1 + bm1 - 07 & 17:49:34.58 & -29:27:14.82 & 12.22 & 11.44 & 10.95 & 10.66 & 2400 & 3873 & 1.34 & 0.08 & 4.2 + bm1 - 08 & 17:49:34.34 & -29:26:57.98 & 11.22 & 10.40 & 9.95 & 9.62 & 1200 & 3650 & 0.89 & 0.17 & 4.1 + bm1 - 10 & 17:49:34.45 & -29:26:48.68 & 10.92 & 10.12 & 9.65 & 9.34 & 1920 & 3787 & 0.80 & -0.23 & 5.0 + bm1 - 11 & 17:49:32.57 & -29:26:30.75 & 11.26 & 10.41 & 10.00 & 9.63 & 1200 & 3812 & 0.92 & 0.12 & 4.6 + bm1 - 13 & 17:49:37.12 & -29:26:40.24 & 10.91 & 10.25 & 9.65 & 9.47 & 1080 & 3721 & 0.84 & -0.94 & 4.9 + bm1 - 17 & 17:49:37.08 & -29:26:21.67 & 11.70 & 10.94 & 10.43 & 10.16 & 1800 & 3775 & 1.13 & -0.83 & 5.0 + bm1 - 18 & 17:49:37.83 & -29:26:19.19 & 12.23 & 11.42 & 10.96 & 10.64 & 1800 & 3780 & 1.32 & 0.22 & 2.7 + bm1 - 19 & 17:49:36.93 & -29:26:10.51 & 12.11 & 11.34 & 10.84 & 10.56 & 3600 & 3958 & 1.32 & 0.18 & 4.0 + + bm2 - 01 & 17:53:29.06 & -29:57:46.22 & 11.44 & 11.11 & 11.08 & 10.89 & 2400 & 3946 & 1.00 & 0.14 & 4.2 + bm2 - 02 & 17:53:24.59 & -29:59:09.48 & 10.78 & 10.44 & 10.41 & 10.21 & 1920 & 4013 & 1.19 & -0.48 & 5.6 + bm2 - 03 & 17:53:27.61 & -29:58:36.39 & 11.30 & 10.93 & 10.94 & 10.70 & 2400 & 3668 & 1.33 & 0.26 & 3.5 + bm2 - 05 & 17:53:33.20 & -29:57:25.88 & 10.07 & 9.51 & 9.70 & 9.28 & 2280 & 3450 & 0.72 & 0.01 & 4.2 + bm2 - 06 & 17:53:30.68 & -29:58:15.75 & 10.01 & 9.71 & 9.65 & 9.48 & 1200 & 4208 & 0.92 & -1.17 & 5.6 + bm2 - 11 & 17:53:31.50 & -29:58:28.51 & 10.20 & 9.91 & 9.83 & 9.68 & 1680 & 4005 & 0.98 & -0.91 & 5.8 + bm2 - 12 & 17:53:31.74 & -29:58:22.94 & 10.67 & 10.32 & 10.30 & 10.09 & 480 & 4003 & 1.14 & -0.11 & 6.2 + bm2 - 13 & 17:53:31.14 & -29:57:32.76 & 10.86 & 10.52 & 10.50 & 10.29 & 2400 & 3727 & 1.18 & -0.16 & 5.2 + bm2 - 15 & 17:53:30.23 & -29:56:42.74 & 10.43 & 9.96 & 10.07 & 9.74 & 1200 & 3665 & 1.01 & 0.22 & 4.5 + bm2 - 16 & 17:53:29.54 & -29:57:22.71 & 10.93 & 10.57 & 10.56 & 10.35 & 1200 & 3886 & 1.22 & 0.10 & 3.2 + + @xmath0 boo & 14:15:39.67 & @xmath2419:10:56.7 & @xmath252.81 & @xmath252.91 & & & @xmath26 & 4286 & 1.66 & @xmath250.52 & 6.3 + bd-012971 & 14:38:48.04 & @xmath2502:17:11.5 & 4.59 & 4.30 & & & 16 & 3573 & 0.50 & @xmath250.78 & 9.3 + 142173@xmath27 & 00:32:12.56 & @xmath2538:34:02.3 & 9.40 & 9.29 & & & 240 & 4330 & 1.50 & @xmath250.77 & 9.3 + 171877@xmath27 & 00:39:20.23 & @xmath2531:31:35.5 & 8.25 & 8.12 & & & 240 & 3975 & 1.10 & @xmath250.93 & 7.0 + 225245@xmath27 & 00:54:46.38 & @xmath2527:35:30.4 & 8.90 & 8.79 & & & 240 & 4031 & 0.65 & @xmath251.16 & 6.6 + 313132@xmath27 & 01:20:20.66 & @xmath2534:09:54.1 & 7.22 & 7.04 & & & 120 & 4530 & 2.00 & @xmath250.20 & 8.7 + 343555@xmath27 & 01:29:42.01 & @xmath2530:15:46.4 & 8.59 & 8.46 & & & 240 & 4530 & 2.25 & @xmath250.74 & 6.8 + hd 787 & 00:12:09.99 & @xmath2517:56:17.8 & 2.03 & 1.86 & & & 4 & 4020 & 0.85 & @xmath250.16 & 8.2 + during the three nights of june , 27 - 29 , 2012 , we observed 28 bulge giants within 2 degrees ( approximately 300 pc in projected distance ) from the galactic center , along the southern minor axis , using the _ very large telescope , vlt , _ spectrometers crires @xcite and isaac @xcite . we observed 9 stars at the galactic center ( gc ) , 9 stars at @xmath28 , and 10 stars at @xmath29 , apart from 7 disk stars which are used for comparison . all the stars are m giants ( @xmath30@xmath31 k and 0.7 @xmath32 2.25 ) , for which we will determine the metallicities , [ fe / h ] , and the abundances of the @xmath0 elements mg and si . typical magnitudes of our bulge stars are @xmath33 . the stars are presented in table [ starsall_tab ] , as well as the six local thick - disk giants , the thin - disk giant bd-012971 , and the reference star @xmath0 boo , which we also will analyse in the same way as for the bulge stars . the @xmath34 and @xmath35 magnitudes of the gc stars are from nishiyama ( private comm . ) and the stars in the @xmath36 and @xmath37 fields , as well as the disk stars , are from 2mass @xcite . our bulge stars were selected in the following way : for the gc and the @xmath36 fields , we used the @xcite data set while for the @xmath37 field the 2mass data set @xcite . the stars are chosen in colour - magnitude diagrams that were dereddened using the high - resolution 3d interstellar extinction map of @xcite and assuming the extinction law of @xcite . the three panels in figure [ cmdall ] show the corresponding dereddened colour magnitude diagram for the gc , @xmath36 and @xmath37 fields , respectively . note that due to the extreme interstellar extinction in the galactic center , the errors in the dereddening procedure remain large which is seen in the larger dispersion of the cmd diagram of the galactic center compared to the @xmath36 and @xmath37 fields . in addition , due to the high interstellar absorption in the galactic center , we use the @xmath38 vs. @xmath39 diagram , since even the @xmath40 magnitudes are not useful due to extinction . our stars were chosen such , that they cover the full colour range in @xmath41 or @xmath42 across the rgb / agb to avoid any biases against metal - rich or metal - poor stars . however , in the galactic center region , small - scale differential extinction not resolved by extinction maps , could lead to possible biases against metal - poor stars . these biases are difficult to quantify , but should be kept in mind . in figure [ cmdall ] our selected stars are shown with red dots superimposed by a few padova isochrones @xcite . evidently , the @xmath43 colour is insensitive to the age of the population , prohibiting an age constrain and therefore a constraint on the mass of our stars . the location of our stars in the cmd shows that we are dealing with m giants and not supergiants . to ensure that our sources are located in the bulge , we placed our stars on the 3d extinction plots of @xcite and found that our stars lie between @xmath44kpc which rules out any possible foreground contamination . furthermore , our derived radial velocity dispersion is about @xmath45 , which is consistent with a typical velocity dispersion of the bulge kinematics ( e.g. * ? ? ? figure [ cmd ] shows the hertzsprung - russell diagram of our selected sources as a function of metallicity . it is based on our derived effective temperatures and surface gravities , @xmath46 , see section [ param ] . it represents nicely the rgb branch covering a wide range of metallicities ( @xmath47[fe / h]@xmath48 ) . , @xmath46 , and metallicity are determined from our spectra , see sect . [ param ] . ] to overcome the extreme extinction we thus observed the giants with the near - ir spectrometers crires @xcite mounted on ut1 ( antu ) , and isaac @xcite on ut3 ( melipal ) of the _ vlt_. every star was observed with both spectrometers . the isaac spectrometer , with a spectral resolution of @xmath49 and a wavelength range between @xmath50@xmath2 , was used to determine the effective temperature of our stars . typical integration times for the high - resolution crires spectra ( @xmath51 , resulting from the @xmath52 wide slit ) are @xmath53 hour per star ( see table [ starsall_tab ] ) , giving a signal - to - noise ratios per pixel of typically 50 - 100 . crires uses , following standard procedures @xcite , nodding on the slit and jittering to reduce the sky background and the adaptive optics ( ao ) macao system to enhance the amount of light captured by the slit . for the crires observations , we used a standard setting ( @xmath54 , order=27 ) with an unvignetted spectral range covering @xmath55 , with three gaps ( 20 ) between the four detector arrays . the reductions of the isaac and crires observations were done by following standard recipes @xcite . for the reduction of the crires spectra , we used gasgano @xcite and , subsequently , iraf @xcite to normalise the continuum , eliminate obvious cosmic hits and correct for telluric lines ( with telluric standard stars ) . two examples of crires spectra are shown in figure [ spectra ] . we have derived the metallicity and the abundances of the @xmath0 elements mg and si for all our stars . here , we describe how we determine the stellar abundances and the important stellar parameters . i.e. the effective temperature ( @xmath30 ) , the surface gravity ( @xmath46 ) , the metallicity ( [ fe / h ] ) , and the microturbulence ( @xmath56 ) , which are needed in order to derive the abundances . we have determined the effective temperatures of our stars in the same manner as in @xcite , using the fact that the co - band - head at @xmath57@xmath2 , in our type of stars , observed at low - resolution , is very temperature sensitive , more so than to any other parameter . @xcite , @xcite , and @xcite studied the behaviour of the @xmath58co band - head with low - resolution , k - band spectra and found a remarkably tight relation between the equivalent width of the co band - head and the effective temperature for m giants . we have thus used this property and obtained relevant low - resolution spectra for all our stars , using the isaac spectrometer . we used the band passes of blum et al . ( 2003 ) as well as their continuum points . effective temperatures were determined by using the following relation @xmath59 ( see blum et al . 2003 ) , with uncertainties of the order of 100k . we apply this method for all our bulge stars . for the three warmer thick - disk stars , we used the temperature determination from @xcite , which is also where we chose our thick disk stars . they determined @xmath30 by imposing an excitation balance for fei lines . surface gravities were determined in the same manner as in @xcite , by using @xmath34 and @xmath35 band photometry and adopting a mean distance of 8.4kpc to the bulge @xcite . in this calculation , we used @xmath34 and @xmath35 band photometry from 2mass ( * ? ? ? * for the @xmath36 and @xmath37 fields ) , nishiyama ( private comm . ) and the vvv survey ( * ? ? ? * for the gc field ) , extinction values from @xcite , and the bolometric corrections from @xcite . we used @xmath60 = 5770 k , @xmath61 = 4.44 , @xmath62 = 4.75 , and assumed that our giants are of 1.0 m@xmath63 . @xcite provide @xmath46 for the disk stars , except for the thin disk star bd-012971 , whose surface gravity was taken from @xcite . the metallicities and microturbulence parameters are determined from our high - resolution , crires spectra , see next paragraph [ stellarabund ] . our derived stellar parameters are given in table [ starsall_tab ] . the fundamental parameters of our comparision star @xmath0 boo are taken from @xcite . with the stellar parameters ( or the fundamental parameters of the star ) determined , we can now obtain the stellar abundances of the elements we are interested in from the high - resolution crires spectra . we analyse these using the software _ spectroscopy made easy , sme _ @xcite , which is a spectral synthesis program that uses a grid of model atmospheres in which it interpolates for a given set of fundamental parameters of the analysed star . in our case we used a grid of marcs spherical - symmetric , lte model - atmospheres @xcite . the spectra that are analysed with sme are pre - normalized and the spectral lines which are of interest are marked with masks that determine what parts of the lines ( i.e. the entire line ) , that will be modelled with synthetic spectra . continuum masks are used to define regions that sme should treat as continuum regions for an extra linear rectification in predefined , narrow windows around the lines analysed . sme then iteratively synthesise spectra for a given set of free variables , which could for example be the searched abundance , under a scheme to minimise the @xmath64 when comparing with the observed spectra . the best fit will provide the abundance of the element from the observed spectral lines . for details , see @xcite . in order to obtain accurate abundances , the atomic data of the spectral lines used have to be known . an atomic line - list based on an extraction from the vald database @xcite was therefore constructed . due to inaccurate atomic data , and the lack of laboratory measurements , we had to fit the atomic lines to the solar intensity spectrum of @xcite , by our determining astrophysical @xmath65 values for , most importantly , fe and si lines . to test these lines we then derive the metallicity and si abundance for @xmath0 boo , which we find to be within 0.05 dex of the values determined by @xcite . the stellar parameters we use for @xmath0 boo are those from the detailed investigation of these authors : @xmath30@xmath66k , @xmath67 ( cgs ) , [ fe / h]@xmath68 , @xmath69kms@xmath70 , and @xmath71kms@xmath70 . furthermore , we used @xmath72 as determined by @xcite and c , n , and o abundances as derived in @xcite and @xcite . these are important in order to synthesise the molecular lines in the k band properly . since the mg lines are impossible to fit in the solar spectrum , we have determined their line strengths such that they fit the standard giant star arcturus flux spectrum @xcite instead , with a mg abundance from @xcite ; [ mg / fe]@xmath73 . in the syntheses of our stars we have also included a line list of the only molecules affecting this wavelength range , namely @xmath58cn and @xmath74cn ( jrgensen & larsson , 1990 ) . in the abundance determination , the microturbulence , @xmath56 , is important for the spectral syntheses , affecting saturated lines . we have chosen to use a typical value of @xmath75 kms@xmath70 found in detailed investigations of red giant spectra in the near - ir by @xcite , see also the discussion in @xcite . finally , in order to match our synthetic spectra with the observed ones , we also convolve the synthetic spectra with a ` macroturbulent ' broadening , @xmath76 , which takes into account the macroturbulence of the stellar atmosphere and instrumental broadening . @xmath76 is found by using sme with this parameter set free , for a set of well determined lines , for every star . in the same manner as described in @xcite , the iron ( giving the metallicity ) , mg and si abundances were then determined for all our stars by @xmath64 minimisation between the observed and synthetic spectra for , depending on star , @xmath77 fe , @xmath78 mg , and @xmath79 si lines . the iron lines have excitation energies between @xmath80ev and line strengths of @xmath81 to @xmath82 . in contrast to @xcite , we do not use the ca abundances due to unknown systematic uncertainties arising from using the ca lines in our wavelength region . to fit the cn lines that might affect the lines we are interested in , we used typically @xmath83 unblended cn lines and fit these by letting the either the c or n abundance free ( see table [ cn ] ) . we then assume a given typical value of the n enrichment or c depletion , respectively , in a heavily cn - cycled red - giant atmosphere , which has experienced the first dredge - up along the subgiant - giant branch , converting c into n : [ n / fe]@xmath84 or [ c / fe]@xmath85 ( for details see * ? ? ? the final metallicities and @xmath0-element abundances we derive , are presented in table [ abund ] and the synthesised spectra are shown with the observed ones for a few typical stars in figure [ spectra ] . the uncertainties in the determination of the abundance ratios , for typical uncertainties in the stellar parameters , are all modest , less than 0.15 dex , ( see table [ errors ] and the discussion in * ? ? ? systematic uncertainties include , for example , non - lte effects and the continuum placement , but are difficult to estimate . the latter we estimate to of the order of 0.1 dex . we thus estimate a total uncertainty of @xmath86 dex . to show the sensitivity of the cn lines to uncertainties in the stellar parameters , we show the uncertainty in the nitrogen abundance for a given c abundance in table [ errors ] , which thus is the abundance change needed to fit the cn lines . lcccc @xmath87 & @xmath240.02 & @xmath250.02 & @xmath250.10 & @xmath250.19 + @xmath88 & @xmath240.02 & @xmath250.02 & @xmath89 & @xmath90 + @xmath91 } = + 0.1 $ ] & & @xmath250.08 & @xmath250.09 & @xmath250.07 + @xmath92 & @xmath250.10 & @xmath250.07 & @xmath250.03 & @xmath250.05 + lccc gc1 & 0.15 & 0.04 & 0.12 + gc20 & 0.14 & 0.10 & 0.04 + gc22 & 0.04 & 0.01 & 0.05 + gc25 & @xmath250.20 & 0.20 & 0.20 + gc27 & 0.23 & 0.01 & 0.15 + gc28 & @xmath250.04 & 0.07 & @xmath250.01 + gc29 & 0.12 & 0.05 & @xmath250.06 + gc37 & @xmath250.08 & 0.06 & 0.18 + gc44 & 0.18 & 0.03 & @xmath250.18 + bm1 - 06 & 0.29 & 0.20 & 0.10 + bm1 - 07 & 0.08 & 0.10 & @xmath250.03 + bm1 - 08 & 0.18 & 0.01 & 0.01 + bm1 - 10 & @xmath250.22 & 0.25 & 0.10 + bm1 - 11 & 0.12 & @xmath250.04 & 0.05 + bm1 - 13 & @xmath250.95 & 0.48 & 0.43 + bm1 - 17 & @xmath250.83 & 0.59 & 0.54 + bm1 - 18 & 0.22 & @xmath250.01 & @xmath250.03 + bm1 - 19 & 0.18 & 0.16 & 0.06 + bm2 - 01 & 0.15 & 0.23 & 0.11 + bm2 - 02 & @xmath250.48 & 0.40 & 0.25 + bm2 - 03 & 0.26 & 0.07 & @xmath250.01 + bm2 - 05 & 0.01 & 0.10 & @xmath250.09 + bm2 - 06 & @xmath251.17 & 0.36 & 0.32 + bm2 - 11 & @xmath250.91 & 0.42 & 0.36 + bm2 - 12 & @xmath250.11 & 0.07 & @xmath250.06 + bm2 - 13 & @xmath250.16 & 0.26 & 0.06 + bm2 - 15 & 0.22 & 0.04 & 0.04 + bm2 - 16 & 0.10 & 0.15 & 0.04 + @xmath0 boo & @xmath250.53 & 0.37@xmath93 & 0.33 + bd-012971 & @xmath250.81 & 0.33 & 0.29 + 142173@xmath94 & @xmath250.77 & 0.27 & 0.26 + 171877@xmath94 & @xmath250.92 & 0.45 & 0.45 + 225245@xmath94 & @xmath251.10 & 0.44 & 0.33 + 313132@xmath94 & @xmath250.28 & 0.29 & 0.27 + 343555@xmath94 & @xmath250.72 & 0.44 & 0.43 + hd 787 & @xmath250.22 & 0.18 & 0.17 + in table [ abund ] we present our abundances : metallicities , [ fe / h ] , and the @xmath0-element abundances of [ mg / fe ] and [ si / fe ] . we have normalised our derived abundances to the solar abundances of @xcite : @xmath95 , @xmath96 , and @xmath97 . we have redetermined the abundances for the galactic center stars of @xcite , in order to be homogenous with the other fields . they only change slightly , in most cases by much less than 0.1 dex , but in a few cases by more than that , which demonstrates that systematic uncertainties can be of that order for stars of low signal - to - noise , or heavy line density in the spectrum . lccc gc1 & 0.19 & & 1.72 + gc20 & 0.00 & & 0.96 + gc22 & 0.12 & & 1.49 + gc25 & 0.34 & & 1.71 + gc27 & 0.16 & & 1.58 + gc28 & 0.06 & & 1.29 + gc29 & 0.14 & & 1.55 + gc37 & 0.15 & & 1.41 + gc44 & 0.08 & & 1.57 + bm1 - 06 & 0.04 & & 1.53 + bm1 - 07 & @xmath250.45 & & 1.06 + bm1 - 08 & @xmath250.06 & & 0.96 + bm1 - 10 & 0.01 & & 1.18 + bm1 - 11 & @xmath250.01 & & 1.20 + bm1 - 13 & 0.00 & & 0.71 + bm1 - 17 & 0.21 & & 1.27 + bm1 - 18 & @xmath250.00 & & 1.29 + bm1 - 19 & 0.09 & & 0.90 + bm2 - 01 & 0.06 & & 1.03 + bm2 - 02 & @xmath250.03 & & 1.02 + bm2 - 03 & 0.17 & & 1.48 + bm2 - 05 & 0.11 & & 1.67 + bm2 - 06 & @xmath250.30 & & 0.49 + bm2 - 11 & @xmath250.10 & & 0.85 + bm2 - 12 & 0.06 & & 1.22 + bm2 - 13 & 0.05 & & 1.17 + bm2 - 15 & 0.04 & & 1.31 + bm2 - 16 & 0.25 & & 1.58 + bd-012971 & @xmath250.03 & & 0.95 + 142173 & 0.02 & & 0.99 + 171877 & 0.05 & & 0.97 + 225245 & 0.08 & & 1.02 + 313132 & 0.38 & & 1.52 + 343555 & @xmath250.01 & & 0.96 + hd 787 & @xmath250.08 & & 0.82 + we have not determined the c and n - abundances for our stars since we do not have any co lines in our wavelength range . however , we have numerous cn lines that we need to fit in the spectral synthesis in order to take blends into account properly . we have therefore determined the c and n abundances such that these are synthesised properly . since the c and n abundances are degenerate , we have determined one of them for a given typical abundance of the other . this means that our carbon or nitrogen abundances are not necessarily the stellar abundances , but rather merely the values we need in combination with the given n or c abundances , respectively , to fit the cn lines . in the second column in table [ cn ] we thus provide the [ c / fe ] abundance that is required to fit the cn lines for a given typical n enhancement for heavily cn - cycled red giants , [ n / fe]@xmath84 @xcite . we see that the carbon abundances , in this case , are more than scaled solar , which is not expected for these type of giants @xcite . in the third column in the table , conversely , the [ n / fe ] abundance that is required to fit the cn lines for a given typical c depletion , [ c / fe]@xmath85 @xcite , are provided . in order to fit the cn lines , in this case , unexpectedly high intrinsic nitrogen abundances are needed . there thus seems to be an indication that we need a high [ c+n / fe ] abundance in these bulge stars , but also our thick disk stars . we note that @xcite measure the c , n , and o abundances in m giants in the inner bulge and find typical red giant values . in @xcite the c , n , and o abundances are also measured from co , cn , and oh lines , which we can not do in the present spectra , and they find only slightly enhanced [ c+n / fe ] for k giants . we find a more extreme change of the c and n abundances , more resembling arp4203 in @xcite , which is a star more resembling those in this paper . we find no correlation with @xmath98 magnitude , but there is a clear correlation with @xmath30 , in the sense that the lower the temperature , the higher the c or n enhancement . in table [ errors ] we show that this large discrepancy can not be explained by an uncertainty in , e.g. , @xmath30 . we do not discuss this further , since we have not measured these abundances , but merely infer them from forcing the cn lines to fit the observations . determining the c , n , and o abundances of these stars or more inner bulge stars would be rewarding . we plot the [ mg / fe ] and [ si / fe ] trends as a function of the metallicity , [ fe / h ] , in figure [ afe1 ] separately for each field : the galactic center , the @xmath99 , and for the @xmath100 fields . in figure [ afe2 ] , we show the trends for all our observed stars in the entire region ( within a projected distance of 300 pc ) together . in this figure , we also include the mean abundance ratios of the two populations of terzan 5 at @xmath101,@xmath102 from @xcite . from the figures we see that the @xmath0-abundance trends , here represented by the [ mg / fe ] and the [ si / fe ] versus [ fe / h ] trends , are very similar in the three inner bulge fields . plotting the trend for all our stars in the inner 300 pc region strengthens this finding . in the figures showing the inner bulge trends , we have also included the trends outlined by the abundances derived from micro - lensed dwarfs from the ` outer ' bulge @xmath103 ) as derived by @xcite . the [ mg / fe ] and [ si / fe ] trends in all three inner regions , from the galactic center to a latitude of @xmath37 , follow the outer bulge trends , within uncertainties , with low values of [ @xmath0/fe ] at [ fe / h]@xmath104 . from these plots it is evident that we can not claim that our [ @xmath0/fe ] trend in the inner 300 pc is particular compared to the ` outer ' bulge , @xmath105 . our thick disk trend also follow the micro - lensed dwarf trend , which gives confidence in the analysis method and shows that we are on the same scale , with room for only small systematic uncertainties . the metallicity of the @xmath37 , @xmath36 , and galactic center fields ranges from @xmath6}<+0.3 $ ] , @xmath106}<+0.3 $ ] , and @xmath7}<+0.3 $ ] , respectively . the two southern fields start picking up some metal - poorer stars , some at very low metallicities ( even below [ fe / h]@xmath107 ) . the spread in metallicities we find are larger than what has been found earlier : @xcite find a dispersion of approximately 0.1 dex around [ fe / h]=@xmath108 to @xmath109 and @xcite find a total spread of @xmath110 dex around [ fe / h]=@xmath111 for the luminous giants and supergiants in the central cluster located within 2.2 pc of the galactic center . the metallicity distributions for the different fields we find in this investigation , are presented in figure [ mdf ] , with the upper three panels showing the three inner fields , and the lowest panel showing the combined metallicity - distribution histogram . the global mean - metallicities of all the observed stars in each of the inner bulge regions are , from the galactic center and outwards , @xmath112 > = + 0.06 $ ] , @xmath113 , and @xmath114 , respectively . for comparison , we can compare our mean metallicities with @xcite who find a mean metallicity of approximately @xmath115[fe / h@xmath116@xmath117 for their fields at @xmath118 and @xmath119 ) , i.e. at slightly lower latitudes . our mean metallicities within the inner @xmath11 gives a formal gradient of @xmath120 or @xmath121dex / degree . this number should be handled with caution due to the very low number of stars . also , there might be undetermined selection effects , which prevents us from selecting metal - poor stars , due to the large and variable extinction in the galactic center field . more stars analysed in the same maner as done here are needed to determine the metallicity distributions in these fields , in order to derive a gradient in the inner bulge . here , we will discuss our results concerning the @xmath0-abundance trends , and the metallicity distributions and gradients in the inner bulge . looking at the mg and si trends , we observe very similar [ @xmath0/fe ] versus [ fe / h ] trends for all three inner fields , given the uncertainties . they are also very similar to the trends found by , for example , @xcite , who analysed micro - lensed dwarfs further out in the bulge ( @xmath105 ) . this is the sample we have compared with in the figures . this suggests a quite homogeneous bulge when it comes to the star - formation history . we can conclude that there are no detectable gradients , to the level of the uncertainties of the present observations , in the [ @xmath0/fe ] levels with latitude . it is interesting to note that @xcite find an identical [ @xmath0/fe ] enhancement trend in plaut s window at @xmath123 ) as in baade s window at @xmath124 , indicating a lack of an [ @xmath0/fe ] gradient in the outer bulge ( @xmath125 . @xcite also find no difference in the [ @xmath0/fe ] in fields at @xmath126 . @xcite also find no gradient from @xmath127 to @xmath14 . jnsson et al . ( in prep . ) find a similar conclusion for fields between @xmath128 . these latter trends are also similar to those in our inner fields @xmath22 presented here . _ we , therefore , conclude that the lack of an [ @xmath0/fe ] gradient further out extends all the way into the galactic centre_. in figure [ afe1 ] we also plot the mean [ mg / fe ] and [ si / fe ] abundances of the two populations in the complex globular cluster terzan 5 located in the inner bulge , @xmath129 degrees north of the galactic plane , from @xcite . the [ mg / fe ] abundances follow our trend identically , whereas the [ si / fe ] abundances are close , with the low - metallicity component being marginally higher than our trend . however , the metal - rich population with a mean at [ fe / h]@xmath130 lies at the high - metallicity end of our distribution . we also note that , in contrast to terzan 5 , we clearly miss a distinct metal - poor population in our galactic centre field . @xcite suggest that there might be a common origin and evolution of field populations in the bulge and terzan 5 , the latter perhaps being a relic building block of the bulge . the similarity in the [ @xmath0/fe ] trends are not inconsistent with this idea . from figure [ afe2 ] we see that the spread in metallicities in the two outer fields span the range similar to what has been found in the ` outer ' bulge , ( see e.g. * ? ? ? * ; * ? ? ? * ) , except that we do not find as high metallicity stars as they do . we do not know the cause for the difference in the high - metallicity end . one possibility is the large difficulty in measuring metallicities , due to the increased number of blends , known and unknown . not taking these into account will systematically lead to too high metallicities . the near - ir wavelength region is not affected as much due to a lower line density . we note that @xcite also find few m - type giants with super - solar metallicities . our galactic centre field , however , shows only the metal - rich component , also found by @xcite and @xcite , even though they found a narrower distribution . we thus find that the metal - poor component seems to disappear at the galactic centre . this could either reflect a true absence of metal - poor stars or it could be due to a selection bias . the interstellar extinction in the galactic centre is extreme and variable and its estimates are not accurate enough , which means that it is very difficult to define an metallicity - unbiased sample of stars . due to this and due to the low number of stars we , therefore , do not want to draw too much conclusions from this fact . @xcite find that the metal - rich component is more concentrated to the plane and progressively disappears with increasing latitude , whereas the metal - poor component shows the same trends and distributions , and kinematics at all the regions investigated , namely @xmath126 . we thus find that this is not the case in the galactic centre . before , discussing the metallicity gradients in the inner bulge further , we will discuss our galactic chemical evolution modelling of the inner bulge . we have modelled the galactic chemical evolution of the inner bulge by using the classical - bulge model adopted in @xcite for the so - called metal - poor population , recently used to model the galactic center by ( * ? ? ? * their model _ ( ii ) _ ; see their figure 8) . in this model we assume a very efficient star formation ( with efficiency larger than a factor of 20 relative to the solar vicinity ) , an imf with more massive stars than normally adopted for the solar neighbourhood @xcite . we have assumed that the bulk of bulge stars formed during a fast episode of gas infall occurring on a timescale of @xmath131 gyr . the adopted stellar yields are the same as in @xcite for massive , low- and intermediate mass stars and supernovae type ia . these yields have been tested on solar neighbourhood stars and can well reproduce the observed [ mg / fe ] and [ si / fe ] trends in those stars . in figure [ vale ] we show the predictions of this model for the inner 500 pc relative to the abundance ratios [ mg / fe ] and [ si / fe ] . we report two curves where the only difference are the yields of mg and si from massive stars . one curve represents the best model of @xcite where the yields of mg and si from massive stars are taken from @xcite , whereas the other curve represents a model where the kobayashi et al . yields of mg and si have been decreased by a factor of 1.35 . this is allowed by the fact that theoretical yields have uncertainties . therefore , the two curves mark the theoretical uncertainties in the stellar yields . the agreement with the [ @xmath0/fe ] vs. [ fe / h ] trend is good for both elements , confirming our previous conclusions in @xcite about the fast formation of the bulk of bulge stars . it is worth noting that the fraction of sne ia in the imf ( @xmath132 ) , adopted in the model , well reproduces the sne ia rate quoted by @xcite . in particular , our predicted rate is @xmath133 sne ia per century . this value is larger than previous estimates ( e.g. * ? ? ? * ) , who quoted 0.03 sne ia per century but pointed out that their value for the type ia sn rate is an order of magnitude lower than expected to explain the large positron injection rate into the galactic centre region , as observed by integral ( * ? ? ? * and references therein ) . the more recent value of @xcite , instead , is between 0.25 and 0.3 sneia per century , in very good agreement with our predicted one . the inner 300 pc includes the central molecular zone ( cmz ) , which is a rich environment with a recent intense star formation ( within 100,000 yrs ; * ? ? ? * ) , massive stars , and is home for three of the most massive young clusters in milky way . however , most stars in this region are nevertheless old , with ages larger than 9 gyr @xcite . to reproduce the star - formation rate ( sfr ) at the present time , @xcite over - impose a recent star burst fitting literature sfr . the origin of the gas can either be from merger processes or from accretion of inner disk gas ( galactic bar ) . this extra burst does however not affect the [ @xmath0/fe ] trends . and @xmath37 fields , below . the lowest panel shows all the stars within @xmath37 of the galactic center in one plot . the global mean of the stars for the three fields are indicated . ] the three top panels of figure [ mdf ] show the metallicity distributions for our three inner bulge fields . the fact that we have very few stars in every region , means that we hardly can talk about metallicity distribution functions . also , there might be a selecition effect against metal - poor stars in the galacitc center field , that we have not quantified . with these caveats in mind , our data shows that the distributions progressively get more metal - rich in the mean , as we approach the galactic centre . the data are consistent with a vertical metallicity gradient . taking the straight mean of the metallicities of the giants in every field , we arrive at a metallicity gradient of @xmath134 dex / degree or @xmath135 dex/100pc corresponds to 140 pc ] . the metal - rich component seems to be stable whereas the metal - poor stars are absent in the galactic centre field . one should , however , be cautious due to the few stars yet available ; given the number of stars in the galactic centre field ( 9 ) and the number of stars with [ fe / h]@xmath136 in the other fields ( 4/19 ) , it is not so unlikely that metal - poor stars were missed in the galactic center field only because of the small number of observed stars . taking the statistics from the two other fields as a basis , we get @xmath137 , which is not a negligibly small probability that no star with [ fe / h]@xmath136 is found if nine stars are being observed . whether there are two components with different proportions at different latitudes or not , is still debated . it should be noted that if there were a clearly distinct metal - poor population with higher scale height and significant numbers , that population should have presented itself clearly in the brava survey @xcite , which is did not . the cylindrical rotation signature of brava is stable through @xmath138 , which argues for a single population with an intrinsic gradient . it is qualitatively well established that there exists a metallicity gradient in the outer bulge along the minor axis from @xmath139 to @xmath140 of the order of @xmath141 to @xmath113 dex / deg ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . also , recent results from the gibs survey show a clear gradient @xcite . more specifically , @xcite find , for example , a gradient of @xmath141 dex / degree from a global photometric metallicity map of the entire @xmath142 bulge . this is close to what @xcite predicted based on a chemodynamic model after a initial collapse , of the inner 500 pc region of the bulge . @xcite study 1200 bulge stars from the _ gaia_-eso survey in 5 fields , the closest being at a latitude of @xmath10 and find a gradient of @xmath108dex / degree ( @xmath143 ) , which however vanishes at @xmath144 . when extrapolating their variation of the two main components from @xmath10 inward , they predict a dominance of the metal - rich component and therefore a weakening of the gradient , which mainly reflects the relative strengths of the two components . this is also seen in the investigation by @xcite of field stars around terzan 5 ( @xmath16 ) . furthermore , @xcite determine a gradient of between @xmath145 and @xmath113dex / degree for the region @xmath146 and @xcite determined a gradient of the order of @xmath121dex / degree for the region @xmath144 . this is also what we find , for our inner fields ( @xmath147 ) . based on our data and these numbers there could be a metallicity gradient of the same order of magnitude extending all the way into the centre . however , in order to state something firmly , we would need many more stars analysed in the same maner and we would need to quantify the selection effect against metal - poor stars , if any . thus , a gradient may be present all the way to the center , but this finding requires further confirmation . some earlier investigations of the metallicity gradient within 4 degrees found evidence for a flattening - out of the gradient , with a narrow peak at [ fe / h]@xmath148 to @xmath149 . @xcite and @xcite find no evidence of a gradient along the minor axis . furthermore , @xcite studied detailed abundances of m - giants in a field at @xmath150 , while @xcite studied fields located at lower latitudes : @xmath151 and @xmath152 . they find a narrow iron - abundance distribution of [ fe / h]=@xmath153 for all three fields , indicating a lack of any major vertical abundance - gradient . however , at the galactic centre , @xcite and @xcite find a narrow metal - rich population at [ fe / h]@xmath18 based on k band spectra . recently , @xcite also find a metallicity peak in the galactic centre at [ fe / h]@xmath18 , but with a slightly larger spread , also from k - band spectra . these are the same data as we present here , however , slightly updated . @xcite , @xcite , and @xcite analysed high - resolution spectra of supergiant stars in the galactic centre finding near - solar metallicity . going north , @xcite find a double peak distribution with peaks at slightly sub - solar and super - solar metallicities with a spread of @xmath154 dex at @xmath16 . recently , @xcite find metal - poor stars beyond 50 pc from the galactic center , indicating the presence of a metal - poor population . more measurements are clearly necessary to establish the nature of the metallicity gradient , if there is one in the inner regions . the detailed studies of @xcite of m giants at @xmath155 in the the @xmath156 , mentioned above , probe the same region as our and uses the same type of objects , even though our m giants lie at @xmath157 in the galactic center and @xmath158 in the southern fields , whereas theirs lie a bit brighter at @xmath159 . they find a narrow iron - abundance distribution of [ fe / h]=@xmath153 for all three fields , but we find a broader distribution . this discrepancy has to be investigated . in order to understand the origin of this difference , we are currently making an differential analysis of these two samples , and investigating the selection of the observed giants in each paper , including how interstellar extinction is traced . what causes the metallicity gradient in the bulge is a matter of debate . early on , a gradient was though to be an evidence for a classical bulge , formed rapidly by an initial collapse , or through accretion of substructures and/or large violent merger events in the @xmath8cdm scenario . however , it has been shown that a gradient can occur also in a purely secular formation scenario ; an intrinsic metallicity gradient could , for example , be produced by the bar due to an original disk gradient @xcite , or may be caused by winds . it is interesting to note that other bulges in spiral galaxies have measured vertical metallicity gradients ( see discussion in * ? ? ? * ; * ? ? ? * ; * ? ? ? how large spiral galaxies and their various structures are formed is not know today ( see e.g. * ? ? ? . an important step on the way to the clarifying this issue , is to study the milky way bulge , which is a major component of our galaxy , a galaxy which we can study in detail . important observables in this context are the [ @xmath0/fe ] trends as a function of metallicity and the latitudinal metallicity - gradients . in this paper , we have presented new abundances and gradients of 28 m - giants located in the inner 300 pc of the milky way bulge . we observed the giants at high spectral resolution in the k band , in order to avoid the extreme visual extinction . our [ mg / fe ] and [ si / fe ] abundances as a function of metallicity show similar behaviour in all three fields , from the galactic centre out to a latitude of @xmath37 . the abundance trends also follow the trends of stars at lower latitudes , in the ` outer ' bulge : the abundance ratios are constant up to [ fe / h]@xmath160 , after which they turn down reaching close to solar - values at [ fe / h]=0 . the ratios then level off at values slightly above scaled - solar values , up to a metallicity of [ fe / h]@xmath161 . we thus find a lack of an [ @xmath0/fe]-gradient in the inner bulge region , similarly to what has been found for the rest of the bulge . we thus conclude that the lack of an [ @xmath0/fe ] gradient further out extends all the way into the galactic centre , which means a rapid formation scenario and a homogeneity of the enrichment process @xcite . our galactic chemical evolution modelling of the inner 300 pc of the bulge shows that our observed abundance ratios and the metallicity distributions , are compatible with this region having experienced a main early , rapid ( @xmath162 gyr ) , strong burst of star formation , with a high star - formation efficiency . also , it indicates a need for an imf skewed to relatively more massive stars compared to the solar vicinity . the inner 300 pc includes the central molecular zone ( cmz ) , with a recent ( second burst of ) intense star formation . this second burst can not have been trigged by more than a modest episode of gas infall or accretion of gas induced by the bar @xcite . this episode does not affect the abundance patterns , however . our data shows a trend where the fraction of metal - poor stars increases the further away from to the galactic centre one looks . in the light of the fact that @xcite find no metal - poor stars in the inner @xmath163 , and since we only have very few stars yet observed in every region , and since there might be selection effects in the galacitc center field due to extreme and variable extinction , often not precisely known , we can not conclude anything about the inner metallicity gradient . more stars in every field is thus needed to draw firm conclusions , which we retrain from doing yet . a metallicity gradient is well established in the bulge at @xmath142 from several different studies . whether this gradient continues all the way into the galactic centre or not , our data can not tell for certain . what we do find is a quite a large range of metallicities in the inner bulge . to summarise , we have presented the first study to connect old stars in the galactic center with the bulge , using high - resolution spectroscopy , stars of similar luminosities , temperatures , and gravities , and using the same analysis methods , including the same spectral region . it firmly argues for the center being in the context of the bulge rather than very distinct . one may tentatively conclude that the very center was not built by the infall of halo or even bulge - like globular clusters , based on metallicity . the @xmath0-element trends with metallicity are very similar for the different regions . only the metallicity distribution seems to change with latitude . this could be expected if the metal - rich component is defined by the boxy / peanut bulge , which increases its fractional importance toward the centre . the boxy / peanut bulge is the inner structure of the bar , originating from vertical instabilities in the disk . this stellar population is , therefore , expected to be similar to the inner disk @xcite . however , as mentioned earlier , a distinct metal - poor population should have presented itself clearly in the brava survey @xcite , which it did not . we would like to thank livia origlia for very fruitful and insightful discussions . acknowledges support from the swedish research council , vr ( project number 621 - 2014 - 5640 ) , funds from kungl . fysiografiska sllskapet i lund . ( stiftelsen walter gyllenbergs fond and mrta och erik holmbergs donation ) . f.m . and v.g . acknowledge financial support from prin miur 2010 - 2011 , project the chemical and dynamical evolution of the milky way and local group galaxies , prot . 2010ly5n2 t . rmr acknowledges support from grant ast-1413755 from the national science foundation , nsf . this work is based on observations collected at the european southern observatory , chile , program number 089.b-0312(a)/vm / crires and 089.b-0312(b)/vm / isaac . this publication makes use of data products from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . _ facilities : _ , . , h .- u . , ballester , p. , biereichel , p. , et al . 2004 , in presented at the society of photo - optical instrumentation engineers ( spie ) conference , vol . 5492 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . a. f. m. moorwood & m. iye , 12181227 , w. , & wallace , l. 1991 , an atlas of the solar spectrum in the infrared from 1850 to 9000 cm-1 ( 1.1 to 5.4 micrometer ) ( nso technical report , tucson : national solar observatory , national optical astronomy observatory , 1991 )

we introduce an architecture for robust and scalable quantum computation using both stationary qubits ( e.g. single photon sources made out of trapped atoms , molecules , ions , quantum dots , or defect centers in solids ) and flying qubits ( e.g. photons ) . our scheme solves some of the most pressing problems in existing non - hybrid proposals , which include the difficulty of scaling conventional stationary qubit approaches , and the lack of practical means for storing single photons in linear optics setups . we combine elements of two previous proposals for distributed quantum computing , namely the efficient photon - loss tolerant build up of cluster states by barrett and kok [ phys . rev . a * 71 * , 060310 ( 2005 ) ] with the idea of repeat - until - success ( rus ) quantum computing by lim _ et al . _ [ phys . rev . lett . * 95 * , 030505 ( 2005 ) ] . this idea can be used to perform eventually deterministic two - qubit logic gates on spatially separated stationary qubits via photon pair measurements . under non - ideal conditions , where photon loss is a possibility , the resulting gates can still be used to build graph states for one - way quantum computing . in this paper , we describe the rus method , present possible experimental realizations , and analyse the generation of graph states .

quant-ph0508218

quantum computing offers a way to realize certain algorithms exponentially more efficiently than with the best known classical solutions @xcite . a substantial effort has therefore been made to develop the corresponding quantum technologies . proof - of - principle experiments demonstrating the feasibility of quantum computing have already been performed : using nuclear magnetic resonance techniques , vandersypen _ et al . _ @xcite realized a simple instance of shor s algorithm by factoring @xmath0 . a two qubit gate has been implemented in a color center in diamond , utilizing the electron spin state of the nitrogen - vacancy defect center together with a nearby nuclear spin as qubits @xcite . groups in innsbruck and boulder implemented a universal two - qubit gate in an ion trap @xcite , and the three - qubit teleportation protocol @xcite . adding more qubits to this `` proto quantum computer '' will increase the density of the motional states used for the two - qubit interaction . consequently , it will become even harder to implement clean two - qubit gates . scaling ion trap quantum computers much further therefore seems to require some form of distributed quantum information processing , possibly involving ion transport @xcite . the schemes mentioned above are based on manipulating _ stationary _ qubits such as atoms , molecules , or trapped ions . an alternative route to finding a feasible and scalable technology for building quantum computers is based on _ flying _ qubits , such as photons . the main advantage of photons is their extremely long coherence time : in vacuum and in simple dielectric media , photons do not interact with their environment , and hence do not lose their quantum information . this is why photons are usually the qubits of choice for quantum communication @xcite . however , at the same time this lack of interaction means that it is very hard to create two - photon entangling gates . it therefore came as a surprise that the bosonic symmetry requirement of the electromagnetic field , together with photon counting and proper single - photon sources , is sufficient for implementing scalable quantum computing @xcite . the overhead cost for linear optical quantum computing ( loqc ) has subsequently been brought down significantly . in particular , the one - way or cluster - state model for quantum computing @xcite has allowed for drastic improvements in the scalability @xcite . recently , a four - qubit cluster state was realized experimentally by walther _ et al_.@xcite . the main drawbacks of loqc are the difficulties of maintaining interferometric stability , the lack of practical ` on demand ' single - photon sources and the lack of quantum memories for photonic qubits @xcite . in this paper , we consider the practical advantage of combining stationary and flying qubits for the realisation of scalable quantum computing . the stationary qubits ( single photon sources ) are arranged in a network of nodes with each node processing and storing a small number of qubits . to achieve scalability , the concept of distributed quantum computing was introduced and it was proposed that distant qubits communicate with each other through the means of flying qubits ( i.e. photons ) @xcite . initial schemes for the implementation of this idea relied on entangled ancillas as a resource @xcite . others required that the photon from one source is fed into another source @xcite or a photon - mediated interaction between two fiber - coupled distant cavities needed to be established @xcite . more hybrid approaches to quantum computing can be found in refs . @xcite . other authors developed schemes for the probabilistic generation of highly entangled states between distant single photon sources @xcite . in these schemes , one generates a photon in each of the sources and then performs an entangling photon measurement . by virtue of entanglement swapping , this results in entangled stationary qubits . it has been shown that similar ideas can also result in the implementation of probabilistic remote two - qubit gates @xcite . at this point , it was believed that scalable quantum computing with distant photon sources requires additional resources such as local entangling gates @xcite or entangled ancillas in order to become deterministic . the concrete setup that we consider in this paper has recently been introduced by lim _ _ @xcite and allows for the more efficient implementation of universal two - qubit gates than previous proposals . the presented scheme consists of a network of single stationary qubits ( like trapped atoms , molecules , ions , quantum dots and nitrogen vacancy color centers ) inside optical cavities , which act as a source for the generation of single photons on demand . read - out measurements and single qubit rotations can be performed on the stationary qubits using laser pulses and standard quantum optics techniques as employed in the recent ion trap experiments in innsbruck and boulder @xcite . the main building block for the realization of a two - qubit gate , which qualifies the setup for universal quantum computing , is shown in figure [ moon ] . it requires the simultaneous generation of a photon in each source involved in the operation . afterwards the photons should pass through a linear optics setup , where a pair measurement is performed in the output ports . this photon pair measurement results either in the completion of the gate or indicates the presence of the original qubits . in the later event , the gate should be repeated . the qubits are never lost in the computation and the presented scheme has therefore been called _ repeat - until - success _ quantum computing @xcite . under realistic conditions , i.e. in the presence of finite detector efficiencies and finite success rates for the generation of a single photon on demand , the setup in figure 1 can still be used for the implementation of probabilistic gates with a very high fidelity . as shown recently by barrett and kok @xcite , it is possible to use probabilistic gates to efficiently generate graph states for one - way quantum computing @xcite . both schemes , @xcite and @xcite overcome the limitations to scalable quantum computing faced before when using the same resources . in ref . @xcite this is achieved with an eventually deterministic gate . @xcite introduced a so - called double - heralding scheme , in which the entangling photo - detection stage was employed twice to eliminate unwanted separable contributions to the density matrix . in this paper , we combine the ideas presented in our previous work @xcite . in this way , we obtain a truly scalable design for quantum computing , i.e. even in the presence of imperfect components , with several key advantages : 1 . since our system uses _ no direct qubit - qubit interactions _ , the qubits can be well isolated . not only does this allow us to address the individual qubits easily , it also means that there are fewer decoherence channels and hence fewer errors in the computation . we achieve _ robustness to photon loss_. in the presence of photon loss , the two qubit gates become non - deterministic . however , the gate failures are heralded , and so the gates can still be used to build high - fidelity entangled states , albeit in a non - deterministic manner . photon loss thus increases the overall overhead cost associated with the scheme , but does not directly reduce the fidelity of the computation . when realistic photo - detectors and optical elements are used , photon loss is inevitable and this built - in robustness is essential . our scheme largely relies only on _ components that have been demonstrated in experiments _ like atom - photon entanglement @xcite . apart from linear optics , we require only relatively good sources for the generation of single photons on demand @xcite , preferrably at a high rate @xcite , and relatively efficient but not necessarily number resolving photon detectors @xcite . combining these in a working quantum computer will be challenging , but the basic physics has been shown to be correct . 4 . the photon pair measurement is _ interferometrically stable_. since each generated photon contributes equally to the detection of a photon in the linear optics setup , fluctuations in the length between the photon source and the detectors can at most result in an overall phase factor with no physical consequences . this constitutes a significant advantage compared to previous schemes based on one - photon measurements ( the only interferometrically stable schemes are @xcite ) , since the photons do not need to arrive simultaneously in the detectors as long as they overlap within their coherence time in the setup . the basic ideas presented in this paper are _ implementation independent _ and the stationary qubits can be realised in a variety of ways . any system with the right energy - level structure and able to produce encoded flying qubits may be used . our scheme is inherently _ distributed_. hence , it can be used in applications which integrate both quantum computation and quantum communication . we show that entanglement can be generated directly between any two stationary qubits in the physical quantum computer . this significantly reduces the computational cost compared to architectures involving only nearest - neighbor interactions between the qubits @xcite . this paper is organized as follows . in the next section , we give an overview on the basic principles of measurement - based quantum computing , since the described hybrid approach to quantum computing constitutes a novel implementation of these ideas . section iii details the general principle of a remote two - qubit gate implementation . in section [ rea ] , we discuss possible gate implementations with polarization and time - bin encoded photons . in section [ cluster ] , we describe how to overcome imperfections of inefficient photon generation and detection with the help of pre - fabricated graph states . finally , we state our conclusions in section [ conc ] . one condition for the successful implementation of a measurement - based quantum gate is that the measurement outcome is _ mutually unbiased _ @xcite with respect to the computational basis . in this way , an observer does not learn anything about the state of the qubits and the information might remain stored inside the computer . to avoid the destruction of qubits , it is not allowed to measure on the qubits directly . measurements should only be performed on ancillas , which have interacted and therefore share entanglement with the qubits . these ancillas can be of the same physical realisation as the computational qubits @xcite but they might also be realised differently . if the stationary qubits are atoms , the ancilla can be the quantised field mode inside an optical cavity @xcite , a common vibrational mode @xcite or newly generated photons , as in the setup considered here . vice versa , it has been found advantageous to use collective atomic states as ancillas for photonic qubits @xcite . let us now briefly describe the principles of measurement - based quantum computing in a more formal way . using the terms ` qubits ' and ` ancillas ' provides a convenient picture , which is especially suited for the description of hybrid approaches , where the qubits may remain encoded in the same physical qubits instead of being assigned dynamically as the computation proceeds . as in ref . @xcite , we consider two systems , @xmath1 and @xmath2 , that are initially in the state ( @xmath3 ) @xmath4 after some interaction , the joint system evolves into @xmath5 where the @xmath6 are the eigenstates of an observable @xmath7 . we can then measure @xmath7 , which will reveal the state of the system @xmath1 . this can be interpreted as a quantum non - demolition measurement of @xmath1 but this is not what we are interested in here . in this article we will instead consider measurements of an observable @xmath8 , as shown in figure [ minkslevel ] , that is complementary to @xmath7 . in other words , the eigenvectors of @xmath7 and @xmath8 form a so - called mutually unbiased basis of the hilbert space of system @xmath2 . a specific outcome labelled @xmath9 of such a measurement corresponds to the application of the projection operator @xmath10 ( associated with the @xmath11 eigenvector of @xmath8 ) , and the state of system @xmath1 is then given by @xmath12}{\text{tr}_{sa } \left [ \langle\phi| \unity \otimes \hat{b}_k |\phi\rangle \right ] } \ , .\ ] ] this can be generalized to situations where @xmath10 is a multi - rank projector or a positive operator valued measure ( povm ) . the conditions for the evolution @xmath13 to be a unitary transformation on system @xmath1 are presented in lapaire _ if @xmath1 describes the qubits and @xmath2 the ancilla , they guarantee , as mentioned above , that the detection of @xmath10 does not reveal any information about the qubits . especially , in the setup considered in this paper the system @xmath1 consists of a set of @xmath14 stationary qubits occupying a hilbert space of size @xmath15 , and system @xmath2 consists of @xmath14 flying quantum systems occupying a hilbert space of dimension @xmath16 . a measurement of the observable @xmath8 on the flying qubits will result in a multi - qubit ( entangling ) operation on the stationary qubits . we are interested in the case where the projector @xmath10 induces a _ unitary _ transformation on the stationary qubits , @xmath17 which means that @xmath10 is a proper projector . there are two interesting cases to consider : 1 . the set @xmath18 corresponds to a basis of states that induces a complete set @xmath19 of entangling quantum gates . as a result , finding any measurement outcome @xmath9 will induce a unitary entangling gate operation on the stationary qubits . the set @xmath18 corresponds to a basis that can be divided into two sets of states : some of the projectors will induce a unitary entangling gate @xmath20 on the stationary qubits , while the remaining projectors induce a transformation that is locally equivalent to the identity map @xmath21 . [ r]@xmath22 [ r]@xmath23 the second setup is interesting for the following reason : suppose that system @xmath1 consists of @xmath14 non - interacting ( e.g. , well - separated ) stationary qubits with long decoherence times . if this system can generate flying qubits in the manner described above , we can perform a measurement of the observable @xmath8 and entangle the non - interacting stationary qubits . when not all measurement outcomes produce an entangling gate on the stationary qubits ( some yield instead the identity operator ) , then the unitary gate is applied only part of the time . however , due to the fact that a gate failure corresponds to an identity operation ( or something locally equivalent ) , we can again prepare the flying qubits in the state @xmath23 . this allows us to repeat the protocol until the entangling gate is applied successfully , which is why we called this idea repeat - until - success quantum computing @xcite . one of the requirements for universal quantum computing is the ability to perform an entangling two - qubit gate operation , like a controlled phase gate @xcite . in this section we describe the general concept for the implementation of this gate between two distant single photon sources . note that our method of distributed quantum computing allows to realize entangling operations , since the measurement on a photon pair can imprint a phase on the state of its sources although it can not change the distribution of their populations . the first step for the implementation of a two - qubit gate is the generation of a photon within each respective source , which encodes the information of the stationary qubit . let us denote the states of the photon sources , which encode the logical qubits @xmath24 and @xmath25 as @xmath26 and @xmath27 , respectively . an arbitrary pure state of two stationary qubits can then be written as @xmath28 where @xmath29 , @xmath30 , @xmath31 and @xmath32 are complex coefficients with @xmath33 leads to the creation of one photon in state @xmath34 , while source 2 prepared in @xmath35 leads to the creation of one photon in state @xmath36 , @xmath37 the simultaneous creation of a photon in both sources then transfers the initial state ( [ original ] ) into @xmath38 note that the generation of photons whose state depends on the states of the stationary qubits is a highly non - linear process . the preparation of the generally entangled state ( [ theencoding ] ) is indeed the key step which allows the completion of an eventually deterministic two - qubit gate with otherwise nothing else than linear optics and photon pair measurements . the way the encoding step ( [ enc ] ) can be realized experimentally is discussed in section [ rea ] . in this section we focus on the general ideas underlying repeat - until - success quantum computing . once the photons have been created , an entangling phase gate can be implemented by performing an absorbing measurement on the photon pair . thereby , it is important to choose the photon measurement such that none of the possible outcomes reveals any information about the coefficients @xmath29 , @xmath30 , @xmath31 and @xmath32 , as mentioned in section ii . this can be achieved with a photon pair measurements in a basis mutually unbiased @xcite with respect to the computational basis given by the states @xmath39 . more concretely , all possible outcomes of the photon measurement should be of the form @xmath40 \ , .\ ] ] as we see below , a complete set of basis states of this form can be found . any bias in the amplitudes would yield information about @xmath29 , @xmath30 , @xmath31 , and @xmath32 , and would therefore not induce a unitary gate on the stationary qubits . detecting the state ( [ unbiased ] ) and absorbing the two photons in the process transfers the encoded state ( [ theencoding ] ) into @xmath41 note that the output state ( [ out ] ) differs from the initial state ( [ original ] ) by a two - qubit phase gate . let us now consider the angle @xmath42 in this case , the state @xmath43 is a product state and the output ( [ out ] ) differs from the initial state ( [ original ] ) only by local operations . however , if @xmath44 the state ( [ unbiased ] ) becomes a maximally entangled state , as it becomes obvious when writing @xmath43 as @xmath45 \ , .\ ] ] the detection of a photon pair in this maximally entangled state results in the completion of a phase gate with maximum entangling power on the stationary qubit . vice versa , maximum entanglement of the state ( [ unbiased ] ) also automatically implies eq . ( [ phasecondition ] ) as one can show by calculating the entanglement of formation of the state ( [ unbiased ] ) . in the following , we denote the states of the measurement basis , i.e. the mutually unbiased basis , by @xmath46 . in order to find a complete bell basis with all states of form ( [ unbiased ] ) , we define @xmath47 \ , , \nonumber \\ { |\phi_2\rangle } & \equiv & { \textstyle { 1 \over \sqrt{2 } } } \big [ { |{\sf a}_1 { \sf b}_1\rangle}-{|{\sf a}_2 { \sf b}_2\rangle } \big ] \ , , \nonumber \\ { |\phi_3\rangle } & \equiv & { \textstyle { 1 \over \sqrt{2 } } } \big [ { |{\sf a}_1 { \sf b}_2\rangle}+{|{\sf a}_2 { \sf b}_1\rangle } \big ] \ , , \nonumber \\ { |\phi_4\rangle } & \equiv & { \textstyle { 1 \over \sqrt{2 } } } \big [ { |{\sf a}_1 { \sf b}_2\rangle}-{|{\sf a}_2 { \sf b}_1\rangle } \big ] \ , , \end{aligned}\ ] ] where the states @xmath48 describe photon 1 and the @xmath49 describe photon 2 . assuming orthogonality , i.e. @xmath50 and @xmath51 , one can write the photon states on the right hand side of eq . ( [ base ] ) without loss of generality as @xmath52 with @xmath53 inserting this into eq . ( [ base ] ) , we find @xmath54 , \nonumber \\ { |\phi_2\rangle } & = & { \textstyle { 1 \over \sqrt{2 } } } \big [ \big(c_1c_2-{\rm e}^{-{\rm i}(\vartheta_1+\vartheta_2 ) } { \rm e}^{-{\rm i}(\xi_1+\xi_2 ) } s_1s_2 \big){|{\sf x}_0{\sf y}_0\rangle } + \big({\rm e}^{{\rm i}\vartheta_2}c_1s_2+{\rm e}^{-{\rm i}\vartheta_1 } { \rm e}^{-{\rm i}(\xi_1+\xi_2 ) } s_1c_2 \big ) { |{\sf x}_0{\sf y}_1\rangle } \nonumber \\ & & + \big ( { \rm e}^{{\rm i}\vartheta_1}s_1c_2+{\rm e}^{-{\rm i}\vartheta_2 } { \rm e}^{-{\rm i}(\xi_1+\xi_2 ) } c_1s_2 \big ) { |{\sf x}_1{\sf y}_0\rangle}+ \big ( { \rm e}^{{\rm i}(\vartheta_1+\vartheta_2)}s_1s_2-{\rm e}^{-{\rm i}(\xi_1+\xi_2 ) } c_1c_2 \big ) { |{\sf x}_1{\sf y}_1\rangle } \big ] , \nonumber \\ { |\phi_3\rangle } & = & { \textstyle { 1 \over \sqrt{2}}}\big [ \big({\rm e}^{-{\rm i}\vartheta_2 } { \rm e}^{-{\rm i}\xi_2}c_1s_2+{\rm e}^{-{\rm i}\vartheta_1 } { \rm e}^{-{\rm i}\xi_1 } s_1c_2 \big){|{\sf x}_0{\sf y}_0\rangle}- \big({\rm e}^{-{\rm i}\xi_2 } c_1c_2-{\rm e}^{-{\rm i}(\vartheta_1-\vartheta_2 ) } { \rm e}^{-{\rm i}\xi_1}s_1s_2 \big ) { |{\sf x}_0{\sf y}_1\rangle } \nonumber \\ & & + \big({\rm e}^{{\rm i}(\vartheta_1-\vartheta_2 ) } { \rm e}^{-{\rm i}\xi_2 } s_1s_2- { \rm e}^{-{\rm i}\xi_1 } c_1c_2 \big ) { |{\sf x}_1{\sf y}_0\rangle } - \big({\rm e}^{{\rm i}\vartheta_1 } { \rm e}^{-{\rm i}\xi_2 } s_1c_2+{\rm e}^{{\rm i}\vartheta_2 } { \rm e}^{-{\rm i}\xi_1}c_1s_2 \big ) { |{\sf x}_1{\sf y}_1\rangle } \big ] , \nonumber \\ { |\phi_4\rangle } & = & { \textstyle { 1 \over \sqrt{2}}}\big [ \big({\rm e}^{-{\rm i}\vartheta_2 } { \rm e}^{-{\rm i}\xi_2 } c_1s_2-{\rm e}^{-{\rm i}\vartheta_1 } { \rm e}^{-{\rm i}\xi_1}s_1c_2 \big ) { |{\sf x}_0{\sf y}_0\rangle } - \big({\rm e}^{-{\rm i}\xi_2 } c_1c_2+{\rm e}^{-{\rm i}(\vartheta_1-\vartheta_2 ) } { \rm e}^{-{\rm i}\xi_1 } s_1s_2 \big ) { |{\sf x}_0{\sf y}_1\rangle } \nonumber \\ & & + \big({\rm e}^{{\rm i}(\vartheta_1-\vartheta_2 ) } { \rm e}^{-{\rm i}\xi_2 } s_1s_2 + { \rm e}^{-{\rm i}\xi_1 } c_1c_2 \big ) { |{\sf x}_1{\sf y}_0\rangle}- \big({\rm e}^{{\rm i}\vartheta_1 } { \rm e}^{-{\rm i}\xi_2 } s_1c_2-{\rm e}^{{\rm i}\vartheta_2 } { \rm e}^{-{\rm i}\xi_1 } c_1s_2 \big ) { |{\sf x}_1{\sf y}_1\rangle } \big ] . \nonumber \\\end{aligned}\ ] ] these states are of the form ( [ unbiased ] ) , if the amplitudes are all of the same size , which yields the conditions @xmath55 and @xmath56 the only solution of the constraints ( [ constraint1 ] ) and ( [ constraint2 ] ) is @xmath57 provided that neither @xmath58 nor @xmath59 equals 1 . in the special case , where either @xmath60 or @xmath61 , condition ( [ con ] ) simplifies to @xmath62 with no restrictions in the angles @xmath63 , @xmath64 , @xmath65 and @xmath66 . one particular way to fulfill the restrictions ( [ con ] ) is to set @xmath67 which corresponds to the choice ( c.f . @xcite ) @xmath68 and yields @xmath69 \ , , \nonumber \\ { |\phi_2\rangle } & = & { \textstyle { 1 \over 2}}{\rm e}^{-{\rm i}\pi/4 } \big [ { |\sf x_0y_0\rangle}+{\rm i}{|\sf x_0y_1\rangle}+{\rm i}{|\sf x_1y_0\rangle}+{|\sf x_1y_1\rangle } \big ] \ , , \nonumber \\ { |\phi_3\rangle } & = & { \textstyle { 1 \over 2}}{\rm e}^{{\rm i}\pi/4 } \big [ { |\sf x_0y_0\rangle}-{\rm i}{|\sf x_0y_1\rangle}+{\rm i}{|\sf x_1y_0\rangle}-{|\sf x_1y_1\rangle } \big ] \ , , \nonumber \\ { |\phi_4\rangle } & = & -{\textstyle { 1 \over 2}}{\rm e}^{-{\rm i}\pi/4 } \big [ { |\sf x_0y_0\rangle}+{\rm i}{|\sf x_0y_1\rangle}-{\rm i}{|\sf x_1y_0\rangle}-{|\sf x_1y_1\rangle } \big ] \ , . \nonumber \\\end{aligned}\ ] ] to find out which gate operation the detection of the corresponding maximally entangled states ( [ base ] ) combined with a subsequent absorption of the photon pair results into , we write the input state ( [ theencoding ] ) as @xmath70 and determine the states @xmath71 of the stationary qubits . using the notation @xmath72 for the controlled two - qubit phase gate ( the cz gate ) and the notation @xmath73 for the local controlled - z gate on photon source @xmath74 @xcite , we find @xmath75 from this we see that one can indeed obtain the cz gate operation ( [ cz ] ) up to local unitary operations upon the detection of any of the four bell states @xmath76 , as it has been pointed out already by protsenko _ _ @xcite . when implementing distributed quantum computing with photons as flying qubits , the problem arises that it is impossible to perform a complete deterministic bell measurement on the photons using only linear optics elements . as it has been shown @xcite , in the best case , one can distinguish two of the four bell states . since the construction of efficient non - linear optical elements remains experimentally challenging , the above described phase gate could therefore be operated at most with success rate @xmath77 . what must be done to solve this problem is to choose the photon pair measurement basis @xmath78 such that two of the basis states are maximally entangled while the other two basis states are product states . most importantly , all basis states must be mutually unbiased with respect to the computational basis and information will not be destroyed at any stage of the computation . in the following we choose @xmath79 and @xmath80 as in eq . ( [ base ] ) and @xmath81 and @xmath82 as product states such that @xmath83 \ , , \nonumber \\ { |\phi_4\rangle } & \equiv & { \textstyle { 1 \over \sqrt{2 } } } \big [ { |{\sf a}_1 { \sf b}_2\rangle}-{|{\sf a}_2 { \sf b}_1\rangle } \big ] \ , .\end{aligned}\ ] ] the aim of this is ( see section [ sectionii ] ) that in the event of the `` failure '' of the gate implementation ( i.e. in case of the detection of @xmath81 or @xmath82 ) the system remains , up to a local phase gate , in the original qubit state . this means that the initial state ( [ original ] ) can be restored and the described protocol can be repeated , thereby eventually resulting in the performance of the universal controlled phase gate ( [ cz ] ) . the probability for the realization of the gate operation within one step equals @xmath77 and the final completion of a quantum phase gate therefore requires on average _ two _ repetitions of the above described photon pair generation and detection process . let us now determine the conditions under which the states @xmath84 are of the form ( [ unbiased ] ) . proceeding as above , we find that the angles @xmath85 , @xmath86 and @xmath87 in eq . ( [ ab ] ) should fulfill , for example , eq . ( [ angels ] ) . in analogy to eqs . ( [ constraint1 ] ) and ( [ constraint2 ] ) , we find that @xmath81 and @xmath82 are mutually unbiased , if @xmath88 which also holds for the parameter choice in eq . ( [ angels ] ) . using eq . ( [ angelencoding ] ) , one can easily verify that with the above choice the basis ( [ twoproduct ] ) becomes @xmath89 \ , , \nonumber \\ { |\phi_2\rangle}&= & { \textstyle { { \rm i } \over 2 } } \big [ { |\sf x_0y_0\rangle}-{|\sf x_0y_1\rangle}-{|\sf x_1y_0\rangle}+{|\sf x_1y_1\rangle } \big ] \ , , \nonumber \\ { |\phi_3\rangle } & = & { \textstyle { 1 \over 2}}{\rm e}^{{\rm i}\pi/4 } \big [ { |\sf x_0y_0\rangle}-{\rm i}{|\sf x_0y_1\rangle}+{\rm i}{|\sf x_1y_0\rangle}-{|\sf x_1y_1\rangle } \big ] \ , , \nonumber \\ { |\phi_4\rangle } & = & -{\textstyle { 1 \over 2}}{\rm e}^{-{\rm i}\pi/4 } \big [ { |\sf x_0y_0\rangle}+{\rm i}{|\sf x_0y_1\rangle}-{\rm i}{|\sf x_1y_0\rangle}-{|\sf x_1y_1\rangle } \big ] \ , . \nonumber \\\end{aligned}\ ] ] choosing the states @xmath48 and @xmath90 as in eq . ( [ angelencoding ] ) allows to implement the gate operation ( [ cz ] ) eventually deterministically . finally , we determine the gate operations corresponding to the detection of a certain measurement outcome @xmath91 . to do this , we decompose the input state ( [ theencoding ] ) again into a state of the form ( [ bla ] ) . proceeding as in the previous subsection , we find @xmath92 again one obtains the cz gate operation ( [ cz ] ) up to local unitary operations upon the detection of either @xmath93 or @xmath94 . in the event of the detection of the product states @xmath81 or @xmath82 , the initial state can be restored with the help of one - qubit phase gates , which then allows us to repeat the operation until success . it should be emphasized that there are other possible encodings that yield a universal two - qubit phase gate upon the detection of a bell - state , but where the original state is destroyed upon the detection of a product state ( see e.g. @xcite ) . this happens when the product states are not mutually unbiased and their detection erases the qubit states in the respective photon sources . to achieve the effect of an _ insurance _ against failure , the encoding ( [ enc ] ) should be chosen as described in this section . possible experimental realizations of the above described eventually deterministic quantum phase gate consist of two basic steps . firstly , the information of the stationary qubits involved in the operation has to be redundantly encoded in the states of two newly generated ancilla photons . afterwards , a measurement is performed on the photon pair resulting with probability @xmath95 in the desired gate operation . depending on the type of the photon source , one can choose different types of encoding . there are also different possibilities how to perform the photon pair measurement . examples are given below . in order to obtain robust qubits , the states @xmath96 and @xmath97 should be two different longliving ground states of the single photon source . each photon source carries one qubit . depending on its level structure ( see figure [ photongun ] ) , it might be advantageous to realise the encoding step ( [ theencoding ] ) either by generating photons with different polarisations ( polarisation encoding ) or photons that agree in all degrees of freedom apart from their creation time ( time bin encoding ) . note that different encodings can easily be transformed into each other using linear optics elements like a polarising beam splitters and delaying photons in time . _ polarization encoding . _ suppose , the photon source contains an atomic double @xmath98 level configuration as shown in figure [ photongun](a ) ( see also ref . a single photon can then be created by simultaneously applying a laser pulse with increasing rabi frequency to the @xmath99-@xmath100 transition and the @xmath101-@xmath102 transition of the atomic system . thereby , the atom goes to the ground state @xmath103 and @xmath104 , respectively , depending on whether its initial state equalled @xmath96 or @xmath97 due to the coupling of the @xmath100-@xmath105 transition and the @xmath102-@xmath106 transition to the cavity mode . it has been shown in the past that this technique @xcite is very well suited to place exactly one excitation into the field of an optical resonator , from where it can leak out @xcite . if the two transitions , @xmath100-@xmath105 and @xmath102-@xmath106 , couple to the two different polarisation modes @xmath107 and @xmath108 , in the cavity field , the photon generation results effectively , for example , in the operation @xmath109 once atom @xmath74 has been repumped into its initial state @xmath110 and @xmath111 , respectively . finally we remark that the encoding does not affect the coefficients @xmath29 , @xmath30 , @xmath31 and @xmath32 of the initial state ( [ original ] ) . as long as no measurement is performed on the system , all coherences are preserved . _ time - bin encoding . _ alternatively , if the photon sources possess a level structure like the one shown in figure [ photongun](b ) , one can redundantly encode the information contained in the qubits into time bin encoded photons , @xmath112 this encoding is simpler and may therefore find realizations not only in atoms but also in quantum dots and nitrogen vacancy color centers . in eq . ( [ higgs ] ) , @xmath113 and @xmath114 denote a single photon generated at an _ early _ and a _ later _ time , respectively . the above operation can be achieved by first coupling a laser field with increasing rabi frequency to the @xmath115 transition , while the cavity mode couples to the @xmath116 transition . once the excitation has been placed into the cavity mode and leaked out through the outcoupling mirror , the atom can be repumped into @xmath26 . afterwards , one should swap the states @xmath117 and @xmath118 and repeat the process . this results in the generation of a late photon , if the system was initially prepared in @xmath27 . to complete the encoding , the states @xmath117 and @xmath118 have to be swapped again . we now give two examples how to perform a photon pair measurement of the mutually unbiased basis ( [ xxx ] ) . the first method is suitable for polarization encoded photons , the second one for dual - rail encoded photons . if the qubits have initially been time bin - encoded , their encoding should be transformed first using standard linear optics techniques . _ polarization encoding . _ it is well known that sending two polarization encoded photons through the different input ports of a 50:50 beam splitter together with polarization sensitive measurements in the @xmath119-basis in the output ports would result in a measurement of the states @xmath120 , @xmath121 and @xmath122 . to measure the states ( [ twoproduct ] ) , we therefore propose to proceed as shown in figure [ moon3](a ) @xcite and to perform the mapping @xmath123 on the photon coming from source @xmath74 . using eq . ( [ angelencoding ] ) , we see that this corresponds to the single qubit rotations @xmath124 \ , , \nonumber \\ u_2 & = & { \textstyle { 1 \over \sqrt{2 } } } \ , \big [ \ , |{\sf h } \rangle \big ( \langle { \sf h}| + \langle { \sf v}| \big ) - { \rm i } \ , |{\sf v } \rangle \big ( \langle { \sf h}| - \langle { \sf v}| \big ) \ , \big ] \ , .\end{aligned}\ ] ] after leaving the beam splitter , the photons should be detected in the @xmath119-basis . a detection of two @xmath107 and two @xmath125 polarized photons indicates a measurement of @xmath126 and @xmath127 , respectively . finding two photons of different polarization in the same or in different detectors corresponds to a detection of @xmath128 or @xmath129 . _ dual - rail encoding . _ alternatively , one can redirect the generated photons to the different input ports of a @xmath130 bell multiport beam splitter as shown in figure [ moon3](b ) . if @xmath131 and @xmath132 denotes the creation operator for a photon in input and output port @xmath133 , respectively , the effect of the multiport can be summarized as @xcite @xmath134 with @xmath135 a bell multiport redirects each incoming photon with equal probability to any of the possible output ports , thereby erasing the which - way information of the incoming photons . one way to measure in the mutually unbiased basis ( [ xxx ] ) is to direct the @xmath136 photon from source 1 to input port 1 , the @xmath137 photon from source 1 to input port 3 and to direct the @xmath138 photon and the @xmath139 photon from source 2 to input port 2 and 4 , respectively . if @xmath140 denotes the state with no photons in the setup , this results in the conversion @xmath141 ) one can show that the network transfers the basis states ( [ xxx ] ) according to @xmath142 finally , detectors measure the presence of photons in each of the possible output ports . the detection of two photons in the same output port , namely in 1 or 3 and in 2 or 4 , corresponds to a measurement of the state @xmath126 and @xmath127 , respectively . the detection of a photon in ports 1 and 4 or in 2 and 3 indicates a measurement of the state @xmath79 , while a photon in the ports 1 and 2 or in 3 and 4 indicates the state @xmath80 . [ moon2 ] any unknown fixed ( or slowly varying with respect to the coherence length of the photon pulse ) phase factor introduced along the photon paths contributes at most to a global phase factor to the input state ( [ theencoding ] ) , which is also a feature of the schemes outlined in refs . the implementation of repeat - until - success quantum computing therefore does not require interferometric stability . it requires only overlapping of the photons within their coherence length within the linear optics setup . in this section , we discuss the possibility of implementing scalable quantum computation using the repeat - until - success quantum gate described in the previous sections . the implementation of this gate requires the generation of single photons on demand and linear optical elements together with absorbing quantum measurements . in the limit of perfect photon emission , collection , and detection efficiency , two - qubit cz gates can be performed deterministically , as described above . in real systems however , photon emission , collection , and detection is not perfect @xcite . in existing experiments , all of these processes have significant inefficiencies , which means that there is a finite probability that two photons will not be observed in the photon measurement . the failure to observe two photons in an attempted cz operation means that the static qubits are left in an unknown state , which constitutes a correlated two qubit error . if such losses are sufficiently small ( e.g. less than @xmath143 ) , the resulting gate failures can be dealt with using existing fault tolerance techniques @xcite . recently , much higher fault tolerance levels of up to 50% were found in linear optical quantum computing @xcite . more concretely , the highest reported photon detection efficiency for single photon detection with photon number resolution is about @xmath144 @xcite . a recent experiment by mckeever _ et al . _ @xcite involving an atom - cavity system for the generation of single photons yields a photon generation efficiency of nearly 70% , limited only by passive cavity loss . the lifetime of the atom in the cavity was @xmath145s , allowing for as many as @xmath146 photon generation events . moreover , legero _ et al_. @xcite demonstrated perfect time - resolved interference with two photons of different frequencies . time - resolved detection acts as a temporal filter to erase the which - way information which is important to any scheme involving photon interference . this suggests that strictly identical single photon sources are not required for attaining high fidelities in the state preparation . the cost of this high fidelity is a lower probability of success . fortunately , scalable quantum computing is possible , even in the presence of large errors , as long as no errors imply a very high fidelity and the occurrence of an error is _ heralded _ : if fewer than two photons are detected , we know that the attempted cz operation has failed . only when the detectors have a substantial amount of dark counts , we can not rely on this error detection mechanism . however , comercially available silicon avalanche photodetectors are available with a detection efficiency of 65% , and a dark count rate of @xmath147s@xmath148 @xcite . assuming , a photon regeneration rate of @xmath149s@xmath148 gives a clock time of @xmath150 s. the total dark count probability is then @xmath151 per clock cycle , which is small enough to be dealt with using existing error correction techniques . moreover , if one could experiment with detectors like the one reported by rosenberg _ et al_. @xcite , dark count rate effects would be negligible . in the case of an error , the state of the static qubits can be determined by subsequently performing measurements on the sources , which allows the sources to be re - prepared in a known state . in earlier work , we have shown that scalable quantum computation can be performed in the presence of significant heralded error rates , by first using a non - deterministic entangling operation to create _ cluster states _ of many qubits @xcite , and subsequently implementing scalable quantum computation via the ` one - way quantum computer ' @xcite . given cluster states of many qubits , the one - way quantum computer can be implemented by single qubit measurements alone . this technique permits fully scalable quantum computation , albeit with a fixed overhead per two qubit gate in the algorithm , which we calculate below . we briefly review how one way quantum computing can proceed within our scheme , and then provide an estimate of the overhead costs involved . one way quantum computation @xcite proceeds by first creating a graph state of many qubits , and subsequently performing single qubit measurements on the graph state @xcite . graph states may be represented as a graph comprising set of qubit ` nodes ' connected by ` edges ' which may be understood as ` bonds ' between the qubits . the quantum state corresponding to such a graph may be defined ( and also implemented ) by the following procedure : ( i ) prepare each qubit in the state @xmath152 ) between the relevant qubits . in this work , we will restrict our attention to the rectangular lattice graph states of the form shown in figure [ fig : cluster1 ] ( hereafter referred to simply as _ cluster states _ ) , which are sufficient for simulating arbitrary logic networks , and hence universal quantum computations @xcite . it is worth noting however that straightforward generalizations of the procedure described below allow us to scalably generate _ arbitrary _ graph states . this may be useful in that it might result in reduced costs for implementing certain algorithms . in these clusters , each horizontal row of physical qubits represents a single logical qubit in the logic network being simulated . two qubit operations are implemented by the vertical bonds acting between rows . we also permit bonds between non - adjacent rows , which permits highly non - local two qubit gates to be implemented . note also that the location of the qubits within the cluster is notional , and need not correspond to the physical location of the static qubit ( the mapping between the notional qubit positions within the cluster , and the actual physical location of the qubits can be stored in a classical computer ) . after making the state , quantum computation proceeds by performing a sequence of single qubit measurements on the static qubits , with each measurement performed in a particular basis so as to implement a given sequence of one- and two - qubit gates @xcite . at each time step , a whole column of physical qubits in the cluster is measured . the measurements are performed in order , starting with the column at left side of the cluster , and proceeding rightwards across the cluster . in general , the basis of the measurements made at a given time step will depend on the outcomes of earlier measurements . once a physical qubit has been measured , that qubit is disentangled from the cluster state and so may be re - initialized in a particular state and subsequently used later in the computation . we assume that single - shot single qubit measurements and single qubit unitary operations on the static qubits can be implemented using standard techniques . implementing one - way quantum computation in our scheme therefore reduces to the problem of scalably generating cluster states using the heralded , non - deterministic cz operation . we outline the general procedure here , and give a more detailed description in the subsequent section . in our scheme , cluster states can be generated by attempting to bond qubits using the non - deterministic cz operation . this operation has three possible outcomes : ` success ' , ` insurance ' , or ` failure ' . in the case of observing two photons , one of the gates of eq.([possibleoutcomesofcz ] ) is implemented , and subsequent application of appropriate single qubit unitaries implements either the cz operation ( denoting a ` success ' ) , or the identity operation ( denoting ` insurance ' ) . in the case of insurance , the cz operation can simply be reattempted . observing fewer than two photons denotes a failure . in this case , the static qubits are left in an unknown state . however , this damage can be repaired as follows . firstly , each of the two qubits involved in the failed gate can be measured in the computational basis to determine the nature of the error . if either qubit was already part of a cluster state , the bonds to its neighbors within the cluster are also destroyed . however , the remainder of the cluster state can be recovered by applying appropriate single qubit unitary operations to these neighboring qubits , conditional on the outcome of the measurement on the qubit involved in the failed cz gate . therefore , the cluster state can grow , shrink , or remain the same size , depending whether the cz operation was successful , failed , or failed with insurance . the key to scalably generating cluster states is to attempt cz operations between qubits in a sequence order such that the cluster state grows on average . we give such a sequence in sec . [ overhead costs ] . we conclude this section by noting that it is not necessary to build the whole cluster required for simulating a particular algorithm before commencing the single qubit measurement part of the computation . it is possible to build a partial cluster , and then to simultaneously perform single qubit measurements on one part of the cluster , while adding new qubits to another region in the cluster . in this approach to one - way quantum computing , one can think of the cluster as being split into three regions , as shown in figure [ fig : cluster2 ] . the _ active region _ , to the left of the cluster , contains the part of the cluster where the logic gate networks are being simulated via single qubit measurements . at the right of the cluster , the _ connection region _ comprises of several horizontal dangling linear chains which extend from the right edge of the main cluster , each corresponding to a logical qubit . in this region , nondeterministic cz operations are applied in order to add further cluster sections to the main section . these additional sections are manufactured separately , as described in sec . [ overhead costs ] . between the active region and the connection region , the _ buffer region _ comprises a quiescent region which suffices to protect the active region in the event of a long sequence of failed cz operations ; this would lead to the right edge of the cluster running back into the active region , damaging the logical computation . the depth of the buffer region should be chosen such that the probability of erasing a logical qubit is sufficiently small that it can be handled with existing fault tolerance techniques @xcite . there are several advantages to this approach . firstly , fewer physical qubits are needed , because qubits that have already been measured at the left edge of the cluster can be recycled and added to the right hand side of the cluster . secondly , preparing the whole cluster initially means that some of the qubits will spend a lot of time in an ` idle ' state before they are involved in the computation ; any errors accumulated in these idle qubits due to decoherence will degrade the fidelity of the computation @xcite . this is crucial if fault tolerant quantum computation is to be implemented within the cluster model , as such schemes require a source of fresh ancilla qubits throughout the algorithm . thirdly , the overhead costs for this approach can be reduced , because it is not necessary to prepare the _ whole _ cluster with a total success probability close to one ; the probability for erasing a given logical qubit need only be made smaller than the error threshold required for fault tolerance . a number of authors have considered efficient cluster state generation using non - deterministic , but heralded , entangling operations ( eos ) @xcite . references @xcite calculated explicit costs for making cluster states of optical qubits in the ideal case ( i.e. neglecting photon loss ) . subsequently , barrett and kok @xcite showed that , in the case of hybrid matter - optical systems ( such as those considered in this work ) , arbitrarily small eo success probabilities could be tolerated . they provided a ` divide and conquer ' algorithm for building linear clusters , which has moderate costs even for small success probability . an efficient algorithm for building two dimensional clusters , capable of simulating arbitrary logic networks was also given in @xcite . more recently , in ref . @xcite , a similar algorithm for building linear clusters was proposed , which made more use of recycling , and hence has a lower overhead cost . @xcite also gives an alternative algorithm for making 2-dimensional clusters , and explicitly calculates the associated overhead costs . in ref . @xcite , some elegant cost reducing improvements to the scheme proposed in ref . @xcite were suggested , utilizing the redundantly encoded qubits inherent in the original scheme . in this work , we will combine elements of the approaches taken in refs . @xcite to provide a simple upper bound for the scaling costs for building cluster states using our scheme . this estimate is based on an explicit procedure , and we do not claim that it is optimal ; an improved algorithm may yield substantially reduced costs . nevertheless , the procedure given here allows a straightforward calculation of the overhead costs . despite its apparent similarity to refs . @xcite , there is a crucial difference : in the scheme under consideration in this paper , there is the possibility of obtaining the ` insurance ' outcome . in general , this leads to a reduction in costs relative to schemes in which there is no insurance outcome . in the presence of imperfect photon emission , detection , and collection , the performance of the cz operation can be characterized by three probabilities : * the probability of successfully implementing the cz operation on the input qubits ( up to local operations ) , @xmath153 ; * the probability of obtaining the ` insurance ' outcome in which known local operations are applied to the qubits , @xmath154 ; * the probability of failure due to failing to emit , collect , or detect one or more photons during the remote gate operation , @xmath155 . these probabilities are determined by the physics of the sources and detectors . calculating the total cost of growing cluster states can be simplified by noting that , in the case of obtaining the ` insurance ' outcome , after applying the necessary single qubit corrections , one simply attempts the gate operation again . this process is repeated until a definite outcome ( success or failure ) is obtained . thus , we can define _ total _ success and failure probabilities , @xmath156 and @xmath157 , of the corresponding definite outcomes after an ( arbitrarily long ) sequence of ` insurance ' outcomes . these probabilities are given by @xmath158 and @xmath159 . the average number of attempted cz operations required before we obtain a definite outcome is @xmath160 . the overhead cost for making cluster states is then found using similar calculations to those presented in refs . we first calculate the cost ( i.e. the number of attempted cz operations per qubit in the final cluster ) of generating linear clusters . if a cz gate is repeatedly applied between the end qubits of two linear chains , each of length @xmath161 , either the gate is ( ultimately ) successful , in which case the total length of the new cluster is @xmath162 , or the gate ( ultimately ) fails , in which case , the length of the original clusters shrinks by one qubit each . repeatedly applying this procedure until a successful outcome is obtained ( or until both original clusters are destroyed ) @xcite gives the expected length @xmath163 . denoting the average number of attempts to create a chain of length @xmath161 by @xmath164 , we also have @xmath165 . solving these recursion relations gives a total cost @xmath166 where @xmath167 denotes the cost of growing a short cluster of length @xmath168 . note that for the average cluster length to grow on each round of the protocol , we require @xmath169 , which implies that the length of the short chains should satisfy @xmath170 . chains of fixed length @xmath168 can be grown independently using the probabilistic cz operation , by joining sub - chains together . growing these short chains adds a constant overhead cost to the cluster generation process . we use a ` divide and conquer ' approach to making these short chains @xcite , in which , on each round of the protocol , we attempt to join equal length pairs of linear clusters using the probabilistic cz operation . if we obtain the ` insurance ' outcome on any such attempt , we try the operation again , whereas if we fail , we assume ( for ease of calculation ) that the short chains are discarded . on the @xmath9th round of this protocol , the length of the chains is @xmath171 , and the number of attempted cz operations is given by the recursion relation @xmath172 . solving these relations gives @xmath173 combining eq . ( [ offlinecost ] ) and eq . ( [ totalcost ] ) , one can calculate the total cost of growing linear clusters for given values of @xmath155 , @xmath153 and @xmath154 . for instance , taking @xmath174 , @xmath175 , we require @xmath176 . taking @xmath177 , the total cost for making a linear cluster of length @xmath178 is found to be @xmath179 attempted cz operations . a moderate increase in success probability can dramatically decrease the cost : taking @xmath180 , @xmath181 , we require @xmath182 , and taking @xmath183 , we find the total cost to be @xmath184 . note that the negative constant term in these expressions is an artefact of joining small numbers of chains together to make an isolated chain of length @xmath178 . this is an ` edge effect ' which should be neglected when considering the asymptotic cost of making long chains . .the average number of entangling operations per vertical bond , given by @xmath185 . here @xmath153 , @xmath155 and @xmath154 are the success , failure and insurance probabilities , respectively . @xmath186 is the length of the chain that is needed to obey the growth requirement , and @xmath167 is the number of eos needed to achieve this length . @xmath187 is the average cluster chain consumed by the forging of a vertical bond , and @xmath188 is the number of eos needed to achieve this length . [ cols="<,<,<,^,>,<,^,>",options="header " , ] linear clusters are not sufficient for simulating arbitrary logic networks @xcite , and therefore it is necessary to generate more general graph states . a variety of techniques for making such states using probabilistic entangling operations have been proposed , which include linking linear clusters using independently prepared ` i ' shaped clusters @xcite , using micro - clusters @xcite , using redundantly encoded qubits @xcite , or by making use of ` + ' shaped clusters @xcite . here , we propose a relatively efficient method for creating vertical bonds between linear cluster chains . we employ a technique based on that introduced by browne and rudolph @xcite , which involves four steps as shown in figure [ fig : cluster3 ] . 1 . first , we assume that we have sufficiently long linear cluster chains . these can be produced efficiently in the manner outlined above . in order to establish the amount of resources needed to create a vertical bond , we will count the number of qubits that are utilized on average in this process , as well as the average number of entangling operations . secondly , we identify the two qubits that we wish to entangle with a vertical bond ( in figure [ fig : cluster3 ] the two left - most qubits ) . the qubits directly on the right of these qubits are then measured in the @xmath189 basis . a hadamard operation on the third qubit in each chain returns the overal state to a graph state . 3 . this will result dangling bonds , or _ cherries _ @xcite hanging from the two qubits that are to be connected . this is a form of redundant encoding , and it allows us to apply the entangling operation to the two cherries . in case of a failure , the entangling operation will _ not _ break the linear cluster chains . it will destroy only the cherries and as a result both chains are shortened by two qubits . steps ( b ) and ( c ) can then be repeated . when the entangling operation succeeds , we have forged a vertical bond between the two qubits chosen in step a ) . the vertical link is itself a chain of two qubits . these are typically not wanted , so we can remove one of them with a @xmath189 measurement , creating another cherry in the other qubit in the chain . this redundancy can be pruned , but may also be useful for creating additional bonds , or may even be useful for error correction . we will now estimate the cost of this procedure . since two qubits are burnt in each step , and we need to repeat the process @xmath190 times , the average length of each chain that is consumed in the bonding process is @xmath191 where the extra @xmath192 counts the qubits that will establish the vertical link . the number of entangling operations needed to make a vertical bond is then @xmath193 where the extra @xmath190 takes into account the number of entangling operations that are needed to link the cherries together into a vertical bond . in table [ tab : cost ] , we calculated the number of entangling operations that are needed to forge a vertical bond , given several specific values for the success , failure , and insurance probabilities . we analysed a hybrid architecture for quantum computing using stationary and flying qubits , which is based on our earlier work @xcite , in detail . it was shown that this new approach solves some of the most pressing problems that arise in non - hybrid architectures . our system is scalable , even with non - ideal components , and more importantly , it uses no direct qubit - qubit interactions . this means that the qubits will be subject to less decoherence and fewer control errors . when realistic photo - detectors are used , photon loss will affect only the efficiency of the scheme . furthermore , our system relies on components that have been demonstrated in experiment , and is largely implementation independent . despite the no - go theorem for optical bell - state measurements , it is in principle possible to implement a deterministic gate between distant qubits . however , when losses are taken into account , the gate becomes necessarily probabilistic . in order to achieve robustness against general decoherence and to guarantee high fidelities , we showed how to construct cluster- or graph - states using the two - qubit gate . our entangling operation , which produces the bonds in the graph states , is not limited to physically adjacent matter qubits . as a consequence , no extensive swapping operations need to be taken into account in the production of nontrivial graph states . this architecture for quantum computation is inherently distributed , and hence can be used for integrated quantum computation and communication purposes . + _ acknowledgment . _ yll , ab and lck thank h.j . briegel , a. browaeys , d.e . browne , t. durt , a.k . ekert , p. grangier , m. jones , and p.l . knight for stimulating discussions . pk and sdb thank s. benjamin , j. eisert , b. lovett , m.a . nielsen , t. stace for fruitful discussions and particularly d. browne for giving them a deeper insight into cluster state generation . yll acknowledges funding from the dso national laboratories in singapore and ab acknowledges support from the royal society and the gchq . this work was supported in part by the european union ( ramboq , qgates ) and the uk engineering and physical sciences research council ( qip irc ) . o. benson , c. santori , m. pelton , and y. yamamoto , phys . * 84 * , 2513 ( 2000 ) ; m. pelton , c. santori , j. vuckovic , b. zhang , g.s . solomon , j. plant , and y. yamamoto , phys . lett * 89 * , 233602 ( 2002 ) .

to capture important physical properties of a spacetime we construct a new diagram , the card diagram , which accurately draws generalized weyl spacetimes in arbitrary dimensions by encoding their global spacetime structure , singularities , horizons , and some aspects of causal structure including null infinity . card diagrams draw only non - trivial directions providing a clearer picture of the geometric features of spacetimes as compared to penrose diagrams , and can change continuously as a function of the geometric parameters . one of our main results is to describe how weyl rods are traversable horizons and the entirety of the spacetime can be mapped out . we review weyl techniques and as examples we systematically discuss properties of a variety of solutions including kerr - newman black holes , black rings , expanding bubbles , and recent spacelike - brane solutions . families of solutions will share qualitatively similar cards . in addition we show how card diagrams not only capture information about a geometry but also its analytic continuations by providing a geometric picture of analytic continuation . weyl techniques are generalized to higher dimensional charged solutions and applied to generate perturbations of bubble and s - brane solutions by israel - khan rods . this paper is a condensed and simplified presentation of the card diagrams in hep - th/0409070 . + .5 in * weyl card diagrams * 0.3 in gregory c. jones and john e. wang 0.15 in _ department of physics , harvard university , cambridge , ma 02138 _ + _ department of physics , national taiwan university + taipei 106 , taiwan _ + [ 0.3 in ] .3 in

hep-th0506023

spacetimes are geometrical objects , independent of the coordinates with which we describe them . however , spacetimes are typically presented and visualized in a specific coordinate system . if the coordinates are poorly chosen , many properties of the spacetime such as horizons , causally connected spacetime points , maximal extensions and null infinity are not readily apparent . a simplification occurs if a @xmath4 dimensional lorentzian spacetime has enough fibered directions ( like a @xmath5-sphere ) or other ignorable directions . one can draw two dimensional diagrams for the remaining directions and such lorentzian @xmath6 signature spacetime slices can be conformally compactified leading to penrose diagrams . penrose diagrams are quite useful in understanding spacetime geometry and successful especially in understanding causal structure although there are some limitations to this approach . for instance just knowing the penrose diagram for the subextremal @xmath7 reissner - nordstrm black hole does not tell us what happens to the spacetime structure in the chargeless or extremal limits . for more complicated spacetimes , penrose diagrams ( which assume symmetry or fibering ) can only draw a slice of the spacetime . as a known example , the penrose diagram for a kerr black hole does not clearly depict the ring singularity and the possibility of crossing through the interior of the ring into a second universe . in addition , recently analytic continuation has been applied to black hole solutions to yield bubble - type @xcite or s - brane @xcite solutions . oftentimes this is done in boyer - lindquist type coordinates which are hard to visualize . again we are not left with a clear picture of the resulting spacetime and the penrose diagrams are missing important noncompact spatial directions . it is useful to have an alternative diagram which can also capture important features of a spacetime . for this reason in this paper we expand the notion of drawing spacetimes in weyl space @xcite . because our diagrams have the appearance of playing cards glued together we will dub them weyl card diagrams . to understand the construction of a card diagram we recall that in @xmath8 dimensions a weyl solution in canonical coordinates @xcite is written as @xmath9\ ] ] where @xmath10 and @xmath11 are functions of @xmath12 . the original weyl class requires two commuting orthogonal killing fields @xmath13 in four dimensions @xcite , or @xmath14 fields for general @xmath4 dimensions @xcite . are discussed in @xcite . ] sometimes weyl solutions are called axially - symmetric gravitational solutions although they in fact are more general . we also include the weyl - papapetrou class for 2 commuting killing vectors in @xmath8 @xcite , and allow charged static solutions in @xmath15 ( see the appendix to this paper ) . furthermore stationary vacuum solutions in @xmath15 are covered with the recent work of @xcite . in four and five dimensions this generalized weyl class includes spinning charged black holes and rings @xcite as well as various arrays @xcite of black holes , spacelike - branes , and includes backgrounds like melvin fluxbranes @xcite and spinning ergotubes @xcite . when constructing card diagrams , we will draw only weyl s canonical coordinates @xmath16 , or coordinates related to them via a conformal transformation . the killing coordinates are ignorable and so this diagram is efficient and will show all details of the spacetime . since there are only two nontrivial coordinates , card diagrams are two dimensional and easy to draw like penrose diagrams . the difference however is that while penrose diagrams are truly two - dimensional , card diagrams are drawn as if embedded in three dimensions . when a @xmath16 region of the spacetime has euclidean @xmath17 signature , we draw the two coordinates @xmath16 horizontally ; and this makes a horizontal card . for lorentzian signature @xmath6 regions we use @xmath18 or @xmath19 , and draw the timelike coordinate vertically ; this makes a vertical card . causal structure is automatically built into the vertical cards since for example the directions @xmath19 appear conformally only through the combination @xmath20 . horizontal cards and vertical cards are attached together at killing horizons and so card diagrams resemble a gluing - together of a house of playing cards . in this paper we will present card diagrams for the familiar spacetimes of black holes , as well as expanding bubbles , s - branes , and black rings . many other spacetimes including the s - dihole , infinite periodic universe , c - metric , and multiple - rod solutions in 4 and 5 dimensions are presented in @xcite , and spacetimes derived from 4 and 5 dimensional kerr geometries will be presented in @xcite . in section 2 we review the schwarzschild black hole in the usual coordinates and in weyl coordinates . by extending through the horizon and properly representing the interior of the black hole we construct the first card diagram . we emphasize the construction of the interior of the black hole as a vertical card comprised of four triangles unfolded across special null lines . we then discuss general card diagram properties such as null lines , list the available card types , and geometrically describe the @xmath11-flip analytic continuation procedure . in section 3 , we discuss the sub / super / extremal reissner - nordstrm black hole card diagrams , the kerr black hole , and the black ring / c - metric card diagrams . we then show that a spacetime can have multiple card diagrams and as examples present the elliptic , hyperbolic , and parabolic representations of the charged witten bubble and charged spacelike brane which we also call s - reissner - nordstrm . finally , as a newer example we discuss the ( twisted ) s - kerr solution @xcite . we conclude with a discussion in section 4 . we give an appendix on perturbing witten bubbles and s - branes by introducing israel - khan rods , in their hyperbolic or elliptic representations . we also give an appendix on how the higher dimensional vacuum weyl ansatz can be extended to include electromagnetic fields . this paper is a condensed presentation of the card diagrams in @xcite . in this section we review the schwarzschild black hole as an example of a weyl spacetime and then we explain the construction of its associated weyl card diagram . general features and properties of card diagrams are also developed . up to now if a solution such as the schwarzschild black hole had horizons , then only the regions outside the horizons have been drawn in weyl coordinates @xcite . to go through a nonextremal horizon , the weyl coordinate @xmath21 must be allowed to take imaginary values . we discuss how the horizon can be represented as a junction of four regions which we call four cards . the regions outside the horizon will be drawn as two horizontal cards while the regions between the horizon and the singularity will be drawn as two vertical cards . the interior vertical cards naively are problematic and have fourfold - covered triangles bounded by ` special null lines ' . however the triangles can be unfolded and glued together into a square along the ` special null lines ' to achieve a singly covered representation of the spacetime in weyl coordinates by properly choosing branches of a square root in the solution . note that card diagrams represent the spaces on which we solve the laplace equation ( horizontal card ) or wave equation ( vertical card ) to find a weyl metric . for example , the schwarzschild black hole has a finite length uniform density rod source along the @xmath22-axis generating the potential @xmath23 , and the remainder of its @xmath22-axis encodes the vanishing of the @xmath24-circle . thus card diagrams give a full account of the boundary conditions necessary to specify the spacetime . this section will describe the construction of the schwarzschild black hole card diagram . the penrose and card diagrams are compared in fig . [ penrose - weyl - comparison ] . = 6.5 in the schwarzschild metric in spherically symmetric ( schwarzschild ) coordinates is @xmath25 there is a horizon at @xmath26 and a curvature singularity at @xmath27 . on the other hand , in weyl s canonical coordinates the metric ansatz is @xcite it is @xmath28 where @xmath10 and @xmath11 are functions of the coordinates @xmath21 and @xmath22 : @xmath29 previously attention was restricted to the h1 half - plane @xmath30 , @xmath31 , known as weyl space , which describes the exterior of the black hole and whose horizon is represented by a ` rod ' line segment @xmath32 , @xmath33 ; see fig . [ weylrod ] . note that the non - killing 2-metric is conformal to the euclidean flat space @xmath34 . the coordinate transformation between schwarzschild and weyl coordinates is @xmath35 now we wish to ask how weyl s coordinates draw the spacetime inside the horizon . the schwarzschild coordinates ( [ coordinatetrans ] ) tell us that for @xmath36 , @xmath21 is imaginary and so we set @xmath37 . ( in general we must perform an analytic continuation of weyl coordinates to go through a horizon which are at the zeros of the weyl functions @xmath38 . ) the analytic continuation gives a region with a conformally minkowskian metric @xmath39 and we will draw this region as being vertical and perpendicularly attached to the horizontal card at the horizon @xmath33 . the vertical direction is always timelike in card diagrams . of course the schwarzschild horizon structure is more complicated than just having two cards joined together . for example we know the penrose diagram in the @xmath40 coordinates is divided into four regions that meet in a @xmath41-horizon structure . in weyl coordinates , we already saw that we can extend @xmath42 or go to negative @xmath21 . this gives us the four regions of the schwarzschild black hole in weyl coordinates . two regions will be horizontal at real values of @xmath21 and two regions will be vertical with imaginary values of @xmath21 . so in addition to the first horizontal card in front of the horizon , we also have a copy of the horizontal external universe behind the horizon and attach two vertical cards , one above and one mirror copy below the horizontal cards ( see fig . [ penrose - weyl - comparison ] ) . all together , four different regions attach together at the same @xmath33 rod horizon . the four regions labelled h1 , h2 , v1 and v2 in the penrose diagram map to the similarly labelled four regions on the card diagram in fig . [ penrose - weyl - comparison ] . note that the weyl cards draw the @xmath43 coordinates which is different from the @xmath44 coordinates of the penrose diagram . however the fact that the radial coordinate @xmath45 describes four distinct regions , two where @xmath46 is spacelike and two where it is timelike , is still apparent in the weyl card diagram . so while a penrose diagram always has a lorentzian @xmath6 signature , a card diagram will flip from being euclidean @xmath17 to lorentzian @xmath6 across a nonextremal killing horizon . let us now examine the construction of the upper vertical card extended in the @xmath47 directions . looking at an @xmath45-orbit for @xmath48 $ ] on the vertical card , we see that in weyl coordinates @xmath49 . the bounding lines where @xmath50 , we will call ` special null lines , ' and they are a general feature of vertical cards with focal points ( the rod endpoints @xmath51 ) . here , special null lines are the envelope of the @xmath45-orbits as we vary @xmath52 . inside the horizon the schwarzschild coordinates apparently fill out a 45 - 45 - 90 degree weyl triangle with hypotenuse length @xmath53 a total of four times , as shown in figure [ 4xcoverfig ] . special null lines play an important role in weyl card diagrams so let us explain their significance . keep in mind that we have already broken the manifest spherical symmetry when we have written the schwarzschild metric in weyl coordinates , so the existence of preferred special null lines is relative to this chosen axis . consider the two 3-surfaces [ blr ] r_=r - mm = 0,which are drawn in fig . [ cardioids ] . these surfaces bound the trajectories of light rays that do not move in the killing directions . the surfaces intersect at @xmath54 and partition the black hole interior into four subregions . these regions correspond to the four weyl triangles . and height @xmath55 . as we illustrate , the triangle interior is covered four times by orbits of @xmath45 at four different values of @xmath52.,width=302 ] into four subregions.,width=302 ] it is clear from ( [ blr ] ) that @xmath56 is positive outside the horizon and there is no difficulty going to negative values inside the horizon . on the other hand in terms of weyl coordinates , the functions @xmath57 are the square root of a positive number when @xmath58 and imaginary if @xmath59 . clearly , instead of dealing with imaginary values of @xmath60 , the way to go ` beyond ' the special null line @xmath61 is to keep @xmath58 but use the other square root branch for @xmath60 as it enters in the weyl functions @xmath10 and @xmath11 by explicitly replacing @xmath62 . since we can pass @xmath63 and/or pass @xmath64 , it is clear that the four different branches of the square root functions differentiate the four copies of the weyl triangle . from ( [ schwarzschildweylrep ] ) , passing each @xmath61 null line changes the sign of @xmath65 and hence exchanges the timelike and spacelike nature of @xmath66 and @xmath22 . since the vertical direction on a vertical card always represents time , two of the triangles are drawn turned on their sides , their hypotenuses vertical and so timelike . the four triangles describing the interior of the black hole @xmath67 glue together along the special null lines to fit neatly into a square ; see fig . [ buildcard6 ] . because of the unfolding of the triangles , the positive @xmath22-direction on the top triangle ( and any attached horizontal cards ) points in the opposite direction compared to that on the original horizontal card . the upper vertical cards is thus a square of length @xmath53 . the bottom of the upper card v1 is the black hole horizon which connects to three other cards in a four card junction . the right and left edges of this vertical card correspond to @xmath68 and are the boundaries where @xmath69 the @xmath24-circle vanishes . the top edge of the card represents the @xmath27 curvature singularity . the second vertical card v2 is built in analogous fashion except the square is built in a downwards fashion towards negative values of @xmath66 . additionally there is a second horizontal card plane , h2 , identical to h1 at negative values of the @xmath21 coordinate attached to the same horizon along @xmath70 $ ] on the @xmath22-axis . one typically stops the construction of the schwarzschild spacetime with the above four regions , and considers the @xmath71 part of the metric to be a separate spacetime . however for reasons which become clear when we look at the reissner - nordstrm and kerr black holes in sections [ rnsubsub ] and [ kerrsubsec ] , we continue the card diagram past the singularity and attach two horizontal half - plane corresponding to negative - mass ( or @xmath71 ) universes h3 & h4 , and further vertical card above the singularity which is identical to v2 . the two new horizontal cards each represent negative mass - universes with no horizon and a naked singularity along @xmath33 . note that although penrose diagrams for the @xmath72 and @xmath71 regions of schwarzschild can not attach together since the singularity is spacelike in one region and timelike in another , cards can naturally attach at this singularity . the extended card diagram for schwarzschild , shown in fig . [ fullschwarzcardfig ] , is an infinite array of repeated cards representing positive ( h1,h2 ) and negative ( h3,h4 ) mass universes and inside - horizon regions . having described the construction of the card diagram for schwarzschild we now turn to general remarks about these card diagrams . card diagrams can be constructed in general d - dimensions starting from a generalized weyl ansatz including the charged cases discussed in appendix [ eweylapp ] . horizontal cards are conformally euclidean and represent stationary regions . vertical cards are always conformally minkowskian and represent regions with ( @xmath14 ) spacelike killing fields . on a vertical card time is always in the vertical direction and a spacetime point s causal future lies between @xmath73 null lines on the card . weyl s coordinates certainly go bad at horizons , so these diagrams are not a full replacement for penrose diagrams at understanding causal structure or particle trajectories . however , it is clear for example from fig . [ penrose - weyl - comparison ] that two vertical and two horizontal cards attach together in a orthogonal @xmath74-configuration in precisely the sense of the @xmath41-horizon structure of the penrose diagram . the prototypical horizon is that of the two rindler and two milne wedges of flat space ; this spacetime has two horizontal half - plane cards and two vertical half plane cards that meet along the horizon being the whole @xmath22-axis . zooming in on a non - extremal horizon of any card diagram yields this rindler / milne picture . a rod endpoint @xmath75 such as @xmath51 for schwarzschild is a ` focus ' for the weyl diagram and often represents the edge of a black hole horizon or the end of an acceleration horizon on a horizontal card . generally , multi - black hole weyl spacetimes depend only on distances to foci , such as @xmath60 in the case of schwarzschild @xcite . to understand that in general it is natural to change the branch of the distance functions , @xmath76 , when crossing the special null line that emanates from the foci at @xmath77 , take some weyl spacetime and imagine moving upwards in a vertical card to meet the special null line where @xmath78 by increasing time @xmath79 for fixed spatial @xmath22 . rearranging @xmath80 as the semi - ellipse @xmath81 we see that a smooth traversal of this semi - ellipse across @xmath78 requires a change in the sign of @xmath76 . in many of our solutions , special null lines are used to reflect vertical triangular cards to create full , rectangular cards . however , in the bonnor - transformed s - dihole geometry of @xcite as well as double killing rotated extremal geometries and parabolic representations of the bubble and s - schwarzschild in sec . [ thirdcard - poincare ] , the special null lines will serve as conformal boundaries at null infinity @xmath82 . boundaries of cards indicate where the metric coefficient along a killing ( circle ) direction vanishes . which circle vanishes is constant over a connected piece of the boundary , even when the boundary turns a right angle onto a vertical card . furthermore the periodicity to eliminate conical singularities is constant along connected parts of the boundary . for schwarzschild , the @xmath24-circle vanishes on both connected boundaries and has periodicity @xmath83 . although not a full replacement for understanding causal structure , it is interesting to consider geodesic trajectories on card diagrams . for example when a light ray is incident from a horizontal card onto a horizon ( to enter the upper vertical card ) , it must turn and meet that horizon perpendicularly . it then appears on the vertical card , again perpendicular to the horizon . only those light rays which go from the lower vertical card to the upper vertical card directly can meet the horizon rod at a non - right angle ; these rays would touch the vertex of the @xmath41 in a schwarzschild penrose diagram . when a light ray on a vertical card hits a boundary where a spacelike circle vanishes , it bounces back at the same angle as drawn on the card relative to the perpendicular . spacetimes with a symmetry group larger than the minimal weyl symmetry can have more than one card diagram representation . multiple diagrams exist when there is more than one equivalent way to choose @xmath5 killing congruences on the spacetime manifold . examples we explicitly discuss in sec . [ bubble - sbrane - subsec ] are the 4d witten bubble and the 4d s - reissner - nordstrm ( s - rn ) which have three card diagrams corresponding to the three types of killing congruences on ds@xmath84 and @xmath85 . these different card diagrams are associated for example with global , patched , and poincar coordinates for ds@xmath84 and the different representations will have different applications and reveal different information . on the other hand the s - kerr solution , whose card diagram is discussed in sec . [ kerrsubsec ] , has symmetry group @xmath86 and its unique card representation looks like the ` elliptic ' representation of s - rn . all spacetimes , new and old , in this paper are built from the following card types . horizontal cards are always half - planes . they may however have one or more branch cuts which may be taken to run perpendicular to the @xmath22-axis . undoing one branch cut leads to a horizontal strip with two boundaries ; multiple branch cuts lead to some open subset , with boundary , of a riemann surface . vertical cards may be noncompact : full planes with or without a pair of special null lines ; half - planes with vertical or horizontal ( horizon ) boundary ; or quarter planes at any @xmath73 orientation . vertical cards may also be compact : squares with a pair of special null lines ; or 45 - 45 - 90 triangles at any @xmath73 orientation . all horizontal and vertical boundaries represent where a killing circle vanishes and hence the end of the card . all @xmath73 null boundaries represent instances of @xmath87 . it is satisfying that for a variety of spacetimes including those in @xcite , the cards are always of the above rigid types . there is one basic procedure which can be performed on vertical cards and their corresponding weyl solutions . it is the analytic continuation @xmath88 , which is allowed since @xmath11 is determined by first order pdes and @xmath65 is real of either sign on vertical cards . this continuation is equivalent to multiplying the metric by a minus sign and then analytically continuing the @xmath14 killing directions . for charged generalizations of weyl solutions ( see appendix b ) , this procedure does not affect the reality of the 1-form gauge field . we call this analytic continuation a @xmath11-flip since the way it acts on a card is to geometrically flip it about a @xmath73 null line ( for example , look at the vertical cards in figs . [ s - rn-3 ] and [ bubblecard3 ] ) . in this section we construct the card diagrams for a wide assortment of solutions including black holes , bubbles and s - branes . the card diagrams are shown to be useful in representing continuous changes in the global spacetime structure such as how reissner - nordstrm black holes change as we take their chargeless and extremal limits . for the superextremal black holes we discuss how to deal with branch points and cuts on horizontal cards . the card diagram also clearly represents the kerr ring singularity and how traversing the interior of the ring leads to a second asymptotic spacetime . the 5d black ring solution , associated c - metric type solutions and twisted s - branes are also discussed . furthermore analytic continuation has an interesting interpretation in terms of card diagrams . we will describe the effect of analytic continuation on the card diagrams by examining two known analytic continuations of the reissner - nordstrm black hole , the charged bubble and the s0-brane which we also call s - rn . these time dependent spacetimes each have three card diagram representation and two are obtained via different analytic continuations in weyl coordinates . the witten bubble and s - rn are related to each other by what we call a @xmath11-flip which is a geometric realization of analytic continuation . in the usual coordinates the reissner - nordstrm black hole takes the form @xmath89 using the coordinate transformation @xmath90 we find the weyl form of reissner - nordstrm black hole @xmath91 and the card diagram for @xmath7 is shown in fig . [ rn - card ] . the construction of the card diagram proceeds along similar lines to the schwarzschild card diagram . there are two adjacent horizontal half - planes , h1 and h2 , which represent the positive mass asymptotically flat regions . the outer horizon at @xmath92 is represented in weyl space as a rod which lies along the @xmath22-axis for @xmath93 . the vertical cards , v1 and v2 , are squares of length @xmath94 and the diagonal lines connecting opposite corners of the square are special null lines . the top of v2 is the @xmath95 rod which is a four - card inner horizon . the black hole singularity no longer is on the edge of v2 but instead is on the outer boundary of the horizontal h1 and h2 regions , at @xmath96 . now the singularity is timelike and avoidable from the view of an observer on a vertical card . the rest of those horizontal cards , regions h3 and h4 , are @xmath71 or equivalently @xmath97 nakedly singular rn spacetimes . at each horizon , the card diagram is continued vertically to obtain an infinite tower of cards . in fig . [ rn - penrose ] we show the penrose diagram for comparison . although the chargeless @xmath98 limit is hard to understand from penrose diagrams , it is easy to understand using the card diagram in fig . [ rn - card ] ; the vertical card expands to a @xmath99 square and the singularity degenerates to a line segment coinciding with the inner horizon . regions h1 and h2 disappear so the singularity is now ` visible ' from v1 and v2 as well as h3 and h4 ; the singularity is spacelike relative to vertical cards and timelike for horizontal card observers . this achieves the schwarzschild infinite array of cards in fig . [ fullschwarzcardfig ] . starting from the above card diagram we now examine the extremal limit @xmath101 . in this case the vertical cards which represent the regions between the two horizons get smaller and disappear . when @xmath102 , the horizontal cards are now only attached at point - like extremal - horizons and only half of the horizontal cards remain connected , see fig . [ rnextremalfig ] . the region near the point - horizons are anti - de sitter throats although cards themselves can not adequately depict the throat region . the throat is a ` connected ' sequence of points on vertically adjacent horizontal cards . to understand how to go across the horizon , it is important to remember that there are special null lines in @xmath103 . in the extremal case the focal distances are equal ( @xmath104 ) and the null lines @xmath61 degenerate to the origin . so when we pass through the throat to an adjacent @xmath105 horizontal card , the sign changes for all occurrences of @xmath106 . half the cards are now absent relative to the subextremal case since we no longer perform analytic continuation to get the vertical card . we can only go from positive real values of @xmath21 to negative values of @xmath21 alternatively . the singularity appears as a semicircle on the @xmath105 cards . for axisymmetric majumdar - papapetrou solutions , this ` sign change rule ' agrees with that in @xcite . our analysis also applies to non - mp axisymmetric solutions such as the dihole @xcite . the superextremal @xmath107 reissner - nordstrm black hole does not have horizons or vertical cards . its card diagram consists of two horizontal cards , connected along the branch cut @xmath108 , @xmath109 . one card has a semi - ellipse singularity passing through the points @xmath110 and @xmath111 ( see fig . [ superrnfig](a ) ) . these two horizontal cards are connected in the same sense as a branched riemann sheet . by choosing weyl s canonical coordinates ( meaning @xmath112 with @xmath113coef@xmath114coef@xmath115re@xmath116 ) , the solution is no longer accurately represented on the horizontal card . this can be seen by examining the coordinate transformation ( [ rnsoln ] ) from schwarzschild coordinates to weyl coordinates . for fixed @xmath45 and varying @xmath52 , the coordinates from @xmath117 cover the weyl plane in semi - ellipses which degenerate to the segment @xmath118 , which serves as a branch cut ; and @xmath119 is the branch point . the range @xmath120 again covers the half - plane with @xmath27 forming an ellipse singularity . crossing the branch cut means choosing the opposite signs for @xmath121 , and indeed we can think of the superextremal ` rod ' as being complex - perpendicular to the weyl @xmath122-plane . this double cover of the weyl plane can be fixed by taking a holomorphic square root ; this preserves the conformally euclidean character of the card diagram . by choosing a new coordinate @xmath123 , we map both the positive and negative - mass universes into the region @xmath124 ( see fig . [ superrnfig](b ) ) . the image of the @xmath22-axis boundary is a hyperbola where the @xmath24-circle vanishes . the origin @xmath125 is the image of the branch point and the image of the line segment @xmath126 is a line connecting the two hyperbolas and intersecting the origin of the @xmath127-plane . the singular nature of @xmath128 has been fixed by @xmath129 . finally the ` black hole ' singularity is mapped to a curved segment stretching from one hyperbola line to the other , to the left of the branch cut . alternatively one can use a schwarz - christoffel transformation to map the two universes onto the strip @xmath130 . this is useful when the horizontal card boundaries are horizons as it allows for multiple horizontal cards to be placed adjacent to each other at the horizons ( fig . [ srn2fig ] ) . this technique of fixing a horizontal card with a branch point will also be used in more complicated geometries such as the hyperbolic representation of s - rn and multi - rod solutions in four and five dimensions @xcite . written in weyl - papapetrou form , the kerr black hole is @xmath131 where @xmath132 . the transformation to boyer - lindquist coordinates is @xmath133 , @xmath134 . for @xmath135 the kerr black hole card diagram ( see figure [ kerrcard ] ) is similar to reissner - nordstrm except that the singularity is now a point and lies at @xmath136 , @xmath109 on each horizontal negative - mass card . the outer and inner ergospheres lie on the positive - and negative - mass cards and are both described by the curve @xmath137 which intersects the rod endpoints at @xmath138 . the boundary of the region with closed timelike curves is also described by a quartic polynomial in weyl coordinates . once again the vertical cards have two special null lines where @xmath60 change sign . the @xmath27 surface in bl coordinates is a semi - ellipse @xmath139 on the negative - mass card ; but it is not a distinguished locus on the card diagram . attempting to make one loop around the ring in the kerr - schild picture clearly does not make a closed loop in weyl space , whereas two loops in the kerr - schild picture will form a single loop around the ring singularity on the card diagram . it is also clear that it is possible to find classical trajectories which avoid the singularity and which safely escape into a second asymptotically flat region . a card diagram for a charged kerr - newman solution can similarly be constructed , with @xmath140 . the extremal kerr(-newman ) solution has a card diagram like fig . [ rnextremalfig ] , but the ring singularity is just a point at @xmath109 and @xmath141 on negative - mass cards . again , @xmath104 and the special null lines degenerate to the origin ; crossing the origin ( which is a twisted ads - type throat ) entails changing the sign of @xmath142 . the superextremal kerr(-newman ) solution is similar to the superextremal rn ( fig . [ superrnfig](b ) ) except that the curved - segment singularity is replaced by a point ( @xmath109 ) , and the ergospheres map to an @xmath143-looking locus centered at @xmath125 . = 3.5 in the 5d static black ring solution of @xcite is @xmath144\nonumber\end{aligned}\ ] ] where @xmath145 , @xmath146 , and @xmath147 . the coordinates @xmath148 , @xmath149 are 4-focus ( including @xmath143 ) or c - metric @xcite coordinates that parametrize a half - plane of weyl space @xmath150 , @xmath31 : @xmath151 the foci are on the @xmath22-axis at @xmath152 and @xmath153 . the black ring horizon is also on the @xmath22-axis along @xmath154 . the @xmath24-circle vanishes along @xmath155 and @xmath156 , and the @xmath157-circle vanishes along @xmath158 . curves of constant @xmath149 degenerate to the horizon line segment as @xmath159 , and degenerate to the @xmath160 ray ( better pictured with a conformally equivalent disk ) for @xmath161 . curves of constant @xmath148 degenerate to the vanishing @xmath24-circle line segment for @xmath162 and to the ray @xmath163 for @xmath164 . the card diagram is easy to construct and is not much different from the four dimensional schwarzschild case . past @xmath165 we can go to @xmath166 and hence imaginary @xmath167 , and move up a @xmath168 square with two special null lines . at the top of the square , at @xmath169 we have the curvature singularity . continuing again to real @xmath21 and running @xmath149 down to @xmath170 , we map out a ( negative - mass ) horizontal card . the locus @xmath171 is the ray @xmath172 . the space closes off here as the @xmath157-circle vanishes , but we formally continue to illustrate how c - metric coordinates run on noncompact vertical cards this is useful in several applications , such as the plebaski - demiaski solution @xcite . past @xmath171 , we see that for fixed @xmath148 , reducing @xmath149 down to @xmath148 makes a topological half - line in a vertical card with a special null line . then for @xmath173 , we traverse another vertical card , which we could attach to our original positive - mass horizontal card along @xmath174 ( see fig . [ blackringfig ] ) . = 3.7 in note that when passing through the black ring horizon at @xmath165 , the weyl conformal factor @xcite @xmath175 stays real ; @xmath176 and @xmath177 go negative . as we pass the special null lines , explicit appearances of @xmath178 and @xmath179 in the weyl functions @xmath180 , @xmath181 change sign . the charged ring of @xcite is generated by a functional transformation and hence inherits a card diagram structure . in fact , any geometry with @xmath14 killing directions written in c - metric coordinates has a card diagram . the schwarzschild black hole can be analytically continued to two different time - dependent geometries , the witten bubble of nothing @xcite with a ds@xmath84 element and s - schwarzschild @xcite with an @xmath85 element , and it is instructive to understand how the card diagram changes . unlike black hole geometries with their unique card diagrams , these time dependent geometries can have multiple card diagram representations . both the bubble and s - schwarzschild have three different card diagram representations corresponding to three different ways to select killing congruences . these three types of killing congruences can be understood by representing @xmath85 as the unit disk ( with its conformal infinity being the unit circle ) . the orientation - preserving isometries of @xmath85 are those mbius transformations preserving the disk , @xmath182 @xcite . mbius transformations @xmath183 have two complex fixed points , counted according to multiplicity . in the upper half - plane @xmath184 representation , @xmath185 , @xmath186 , @xmath187 , and @xmath188 are real , so the fixed points are roots of a real quadratic . hence they may be ( i ) distinct on the real boundary ( hyperbolic ) , ( ii ) degenerate on the real boundary ( parabolic ) , or ( iii ) nonreal complex conjugate pairs , one interior to the upper half - plane @xmath85 ( elliptic ) . prototypes of killing fields are ( i ) @xmath189 for the upper half - plane , ( ii ) @xmath190 for the upper half - plane ; and ( iii ) @xmath191 for the disk @xmath192 . these are the striped , poincar , and azimuthal congruences . in these hyperbolic , parabolic and elliptic representations , the s - reissner - nordstrm ( s - rn ) and the witten bubble each have 0 , 1 , and 2 weyl foci . the bubble of nothing in @xmath4 dimensions has the interpretation as a semi - classical decay mode of the kaluza - klein vacuum . a spatial slice is topologically @xmath193 , where the @xmath194 is a cigar with the asymptotic - kk @xmath195 closing at some fixed @xmath45 , which is an @xmath196 bubble . as time passes , the bubble increases in size and ` destroys ' the spacetime . the solution is obtained as an analytic continuation of a black hole and can be generalized to incorporate gauge fields . the electrically charged bubble of nothing in its elliptic representation is gotten from ( [ rnschwarz ] ) by sending @xmath197 , @xmath198 , and to keep the field strength real we need @xmath199 . the metric is @xmath200 at @xmath68 there are clearly rindler - type horizons about which we analytically continue @xmath52 and obtain the rest of ds@xmath84 , @xmath201 . these six patches will precisely correspond to the six cards of fig . [ bubblecard1 ] . = 4.4 in let us now turn to the effect of the analytic continuation of the reissner - nordstrm black hole in weyl coordinates . in weyl coordinates , the effect of wick rotating @xmath202 turns the horizon of the schwarzschild card into an @xmath203-boundary , while @xmath198 turns the boundaries on the horizontal card into noncompact acceleration horizons along the rays @xmath204 , @xmath32 . at the two horizons sending @xmath205 corresponds to @xmath206 along the @xmath207 rays . we find vertical noncompact one eighth - plane cards with special null lines along @xmath208 for @xmath209 and @xmath210 for @xmath211 . each piece of the plane is a doubly covered triangle and it is necessary to change branches of the function @xmath60 at the null line , as they appear in ( [ rnweyl ] ) to transform the card into a single covered quarter plane card . the elliptic ( and as we will shortly see the hyperbolic ) representations of the witten bubble are simply obtained because their ds@xmath84 killing congruences are trivially obtained from those on @xmath212 . specifically , take the @xmath212 embedding into flat space with @xmath213 , @xmath214 , @xmath215 . sending @xmath198 has the effect @xmath216 so the surface becomes @xmath217 embedded in minkowski space , or ds@xmath84 with @xmath24 as an elliptic ( azimuthal ) congruence . on the other hand sending @xmath218 , has the effect @xmath219 giving @xmath220 , which is ds@xmath84 again but with @xmath24 as a hyperbolic ( striped ) congruence . next the s - brane solution of @xcite @xmath221 can also be gotten from ( [ rnschwarz ] ) by taking @xmath202 , @xmath222 , @xmath223 , and @xmath224 . from ( [ rntrans ] ) we see that in weyl s coordinates this analytic continuation of rn can be implemented by sending @xmath202 , @xmath225 , @xmath224 , up to a real coordinate transformation . the card diagram for elliptic s - rn ( fig . [ chargedsbranecard ] ) has the same structure as the witten bubble ( fig . [ bubblecard1 ] ) except that the 6-segment boundary is now @xmath226 where the @xmath24-circle vanishes . the right and left horizons are at @xmath227 . the @xmath228 singularity is a hyperbola on the @xmath229 horizontal cards parametrized as @xmath230 . any @xmath231 gives the same qualitative diagram . the ` smaller ' connected universe on the card diagram is the negative - mass version of s - rn . either sign of the mass gives a universe that is cosmologically singular . the penrose diagram is given in fig . [ srnpenrosefig ] . in the limit @xmath98 , the hyperbola singularity degenerates to a straight line covering the horizon at @xmath228 . the two @xmath232 vertical cards and two @xmath233 horizontal cards then form a positive - mass s - schwarzschild , while each @xmath234 card forms a negative - mass s - schwarzschild whose singularities begin or end the spacetime . one can alternatively form the elliptic s - schwarzschild from the elliptic witten bubble by performing the @xmath11-flip on any vertical card . this procedure is immediate ; the net continuation from schwarzschild is @xmath222 , @xmath235 , and avoids @xmath223 , @xmath225 , and @xmath236 . note how the card representation of the s - brane is quite different from the black hole card diagram while the penrose diagrams of the two spacetimes are nearly just related by ninety degree rotation . this is because the card diagram shows the compact or noncompact @xmath52 direction . the elliptic form of the card diagrams show that schwarzschild s - brane , witten bubble and schwarzschild solutions have similar structures and in fact they are all related by @xmath11-flips and trivial killing continuations . are considered to be less trivial . ] solutions which are related in this manner may be conveniently drawn together in one diagram which simultaneously displays all of their card diagrams . for example in fig . [ fullsschwarz ] the s - schwarzschild solution comprises regions @xmath237 , the witten bubble regions @xmath238 , and the schwarzschild black hole @xmath239 . regions @xmath240 correspond to a singular witten bubble of negative ` mass . ' in this diagram we also see that the unification of the special null lines as they extend through all three solutions ; we can say @xmath63 is the long @xmath241-null line and @xmath64 is the long @xmath242-null line . the charged reissner - nordstrm bh / bubble / s - brane solutions can not be depicted together on such a diagram because @xmath243 changes @xmath244 to @xmath245 . similar diagrams can be found in @xcite . = 3.5 in the charged witten bubble @xmath246 can alternatively be obtained from the rn black hole ( [ rnschwarz ] ) by taking @xmath247 and @xmath197 , @xmath199 . here , @xmath52 plays the role of time and @xmath226 is the time where the bubble ` has minimum size . ' ( this statement has meaning if we break @xmath248 symmetry . ) to achieve this in weyl s coordinates , we put @xmath225 , @xmath202 , @xmath199 ; the resulting space is equivalent to witten s bubble by the real coordinate transformation @xmath249 thus in weyl coordinates the only difference between the hyperbolic witten bubble and the elliptic s - rn is putting @xmath250 . witten s bubble universe is represented in weyl coordinates as a vertical half - plane card , @xmath251 , @xmath252 , where now the @xmath203-circle , and not the @xmath24-circle , vanishes at @xmath32 . this boundary is where the bubble of nothing begins . note that the vertical card now has minkowski signature and is conformal to @xmath20 . this vertical card does not have special null lines since the foci are at imaginary values @xmath253 , and so the spacetime is covered only once by the schwarzschild coordinates . the bubble does have a rod which is along the imaginary @xmath254 axis and which intersects the card at the @xmath32 , @xmath255 origin . the hyperbolic representation of the charged witten bubble is therefore just a vertical half - plane . in such a case the card diagrams have just as much causal information as a penrose diagram and the boundary of the card easily describes the edge of the bubble of nothing . we obtain a hyperbolic representation for s - rn from ( [ rnschwarz ] ) by sending @xmath256 , @xmath224 , @xmath257 , @xmath218 , and @xmath198 . the fibered directions are now hyperbolic @xmath258 . in weyl coordinates , to maintain reality of the solution , we begin with rn ( [ rnweyl ] ) and send @xmath202 , @xmath198 , @xmath236 , and must explicitly change the branch of the square root introducing the minus sign @xmath259 . this has the interpretation of staying on the same horizontal card and rotating the rod in the complex @xmath22-plane . the foci are then at ( @xmath260 , @xmath32 ) with their special null lines intersecting the real half - plane at @xmath109 , @xmath261 . the half - plane is doubly covered , and we will take @xmath262 as the branch cut . the sign change of @xmath263 has effectively reversed the roles of @xmath45 and @xmath52 so that , after undoing the branch cut with say a square - root conformal transformation , @xmath264 is one boundary - horizon and @xmath265 is the other . the hyperbolic angle @xmath52 is unbounded @xmath266 . the doubly - covered half - plane is physically cut into two by the @xmath27 singularity . at each horizon @xmath267 we have a four - card junction ; the double half - plane horizontal card meets another mirror copy as well as two vertical cards . the vertical cards are at @xmath268 and @xmath31 and have no special null lines or other features . a full card diagram is shown in fig . [ srn2fig ] . note that there are no boundaries of this card diagram where a spacelike killing circle vanishes . = 2.5 in setting @xmath269 and @xmath270 , the explicit form of hyperbolic s - rn on the horizontal card is @xmath271 we can arrive at this spacetime in a simpler way . take the rn black hole and analytically continue to get the hyperbolic charged witten bubbles . these universes are nonsingular for @xmath272 or @xmath273 . they have boundaries at @xmath267 where the @xmath203-circle vanishes . now , turn these universes on their sides with the @xmath11-flip . this allows us to decompactify @xmath203 and @xmath267 are now milne horizons we are looking precisely at the vertical cards of fig . [ srn2fig ] , and they are connected in a card diagram by an @xmath274 card which is now accessible . we see that generally , vertical half - plane cards parametrized in spherical prolate fashion with no special null lines , when turned on their sides , connect to branched horizontal cards . the @xmath98 limit ( hyperbolic s - schwarzschild ) of the card diagram is easy to picture : the singularities of fig . [ srn2fig ] collapse onto the @xmath275 horizon . one may wonder what happens if we take @xmath225 , on the horizontal card for hyperbolic s - rn , and achieves a vertical card sandwiched between special null lines at @xmath276 . this must unfold to give a vertical plane with two intersecting special null lines . this card diagram structure is discussed in @xcite . there is a third way to put a killing congruence on @xmath85 or ds@xmath84 using parabolic or poincar coordinates . parametrizing hyperbolic space ( euclideanized ads@xmath84 ) as @xmath277 , and keeping the schwarzschild s - brane coordinate @xmath45 and the usual @xmath203 we get a poincar weyl representation of s - rn spacetime @xcite . it is @xmath278 in this weyl representation , @xmath66 is timelike on @xmath272 vertical cards which are noncompact @xmath73 wedges , @xmath279 . this connects along @xmath280 to an @xmath281 horizontal card ; @xmath273 vertical cards attaches to @xmath282 . so this is similar to the elliptic representation of s - rn , except the line segment @xmath283 has collapsed and the special null lines are now conformal null infinity ( fig . [ s - rn-3 ] ) . the singularity on the horizontal card is particularly easy to describe in these coordinates . on the first horizontal card it is on a ray @xmath284 and on the second card it is on a ray @xmath285 . wedge vertical cards pointing up and down with each wedge joined to two others at the tip in a ds@xmath84 fashion.,width=302 ] wedge vertical cards pointing up and down with each wedge joined to two others at the tip in a ds@xmath84 fashion.,width=264 ] if we take the @xmath272 ( or @xmath273 ) @xmath73 wedges and turn them on their sides via the @xmath11-flip , we get the parabolic version of the @xmath272 ( or @xmath273 ) charged witten bubble . the line which used to be the horizon in the s - brane card diagram becomes a boundary which is the ` minimum volume ' sphere , at @xmath32 . time is now purely along the @xmath254 direction as in the hyperbolic witten bubble . the special null line is still @xmath87 since @xmath286 corresponds to @xmath287 . the vertex of the triangular card is not the end of the spacetime . these wedge cards only represent @xmath288 and so the card diagram should be extended to @xmath289 . the card diagram is an infinite array of @xmath73 wedge cards pointing up and an infinite number pointing down . the vertex of each upward card is attached to its nearest two downward neighbors ( one to the left and one to the right ) , in the ds@xmath84 fashion as shown in figure [ bubblecard3 ] . one can identify cards so only needs one upward and one downward card with two attachments . although this card diagram is not the most obvious representation of the witten bubble , it is useful in understanding the s - kerr solution of the next subsection as well as the more complicated s - dihole @xmath290 , @xmath291 universes of @xcite . the twisted s - brane @xcite , see also @xcite , is also known as s - kerr , and it is another example of a nonsingular time - dependent solution . twisted s - branes describe the decay of unstable massive strings . they can be obtained from the kerr black hole using the following card diagram method . double killing continue @xmath202 , @xmath292 to achieve a @xmath293 bubble solution , go to the vertical card via @xmath222 , then perform a @xmath11-flip to achieve s - kerr . s - kerr has symmetry group @xmath294 and therefore has a unique card diagram . for the parameter range @xmath135 there are horizons and the card diagram structure is that of the elliptic bubble ( fig . [ bubblecard1 ] ) . the foci are at @xmath295 . there is an ergosphere and ctc region on the horizontal card and has the same qualitative shape as it does for the kerr black hole diagram . in comparison , the penrose diagram showing ( variably twisted ) @xmath203 and the boyer - lindquist coordinate @xmath45 is shown in fig . [ skerrpenrosefig ] . = 2.5 in in the extremal limit @xmath296 , @xmath52-orbits on the vertical card shift up relative to the special null line , and any fixed ( @xmath45 , @xmath52 ) point is sent above the null line . the region below the null line disappears in this limit and the horizontal card collapses to a point . furthermore , those geodesics in the upper - right card can only reach the lower - left card ( and the same with upper - left and lower - right ) , splitting the universe into connected @xmath73 wedges just like the parabolic representation for the witten bubble ( fig . [ bubblecard3 ] ) , where the connections are ds@xmath84-like twisted throats . the case @xmath297 for s - kerr does not have horizons and can be represented as a single vertical half - plane card with @xmath251 and with no special null lines . just like the hyperbolic witten bubble , superextremal s - kerr can be turned on its side with the @xmath11-flip , to yield a new spacetime @xcite . in the limit where @xmath298 , the @xmath52-orbits flatten out on the card , and the solution becomes flat space . the penrose diagrams for extremal and superextremal s - kerr are given in fig . [ skerrextremalpenrosefig ] . in this paper we have examined and extended the utility of the weyl ansatz through the construction of an associated weyl card diagram . weyl coordinates are an excellent choice for clearly understanding geometric features and so drawing the card diagram is useful . the card diagram conveniently captures most of the interesting properties of a spacetime including its singularities , horizons , and some aspects of its causal structure and null infinity . card diagrams for families of solutions such as charged and rotating black holes ( and bubbles and s - branes ) share similar features and can change continuously . they are useful in keeping track of various analytic continuations and mentally partitioning complicated spacetimes into simpler regions . the only technical issues that seem to arise and which we resolved dealt with branches of weyl distance square - root functions , special null lines on vertical cards , and branched horizontal cards . here we summarize the solutions in this paper . the card diagrams correctly capture the different regions of the charged reissner - nordstrm black hole and its various charged and chargeless limits , and its negative mass complement . we also analyzed the kerr black hole and its singularity structure . in particular the safe passage through the interior of the ring to the second asymptotic universe through @xmath27 is clearly depicted . the witten bubbles and s - branes each have three card diagrams representations corresponding to the three different choices of killing congruences on ds@xmath84 or @xmath85 . the elliptic representations had two foci and six cards . s - rn had a cosmological singularity on the horizontal card splitting the card diagram into two connected universes , whereas the charged bubbles have the same card structure but are nonsingular . the hyperbolic representations have no foci : the bubble is a simple vertical half - plane card and hyperbolic s - rn has a branched horizontal card which we fixed by a conformal mapping of the half - plane . finally the parabolic representation of the bubble was an infinite array of @xmath73 triangle wedges connected pointwise while s - rn had a 6-card butterfly shape . this parabolic representation showcased special null lines serving as null infinity . s - reissner nordstrm can be obtained from the bubble in two ways . one may start with the bubble and analytically continue @xmath224 in weyl coordinates . in this case the hyperbolic / elliptic representation of the bubble maps to the elliptic / hyperbolic representation of the s - brane . we also found that these two solutions can be related by what we called the @xmath11-flip , which is conveniently visualized as a flip of the associated cards about a null line . this procedure maintains the number of weyl foci on the card and so maps the elliptic / parabolic / hyperbolic bubble to the elliptic / parabolic / hyperbolic s - brane . the @xmath11-flip provides a simple and geometric way to relate schwarzschild with its two analytic continuations , the bubble of nothing and the s - brane . in fact spacetimes related by @xmath11-flips can be simultaneously drawn together in a complexified @xmath299 ( or @xmath300 ) spacetime diagram . just as we used the @xmath11-flip to turn the hyperbolic witten bubble on its side and got hyperbolic s - rn , we can take the vertical half - plane card diagrams for the kerr bubble , dihole wave , superextremal s - kerr and superextremal s - dihole and apply the @xmath11-flip to yield new spacetimes . these new solutions will be described in @xcite . the card diagram formalism can be further generalized ; the recently developed weyl - papapetrou formalism @xcite for @xmath301 will yield card diagrams and 5d kerr - related solutions will appear in @xcite . furthermore card diagrams do not require weyl s canonical coordinates . spacetimes with weyl - type symmetry and yet where weyl s procedure fails algebraically can still admit card diagrams . an example is the inclusion of a nonzero cosmological constant @xmath302 , where a @xmath11-flip changes the sign of @xmath302 . card diagrams for pure ( a)ds@xmath303 space for @xmath304 are presented in @xcite . constant - curvature black holes obtained by quotienting @xcite will also have card diagrams . we hope that our methods , and their further generalizations , will have even greater applicability than to the multitude of spacetimes already discussed herein . we thank d. jatkar , a. maloney , w. g. ritter , a. strominger , t. wiseman and x. yin for valuable discussions and comments . g. c. j. would like to thank the nsf for funding . j. e. w. is supported in part by the national science council , the center for theoretical physics at national taiwan university , the national center for theoretical sciences and would like to thank the organizers of strings 2004 for support and a wonderful conference where part of this research was conducted . the chargeless schwarzschild black hole is easily perturbed as a weyl solution by adding more rod - horizons , to form israel - khan arrays . we can use these solutions to smoothly perturb the witten bubble in two different ways and also the s - schwarzschild in two different ways . addition of charge can be done via weyl s electrification method @xcite . the analytic continuation in weyl space to obtain the hyperbolic witten bubble is precisely the same as in @xcite , except here the schwarzschild rod crosses @xmath109 . even - in-@xmath22 israel - khan arrays where no rod crosses @xmath109 can be analytically continued to gravitational wave solutions sourced by imaginary black holes , ie rods at imaginary time . we can thus generalize the witten bubble by adding additional waves by symmetrically placing more rods in addition to the one which crosses @xmath109 . we dub such an array the ` hyperbolic - perturbed witten bubble . ' as these additional rods are made to cover more of the @xmath22-axis and are brought closer and closer to the principal rod , the deformed witten bubble solution hangs longer with a minimum - radius @xmath24-circle . in the limit where rods occupy the entire @xmath22-axis , we get a static flat solution , which is minkowski 3-space times a fixed - circumference @xmath24-circle . the hyperbolic - perturbed witten bubble can be turned on its side , yielding a hyperbolic - perturbed s - schwarzschild . it has the card diagram structure of fig . [ srn2fig ] . we can also perturb s - schwarzschild by adding rods before analytically continuing @xmath225 , @xmath236 . we dub this the ` elliptic - perturbed s - schwarzschild . ' it is different than hyperbolic - perturbed s - schwarzschild , and has the card diagram structure of fig . [ chargedsbranecard ] . turned on its side , it yields an elliptic - perturbed witten bubble , with the card diagram structure of fig . [ bubblecard1 ] . it is different than the hyperbolic - perturbed witten bubble . in any of the cases , we can choose to analytically continue the mass parameters of the additional rods or not . additionally , we can displace some rods in the imaginary @xmath22-direction which affects the @xmath254-center of their disturbance . if we do everything in an even fashion , i.e. we respect @xmath305 , the resulting geometry ( at real @xmath254 ) will be real . in particular , rotating a rod at @xmath280 counterclockwise means rotating its image at @xmath282 clockwise . we see in the discussion of the 2-rod example @xcite that there may be several choices for branches . the card diagram techniques allow us to easily construct these two inequivalent families of perturbed witten bubble and perturbed s - schwarzschild solutions . these and other multi - rod , s - dihole , and infinite - periodic - universe solutions are described in @xcite and can not be easily described or understood without weyl coordinates and the construction and language of card diagrams . the nontrivial @xmath225 continuation is essential . the formalism of @xcite can be extended for general @xmath4 to include an electrostatic potential . this is somewhat surprising since the electromagnetic energy - momentum tensor @xmath306 is traceless only in @xmath8 and so einstein s equations are more complicated . nevertheless , a cancellation does occur and one may sum the diagonal killing frame components of the ricci tensor to achieve a harmonic condition . follow the notation of @xcite and add a 1-form potential @xmath307 where @xmath308 is timelike ( @xmath309 ) and all other @xmath310 , @xmath311 are spacelike ( with @xmath312 ) . the metric takes the form @xmath313 from which we extract the frame metric @xmath314.\ ] ] for @xmath315 we have @xmath316 and @xmath317 , all other components vanishing . we compute @xmath318 and @xmath319 the field equations are @xmath320 ; taking the trace , we get @xmath321 and einstein s equations are then @xmath322 form the sum @xmath323 ; the right side of ( [ neweinstein ] ) gives @xmath324 hence ( following ( 2.4)-(2.5 ) of @xcite ) we get @xmath325 the weyl harmonic condition . one can add magnetostatic potentials along spatial killing directions as well . we skip remaining details and give the equations . let us assume @xmath326 is timelike and @xmath310 are spacelike for @xmath311 , the potential is @xmath327 , and the metric is @xmath328 and @xmath329 , @xmath330 . einstein s equations are @xmath331 and @xmath332 maxwell s equations are @xmath333 all laplacians and divergences are with respect to a flat 3d axisymmetric auxiliary space with coordinates @xmath300 . a. sen , `` rolling tachyon , '' jhep * 0204 * , 048 ( 2002 ) , [ arxiv : hep - th/0203211 ] ; `` tachyon matter , '' jhep * 0207 * , 065 ( 2002 ) , [ arxiv : hep - th/0203265 ] ; `` field theory of tachyon matter , '' mod . phys . lett . * a17 * , 1797 ( 2002 ) , [ arxiv : hep - th/0204143 ] ; `` time evolution in open string theory , '' jhep * 0210 * , 003 ( 2002 ) , [ arxiv : hep - th/0207105 ] .

we consider the sampling problem for functional pca ( fpca ) , where the simplest example is the case of taking time samples of the underlying functional components . more generally , we model the sampling operation as a continuous linear map from to , where the functional components to lie in some hilbert subspace of , such as a reproducing kernel hilbert space of smooth functions . this model includes time and frequency sampling as special cases . in contrast to classical approach in fpca in which access to entire functions is assumed , having a limited number of functional samples places limitations on the performance of statistical procedures . we study these effects by analyzing the rate of convergence of an-estimator for the subspace spanned by the leading components in a multi - spiked covariance model . the estimator takes the form of regularized pca , and hence is computationally attractive . we analyze the behavior of this estimator within a nonasymptotic framework , and provide bounds that hold with high probability as a function of the number of statistical samples and the number of functional samples . we also derive lower bounds showing that the rates obtained are minimax optimal .

1109.3336

the statistical analysis of functional data , commonly referred to as functional data analysis ( fda ) , is an established area of statistics with a great number of practical applications ; see the books @xcite and references therein for various examples . when the data is available as finely sampled curves , say in time , it is common to treat it as a collection of continuous - time curves or functions , each being observed in totality . these datasets are then termed `` functional , '' and various statistical procedures applicable in finite dimensions can be extended to this functional setting . among such procedures is principal component analysis ( pca ) , which is the focus of present work . if one thinks of continuity as a mathematical abstraction of reality , then treating functional data as continuous curves is arguably a valid modeling device . however , in practice , one is faced with finite computational resources and is forced to implement a ( finite - dimensional ) approximation of true functional procedures by some sort of truncation procedure , for instance , in the frequency domain . it is then important to understand the effects of this truncation on the statistical performance of the procedure . in other situations , such as in longitudinal data analysis @xcite , a continuous curve model is justified as a hidden underlying generating process to which one has access only through sparsely sampled measurements in time , possibly corrupted by noise . studying how the time - sampling affects the estimation of the underlying functions in the presence of noise shares various common elements with the frequency - domain problem described above . the aim of this paper is to study effects of `` sampling''in a fairly general sense on functional principal component analysis in smooth function spaces . in order to do so , we adopt a functional - theoretic approach by treating the sampling procedure as a ( continuous ) linear operator . this set - up provides us with a notion of sampling general enough to treat both the frequency - truncation and time - sampling within a unified framework . we take as our smooth function space a hilbert subspace @xmath0 of @xmath6 $ ] and denote the sampling operator by @xmath7 . we assume that there are functions @xmath8 , @xmath9 $ ] , in @xmath0 for @xmath10 , generated i.i.d . from a probabilistic model ( to be discussed ) . we then observe the collection @xmath11 in noise . we refer to the index @xmath5 as the number of _ statistical samples _ , and to the index @xmath3 as the number of _ functional samples_. we analyze a natural @xmath4-estimator which takes the form of a regularized pca in @xmath1 , and provide nonasymptotic bounds on the estimation error in terms of @xmath5 and @xmath3 . the eigen - decay of two operators govern the rates , the product of the sampling operator @xmath12 and its adjoint , and the product of the map embedding @xmath0 in @xmath2 and its adjoint . our focus will be on the setting where @xmath0 is a reproducing kernel hilbert space ( rkhs ) , in which case the two eigen - decays are intimately related through the kernel function @xmath13 . in such cases , the two components of the rate interact and give rise to optimal values for the number of functional samples ( @xmath3 ) in terms of the number of statistical samples ( @xmath5 ) or vice versa . this has practical appeal in cases where obtaining either type of samples is costly . our model for the functions @xmath14 is an extension to function spaces of the _ spiked covariance model _ introduced by johnstone and his collaborators @xcite , and studied by various authors ( e.g. , @xcite ) . we consider such models with @xmath15 components , each lying within the hilbert ball @xmath16 of radius @xmath17 , with the goal of recovering the @xmath15-dimensional subspace spanned by the spiked components in this functional model . we analyze our @xmath4-estimators within a high - dimensional framework that allows both the number of statistical samples @xmath5 and the number of functional samples @xmath3 to diverge together . our main theoretical contributions are to derive nonasymptotic bounds on the estimation error as a function of the pair @xmath18 , which are shown to be sharp ( minimax - optimal ) . although our rates also explicitly track the number of components @xmath15 and the smoothness parameter @xmath17 , we do not make any effort to obtain optimal dependence on these parameters . the general asymptotic properties of pca in function spaces have been investigated by various authors ( e.g. , @xcite ) . accounting for smoothness of functions by introducing various roughness / smoothness penalties is a standard approach , used in the papers @xcite , among others . the problem of principal component analysis for sampled functions , with a similar functional - theoretic perspective , is discussed by besse and ramsey @xcite for the noiseless case . a more recent line of work is devoted to the case of functional pca with noisy sampled functions @xcite . cardot @xcite considers estimation via spline - based approximation , and derives mise rates in terms of various parameters of the model . hall et al . @xcite study estimation via local linear smoothing , and establish minimax - optimality in certain settings that involve a fixed number of functional samples . both papers @xcite demonstrate trade - offs between the numbers of statistical and functional samples ; we refer the reader to hall et al . @xcite for an illuminating discussion of connections between fda and lda approaches ( i.e. , having full versus sampled functions ) , which inspired much of the present work . we note that the regularization present in our @xmath4-estimator is closely related to classical roughness penalties @xcite in the special case of spline kernels , although the discussion there applies to fully - observed functions , as opposed to the sampled models considered here . after initial posting of this work , we became aware of more recent work on sampled functional pca . working within the framework of hall et al . @xcite , the analysis of li and hsing @xcite allows for more flexible sample sizes per curve ; they derive optimal uniform ( i.e. , @xmath19 ) rates of convergence for local linear smoothing estimators of covariance function and the resulting eigenfunctions . another line of work @xcite has analyzed sampled forms of silverman s criterion @xcite , with some variations . huang et al . @xcite derive a criterion based on rank - one approximation coupled with scale invariance considerations , combined with an extra weighting of the covariance matrix . xi and zhao @xcite also show the consistency of their estimator for both regular and irregular sampling . the regular ( time ) sampling setup in both papers have an overlap with our work ; the eigenfunctions are assumed to lie in a second order sobolev space , corresponding to a special case of a rkhs . however , even in this particular case , our estimator is different , and it is an interesting question whether a version of the results presented here can be used to show the minimax optimality of these silverman - type criteria . there has also been recent work with emphasis on sampled functional covariance estimation , including the work of cai and yuan @xcite , who analyze an estimator which can be described as regularized least - squares with penalty being the norm of tensor product of rkhs with itself . they provide rates of convergence for the covariance function , from which certain rates ( argued to be optimal within logarithmic factors ) for eigenfunctions follow . as mentioned above , our sampled model resembles very much that of spiked covariance model for high - dimensional principal component analysis . a line of work on this model has treated various types of sparsity conditions on the eigenfunctions @xcite ; in contrast , here the smoothness condition on functional components translates into an ellipsoid condition on the vector principal components . perhaps an even more significant difference is that in this paper , the effective scaling of noise in @xmath1 is substantially smaller in some cases ( e.g. , the case of time sampling ) . this difference could explain why the difficulty of `` high - dimensional '' setting is not observed in such cases as one lets @xmath20 . on the other hand , a difficulty particular to our sampled model is the lack of orthonormality between components after sampling . it not only leads to identifiability issues , but also makes recovering individual components difficult . in order to derive nonasymptotic bounds on our @xmath4-estimator , we exploit various techniques from empirical process theory ( e.g. , @xcite ) , as well as the concentration of measure ( e.g. , @xcite ) . we also exploit recent work @xcite on the localized rademacher complexities of unit balls in a reproducing kernel hilbert space , as well as techniques from nonasymptotic random matrix theory , as discussed in davidson and szarek @xcite , in order to control various norms of random matrices . these techniques allow us to obtain finite - sample bounds that hold with high probability , and are specified explicitly in terms of the pair @xmath18 , and the underlying smoothness of the hilbert space . the remainder of this paper is organized as follows . section [ secbackground ] is devoted to background material on reproducing kernel hilbert spaces , adjoints of operators , as well as the class of sampled functional models that we study in this paper . in section [ secmest ] , we describe @xmath4-estimators for sampled functional pca , and discuss various implementation details . section [ secmain ] is devoted to the statements of our main results , and discussion of their consequences for particular sampling models . in subsequent sections , we provide the proofs of our results , with some more technical aspects deferred to the supplementary material @xcite . section [ secproofsubspace ] is devoted to bounds on the subspace - based error . we conclude with a discussion in section [ secdiscuss ] . in the supplementary material @xcite , section 7 is devoted to proofs of bounds on error in the function space , whereas section 8 provides proofs of matching lower bounds on the minimax error , showing that our analysis is sharp . _ notation_. we will use to denote the hilbert schmidt norm of an operator or a matrix . the corresponding inner product is denoted as @xmath21 . if @xmath22 is an operator on a hilbert space @xmath0 with an orthonormal basis @xmath23 , then @xmath24 . for a matrix @xmath25 , we have @xmath26 , the range as @xmath27 and the kernel as @xmath28 . in this section , we begin by introducing background on reproducing kernel hilbert spaces , as well as linear operators and their adjoints . we then introduce the functional and observation model that we study in this paper , and conclude with discussion of some approximation - theoretic issues that play an important role in parts of our analysis . we begin with a quick overview of some standard properties of reproducing kernel hilbert spaces ; we refer the reader to the books @xcite and references therein for more details . a reproducing kernel hilbert space ( or rkhs for short ) is a hilbert space @xmath0 of functions @xmath29 that is equipped with a symmetric positive semidefinite function @xmath30 , known as the kernel function . we assume the kernel to be continuous , and the set @xmath31 to be compact . for concreteness , we think of @xmath32 $ ] throughout this paper , but any compact set of @xmath33 suffices . for each @xmath34 , the function @xmath35 belongs to the hilbert space @xmath0 and it acts as the _ representer of evaluation _ , meaning that @xmath36 for all @xmath37 . the kernel @xmath38 defines an integral operator @xmath39 on @xmath40 , mapping the function @xmath41 to the function @xmath42 . by the spectral theorem in hilbert spaces , this operator can be associated with a sequence of eigenfunctions @xmath43 , in @xmath0 , orthogonal in @xmath0 and orthonormal in @xmath40 , and a sequence of nonnegative eigenvalues @xmath44 . most useful for this paper is the fact that any function @xmath37 has an expansion in terms of these eigenfunctions and eigenvalues , namely @xmath45 for some @xmath46 . in terms of this expansion , we have the representations @xmath47 and @xmath48 . many of our results involve the decay rate of these eigenvalues : in particular , for some parameter @xmath49 , we say that the kernel operator has eigenvalues with _ polynomial-@xmath50 decay _ if there is a constant @xmath51 such that @xmath52 let us consider an example to illustrate . in the case @xmath32 $ ] and @xmath53 , we can consider the kernel function @xmath54 . as discussed in appendix a of the supplementary material @xcite , this kernel generates the class of functions @xmath55\bigr ) \mid f(0 ) = 0 , f \mbox { absolutely continuous and } f ' \in l^2\bigl([0,1]\bigr ) \bigr\}.\ ] ] the class @xmath0 is an rkhs with inner product @xmath56 , and the ball @xmath16 corresponds to a sobolev space with smoothness @xmath53 . the eigen - decomposition of the kernel integral operator is @xmath57^{-2},\qquad \psi_k(t ) = \sqrt{2 } \sin\bigl ( \mu_k^{-1/2 } t \bigr),\qquad k=1,2,\ldots.\ ] ] consequently , this class has polynomial decay with parameter @xmath53 . we note that there are natural generalizations of this example to @xmath58 , corresponding to the sobolev classes of @xmath50-times differentiable functions ; for example , see the books @xcite . in this paper , the operation of generalized sampling is defined in terms of a bounded linear operator @xmath59 on the hilbert space . its adjoint is a mapping @xmath60 , defined by the relation @xmath61 for all @xmath37 and @xmath62 . in order to compute a representation of the adjoint , we note that by the riesz representation theorem , the @xmath63th coordinate of this mapping namely , @xmath64_j$]can be represented as an inner product @xmath65 , for some element , and we can write @xmath66^t.\ ] ] consequently , we have @xmath67 , so that for any @xmath62 , @xmath68 this adjoint operator plays an important role in our analysis . let @xmath69 be a fixed sequence of positive numbers , and let @xmath70 be a fixed sequence of functions orthonormal in @xmath6 $ ] . consider a collection of @xmath5 i.i.d . random functions @xmath71 , generated according to the model @xmath72 where @xmath73 are i.i.d . @xmath74 across all pairs @xmath75 . this model corresponds to a finite - rank instantiation of functional pca , in which the goal is to estimate the span of the unknown eigenfunctions @xmath76 . typically , these eigenfunctions are assumed to satisfy certain smoothness conditions ; in this paper , we model such conditions by assuming that the eigenfunctions belong to a reproducing kernel hilbert space @xmath0 embedded within @xmath6 $ ] ; more specifically , they lie in some ball in @xmath0 , @xmath77 for statistical problems involving estimation of functions , the random functions might only be observed at certain times @xmath78 , such as in longitudinal data analysis , or we might collect only projections of each @xmath79 in certain directions , such as in tomographic reconstruction . more concretely , in a _ time - sampling model _ , we observe @xmath3-dimensional vectors of the form @xmath80^t + \sigma_0 w_i\qquad\mbox{for $ i = 1 , 2,\ldots , n$},\ ] ] where @xmath81 is a fixed collection of design points , and @xmath82 is a noise vector . another observation model is the _ basis truncation model _ in which we observe the projections of @xmath41 onto the first @xmath3 basis functions @xmath83 of the kernel operator namely , @xmath84^t + \sigma_0 w_i\nonumber\\[-8pt]\\[-8pt ] & & \eqntext{\mbox{for $ i = 1 , 2,\ldots , n$},}\end{aligned}\ ] ] where @xmath85 represents the inner product in @xmath6 $ ] . in order to model these and other scenarios in a unified manner , we introduce a linear operator @xmath86 that maps any function @xmath87 in the hilbert space to a vector @xmath88 of @xmath3 samples , and then consider the linear observation model @xmath89 this model ( [ eqnlinobs ] ) can be viewed as a functional analog of the spiked covariance models introduced by johnstone @xcite as an analytically - convenient model for studying high - dimensional effects in classical pca . both the time - sampling ( [ eqntimesamp ] ) and frequency truncation ( [ eqnbasistrun ] ) models can be represented in this way , for appropriate choices of the operator @xmath86 . recall representation ( [ eqnlinoprep ] ) of @xmath86 in terms of the functions @xmath90 . * for the time sampling model ( [ eqntimesamp ] ) , we set @xmath91 , so that by the reproducing property of the kernel , we have @xmath92 for all @xmath37 , and @xmath93 . with these choices , the operator @xmath86 maps each @xmath37 to the @xmath3-vector of rescaled samples @xmath94^t.\ ] ] defining the rescaled noise @xmath95 yields an instantiation of model ( [ eqnlinobs ] ) which is equivalent to time - sampling ( [ eqntimesamp ] ) . * for the basis truncation model ( [ eqnbasistrun ] ) , we set @xmath96 so that the operator @xmath12 maps each function @xmath37 to the vector of basis coefficients @xmath97^t$ ] . setting @xmath98 then yields another instantiation of model ( [ eqnlinobs ] ) , this one equivalent to basis truncation ( [ eqnbasistrun ] ) . a remark on notation before proceeding : in the remainder of the paper , we use @xmath99 as shorthand notation for @xmath100 , since the index @xmath3 should be implicitly understood throughout our analysis . in this paper , we provide and analyze estimators for the @xmath15-dimensional eigen - subspace spanned by @xmath101 , in both the sampled domain @xmath1 and in the functional domain . to be more specific , for @xmath102 , define the vectors @xmath103 , and the subspaces @xmath104 and let @xmath105 and @xmath106 denote the corresponding estimators . in order to measure the performance of the estimators , we will use projection - based distances between subspaces . in particular , let @xmath107 and @xmath108 be orthogonal projection operators into @xmath109 and @xmath105 , respectively , considered as subspaces of @xmath110 . similarly , let @xmath111 and @xmath112 be orthogonal projection operators into @xmath113 and @xmath106 , respectively , considered as subspaces of @xmath114 . we are interested in bounding the deviations @xmath115 where @xmath116 is the hilbert schmidt norm of an operator ( or matrix ) . one object that plays an important role in our analysis is the matrix @xmath117 . from the form of the adjoint , it can be seen that @xmath118_{ij } = \langle\phi_i , \phi_j \rangle_{\mathcal{h}}$ ] . for future reference , let us compute this matrix for the two special cases of linear operators considered thus far : * for the time sampling model ( [ eqntimesamp ] ) , we have @xmath91 for all @xmath119 , and hence @xmath118_{ij } = \frac{1}{m}\langle{\mathbb{k}}(\cdot , t_i),{\mathbb{k}}(\cdot , t_j)\rangle_{{\mathcal{h } } } = \frac{1}{m } { \mathbb{k}}(t_i , t_j)$ ] , using the reproducing property of the kernel . * for the basis truncation model ( [ eqnbasistrun ] ) , we have @xmath120 , and hence @xmath118_{ij } = \langle\mu_i \psi_i,\mu_j\psi_j\rangle_{{\mathcal{h } } } = \mu_i\delta_{ij}$ ] . thus , in this special case , we have @xmath121 . in general , the matrix @xmath122 is a type of gram matrix , and so is symmetric and positive semidefinite . we assume throughout this paper that the functions @xmath90 are linearly independent in @xmath0 , which implies that @xmath122 is strictly positive definite . consequently , it has a set of eigenvalues which can be ordered as @xmath123 under this condition , we may use @xmath122 to define a norm on @xmath1 via @xmath124 . moreover , we have the following interpolation lemma , which is proved in appendix b.1 of the supplementary material @xcite : [ leminterpolate ] for any @xmath37 , we have @xmath125 , with equality if and only if @xmath126 . moreover , for any @xmath127 , the function @xmath128 has smallest hilbert norm of all functions satisfying @xmath129 , and is the unique function with this property . this lemma is useful in constructing a function - based estimator , as will be clarified in section [ secmest ] . in our analysis of the functional error @xmath130 , a number of approximation - theoretic quantities play an important role . as a mapping from an infinite - dimensional space @xmath0 to @xmath1 , the operator @xmath12 has a nontrivial nullspace . given the observation model ( [ eqnlinobs ] ) , we receive no information about any component of a function @xmath131 that lies within this nullspace . for this reason , we define the width of the nullspace in the @xmath2-norm , namely the quantity @xmath132 in addition , the observation operator @xmath12 induces a semi - norm on the space @xmath0 , defined by @xmath133_j^2.\ ] ] it is of interest to assess how well this semi - norm approximates the @xmath2-norm . accordingly , we define the quantity @xmath134 which measures the worst - case gap between these two ( semi)-norms , uniformly over the hilbert ball of radius one , restricted to the subspace of interest @xmath135 . given knowledge of the linear operator @xmath12 , the quantity @xmath136 can be computed in a relatively straightforward manner . in particular , recall the definition of the matrix @xmath122 , and let us define a second matrix @xmath137 with entries @xmath138 . [ lemdefect ] we have the equivalence @xmath139 where denotes the @xmath140-operator norm . see appendix b.2 of the supplementary material @xcite for the proof of this claim . with this background in place , we now turn to the description of our @xmath4-estimator , as well as practical details associated with its implementation . we begin with some preliminaries on notation , and our representation of subspaces . recall definition ( [ eqdefzfsffs ] ) of @xmath109 as the @xmath15-dimensional subspace of @xmath1 spanned by @xmath141 , where @xmath142 . our initial goal is to construct an estimate @xmath105 , itself an @xmath15-dimensional subspace , of the unknown subspace @xmath109 . we represent subspaces by elements of the stiefel manifold @xmath143 , which consists of @xmath144 matrices @xmath145 with orthonormal columns @xmath146 a given matrix @xmath145 acts as a representative of the subspace spanned by its columns , denoted by @xmath147 . for any @xmath148 , the matrix @xmath149 also belongs to the stiefel manifold , and since @xmath150 , we may call @xmath149 a version of @xmath145 . we let @xmath151 be the orthogonal projection onto @xmath147 . for two matrices @xmath152 , we measure the distance between the associated subspaces via @xmath153 , where @xmath116 is the hilbert schmidt ( or frobenius ) matrix norm . with this notation , we now specify an @xmath4-estimator for the subspace @xmath154 . let us begin with some intuition . given the @xmath5 samples @xmath155 , let us define the @xmath156 sample covariance matrix @xmath157 . given the observation model ( [ eqnlinobs ] ) , a straightforward computation shows that @xmath158 = \sum_{j=1}^rs^2_j z^*_j \bigl(z^*_j\bigr)^t + \sigma_m^2 i_{m}.\ ] ] thus , as @xmath5 becomes large , we expect that the top @xmath15 eigenvectors of @xmath159 might give a good approximation to @xmath160 . by the courant fischer variational representation , these @xmath15 eigenvectors can be obtained by maximizing the objective function @xmath161 over all matrices @xmath162 . however , this approach fails to take into account the smoothness constraints that the vectors @xmath142 inherit from the smoothness of the eigenfunctions @xmath163 . since @xmath164 by assumption , lemma [ leminterpolate ] implies that @xmath165 consequently , if we define the matrix @xmath166 \in{\mathbb{r}}^{m\times r}$ ] , then it must satisfy the _ trace smoothness condition _ @xmath167 this calculation motivates the constraint @xmath168 in our estimation procedure . based on the preceding intuition , we are led to consider the optimization problem @xmath169 where we recall that @xmath170 . given any optimal solution @xmath171 , we return the subspace @xmath172 as our estimate of @xmath109 . as discussed at more length in section [ secimplementation ] , it is straightforward to compute @xmath171 in polynomial time . the reader might wonder why we have included an additional factor of two in this trace smoothness condition . this slack is actually needed due to the potential infeasibility of the matrix @xmath173 for to problem ( [ eqndiscretemest ] ) , which arises since the columns of @xmath173 are not guaranteed to be orthonormal . as shown by our analysis , the additional slack allows us to find a matrix @xmath174 that spans the same subspace as @xmath173 , and is also feasible for to problem ( [ eqndiscretemest ] ) . more formally , we have : [ lemfeasible ] under condition ( [ eqassumpa2 ] ) , there exists a matrix @xmath174 such that @xmath175 see appendix b.3 of the supplementary material @xcite for the proof of this claim . having thus obtained an estimate is any collection of vectors that span @xmath105 . as we are ultimately only interested in the resulting functional `` subspace , '' it does not matter which particular collection we choose . ] @xmath176 of @xmath177 , we now need to construct a @xmath15-dimensional subspace @xmath106 of the hilbert space to be used as an estimate of @xmath178 . we do so using the interpolation suggested by lemma [ leminterpolate ] . for each @xmath102 , let us define the function @xmath179 since @xmath180 by definition , this construction ensures that @xmath181 . moreover , lemma [ leminterpolate ] guarantees that @xmath182 has the minimal hilbert norm ( and hence is smoothest in a certain sense ) over all functions that have this property . finally , since @xmath12 is assumed to be surjective ( equivalently , @xmath122 assumed invertible ) , @xmath183 maps linearly independent vectors to linearly independent functions , and hence preserves dimension . consequently , the space @xmath184 is an @xmath15-dimensional subspace of @xmath0 that we take as our estimate of @xmath113 . in this section , we consider some practical aspects of implementing the @xmath4-estimator , and present some simulations to illustrate its qualitative properties . we begin by observing that once the subspace vectors @xmath185 have been computed , then it is straightforward to compute the function estimates @xmath186 , as weighted combinations of the functions @xmath90 . accordingly , we focus our attention on solving problem ( [ eqndiscretemest ] ) . on the surface , problem ( [ eqndiscretemest ] ) might appear nonconvex , due to the stiefel manifold constraint . however , it can be reformulated as a semidefinite program ( sdp ) , a well - known class of convex programs , as clarified in the following : [ lemimplement ] problem ( [ eqndiscretemest ] ) is equivalent to solving the sdp @xmath187\\[-8pt ] & & \eqntext{\mbox{such that $ { |\!|\!|}x { |\!|\!|}_{{2 } } \leq1 $ , $ { \operatorname{tr}}(x ) = r$ , and $ { \bigl\langle\bigl\langle}{k^{-1}},{x } { \bigr\rangle\bigr\rangle}\leq2 r\rho^2$}}\end{aligned}\ ] ] for which there always exists an optimal rank @xmath15 solution . moreover , by lagrangian duality , for some @xmath188 , the problem is equivalent to @xmath189 which can be solved by an eigen decomposition of @xmath190 . as a consequence , for a given lagrange multiplier @xmath191 , the regularized form of the estimator can be solved with the cost of solving an eigenvalue problem . for a given constraint @xmath192 , the appropriate value of @xmath191 can be found by a path - tracing algorithm , or a simple dyadic splitting approach . in practice where the radius @xmath17 is not known , one could use cross - validation to set a proper value for the lagrange multiplier @xmath191 . a possibly simpler approach is to evaluate @xmath193 for the optimal @xmath194 on a grid of @xmath191 and choose a value around which @xmath193 is least variable . as for the choice of the number of components @xmath15 , a standard approach for choosing it would be to compute the estimator for different choices , and plot the residual sum of eigenvalues of the sample covariance matrix . as in ordinary pca , an elbow in such a plot indicates a proper trade - off between the number of components to keep and the amount of variation explained . `` true '' principal components @xmath195 with signal - to - noise ratios @xmath196 and @xmath197 , respectively . the number of statistical and functional samples are @xmath198 and @xmath199 . subsequent rows show the corresponding estimators @xmath200 obtained by applying the regularized form ( [ eqsdpregver ] ) . ] in order to illustrate the estimator , we consider the time sampling model ( [ eqntimesamp ] ) , with uniformly spaced samples , in the context of a first - order sobolev rkhs [ with kernel function @xmath201 . the parameters of the model are taken to be @xmath202 , @xmath203 , @xmath204 , @xmath199 and @xmath198 . the regularized form ( [ eqsdpregver ] ) of the estimator is applied , and the results are shown in figure [ fig1 ] . the top row corresponds to the four `` true '' signals @xmath101 , the leftmost being @xmath205 ( i.e. , having the highest signal - to - noise ratio ) and the rightmost @xmath206 . the subsequent rows show the corresponding estimates @xmath207 , obtained using different values of @xmath191 . the second , third and fourth rows correspond to @xmath208 , @xmath209 and @xmath210 . one observes that without regularization ( @xmath211 ) , the estimates for the two weakest signals ( @xmath212 and @xmath206 ) are poor . the case @xmath213 is roughly the one which achieves the minimum for the dual problem . one observes that the quality of the estimates of the signals , and in particular the weakest ones , are considerably improved . the optimal ( oracle ) value of @xmath191 , that is , the one which achieves the minimum error between @xmath101 and @xmath214 , is @xmath215 in this problem . the corresponding estimates are qualitatively similar to those of @xmath213 and are not shown . the case @xmath210 shows the effect of over - regularization . it produces very smooth signals , and although it fails to reveal @xmath205 and @xmath216 , it reveals highly accurate versions of @xmath212 and @xmath206 . it is also interesting to note that the smoothest signal , @xmath206 , now occupies the position of the second ( estimated ) principal component . that is , the regularized pca sees an effective signal - to - noise ratio which is influenced by smoothness . this suggests a rather practical appeal of the method in revealing smooth signals embedded in noise . one can vary @xmath191 from zero upward , and if some patterns seem to be present for a wide range of @xmath191 ( and getting smoother as @xmath191 is increased ) , one might suspect that they are indeed present in data but masked by noise . we now turn to the statistical analysis of our estimators , in particular deriving high - probability upper bounds on the error of the subspace - based estimate @xmath105 , and the functional estimate @xmath106 . in both cases , we begin by stating general theorems that apply to arbitrary linear operators @xmath12theorems [ thmsubspaceerr ] and [ thmfuncerr ] , respectively and then derive a number of corollaries for particular instantiations of the observation operator . we begin by stating high - probability upper bounds on the error @xmath217 of the subspace - based estimates . our rates are stated in terms of a function that involves the eigenvalues of the matrix @xmath218 , ordered as @xmath219 . consider the function @xmath220 given by @xmath221^{1/2}.\ ] ] as will be clarified in our proofs , this function provides a measure of the statistical complexity of the function class @xmath222 we require a few regularity assumptions . define the quantity @xmath223 which measures the departure from orthonormality of the vectors @xmath224 in @xmath1 . a straightforward argument using a polarization identity shows that @xmath225 is upper bounded ( up to a constant factor ) by the uniform quantity @xmath136 , as defined in equation ( [ eqndefnphidefect ] ) . recall that the random functions are generated according to the model @xmath226 , where the signal strengths are ordered as @xmath227 , and that @xmath228 denotes the noise standard deviation in the observation model ( [ eqnlinobs ] ) . in terms of these quantities , we require the following assumptions : @xmath229 the first part of condition ( a1 ) is to prevent the ratio @xmath230 from going to zero as the pair @xmath18 increases , where the constant @xmath231 is chosen for convenience . such a lower bound is necessary for consistent estimation of the eigen - subspace corresponding to @xmath232 . the second part of condition ( a1 ) , involving the constant @xmath233 , provides a lower bound on the signal - to - noise ratio @xmath234 . condition ( a2 ) is required to prevent degeneracy among the vectors @xmath142 obtained by mapping the unknown eigenfunctions to the observation space @xmath1 . [ in the ideal setting , we would have @xmath235 , but our analysis shows that the upper bound in ( a2 ) is sufficient . ] condition ( a3 ) is required so that the critical tolerance @xmath236 specified below is well - defined ; as will be clarified , it is always satisfied for the time - sampling model , and holds for the basis truncation model whenever @xmath237 . condition ( a4 ) is easily satisfied , since the right - hand side of ( [ eqassumpa4 ] ) goes to infinity while we usually take @xmath15 to be fixed . our results , however , are still valid if @xmath15 grows slowly with @xmath3 and @xmath5 subject to ( [ eqassumpa4 ] ) . [ thmsubspaceerr ] under conditions for a sufficiently small constant @xmath233 , let @xmath236 be the smallest positive number satisfying the inequality @xmath238 then there are universal positive constants @xmath239 such that @xmath240 \geq1 - \varphi(n , { \epsilon_{m , n}}),\ ] ] where @xmath241 . we note that theorem [ thmsubspaceerr ] is a general result , applying to an arbitrary bounded linear operator @xmath12 . however , we can obtain a number of concrete results by making specific choices of this sampling operator , as we explore in the following sections . let us begin with the time - sampling model ( [ eqntimesamp ] ) , in which we observe the sampled functions @xmath242^t + \sigma_0 w_i\qquad \mbox{for $ i = 1 , 2,\ldots , m$.}\ ] ] as noted earlier , this set - up can be modeled in our general setting ( [ eqnlinobs ] ) with @xmath243 and @xmath244 . in this case , by the reproducing property of the rkhs , the matrix @xmath245 has entries of the form @xmath246 . letting @xmath247 denote its ordered eigenvalues , we say that the kernel matrix @xmath122 has polynomial - decay with parameter @xmath49 if there is a constant @xmath248 such that @xmath249 for all @xmath250 . since the kernel matrix @xmath122 represents a discretized approximation of the kernel integral operator defined by @xmath38 , this type of polynomial decay is to be expected whenever the kernel operator has polynomial-@xmath50 decaying eigenvalues . for example , the usual spline kernels that define sobolev spaces have this type of polynomial decay @xcite . in appendix a of the supplementary material @xcite , we verify this property explicitly for the kernel @xmath251 that defines the sobolev class with smoothness @xmath53 . for any such kernel , we have the following consequence of theorem [ thmsubspaceerr ] : [ cortimeemp ] consider the case of a time - sampling operator @xmath12 . in addition to conditions and , suppose that the kernel matrix @xmath122 has polynomial - decay with parameter @xmath49 . then we have @xmath252 \geq1 - \varphi(n , m),\ ] ] where @xmath253 , and @xmath254 . \(a ) disregarding constant pre - factors not depending on the pair @xmath18 , corollary [ cortimeemp ] guarantees that solving problem ( [ eqndiscretemest ] ) returns a subspace estimate @xmath255 such that @xmath256 with high probability as @xmath257 increase . depending on the scaling of the number of time samples @xmath3 relative to the number of functional samples @xmath5 , either term in this upper bound can be the smallest ( and hence active ) one . for instance , it can be verified that whenever @xmath258 , then the first term is smallest , so that we achieve the rate @xmath259 . the appearance of the term @xmath260 is quite natural , as it corresponds to the minimax rate of a nonparametric regression problem with smoothness @xmath50 , based on @xmath3 samples each of variance @xmath261 . later , in section [ seclower ] , we provide results guaranteeing that this scaling is minimax optimal under reasonable conditions on the choice of sample points ; in particular , see theorem [ thmminimaxemp](a ) . \(b ) to be clear , although bound ( [ eqntimeemp ] ) allows for the possibility that the error is of order _ lower than @xmath261 _ , we note that the probability with which the guarantee holds includes a term of the order @xmath262 . consequently , in terms of expected error , we can not guarantee a rate faster than @xmath261 . proof of corollary [ cortimeemp ] we need to bound the critical value @xmath236 defined in the theorem statement ( [ eqndefnepscrit ] ) . define the function @xmath263 , and note that @xmath264 by construction . under the assumption of polynomial-@xmath50 eigendecay , we have @xmath265 and some algebra then shows that @xmath266 . disregarding constant factors , an upper bound on the critical @xmath236 can be obtained by solving the equation @xmath267 doing so yields the upper bound @xmath268^{{{2 \alpha}/({2 \alpha + 1})}}$ ] . otherwise , we also have the trivial upper bound @xmath269 , which yields the alternative upper bound @xmath270 . recalling that @xmath244 and combining the pieces yields the claim . notice that this last ( trivial ) bound on @xmath271 implies that condition ( a3 ) is always satisfied for the time - sampling model . we now turn to some consequences for the basis truncation model ( [ eqnbasistrun ] ) . [ cortrunemp ] consider a basis truncation operator @xmath12 in a hilbert space with polynomial-@xmath50 decay . under conditions and @xmath272 , we have @xmath273 \geq1 - \varphi(n , m),\ ] ] where @xmath253 , and @xmath274 . we note that as long as @xmath275 , condition ( a3 ) is satisfied , since @xmath276 . the rest of the proof follows that of corollary [ cortimeemp ] , noting that in the last step we have @xmath98 for the basis truncation model . as mentioned earlier , given the consistency of @xmath277 , the consistency of @xmath106 is closely related to approximation properties of the semi - norm @xmath278 induced by @xmath12 , and in particular how closely it approximates the @xmath2-norm . these approximation - theoretic properties are captured in part by the nullspace width @xmath279 and defect @xmath136 defined earlier in equations ( [ eqndefnphinull ] ) and ( [ eqndefnphidefect ] ) , respectively . in addition to these previously defined quantities , we require bounds on the following global quantity : @xmath280 a general upper bound on this quantity is of the form @xmath281 in fact , it is not hard to show that such a bound exists with @xmath282 and @xmath283 using the decomposition @xmath284 . however , this bound is not sharp . instead , one can show that in most cases of interest , the term @xmath285 is of the order of @xmath279 . there are a variety of conditions that ensure that @xmath285 has this scaling ; we refer the reader to the paper @xcite for a general approach . here we provide a simple sufficient condition , namely , @xmath286 for a positive constant @xmath287 . [ lemremain ] under , bound ( [ equpboundrpremain ] ) holds with @xmath288 and @xmath289 . see appendix b.4 of the supplementary material @xcite for the proof of this claim . in the sequel , we show that the first - order sobolev rkhs satisfies condition ( b1 ) . [ thmfuncerr ] suppose that condition holds , and the approximation - theoretic quantities satisfy the bounds @xmath290 and @xmath291 . then there is a constant @xmath292 such that @xmath293 ^ 2 \bigr\}\ ] ] with the same probability as in theorem [ thmsubspaceerr ] . as with theorem [ thmsubspaceerr ] , this is a generally applicable result , stated in abstract form . by specializing it to different sampling models , we can obtain concrete rates , as illustrated in the following sections . we begin by returning to the case of the time sampling model ( [ eqntimesamp ] ) , where @xmath243 . in this case , condition ( b1 ) needs to be verified by some calculations . for instance , as shown in appendix a of the supplementary material @xcite , in the case of the sobolev kernel with smoothness @xmath53 [ namely , @xmath251 ] , we are guaranteed that ( b1 ) holds with @xmath295 , whenever the samples @xmath296 are chosen uniformly over @xmath297 $ ] ; hence , by lemma [ lemremain ] , @xmath289 . moreover , in the case of uniform sampling , we expect that the nullspace width @xmath279 is upper bounded by @xmath298 , and so will be proportional to @xmath299 in the case of a kernel operator with polynomial-@xmath50 decay . this is verified in @xcite ( up to a logarithmic factor ) for the case of the first - order sobolev kernel . in appendix a of the supplementary material @xcite , we also show that , for this kernel , @xmath300 ^ 2 $ ] is of the order @xmath299 , that is , of the same order as @xmath279 . [ cortimefunc ] consider the basis truncation model ( [ eqnbasistrun ] ) with uniformly spaced samples , and assume condition holds and that @xmath301 ^ 2 \precsim m^{-2\alpha}$ ] . then the @xmath4-estimator returns a subspace estimate @xmath106 such that @xmath302 with the same probability as in corollary [ cortimeemp ] . in this case , there is an interesting trade - off between the bias or approximation error which is of order @xmath303 and the estimation error . an interesting transition occurs at the point when @xmath304 , at which : * the bias term @xmath303 becomes of the order @xmath261 , so that it is no longer dominant , and * for the two terms in the estimation error , we have the ordering @xmath305 consequently , we conclude that the scaling @xmath306 is the minimal number of samples such that we achieve an overall bound of the order @xmath261 in the time - sampling model . in section [ seclower ] , we will see that these rates are minimax - optimal . for the basis truncation operator @xmath12 , we have @xmath307 so that condition ( b1 ) is satisfied trivially with @xmath295 . moreover , lemma [ lemdefect ] implies @xmath308 . in addition , a function @xmath309 satisfies @xmath310 if and only if @xmath311 , so that @xmath312 consequently , we obtain the following corollary of theorem [ thmfuncerr ] : [ corbasistrunfunc ] consider the basis truncation model ( [ eqnbasistrun ] ) with a kernel operator that has polynomial-@xmath50 decaying eigenvalues . then the @xmath4-estimator returns a function subspace estimate @xmath313 such that @xmath314 with the same probability as in corollary [ cortrunemp ] . by comparison to corollary [ cortimefunc ] , we see that the trade - offs between @xmath18 are very different for basis truncation . in particular , there is _ no interaction _ between the number of functional samples @xmath3 and the number of statistical samples @xmath5 . increasing @xmath3 only reduces the approximation error , whereas increasing @xmath5 only reduces the estimation error . moreover , in contrast to the time sampling model of corollary [ cortimefunc ] , it is impossible to achieve the fast rate @xmath261 , regardless of how we choose the pair @xmath257 . in section [ seclower ] , we will also see that the rates given in corollary [ corbasistrunfunc ] are minimax optimal . we now turn to lower bounds on the minimax risk , demonstrating the sharpness of our achievable results in terms of their scaling with @xmath18 . in order to do so , it suffices to consider the simple model with a single functional component @xmath315 , so that we observe @xmath316 for @xmath317 , where @xmath318 are i.i.d . standard normal variates . the minimax risk over the unit ball of the function space @xmath0 in the @xmath12-norm is given by @xmath319 where the function @xmath131 ranges over the unit ball @xmath320 of some hilbert space , and @xmath321 ranges over measurable functions of the data matrix @xmath322 . [ thmminimaxemp ] suppose that the kernel matrix @xmath122 has eigenvalues with polynomial-@xmath50 decay and holds . for the time - sampling model , we have @xmath323 for the frequency - truncation model , with @xmath324 , we have @xmath325 note that part ( a ) of theorem [ thmminimaxemp ] shows that the rates obtained in corollary [ cortimefunc ] for the case of time - sampling are minimax optimal . similarly , comparing part ( b ) of the theorem to corollary [ corbasistrunfunc ] , we conclude that the rates obtained for frequency truncation model are minimax optimal for @xmath326 $ ] . as will become clear momentarily ( as a consequence of our next theorem ) , the case @xmath327 is not of practical interest . we now turn to lower bounds on the minimax risk in the @xmath328 norm namely @xmath329 obtaining lower bounds on this minimax risk requires another approximation property of the norm relative to . consider matrix @xmath330 with entries @xmath331 . since the eigenfunctions are orthogonal in @xmath2 , the deviation of @xmath332 from the identity measures how well the inner product defined by @xmath12 approximates the @xmath2-inner product over the first @xmath3 eigenfunctions of the kernel operator . for proving lower bounds , we require an upper bound of the form @xmath333 for some universal constant @xmath334 . as the proof will clarify , this upper bound is necessary in order that the kullback leibler divergence which controls the relative discriminability between different models can be upper bounded in terms of the @xmath2-norm . [ thmminimaxltwo ] suppose that condition holds , and the operator associated with kernel function @xmath38 of the reproducing kernel hilbert space @xmath0 has eigenvalues with polynomial-@xmath50 decay . for the time - sampling model , the minimax risk is lower bounded as @xmath335 for the frequency - truncation model , the minimax error is lower bounded as @xmath336 verifying condition ( b2 ) requires , in general , some calculations in the case of the time - sampling model . it is verified for uniform time - sampling for the first - order sobolev rkhs in appendix a of the supplementary material @xcite . for the frequency - truncation model , condition ( b2 ) always holds trivially since @xmath337 . by this theorem , the @xmath338 convergence rates of corollaries [ cortimefunc ] and [ corbasistrunfunc ] are minimax optimal . also note that due to the presence of the approximation term @xmath299 in ( [ eqminimaxl2freq ] ) , the @xmath12-norm term @xmath339 is only dominant when @xmath340 implying that this is the interesting regime for theorem [ thmminimaxemp](b ) . we now turn to the proofs of the results involving the error @xmath341 between the estimated @xmath105 and true subspace @xmath109 . we begin by proving theorem [ thmsubspaceerr ] , and then turn to its corollaries . we begin with some preliminaries before proceeding to the heart of the proof . let us first introduce some convenient notation . consider the @xmath342 matrices @xmath343^t \quad\mbox{and}\quad w:={\left}[\matrix { { w_{1 } } & { w_{2 } } & \cdots & { w_{n } } } { \right}]^t,\ ] ] corresponding to the observation matrix @xmath344 and noise matrix @xmath345 , respectively . in addition , we define the matrix @xmath346 , and the diagonal matrix @xmath347 . recalling that @xmath348 , the observation model ( [ eqnlinobs ] ) can be written in the matrix form @xmath349 . moreover , let us define the matrices @xmath350 and @xmath351 . using this notation , some algebra shows that the associated sample covariance @xmath352 can be written in the form @xmath353 where @xmath354 $ ] and @xmath355 . lemma [ lemfeasible ] , proved in appendix b.3 of the supplementary material @xcite , establishes the existence of a matrix @xmath356 such that @xmath357 . as discussed earlier , due to the nature of the steifel manifold , there are many versions of this matrix @xmath358 , and also of any optimal solution matrix @xmath171 , obtained via right multiplication with an orthogonal matrix . for the subsequent arguments , we need to work with a particular version of @xmath358 ( and @xmath171 ) that we describe here . let us fix some convenient versions of @xmath359 and @xmath171 . as a consequence of cs decomposition , as long as @xmath360 , there exist orthogonal matrices @xmath361 and an orthogonal matrix @xmath362 such that @xmath363 where @xmath364 and @xmath365 such that @xmath366 and @xmath367 . see bhatia @xcite , theorem vii.1.8 , for details on this decomposition . in the analysis to follow , we work with @xmath368 and @xmath369 instead of @xmath359 and @xmath171 . to avoid extra notation , from now on , we will use @xmath359 and @xmath171 for these new versions , which we refer to as _ properly aligned_. with this choice , we may assume @xmath370 in the cs decomposition ( [ eqcsdecompconseq ] ) . the following lemma isolates some useful properties of properly aligned subspaces : [ lemmanageable ] let @xmath359 and @xmath171 be properly aligned , and define the matrices @xmath371 in terms of the cs decomposition ( [ eqcsdecompconseq ] ) , we have : @xmath372 from the cs decomposition ( [ eqcsdecompconseq ] ) , we have @xmath373 from which relations ( [ eqpertmidproof2 ] ) and ( [ eqpertmidproof3 ] ) follow . from decomposition ( [ eqcsdecompconseq ] ) and the proper alignment condition @xmath370 , we have @xmath374\\[-8pt ] & = & 2 \sum_{i=1}^r(1 - \widehat{c}_i ) \le2 \sum_{i=1}^r \bigl(1 - \widehat{c}_i^2\bigr ) = 2 \sum _ { i=1}^r\widehat{s}_i^2 = { |\!|\!|}\widehat{p}{|\!|\!|}_{\mathrm{hs}}^2,\nonumber\end{aligned}\ ] ] where we have used the relations @xmath375 , @xmath376 $ ] and @xmath377 . using the notation introduced in lemma [ lemmanageable ] , our goal is to bound the hilbert schmidt norm @xmath378 . without loss of generality we will assume @xmath379 throughout . recalling definition ( [ eqnsamsplit ] ) of the random matrix @xmath380 , the following inequality plays a central role in the proof : [ lemperturbation ] under condition and @xmath381 , we have @xmath382 with probability at least @xmath383 . we use the shorthand notation @xmath384 for the proof . since @xmath359 is feasible and @xmath171 is optimal for problem ( [ eqndiscretemest ] ) , we have the basic inequality @xmath385 . using the decomposition @xmath386 and rearranging yields the inequality @xmath387 from definition ( [ eqnsamsplit ] ) of @xmath388 and @xmath389 , the left - hand side of the inequality ( [ eqpertmidproofinequ ] ) can be lower bounded as @xmath390 where we have used ( 90 ) and ( 91 ) of appendix i several times ( cf . the supplementary material @xcite ) . we note that @xmath391 and @xmath392 provided @xmath393 ; see equation ( 70 ) . to bound the minimum eigenvalue of @xmath394 , let @xmath395 denote the minimum singular value of the @xmath396 gaussian matrix @xmath397 . the following concentration inequality is well known ( cf . @xcite ) : @xmath398 \le \exp\bigl(-t^2/2\bigr)\qquad\mbox{for all $ t > 0$.}\ ] ] since @xmath399 , we have that @xmath400 with probability at least @xmath401 . assuming @xmath402 and setting @xmath403 , we get @xmath404 with probability at least @xmath405 . putting the pieces together yields the claim . inequality ( [ eqnbasic ] ) reduces the problem of bounding @xmath406 to the sub - problem of studying the random variable @xmath407 . based on lemma [ lemperturbation ] , our next step is to establish an inequality ( holding with high probability ) of the form @xmath408 where @xmath409 is some universal constant , @xmath233 is the constant in condition ( a1 ) and @xmath236 is the critical radius from theorem [ thmsubspaceerr ] . doing so is a nontrivial task : both matrices @xmath410 and @xmath380 are random and depend on one another , since the subspace @xmath171 was obtained by optimizing a random function depending on @xmath380 . consequently , our proof of bound ( [ eqninitialbound ] ) involves deriving a uniform law of large numbers for a certain matrix class . suppose that bound ( [ eqninitialbound ] ) holds , and that the subspaces @xmath359 and @xmath171 are properly aligned . lemma [ lemmanageable ] implies that @xmath411 , and since @xmath412 is a nondecreasing function , inequality ( [ eqninitialbound ] ) combined with lemma [ lemperturbation ] implies that @xmath413 from which the claim follows as long as @xmath233 is suitably small ( e.g. , @xmath414 suffices ) . accordingly , in order to complete the proof of theorem [ thmsubspaceerr ] , it remains to prove bound ( [ eqninitialbound ] ) , and the remainder of our work is devoted to this goal . given the linearity of trace , we can bound the terms @xmath415 and @xmath416 separately . let @xmath417 , @xmath418 and @xmath419 and @xmath420 denote the columns of @xmath171 , @xmath359 , @xmath421 and @xmath422 , respectively , where we recall the definitions of these quantities from equation ( [ eqnsamsplit ] ) and lemma [ lemmanageable ] . note that @xmath423 . in appendix c.1 of the supplementary material @xcite , we show that @xmath424 consequently , we need to obtain bounds on quantities of the form @xmath425 , where the vector @xmath426 is either fixed ( e.g. , @xmath427 ) or random ( e.g. , @xmath428 ) . the following lemmas provide us with the requisite bounds : [ lemdeltaoneellipse ] we have @xmath429 with probability at least @xmath430 . [ lemdeltaonefixed ] we have @xmath431 \geq1 - r^2 \exp\bigl(- \kappa^2 r^{-2 } { n}/2 \sigma^2\bigr).\ ] ] see appendices c.2 and c.3 in the supplementary material @xcite for the proofs of these claims . recalling definition ( [ eqnsamsplit ] ) of @xmath432 and using linearity of the trace , we obtain @xmath433 since @xmath434 , we have @xmath435\\[-8pt ] & \leq & \sigma^2 \sum_{j=1}^r \biggl\ { 2 \underbrace{\bigl(\widetilde{z}^*_j\bigr)^t \biggl ( \frac{1}{{n}}w^tw- i_r \biggr ) \widehat{e}_j}_{t_1(\widehat{e}_j ; \widetilde{z}^*_j ) } + \underbrace{\frac{1}{{n } } \| w\widehat{e}_j \|_2 ^ 2}_{t_2(\widehat{e}_j ) } \biggr\},\nonumber\end{aligned}\ ] ] where we have used the fact that @xmath436 = 2\sum_j ( \widehat{c}_j - 1 ) = -{|\!|\!|}\widehat{e}{|\!|\!|}_{\mathrm{hs}}^2 \le 0 $ ] . the following lemmas provide high probability bounds on the terms @xmath437 and @xmath438 . [ lemtermone ] we have the upper bound @xmath439 with probability @xmath440 . [ lemtermtwo ] we have the upper bound @xmath441 with probability at least @xmath442 . see appendices c.4 and c.5 in the supplementary material @xcite for the proofs of these claims . we studied the problem of sampling for functional pca from a functional - theoretic viewpoint . the principal components were assumed to lie in some hilbert subspace @xmath0 of @xmath338 , usually a rkhs , and the sampling operator , a bounded linear map @xmath443 . the observation model was taken to be the output of @xmath12 plus some gaussian noise . the two main examples of @xmath12 considered were time sampling , @xmath444_{j } = f(t_{j})$ ] and ( generalized ) frequency truncation @xmath444_{j } = \langle\psi_{j } , f \rangle_{l^{2}}$ ] . we showed that it is possible to recover the subspace spanned by the original components , by applying a regularized version of pca in @xmath445 followed by simple linear mapping back to function space . the regularization involved the `` trace - smoothness condition '' ( [ eqntracesmooth ] ) based on the matrix @xmath446 whose eigendecay influenced the rate of convergence in @xmath445 . we obtained the rates of convergence for the subspace estimators both in the discrete domain , @xmath445 , and the function domain , @xmath338 . as examples , for the case of a rkhs @xmath0 for which both the kernel integral operator and the kernel matrix @xmath122 have polynomial-@xmath50 eigendecay ( i.e. , @xmath447 ) , the following rates in @xmath448-projection distance for subspaces in the function domain were worked out in detail : the two terms in each rate can be associated , respectively , with the estimation error ( due to noise ) and approximation error ( due to having finite samples of an infinite - dimensional object ) . both rates exhibit a trade - off between the number of statistical samples ( @xmath5 ) and that of functional samples ( @xmath3 ) . the two rates are qualitatively different : the two terms in the time sampling case interact to give an overall fast rate of @xmath261 for the optimal trade - off @xmath449 , while there is no interaction between the two terms in the frequency truncation ; the optimal trade - off gives an overall rate of @xmath450 , a characteristics of nonparametric problems . finally , these rates were shown to be minimax optimal .

we show that fermi repulsion can lead to cored density profiles in dwarf galaxies for sub - kev fermionic dark matter . we treat the dark matter as a quasi - degenerate self - gravitating fermi gas and calculate its density profile assuming hydrostatic equilibrium . we find that suitable dwarf galaxy cores of size pc can be achieved for fermion dark matter with mass in the range 70 ev 400 ev . while in conventional dark matter scenarios , such sub - kev thermal dark matter would be excluded by free streaming bounds , the constraints are ameliorated in models with dark matter at lower temperature than conventional thermal scenarios , such as the flooded dark matter model that we have previously considered . modifying the arguments of tremaine and gunn we derive a conservative lower bound on the mass of fermionic dark matter of 70 ev and a stronger lower bound from lyman clouds of about 470 ev , leading to slightly smaller cores than have been observed . we comment on this result and how the tension is relaxed in dark matter scenarios with non - thermal momentum distributions . cosmology : theory , dark matter , elementary particles galaxies : dwarf .

1611.04590

the cold dark matter paradigm ( @xmath2cdm ) provides a remarkably good description of cosmology and astrophysics . however , observations connected with small scales might be in tension with this framework . in particular , structure formation simulations to date assuming @xmath2cdm predict a greater number of galaxy satellites and suggest that the density profiles of dwarf spheroidal galaxies ( dsphs ) should exhibit cusps , in contrast to observations @xcite . some cores might be explained by baryonic feedback e.g. @xcite , but this is still undecided @xcite , indicating the possibility of a more fundamental discrepancy . it is of interest to look at potential resolutions from dark matter effects beyond @xmath2cdm , such as warm dark matter @xcite , self - interactions @xcite , or boson degeneracy @xcite . an interesting alternative resolution was proposed by @xcite as well as @xcite ( see also @xcite ) , who observed that a sub - kev fermion dark matter particle could resolve the core - cusp problem when quantum pressure due to fermi repulsion provides the dominant support in the central regions of dwarf galaxies . in this scenario , the cores of dwarf galaxies are highly analogous to white dwarfs or neutron stars , both of which are supported by degeneracy pressure . this proposal would have the remarkable implication that the structure of dwarf galaxies arises due to a quantum effect on a cosmic scale . however , dark matter composed of a thermal relic lighter than a few kev would erase small scale density perturbations , in conflict with observations @xcite . thus the prospect of fermi pressure playing a role in the structure of dwarf galaxies is lost for dark matter at the baryonic temperature . however , this scenario might be realized when dark matter is not in thermal contact with baryons so it is cooler when dark matter is decoupled and is not reheated by the standard model , for example . the dark matter would then be non - relativistic earlier than a thermal relic , mitigating free streaming bounds . for models that realize such a scenario see for instance : @xcite another possibility is that the lower limit on the dark matter mass is relaxed by a late - entropy release , as with `` flooded dark matter '' ( fdm ) @xcite . in this class of models the standard model undergoes late - time reheating via entropy injections from long - lived heavy states decaying to the visible sector . we provide a brief outline of the minimal model of flooded dark matter in appendix [ sec : fdm ] . we recast the warm dark matter mass bounds derived from lyman-@xmath1 data into constraints on models such as this one with colder temperature and find the following mass limit for flooded dark matter m_fdm 470 ev n_f^- , in terms of @xmath3 ` flavors ' of majorana fermion dark matter . this value is borderline for resolving the core - cusp problem but might work well with baryonic feedback in explaining the variability in core sizes that is observed so far . the full set of constraints we present is : * lower mass bound from fit to cores : @xcite find a preference for a narrow dark matter mass range @xmath4 ev @xmath5 ev . once we include the effects of a thermal envelope these constraints reduce to @xmath6 ev , with no upper bound on the mass . * upper mass bound from core size : both @xcite and @xcite find a two sigma constraint of @xmath7 parsec , which would imply @xmath8 ev . when we partially redo their analysis with a burkert profile for dark matter we obtain smaller preferred values @xmath9 kpc . our limited analysis does not recover their result but when we fit their data using a fermi core with two scales ( core radius and scale radius ) , we obtain an even smaller core @xmath10 kpc , consistent with a larger fermion mass of @xmath11 ev . we can conclude , based on comparing our one - scale and two - scale results , that that the profile with the fermi - degenerate core ( and hence two scales ) should reduce the lower bound on the core size and increase the upper bound on the mass of the fermion to @xmath12 ev . * lower mass bound from largest allowed core size : @xcite also put an upper bound on core size at @xmath13 level @xmath14 kpc for the fornax dsph which implies @xmath15 ev lower mass bound . * lower mass bounds from black hole formation : we show that @xmath16 ev suffices to avoid forming black holes out of degenerate fermi gas cores of dwarf galaxies . * lyman-@xmath1 : for constant @xmath17 @xcite found @xmath18 kev , corresponding to @xmath19 ev as we will show in eq . [ eq : wdmfdmeq ] . the most conservative , even weaker , bound for warm dark matter was @xmath20 kev , corresponding to @xmath21 ev . we later argue that using a flavored dark matter model we can relax the lyman-@xmath1 bounds arbitrarily , at the cost of more complicated dark matter model . alternatively , while the lyman-@xmath1 constraint ( @xmath21 ev ) is in tension with the mass range for which appropriately large cores are obtained ( @xmath22 ev ) , this bound can be reconciled if the dark matter has a non - thermal momentum distribution . we will discuss one way to achieve such a nonthermal scenario . in any case , the coring mechanism based on fermi degeneracy operates in conjunction with baryonic feedback and it is quite likely that baryonic feedback creates a large core in the fornax galaxy for example . as a result it may be the case that only _ smaller _ cores in systems such as draco need to be resolved by new physics , in which case the lyman-@xmath1 constraints become irrelevant . in what follows , we extend and improve the analysis of @xcite . the earlier study made the simplifying assumption of a completely degenerate fermi gas , which is not appropriate for realistic scenarios . we treat the dark matter as a quasi - degenerate fermi gas surrounded by a thermal envelope . furthermore , we highlight concerns regarding the criteria used in matching to observed data and clarify the condition under which dsphs would be successfully cored through fermi pressure . the structure of this paper is as follows : in section [ sec : fermirep ] we outline the physics of self gravitating fermi gases and the role of fermi repulsion in determining the density profile . we argue that even with sharp power law density profiles dwarf galaxies can not accommodate arbitrarily large central densities and that fermi repulsion is generally important in the central region . in section [ sec : fits ] , using the empirically motivated criterion that dwarf spheroidal galaxies should feature a core of order a few hundred pc , we determine the parameters that yield appropriate sized cores . we point out essential details , such as the importance of stellar anisotropy degeneracy with reference to @xcite . we also supplement with newer data to reproduce observed properties of the classical milky way dwarf galaxies . in section [ sec : lowermassbounds ] , we derive a number of lower bounds on the mass of fermion dark matter . in section [ sec : chandra ] , we discuss the lower bound on the mass of fermion dark matter from the requirement that dwarf cores do not collapse into black holes . appendix [ apb ] presents certain definitions and derivations that we use in the main text and appendix [ apmodels ] gives more details about the models we discuss in this paper . we first argue that in the absence of a central black hole , sufficiently light fermionic dark matter manifests fermi repulsion that would be relevant in the central region of dwarf galaxies . assuming a cuspy profile as a starting point , there would always exist a central region with density high enough that the fermi velocity would exceed the escape velocity so the density in the central region would be inconsistent with the pauli exclusion principle and would have to be reduced . we later show that full solutions including fermi pressure lead to cored density distributions . a fermionic gas of density @xmath23 is degenerate at temperatures below t_deg&~ ( ) ^ , [ eq : deg ] where @xmath24 is the mass of the fermion . in this regime the highest occupation level is @xmath25 ( see e.g. @xcite ) p_f = m v_f = h ( ) ^ , leading to the fermi pressure in the non - relativistic limit : p_f=_0^p_fdp ( ) = ( ) ^ ^ , [ eq : pf ] which at high density and low temperature dominates over the classical pressure ( we choose units such that @xmath26 ) p_cl= ( ) . [ eq : cl ] a typical temperature at any point inside this ball is determined by the virial condition t(r)=. therefore , the classical pressure can be estimated to be p_cl=. [ eq : cl2 ] when @xmath27 the classical pressure term dominates . the pressure of the gas is then @xmath28 we now argue that systems described by integrable power - laws will generically become quasi - degenerate . it is useful to parameterize the density as = _ 0()^n . [ den ] integrable profiles satisfy the condition > -3 r0 . [ eqr ] we claim that for an integrable density profile one can always find a radius below which the fermi velocity is larger than escape velocity @xmath29 , leading to an inconsistent density profile , because states with @xmath30 will escape the gravitational potential , thus reducing the central density . hence , the system would dynamically relax through evaporative cooling to a quasi - degenerate fermi gas even from an initially non - degenerate configuration . observe that eq . ( [ den ] ) implies a mass profile of the form m(r)=4_0^r r^2(r ) dr = 4()r^3-n . [ n<3 ] it follows that the virial velocity can be written v_vir^2 4 g r^2-n ( ) . [ vir ] for the given local density @xmath31 , the fermi surface lies at p_f= ( ) ^ = ( ) ^ , which determines the fermi velocity v_f = p_f / m ^. there is some critical radius @xmath32 for which @xmath33 . this roughly coincides with the point at which @xmath34 . for an integrable profile ( @xmath35 ) r_*^(n)=r_0^n/(6-n ) ( ) ^1/(6-n ) . [ rc0 ] we trade the unknown parameters @xmath36 and @xmath37 for the measured mass @xmath38 within a half - light radius @xmath39 , defined by 4()= , [ rho-0 ] in order to obtain r_*^(n)=^1/(6-n ) . [ rc1 ] degeneracy effects are important in the region @xmath40 . for example for @xmath41 : r_*^(2 ) 160 pc()^2 ( ) ^ ( ) ^. similarly for @xmath42 : r_*^(0)=130 pc()^4/3 ( ) ^ ( ) ^. [ rc2 ] more generally , for @xmath43 solving consistently gives a core of @xmath44 pc for dark matter masses of a few hundred ev , for typical values of @xmath38 and @xmath39 . having shown that a integrable profile always leads to fermi degeneracy ( for a sufficiently small mass - dependent region ) , we now consider non - integrable profiles ( which do not satisfy the inequality of eq . ( [ eqr ] ) ) . since these profiles lead to an unphysical infinite total mass in the center of the distribution they must be cut off at some small radius @xmath45 . below the cutoff @xmath46 the profile transitions to an integrable profile . rather than focus on the outer region , we consider the central cutoff region , which we parameterize solely by the mass @xmath47 enclosed and the cut - off radius @xmath46 . we determine the minimum radius required to fit a typical dwarf galaxy mass inside a given size as a function of the fermion mass , where we require that the escape velocity is larger than the fermi velocity everywhere inside the gas . this happens for v_esc^2 = > v_f^2 = ( 3 ^ 2 ) ^ . the least stringent requirement is found when the matter is evenly distributed over the whole region , generating the lowest maximum density , in which case the system must satisfy : > ( ) ^ . the above inequality implies a lower bound on the cut off & > ( ) ^ + & 180 ( ) ^ ( ) ^ . therefore for light fermion masses , @xmath46 needs to be similar to , or greater than , the dwarf galaxy size . in other words , the system should be described by an integrable profile over the whole range . one might suppose that rather than a soft profile below the cutoff of a sharp profile one could encounter a sequence of nested singular profiles . however , ultimately this sequence must be cut off at some @xmath45 at which the profile becomes non - singular , and the argument above still applies . having reduced to the previous case , we conclude that a fermi degenerate region must always be present in a physically relevant region . we next solve the hydrostatic equilibrium equations explicitly to derive the density profile accounting for fermi degeneracy to show that the result is a cored central region . we are interested in determining the density profile of a self - gravitating ball of fermions . such a static density profile has to satisfy three equations : + _ i ) _ hydrostatic equilibrium & = -()(r ) , [ eq : hydroeq1 ] pressure equality p(r ) & = p_cl+ p_f + & + ( ) ^ ( r)^ , [ eq : hydroeq2 ] the continuity condition m(r ) & = 4_0^r ( r ) r^2 dr . [ eq : hydroeq3 ] equations ( [ eq : hydroeq1])-([eq : hydroeq3 ] ) with initial conditions @xmath48 , and @xmath49 form a closed system that can be solved to obtain a self - consistent static density profile of a gas of fermions . first we construct solutions supported by degeneracy pressure alone , by setting @xmath50 . in this case ( [ eq : hydroeq1 ] ) , ( [ eq : hydroeq2 ] ) & ( [ eq : hydroeq3 ] ) reduce to ( ) ^ ( ) = -4 g r^2 ( r ) . [ eq : fermialone ] note that for small @xmath51 an approximate solution is given by @xmath52 $ ] , for @xmath53 an appropriately chosen scale r_a^2 = ( ) ^ , [ ra ] in agreement with the approximate analysis in the previous section . we can also see that this scale arise in eq . ( [ rc0 ] ) for the choice @xmath42 . this indicates we should expect a constant density core of size @xmath53 as fermi pressure flattens the expected cusps of density distributions of dwarf galaxies . ev and central density @xmath54 . this implies @xmath55pc ( vertical line ) , the critical radius at which the associated fully degenerate profile terminates , as given in eq . ( [ eq : scalingr0 ] ) . note that at @xmath56 the contribution of each pressure component is significant . [ fig : pclpf],scaledwidth=45.0% ] we now look at the full solution . ( [ eq : fermialone ] ) is an example of a lane - emden equation with index @xmath57 . the solutions to this differential equation are polytropes with index @xmath58 , see e.g. @xcite . indeed , these solutions are constant as @xmath59 , acting as cored profiles . moreover , the density profile of a fully degenerate fermi gas has finite extent with density vanishing at r_0 ^ 2 = _ 1 ^ 2 ( ) ^ , [ eq : scalingr0 ] where @xmath60 is a numerical constant . figure [ fig : pclpf ] illustrates the behavior of the two pressure components for an example with dark matter mass of @xmath61 ev and central density @xmath62 . with these parameter values , @xmath63 ( taking @xmath64 ) . we see that close to @xmath53 , the classical pressure is as important as the fermi pressure . to increase the range of validity of our density profiles we investigate the full solutions to eq . ( [ eq : hydroeq1])([eq : hydroeq3 ] ) including both @xmath65 and @xmath66 . figure [ fig : rhovsr ] shows a set of solutions to these equations including both classical and degeneracy pressure ( we fix the central densities of these solutions such that they reproduce the @xmath38 for the fornax dsph ) . these solutions are constant for @xmath67 just as those supported by the fermi pressure alone . however , the full solutions do not vanish at some finite @xmath51 . in fact @xmath68 is a solution for @xmath69 as the classical pressure term takes over . there is a characteristic length scale associated with the transition from constant density core to isothermal behavior . we can define this scale as the radius @xmath70 at which the @xmath71 . we choose to define the core radius @xmath72 as a radius at which the slope of the density distribution reaches a particular value and we use the definition of @xcite : .|_r_c=- . [ 32 ] this definition allows us to compare our results with the results of others in a model - independent way . notice that @xmath73 , and similarly @xmath74 . this is not surprising as there is only one relevant combination of @xmath75 and @xmath36 with dimensions of length namely eq . ( [ eq : scalingr0 ] ) . for the fornax dsph . observe for 2 kev fermions the profile is decidedly cuspy , whereas the profiles for 100 ev and 200 ev dark matter exhibits flat cores of a few hundred parsecs . the dashed curves correspond to solutions to eq . ( [ eq : fermialone ] ) , and give the distributions corresponding to fully degenerate fermi gases . [ fig : rhovsr],scaledwidth=45.0% ] in the previous section we have shown that including fermi pressure in the hydrostatic equations for a self - gravitating gas of light fermions flattens their density profiles to constant density cores with mass - dependent characteristic sizes . we also showed that including the classical pressure term extends otherwise finite extent solutions to include isothermal tails . in this section we fit these quasi - degenerate profiles to the kinematic properties of the eight classical dwarf galaxies : carina , draco , fornax , leo i , leo ii , sculptor , sextans , and ursa minor , see e.g. @xcite . we first consider the kinematic variables studied in @xcite and demonstrate that due to a degeneracy this method does not suffice to determine the core sizes . one of the main observables that is used to obtain kinematic information is the projected velocity dispersion along the line - of - sight ( los ) @xmath76 , as defined in @xcite : _ los^2(r)=_r^(r)m(r)(r)^2 - 2f(,r , r ) dr , [ slos]where @xmath77 is the projected stellar density , and @xmath78 is the associated 3d stellar density . following e.g. @xcite , we use the plummer profile @xcite , and we give the forms of @xmath77 , @xmath78 and @xmath79 in appendix [ apb ] . as an example , we show the data reported in @xcite for the classical dwarf galaxy fornax in figure [ fig : sigmalos ] . importantly , fits to @xmath76 exhibit a well known degeneracy between values of @xmath80 the stellar anisotropy and deformations in the density profile . the dependence of @xmath76 on the stellar orbital anisotropy @xmath80 arises because we can observe only the line of sight velocities which forces us to project the two different velocity components @xmath81 onto the line of sight : v_los = v_r -v _ . in order to recover the lost information one typically assumes there is some overall relationship between @xmath82 and @xmath83 parametrized by @xmath80 as = 1- . figure [ fig : sigmalos ] gives the observed @xmath76 for fornax , and we can compare this to the prediction for a cored profile ( e.g. 200 ev fermion dark matter ) and also to a decidedly cuspy profile ( e.g. 2 kev fermion dark matter ) . as can be seen both are in good agreement with the data and so @xmath76 does not discriminate between cored and cuspy profiles . of fornax as reported in @xcite . the red curve is the predicted @xmath76 for a cored density profile ( @xmath84 pc ) due to 200 ev fermion dark matter , corresponding to our hydrostatic solution ( rsu ) with fermi gas . in green is a cuspy profile ( @xmath85 pc ) assuming 2 kev fermion dark matter . alongside , in black , we display the result of @xcite for 200 ev dark matter which assumes the fermi gas is fully degenerate ( d&u ) and in brown we show a fully degenerate gas with dark matter mass 2 kev , which is very cuspy.,scaledwidth=45.0% ] this issue is not apparent in @xcite because the analysis considers only fully degenerate fermi distributions . the finite - extent step - like density profile of a degenerate fermi - gas would predict decreasing @xmath76 outside the core region . since the data indicates a nearly constant @xmath76 over the entire observed dwarf , the fully degenerate model is best fit with large core radii . this gives the misleading impression that the core - cusp problem is manifested in the @xmath76 data as can be seen in the @xmath86 analysis in figure [ fig : chisqvsm ] . once the thermal tail is included , the predicted @xmath76 no longer drops outside of the core region ( @xmath87 profiles have flat rotation curves ) and the cores are allowed to be nearly arbitrarily small ( or equivalently the profile arbitrarily cuspy ) by the data . we conclude that the observed @xmath76 can be fit equally well by cuspy and cored profiles by adjusting @xmath80 @xcite in the range @xmath88 . in figure [ fig : chisqvsm ] we display in blue the @xmath86 per degree of freedom for fits to the eight classical dwarf galaxies as the dark matter mass is varied , marginalizing over @xmath80 and @xmath36 . our fits suggest that dark matter with mass @xmath89 matches the kinematic data very well . the flatness of the @xmath86 above @xmath90 ev in figure [ fig : chisqvsm ] indicates that the data does not definitely establish existence of cores due to the @xmath80 degeneracy inherent in fitting to @xmath76 data . in figure [ fig : rhocvsrc ] we show the fitted core densities and core sizes of the eight classical dwarfs that arise as our best fit parameters for different dark matter masses from the @xmath76 data of @xcite . new approaches circumvent the @xmath80-degeneracy @xcite and can be used to more robustly determine core sizes of dwarf spheroidals in which star subpopulations can be separated . using these more recent results we show that there is a range of fermion masses with cores of the right order of magnitude and which fit the stellar kinematics . the approach of @xcite is to split the tracer star population chemically into subpopulations and extract the mass enclosed at radii at which the uncertainties due to the stellar orbital anisotropy are the smallest for each subpopulation . the resulting mass profile is only known at two or three radial points , since the subpopulations have different spatial extent , but is nearly insensitive to @xmath80 . studies utilizing this approach have confirmed the presence of flat cores in fornax and sculptor and are in broad agreement @xcite . @xcite consider a cored and a cuspy density profile to argue that the dsphs fornax and sculptor require core radii of order a few hundred pc at @xmath9195% cl . to match observations we will therefore require a constant density core of magnitude r_c~few100 pc . similarly @xcite find that at @xmath13 confidence level @xmath92 for the fornax dsph . for fits to data of amorisco the red bands bounded by the dashed ( solid ) contours show the 1@xmath93 ( 2@xmath93 ) fits to with a burkert profile with a single scale parameter @xmath72 ( and is therefore independent of @xmath94 ) . the blue bands correspond to a fit with a profile from eq . ( [ eq : twosc ] ) . the grey area corresponds to parameter space @xmath95 . although not in principle invalid , these solutions do not correspond to quasi - degenerate fermi gas . [ fig : twoscale],scaledwidth=40.0% ] for fits to data of amorisco et al . the red bands bounded by the dashed ( solid ) contours show the 1@xmath93 ( 2@xmath93 ) fits to with a burkert profile with a single scale parameter @xmath72 ( and is therefore independent of @xmath94 ) . the blue bands correspond to a fit with a profile from eq . ( [ eq : twosc ] ) . the grey area corresponds to parameter space @xmath95 . although not in principle invalid , these solutions do not correspond to quasi - degenerate fermi gas . [ fig : twoscale],scaledwidth=40.0% ] for fits to data of amorisco et al . the red bands bounded by the dashed ( solid ) contours show the 1@xmath93 ( 2@xmath93 ) fits to with a burkert profile with a single scale parameter @xmath72 ( and is therefore independent of @xmath94 ) . the blue bands correspond to a fit with a profile from eq . ( [ eq : twosc ] ) . the grey area corresponds to parameter space @xmath95 . although not in principle invalid , these solutions do not correspond to quasi - degenerate fermi gas . [ fig : twoscale],scaledwidth=42.0% ] however , both @xcite and @xcite considered profiles that have a steep drop - off in density ( @xmath96 ) outside their core . however , the solution to the hydrostatic equilibrium equation that we derived has an isothermal envelope that behaves like @xmath97 ( with presumably a steeper drop - off further out ) . since both burkert and nfw profiles use only one scale radius , fitting the kinematic properties of the outer parts of the galaxy makes the core radii of @xcite larger than those obtained using an isothermal profile . the fit is sensitive to both the physics responsible for the core and the physics responsible for the asymptotic behavior of the dwarf galaxy . with only one variable to fit , the derived core radii are some sort of geometrical averages of these two scales . because our profile includes an explicit core , our solution requires that we separate the two scales . in order to make this clear we fit the @xmath98 data for three distinct stellar subpopulations from @xcite to a two scale profile ( with one more scale parameter than nfw or burkert profiles ) of the form : _ 2(r ) = . [ eq : twosc ] this profile behaves like a cored profile up to @xmath99 , transitions into an isothermal profile and then asymptotes to a @xmath100 profile when @xmath101 . we used the velocity dispersions of three sub - populations of the fornax galaxy from figure 1 of @xcite to constrain a two scale dark matter density profile in fornax . we fit the data in a following way : for each combination of @xmath102 we sample the four dimensional space of proposed central density @xmath36 of the dark matter density profile and three independent velocity anisotropies of each star sub - population @xmath103 . we marginalized over this four - dimensional parameter space ( central density , three independent @xmath103 for each sub - population ) to obtain the best @xmath86 for a given @xmath104 combination . our fit indicates that allowing for a second scale relaxes the lower bound on the core radius to below 300 pc as can be seen in figure [ fig : twoscale ] . however , when we redo the amorisco _ et al . _ analysis for a burkert profile , we do not reproduce their results : our fit of the burkert profile leads to @xmath105 pc instead of @xmath106 pc from @xcite . we are unable to resolve this discrepancy without a careful reanalysis of the original data . our incomplete reanalysis of amorisco _ et al . _ would lead to a stronger bound which is however less reliable than the original bound . however , based on comparing our one - scale and two - scale results , the two - scale profile tends to reduce the lower bound on core size compared to the burkert profile by about a factor of @xmath107 cf . figure [ fig : twoscale ] . applying this factor , we assume that a full analysis with a two - scale profile would lead to reduced lower bound on the core size somewhere in the proximity of @xmath108 pc @xmath109 pc . this lower bound on core size would correspond to a reduced upper bound of @xmath110 ev . we illustrate these results in figure [ fig : fornax ] . this plot shows our prediction for the core radius in the fornax dsph as function of @xmath24 , by demanding that the profile reproduces the observed half - light mass @xmath111 as given in @xcite . the blue band indicates the core size preferred by @xcite ( @xmath112 ) . should these be relaxed by a two scale fit , we also mark a weaker condition @xmath113 by a green band . our fits to the core radius data of @xcite favor masses of 70 ev 320 ev and extend to 400 ev once we consider the effect of a two - scale fit . some variation in cores size is expected due to differences between dwarfs , e.g. fornax has stellar mass @xmath114 @xcite , while sculptor has @xmath115 @xcite . variations in the core size among dwarfs in this model is a result of differences in their central densities . although fermi repulsion might improve the density profiles of light fermionic dark matter , too low a mass would lead to overly large cored profiles that would disagree with observed structures of dwarf galaxies . too light dark matter would also prevent the formation of structure on small mass scales , leading to disagreement with lyman-@xmath1 constraints on the matter power spectrum . this latter bound depends on the relative temperature of the dark matter , which leads to model dependence . we now derive both bounds . we focus on the dwarf galaxy fornax , for which the core radius has been measured ( albeit with sizable uncertainty ) by @xcite : r_c^fornax = 1^+0.8_-0.4 kpc . fermion masses that would produce cores in excess of @xmath116 kpc are ruled out at 95% cl by this measurement . to the fornax dwarf galaxy , we show the necessary dark matter mass @xmath24 as a function of the desired core radius @xmath72 . the blue part of the curve indicates the preferred set of @xmath72 from @xcite . the green portion of the curve shows a weaker condition @xmath117 .[fig : fornax],scaledwidth=45.0% ] if the density profile remains flat to large radius , the central density of fornax can be self - consistently determined from the half - light radius @xmath118 pc ( note that @xmath119 ) and the mass this encloses is @xmath120 @xcite . taking the central values gives _ 0= 310 ^ -21 kg / m^3 . with this central density , this scenario predicts a core in excess of 2.6 kpc for dark matter mass @xmath121 , cf . fig [ fig : fornax ] . as a result we obtain a bound m70 ev . this argument is similar to the reasoning underlying the tremaine - gunn bound @xcite , in which the authors used an isothermal profile in the milky way galaxy and other structures to put a lower mass bound on dark matter based solely on phase space density considerations , and a slightly weaker bound based on fermi repulsion . they too obtained their bound by observing that with too small a mass for the dark matter particle , objects of a given total mass would exceed their observed sizes ( this statement can be re - expressed in terms of velocity dispersions ) . here we use the cored profile given by fermi repulsion and find that the radius of fornax would exceed its measured value unless a minimum mass bound applies . future observations should provide a better understanding of @xmath72 , @xmath39 and @xmath38 in fornax and other dwarf spheroidals , which will in turn improve the lower mass bound for fermion dark matter . small scale structure growth also gives lower bounds on the dark matter mass . the growth of over - densities in the matter distribution is suppressed when dark matter is relativistic . as a result , if dark matter is too light and relativistic during late cosmic evolution , small scale structure over - densities ( the matter power spectrum for large @xmath122 ) would be suppressed compared to the standard @xmath2cdm scenario . but we know that the matter power spectrum @xmath123 is not suppressed at large @xmath124 because we observe absorption lines in the corresponding spectra of quasars due to small hydrogen clouds in their lines of sight the so called lyman-@xmath1 forests . existence of these lines has been interpreted as a lower bound on the mass of dark matter by studies such as @xcite . the current bounds on dark matter mass comes from @xcite who require that conventional thermal dark matter has to have and @xmath125 constraints @xcite obtain a limit @xmath126 kev . after including additional datasets ( lyman-@xmath1 + planck ( tt + lowp+ te + ee ) + bao ) they set two further bounds , depending on whether running of @xmath17 is incorporated or not , these are @xmath127 kev and @xmath128 kev , receptively . here we adopt the most conservative bound from @xcite @xmath129 kev ] @xmath130 kev at @xmath131 cl . we note that such bounds depend on the dark matter temperature , which is critical to determining which states are relativistic when the lyman-@xmath1 producing clouds are formed . allowing for dark matter that is colder than the standard model sector and that becomes non - relativistic earlier softens the bounds of @xcite . dark matter that freezes out relativistically can have temperature lower than that of the standard model , such as with flooded dark matter @xcite , which we describe in appendix [ sec : fdm ] . here we give a bound on any dark matter sector , assuming only that the sector is thermal with temperature sufficiently high to allow for the correct energy density . even weaker bounds can apply for dark matter with a nonthermal distribution , as we discuss later in this section . in order to map the bounds on warm dark matter onto bounds on models such as flooded dark matter we need to satisfy two conditions . first we require that the flooded dark matter candidate reproduces the observed relic density . this fixes a relationship between the dark matter temperature @xmath132 , mass @xmath24 and degrees of freedom @xmath133 as indicated in appendix [ sec : fdm ] : = ( ) ^1/3 , [ eq : tdm ] with @xmath134 the nucleon mass , and @xmath135 . a second condition states that two dark matter models will have the same effect on the power spectrum if their effective number of relativistic degrees of freedom is the same . once we find a combination of mass @xmath24 and number of degrees of freedom @xmath133 such that we get the correct relic density and effective number of relativistic degrees of freedom identical to that of warm dark matter with mass @xmath136 , we can determine whether it is ruled out by @xcite . the effective number of relativistic degrees of freedom at a given @xmath137 is @xmath138 where @xmath139 is the energy density of neutrinos and @xmath140 is the pressure of the dark matter component defined by @xmath141 to translate the bound of @xcite , we compare the contribution to the number of relativistic degrees of freedom from a standard warm dark matter model with mass @xmath142 at temperature @xmath143 and another dark matter candidate with mass @xmath24 and temperature @xmath144 given by eq . ( [ eq : tdm ] ) . because each model is described by two scales ( temperature and mass ) and the overall density is fixed by @xmath145 , the ratio of these scales fully characterizes these models . as long as these ratios are equal , the models have the same @xmath146 . the equality @xmath147 together with the expression for @xmath148 from eq . ( [ eq : tdm ] ) allows us to establish a correspondence between thermal dark matter and a dark matter model with a temperature given by ( [ eq : tdm ] ) . using equations ( [ eq : tdm ] ) , ( [ eq : eq ] ) and that @xmath149 we can show that to get an equivalent effect on lyman-@xmath1 power spectrum from our model would require m & = ( m_wdm^3 m_b)^ ( ) ^ + & = 235 ( ) ^ ( ) ^ . [ eq : wdmfdmeq ] we have verified the statement of eq . ( [ eq : wdmfdmeq ] ) by computing the matter power spectra for both models using the class software @xcite for a range of equivalent masses . as a result , the @xmath150 kev lower mass bound for conventional thermal dark matter @xcite is weakened : m _ > 470 ( ) ^ . [ eq : eqgdm ] observe that the lyman-@xmath1 limit is even weaker for larger @xmath151 , i.e. for dark matter with more flavors . table [ tab : masses ] shows the explicit bounds as a function of number of flavors of @xmath152 and for two different bounds on warm dark matter from @xcite : the middle column shows a bound assuming @xmath153 , the right column shows a bound for @xmath154 . note that the fits of @xcite and @xcite favor masses of 70 ev 320 ev ( 400 ev in case we include the two - scale effect ) . this mass range is in tension with the lyman-@xmath1 bound , and thus resolving the core - cusp problem might require additional effects , such as baryonic feedback . there is , however , a caveat to this conclusion , as we discuss next . .minimum fdm mass as a function of number of flavors of mass degenerate majorana fermions . the middle column shows a bound assuming @xmath153 , the right column shows bound for @xmath154 [ cols="^,^,^,^",options="header " , ] the lyman-@xmath1 bounds of table [ tab : masses ] are model independent lower bounds on dark matter with a thermal momentum distribution , dependent only on @xmath155 . the reason for this is that one needs to match the relic density , this is determined by eq . ( [ eq : tdm ] ) via matching to late time cosmological observables . although lyman-@xmath1 bounds can be relaxed if @xmath155 is large , cf . ( [ eq : eqgdm ] ) , increasing the internal number of degrees of freedom of the dark matter simultaneously increases the available occupation levels in the fermi gas and therefore shrinks the fermi gas core size ( cf . ( [ eq : scalingr0 ] ) ) . changing the number of flavors of dark matter weakens the bound by @xmath156 , but the core radius also scales as @xmath157 . notably , these effects scale identically . thus for larger @xmath155 one needs increasingly lighter dark matter to maintain appreciable cores , and larger cores ( @xmath158 kpc ) remain in tension with lyman-@xmath1 . this conclusion can be circumvented if the dark matter momentum distribution is not thermal , but skewed to lower energies . in this case the bounds from lyman-@xmath1 can be relaxed , potentially permitting lighter dark matter , and hence larger cores . skewed momenta might arise in models with resonantly produced sterile neutrinos @xcite or preheating @xcite ) . in an appendix we also consider an alternative scenario in which the dark matter has fewer effective degrees of freedom later in the cosmological evolution than when it was first produced . the model has @xmath3 flavors but also has small mass splittings that lead the heavier dark matter particles to decay without heating the remaining population . as a result of the large @xmath3 , the temperature of the hidden sector can be lower while maintaining the relic dm density and at the same time appear to have fewer degrees of freedom during core formation and therefore lead to larger cores . if we choose @xmath159 , we can relax the lyman-@xmath1 mass bound by a factor of two down to @xmath160 ev and completely remove the tension between core sizes and lyman-@xmath1 mass constraints . for more details see appendix [ sec : skewed ] . as one final consideration we look at core collapse into a black hole , and find that it does not further constrain our system . solving the lane - emden equation in the ultra - relativistic limit one obtains the chandrasekhar bound for a given dark matter mass , as derived in @xcite m_ch=8 ^ 2 ^0_3 = 510 ^ 18()^2 m _ , where @xmath161 is a numerical constant . without the thermal envelope the solutions are similar to white dwarfs or neutron stars . these objects expel most of their outer layers during their formation and look like the solutions of @xcite . the condition for the core of a dwarf galaxy of degenerate fermi gas to collapse into a black hole is that the degenerate core is more massive than the chandrasekhar limit for a given fermion mass . the mass enclosed in the core is m_c~_0 r_a^3~ ( ) , where a typical dwarf galaxy central density is of order @xmath162 . a dwarf galaxy core collapses into a black hole if @xmath163 , i.e. when m _ 0 ^ 1/4 0.1 . however , for @xmath164 ev , the mass of the core , @xmath165 , exceeds the mass of any galaxy and therefore dwarf galaxies with degenerate fermion gas cores are never in danger of collapsing into black holes without additional mechanisms . a ( quasi ) degenerate fermi gas in dwarf galaxies can alter small scale structure formation . this can provide an elegant and minimal solution to the core - cusp problem . using the criterion that cores of order a few 100 pc are required to reproduce the observations of dwarf galaxies , we have shown that suitable dwarf galaxy cores can be achieved for fermion dark matter with mass in the range 70 ev ultralight fermions arise in many motivated contexts , such as gravitinos , or sterile neutrinos . modeling the dark matter as a quasi degenerate fermi gas provides a well defined profile , eq . ( [ eq : scalingr0 ] ) for describing the density distribution of dwarf galaxies , as illustrated in figure [ fig : pclpf ] . here we have proposed a two - scale profile for general modeling of dwarf galaxy density profiles . we have argued that the two - scale profile defined in eq . ( [ eq : twosc ] ) may provide more physically motivated fits . this is because with only a single variable scale , as in burkert and nfw profiles , fits to the dwarf galaxy profile are sensitive to the physics responsible for the core and the asymptotic behavior at large radial distances . since the physics of the asymptotic behavior and the coring are likely unrelated it is important to treat these two length scales as independent . employing the burkert profile , it was shown in @xcite that both large and dwarf galaxies obey a scaling relationship @xmath166 , see also @xcite . the assumption of a quasi - degenerate fermi gas profile automatically implies a relationship between the characteristic core size @xmath74 . from eq . ( [ eq : scalingr0 ] ) it follows that @xmath167 . interestingly , for dark matter in the mass range to give realistic core sizes , the scaling relationship identified by @xcite can be accommodated . this scaling relationship is illustrated in figure [ fig : rhocvsrc ] . while sub - kev dark matter is typically excluded for a thermal relic , it is a viable possibility for non - thermal dark matter models @xcite , or flooded dark matter @xcite . dark matter can become non - relativistic earlier in these models which would weaken the limits from small scale structure observations compared to thermal relics @xcite . moreover , we argued that if the dark matter momentum distribution is non - thermal and skewed to lower energies , then the lyman-@xmath1 bounds are relaxed . in the case that free streaming bounds can be ignored , the leading conservative lower bound is @xmath168 ev , derived here in section [ sec : lowermassbounds ] . we have focused on the case that fermi repulsion is entirely responsible for coring dwarf galaxies , and resolving the core - cusp problem . however in principle the cores in dwarfs could emerge due to a combination of fermi repulsion and baryonic feedback . in the case that baryonic feedback plays a role , this might allow for appreciable cores for somewhat heavier dark matter . it would be interesting to combine these effects in a numerical simulation . we are grateful to p. agrawal , n. amorisco , j. binney , f. y. cyr - racine , m. reece , a. urbano , m. walker , e. witten , and i. yavin for useful discussions . this work was supported in part by nsf grants phy-0855591 , phy-1216270 , and phy-1415548 , and the fundamental laws initiative of the harvard center for the fundamental laws of nature . for completeness we give here the profiles and functions used in the definition of @xmath76 in eq . ( [ slos ] ) . the plummer profile @xcite for the projected stellar density @xmath77 is defined , in terms of the total luminosity @xmath169 , as follows i(r)=^-2 ( ) . [ ir ] given eq . ( [ ir ] ) the associated 3d density is @xcite ( r)=^- ( ) . the @xmath80-dependent function which appears in eq . ( [ slos ] ) is f(,r , r)=_r^r dr . cosmological bounds on light dark matter can be relaxed if the dark matter sector is colder than the standard model sector . this scenario arises in flooded dark matter @xcite but includes other scenarios as well . we first derive a bound on the temperature ratio that allows for thermal populations to give the correct energy and number densities that applies to any model . we allow for decoupling but assume that dark matter is nonetheless of thermal origin and its density is @xmath170 , where @xmath148 is its temperature . the density of dark matter today is then _ = m_n_= m_g_t_,0 ^ 3 , where @xmath171 denotes the temperature today . similarly , for the standard model sector _ b = m_n g_t_,0 ^ 3 . here @xmath134 is the mass of a nucleon and @xmath172 is the baryon asymmetry . assuming dark matter froze out while relativistic , the ratios of temperatures of the dark and standard model sectors must be related by : = ( ) ^1/3 . [ eq : tdmtsm ] for dark matter masses above an ev , the dark matter sector can be colder than the standard model , in which case the lyman-@xmath1 bound on the dark matter mass would be weaker . this relation applies to any model with a thermal population of dark matter . to avoid this conclusion requires a nonthermal distribution . this is also the conclusion in the flooded dark matter scenario @xcite . suppose the dark matter constitutes a hidden sector which is essentially decoupled from the visible sector , and that after inflation these sectors are reheated democratically . to reproduce standard cosmology at late time the visible sector must typically receive further entropy injections . a long lived heavy state @xmath173 which decays only to the visible sector can provide an appropriate source of entropy production . the evolution of the light states and @xmath173 are distinct . @xmath173 becomes non - relativistic at early time , at which point its energy density @xmath174 redshifts as matter . provided @xmath173 is sufficiently long - lived , it will come to dominate the energy density of the universe . when @xmath173 decays its energy density is transferred to the standard model states . the decay rate required to match the observed dark matter relic density ( @xmath175 ) is ~ ( ) ^2 . [ eq : ga-0 ] the entropy injection from @xmath173 decays heat up the visible sector and sets the temperature ratio as required by eq . [ eq : tdmtsm ] . diluting the dark matter sufficiently requires small @xmath173 decay rates , as dictated by eq . ( [ eq : ga-0 ] ) . however successful big bang nucleosynthesis sets a lower bound on the reheat temperature @xmath176 . moreover , certain baryogenesis scenarios need higher @xmath177 , in particular mechanisms tied to the electroweak phase transition require @xmath178 gev . see e.g. @xcite for models and constraints . consider @xmath3 dark matter ` flavors ' of 200 - 300 ev dark matter with a thermal momentum distribution at @xmath179 . to avoid the lyman-@xmath1 limit one needs @xmath180 ( cf . table [ tab : masses ] ) . even lighter dark matter can be accommodated with larger @xmath3 . moderate values of @xmath3 could arise due to an adjoint representation of some broken su(@xmath181 ) , and could be connected to ideas of ` flavored dark matter ' @xcite . we envisage that the a weakly gauged dark sector gauge group breaks , yielding a low energy theory consisting of @xmath3 closely spaced fermions with small mass splittings @xmath182 . if the gauge symmetry is broken by high scale operators , cutoff at @xmath2 , then the induced mass splittings are parametrically _ ij~ , where @xmath183 is the lightest state and @xmath184 is the gauge coupling . eventually the ( slightly ) heavier states @xmath185 decay to light hidden sector states @xmath186 , removing the @xmath185 population . in order to relax the lyman-@xmath1 bounds , @xmath185 decays should occur after the dark matter decouples from the lighter species . otherwise the @xmath183 would receive a compensating energy contribution from the @xmath185 decays through interactions with the @xmath186-bath . as a result there is a reduction of the number of degrees of freedom , but due to the small mass splitting , there is no appreciable heating of the @xmath183 population . accordingly the momentum distribution of the @xmath183 states appears with a temperature which is scaled relative to the thermal expectation by a factor of @xmath187 . as indicated above , in order for the heavier @xmath185 states to decay the model must include light or massless hidden sector states @xmath186 . the energy injected into these radiation states will subsequently redshift away . since the hidden sector is cold relative to the visible sector the @xmath186 states will typically not give an appreciable contribution to the number of relativistic species @xmath189 @xcite . generally , this decay should occur with a small coupling to avoid a ) thermalization and hence reheating of the remaining light states , b ) freeze - out , and c ) self - scattering effects which are not ruled out but would greatly alter our core - cusp analysis . specifically , for too large @xmath184 , dark matter annihilations can be significant and there can be a stage of freeze - out dynamics which would impact the dark matter relic density . therefore to avoid @xmath183 freeze - out the annihilations @xmath190 must be inactive down to the decoupling temperature : @xmath191 . given that the annihilation cross section of @xmath183 is parametrically @xmath192 , which should be insignificant down to the decoupling temperature , we require that _ x ( ) ^1/4 ~10 ^ -6 furthermore , to avoid significant self scattering we require @xcite _ x^4/(v^4 m__0 ^ 3 ) 0.1 cm^2/g for velocity dispersions of @xmath9110 km / s , this gives @xmath193 . a. agnello and n. w. evans , _ a virial core in the sculptor dwarf spheroidal galaxy , _ astrophys . j. * 754 * ( 2012 ) l39 [ 1205.6673 ] . p. agrawal , s. blanchet , z. chacko and c. kilic , _ flavored dark matter , and its implications for direct detection and colliders , _ phys . d * 86 * ( 2012 ) 055002 [ 1109.3516 ] . s. alexander and s. cormack , _ gravitationally bound bcs state as dark matter , _ [ 1607.08621 ] . n. c. amorisco , a. agnello and n. w. evans , _ the core size of the fornax dwarf spheroidal , _ mnras * 429 * ( 2013 ) 89 [ 1210.3157 ] . b. batell , j. pradler and m. spannowsky , _ dark matter from minimal flavor violation , _ jhep * 1108 * ( 2011 ) 038 [ 1105.1781 ] . g. battaglia , _ et al . _ _ the kinematic status and mass content of the sculptor dwarf spheroidal galaxy , _ astrophys . j. * 681 * ( 2008 ) l13 [ 0802.4220 ] . j. baur , _ et al . _ _ lyman - alpha forests cool warm dark matter , _ jcap * 1608 * ( 2016 ) no.08 , 012 [ 1512.01981 ] . z. berezhiani , a. dolgov and r. mohapatra , _ asymmetric inflationary reheating and the nature of mirror universe , _ phys . b * 375 * ( 1996 ) 26 [ hep - ph/9511221 ] . j. binney and s. tremaine , _ galactic dynamics _ , 2nd ed . , princeton university press ( 2008 ) . d. blas , j. lesgourgues and t. tram , _ the cosmic linear anisotropy solving system ( class ) ii : approximation schemes , _ jcap * 1107 * ( 2011 ) 034 [ 1104.2933 ] . w. j. g. de blok , _ the core - cusp problem , _ adv . astron . * 2010 * ( 2010 ) 789293 [ 0910.3538 ] . p. bode , j. p. ostriker and n. turok , _ halo formation in warm dark matter models , _ astrophys . j. * 556 * ( 2001 ) 93 [ astro - ph/0010389 ] . t. j. l. de boer _ et al . _ , _ the star formation & chemical evolution history of the sculptor dwarf spheroidal galaxy , _ astron . astrophys . * 539 * ( 2012a ) a103 [ 1201.2408 ] . t. j. l. de boer _ et al . _ , _ the star formation & chemical evolution history of the fornax dwarf spheroidal galaxy , _ astron . astrophys . * 544 * ( 2012b ) a73 [ 1206.6968 ] . j. s. bolton , m. g. haehnelt , m. viel and v. springel , _ the lyman - alpha forest opacity and the metagalactic hydrogen ionization rate at z @xmath91 2 - 4 , _ mnras * 357 * ( 2005 ) 1178 [ astro - ph/0411072 ] . a. burkert , _ the structure and dark halo core properties of dwarf spheroidal galaxies , _ astrophys . j. * 808 * ( 2015 ) 2 , 158 [ 1501.06604 ] . j. j. dalcanton and c. j. hogan , _ halo cores and phase space densities : observational constraints on dark matter physics and structure formation , _ astrophys . j. * 561 * ( 2001 ) 35 [ astro - ph/0004381 ] . f. donato _ et al . _ , _ a constant dark matter halo surface density in galaxies , _ mon . not . * 397 * ( 2009 ) 1169 [ 0904.4054 ] . j. dubinski , i. berentzen and i. shlosman , _ anatomy of the bar instability in cuspy dark matter halos , _ astrophys . j. * 697 * ( 2009 ) 293 [ 0810.4925 ] . n. w. evans , j. an and m. g. walker , _ cores and cusps in the dwarf spheroidals , _ mnras * 393 * ( 2009 ) 50 [ 0811.1488 ] . j. l. feng , h. tu and h. b. yu , _ thermal relics in hidden sectors , _ jcap * 0810 * ( 2008 ) 043 [ 0808.2318 ] . j. goodman , _ repulsive dark matter , _ new astron . * 5 * ( 2000 ) 103 [ astro - ph/0003018 ] . w. hu , r. barkana and a. gruzinov , _ cold and fuzzy dark matter , _ phys . * 85 * ( 2000 ) 1158 [ astro - ph/0003365 ] . l. hui , j. p. ostriker , s. tremaine and e. witten , _ on the hypothesis that cosmological dark matter is composed of ultra - light bosons , _ [ 1610.08297 ] . r. l. jaffe , _ degenerate fermion systems _ , mit 8.322 lecture notes ( 2006 ) . j. jardel and j. sellwood , _ halo density reduction by baryonic settling ? , _ astrophys . j. * 691 * ( 2009 ) 1300 [ 0808.3449 ] . s. u. ji and s. j. sin , _ late time phase transition and the galactic halo as a bose liquid : 2 . the effect of visible matter , _ phys . d * 50 * ( 1994 ) 3655 [ hep - ph/9409267 ] . l. kofman , a. d. linde and a. a. starobinsky , _ towards the theory of reheating after inflation , _ phys . d * 56 * ( 1997 ) 3258 [ hep - ph/9704452 ] . l. landau and e. lifshitz , _ course of theoretical physics : vol . 5 , statistical physics : part 1 _ , 3rd ed . , elsevier , 1980 . f. marinacci , r. pakmor and v. springel , _ the formation of disc galaxies in high resolution moving - mesh cosmological simulations , _ mnras * 437 * ( 2014 ) 2 , 1750 [ 1305.5360 ] . s. mashchenko , j. wadsley and h. m. p. couchman , _ stellar feedback in dwarf galaxy formation , _ science * 319 * ( 2008 ) 174 [ 0711.4803 ] . d. e. morrissey and m. j. ramsey - musolf , _ electroweak baryogenesis , _ new j. phys . * 14 * ( 2012 ) 125003 [ 1206.2942 ] . c. nipoti and j. binney , _ early flattening of dark matter cusps in dwarf spheroidal galaxies , _ mnras * 446 * ( 2015 ) 1820 [ 1410.6169 ] . p. j. e. peebles , _ fluid dark matter , _ astrophys . j. * 534 * ( 2000 ) l127 [ astro - ph/0002495 ] . j. penarrubia , a. pontzen , m. g. walker and s. e. koposov , _ the coupling between the core / cusp and missing satellite problems , _ astrophys . j. * 759 * ( 2012 ) l42 [ 1207.2772 ] . a. h. g. peter , m. rocha , j. s. bullock and m. kaplinghat , _ cosmological simulations with self - interacting dark matter ii : halo shapes vs. observations , _ mnras * 430 * ( 2013 ) 105 [ 1208.3026 ] . h. c. plummer , _ on the problem of distribution in globular star clusters , _ mnras * 71 * , 460 ( 1911 ) . a. pontzen and f. governato , _ how supernova feedback turns dark matter cusps into cores , _ mnras * 421 * ( 2012 ) 3464 [ 1106.0499 ] . l. randall , j. scholtz and j. unwin , _ flooded dark matter and s level rise _ , jhep * 1603 * ( 2016 ) 011 [ 1509.08477 ] . j. a. sellwood , _ bars and dark matter halo cores , _ astrophys . j. * 587 * ( 2003 ) 638 [ astro - ph/0210079 ] . x. d. shi and g. m. fuller , _ a new dark matter candidate : nonthermal sterile neutrinos , _ phys . * 82 * ( 1999 ) 2832 [ astro - ph/9810076 ] . d. n. spergel and p. j. steinhardt , _ observational evidence for selfinteracting cold dark matter , _ phys . rev . * 84 * ( 2000 ) 3760 [ astro - ph/9909386 ] . s. tremaine and j. gunn , _ dynamical role of light neutral leptons in cosmology , _ phys.rev.lett . * 42 * ( 1979 ) 407 . m. viel , g. d. becker , j. s. bolton and m. g. haehnelt , _ warm dark matter as a solution to the small scale crisis : new constraints from high redshift lyman-@xmath1 forest data , _ phys . d * 88 * ( 2013 ) 043502 [ 1306.2314 ] . m. g. walker , _ et al . _ _ a universal mass profile for dwarf spheroidal galaxies , _ astrophys . j. * 704 * ( 2009 ) 1274 [ erratum - ibid . * 710 * ( 2010 ) 886 ] [ 0906.0341 ] . m. g. walker and j. penarrubia , _ a method for measuring ( slopes of ) the mass profiles of dwarf spheroidal galaxies , _ astrophys . j. * 742 * ( 2011 ) 20 [ 1108.2404 ] . d. h. weinberg , j. s. bullock , f. governato , r. k. de naray and a. h. g. peter , _ cold dark matter : controversies on small scales , _ [ 1306.0913 ] . m. d. weinberg and n. katz , _ bar - driven dark halo evolution : a resolution of the cusp - core controversy , _ astrophys . j. * 580 * ( 2002 ) 627 [ astro - ph/0110632 ] .

dwarf irregular galaxies ( dirrs ) including the magellanic clouds in the local universe , in many cases , exhibit an unusually low n / o abundance ratio ( n / o ) in h ii regions as compared with the solar value ( ) . this ratio is broadly equivalent to the average level of extremely metal - poor stars in the galactic halo , suggesting that n released from asymptotic giant branch ( agb ) stars is missing in the present - day interstellar matter of these dirrs . we find evidence for past tidal interactions in the properties of individual dirrs exhibiting low n / o ratios , while a clear signature of interactions is unseen for dirrs with high n / o ratios . accordingly , we propose that the ejecta of massive agb stars that correspond to a major production site of n can be stripped from dirrs that have undergone a strong interaction with a luminous galaxy . the physical process of its stripping is made up of two stages : ( i ) the ejecta of massive agb stars in a dirr are first merged with those of the bursting prompt sne ia and pushed up together to the galaxy halo , and ( ii ) subsequently through tidal interactions with a luminous galaxy , these ejecta are stripped from a dwarf galaxy s potential well . our new chemical evolution models with stripping of agb ejecta succeed in reproducing the observed low n / o ratio . furthermore , we perform n - body + hydrodynamical simulations to trace the fate of agb ejecta inside a dirr orbiting the milky way , and confirm that a tidal interaction is responsible for the efficient stripping of agb ejecta from dirrs . [ firstpage ] galaxies : abundances galaxies : dwarf galaxies : evolution galaxies : irregular .

1309.0526

the chemical abundance of galaxies is an integrated information on the sequential process of star formation , stellar death , and ejection of heavy elements lasting over a galactic age under an individual specific galaxy environment . therefore , an unusual observed chemical feature detected in some galaxies , that is , the feature which is theoretically hard to understand , will tell us some physical key process that drives their chemical evolution and we have overlooked . the low n abundance in h ii regions and very young stellar populations of the magellanic clouds ( mcs ) is one of the long - standing puzzling problem that remains unsolved . the relative n / o ratio in the large mc ( lmc ) by many studies thus far leads to a mean @xmath0 ( n / o ) @xmath3@xmath4 @xmath5 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , while @xmath3@xmath5 @xmath6 for the small mc ( smc ) @xcite . since the solar n / o ratio is @xmath7 @xcite , it turns out that the mcs exibit lower n / o ratio by @xmath8@xmath9@xmath10 dex than the sun . to assess the origin of low n / o , two observational facts are worth being highlighted . first , the observed n / o ratios in the mcs are broadly equivalent to an average n / o ratio of extremely metal - poor ( emp ) stars ( [ fe / h]@xmath11 ) in the galactic halo @xcite . since the chemical composition of emp stars in the galactic halo can be explained in the context of sne - ii nucleosynthesis in massive stellar progenitors ( @xmath12 10 ) ( e.g. , * ? ? ? * ) , it can be interpreted that n released from asymptotic giant branch ( agb ) stars with a time delay , which has lifted the n abundance by @xmath8 0.6 dex in the solar vicinity , is missing in the present - day interstellar matter ( ism ) of the mcs . secondly , the n - deficient feature is in common with many other dwarf irregular galaxies ( dirrs ) in the local universe , as represented by the well - known plateau of the n / o ratio ( @xmath1 , e.g. , * ? ? ? * ) for extragalactic h ii regions at 12 + @xmath0 ( o / h ) 8.0 . this fact may suggest that the mechanism to lower the n abundance is associated with the common properties characteristic to these dirrs . moreover , neutral gas in bursting dirrs ( i.e. , blue compact dwarf ( bcd ) galaxies ) that accounts for a much larger mass in galaxies than ionized one also exhibit low n / o ratios ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? this argument holds for the mcs , as implied from low n / o of young mc stars ( e.g. , * ? ? ? * ; * ? ? ? * ) . theoretically , it is a hard task to predict such a low n / o ratio for dirrs covering a wide metallicity range up to 12 + @xmath0 ( o / h ) @xmath8 8.4 for the lmc as not a passing point but a kind of terminus of galaxy evolution . @xcite show that a low n / o ratio can be identical to the values at the early phase ( @xmath8 a first 1 - 2 gyr ) of predicted evolutionary tracks ( see also * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? on the other hand , @xcite find that a low n / o ratio can be reproduced as a present - day value as long as 12 + @xmath0 ( o / h ) 8.0 , utilizing a very low n yield in massive stars . it should be here of note that a low n / o ratio ( @xmath13 ) extends to 12 + @xmath0 ( o / h ) @xmath3 8.6 - 8.7 , i.e. , a solar metallicity , for dirrs in the virgo cluster @xcite . some models adopting starbursts with burst duration of 0.01 gyr and/or 0.5 gyr yield a low n / o ratio ( * ? ? ? * see also lanfranchi & matteucci 2003 ; henry et al . 2006 ) . in modeling , however , we should be mindful of the star formation history of individual dirrs . many studies clearly reveal that the mcs undergo a continuous star formation over more than 10 gyr ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? in addition , since a deep image provided by _ hubble space telescope _ ( hst ) in particular leads to a color magnitude diagram reaching a very faint magnitude , accurate star formation histories of other dirrs in local volume ( @xmath14 5 mpc ) are now accessible ( e.g. , * ? ? ? indeed , as discussed in the next section , dirrs exhibiting low n / o ratios have a long - term star formation similar to the mcs . the production site of n in agb stars is inclined to massive ones such as 4 in which a hot bottom burning operates at the bottom of an envelope ( e.g. , * ? ? ? * ) . accordingly , agb stars release n with a time delay of @xmath80.1 gyr from their steller birth . this fact implies that the n - rich agb ejecta are likely to interact with those of type ia supernovae ( sne ia ) . recent results regarding the delay time distribution ( dtd ) of sne ia yielded by the studies on the sn ia rate in external galaxies dramatically shorten the sn ia s delay time , compared with its conventional timescale of @xmath81 gyr @xcite . @xcite claim that about 50 % of sne ia explode soon after their stellar birth , and further works reveal that the dtd is proportional to @xmath15 with its peak at around 0.1 gyr extending to @xmath8 10 gyr @xcite . thus , the current view on sne ia is organized such that a majority of sne ia explode promptly after the bursting explosions of sne ii ( prompt sne ia ) , and the rest gradually emerge with a long interval of gyrs ( slow sne ia ) . therefore , the timing of n ejection from agb stars is broadly equivalently to that of the explosion of prompt sne ia . @xcite discuss the origin of the different [ cr , mn , ni / fe ] features between the milky way and the lmc , and conclude that their difference can be nicely explained if the ejecta of prompt sne ia escape from the gravitational potential of the lmc . they also show that the predicted evolution of [ @xmath16-elements / fe ] under the process of chemical enrichment by massive stars together with slow sne ia is in good agreement with the observed trend of the lmc . their finding suggests that the total number of sne ia contributing to the chemical enrichment in the lmc is smaller than that in the milky way . this prediction is fully compatible with the implication from the observed high [ ba / fe ] ratios in the lmc as well as in the fornax dwarf spheroidal ( dsph ) galaxy , which demands about @xmath81/3 of the sn ia rate of the milky way @xcite . in addition , recent observed finding of the chemical feature mainly enriched by massive stars , i.e. , enhanced [ @xmath16/fe ] ratios , of the neutral medium in bcd galaxies @xcite may be suggestive of a preferential removal of sn ia ejecta . this line of evidence suggests that the energetic ejecta of prompt sne ia entrain massive stellar agb ejecta out of the galaxy potential of dwarf galaxies owing to the coincidental occurrence timing between the two . therefore , the question is raised of what determines the fate of agb ejecta associated with the bursting prompt sn ia event . it is no doubt that one primary factor is a galaxy potential . the shallow potential is necessary to strip off agb ejecta but not a sufficient condition . for instance , ic 1613 and ngc 5152 , which are much less luminous galaxies than the smc , show a signature of agb enrichment in hii regions as their elevated n / o ratios @xcite . in search for the second key factor , some witness might be obtainable from a well - studied evolution of the mcs . the mcs have experienced close encounters with the milky way with the orbital period of @xmath8 2.5 gyr @xcite . such interactions cause the mcs to lose a part of dark and stellar halo @xcite . this process is likely to cause the joint stripping of agb and prompt sne ia ejecta . on the other hand , if a dirr is isolated for its overall evolution , agb ejecta seem hard to be stripped since the prompt sne ia are unlikely to retain a sufficient energy to push them out of a galaxy potential . this view is supported by the observations that the speed of sn - driven outflows detected in many dwarf galaxies does not exceed the escape velocity of the host galaxy ( e.g. , * ? ? ? * ; * ? ? ? this paper is organized as follows . we start with a brief review on the properties of dirrs with stress on interactions , classifying them into two groups in terms of the n / o ratio ( 2 ) . then , it is followed by modeling the chemical evolutions of n / o for some dirrs including the lmc ( 3 ) . in 4 , we perform numerical simulations for tracing the fate of agb ejecta in a dirr analogous to the lmc under the interactions with the milky way to investigate the role of an interaction on the stripping of agb ejecta from dirrs . cccccccc galaxy & group & d ( mpc ) & @xmath17 & 12+log ( o / h ) & @xmath18log ( o / h ) & log ( n / o ) & comment + + leo a@xmath19 & local group & 0.79 & -11.70 & 7.38 & @xmath20 & [email protected] & sufficiently small + ngc 6822 & local group & 0.50 & -15.20 & 8.06@xmath22 & + 0.09 & [email protected] & pass through mw@xmath23 + ic 4662@xmath24 & centaurus & 2.44 & -15.22 & 8.09@xmath24 & + 0.12 & [email protected] & recent merger@xmath25 + izw 18 & beyond lv & 19.0@xmath26 & -15.31@xmath27 & 7.21@xmath28 & @xmath29 & [email protected]@xmath28 & dm stripping + ngc 5408@xmath25 & m 83 & 4.81 & -16.12 & 7.99 & @xmath30 & [email protected] & outflow@xmath25 + smc & milky way & 0.06 & -16.31 & 8.13 & 0 & [email protected] & inside mw + ho ii@xmath31 & m 81 & 3.39 & -16.72 & 7.92 & @xmath32 & [email protected] & dm stripping + ngc 2366@xmath33 & m 81 & 3.40 & -17.17 & 7.91 & @xmath34 & -1.61@xmath35 & @xmath36 + ic 2574@xmath31 & m 81 & 4.02 & -17.46 & 7.93 & @xmath37 & [email protected] & @xmath36 + ngc 4395 & cvn & 4.61 & -17.78 & 8.09@xmath38 & @xmath39 & [email protected] & @xmath36 + lmc & milky way & 0.05 & -17.91 & 8.37 & 0 & -1.48@xmath40@xmath41 & inside mw + + pegasus & m 31 & 0.92@xmath42 & -12.60 & 7.93 & + 0.36 & [email protected] & undisturbed outskirts@xmath43 + ic 5152 & field & 2.07 & -14.50 & 7.92 & + 0.06 & [email protected] & isolated + ic 1613 & local group & 0.73 & -14.51 & 7.90@xmath44 & + 0.04 & [email protected] & no association with mw@xmath23 + ddo 82@xmath31 & m 81 & 4.00 & -14.63 & 7.95 & + 0.07 & [email protected] & + ngc 5264 & m 83 & 4.53 & -15.90 & 8.66 & + 0.59 & [email protected] & massive dm + ngc 625@xmath45 & sculptor & 3.89 & -16.30 & 8.14 & + 0.01 & [email protected] & + ngc 5253 & centaurus & 4.00 & -17.38 & 8.30@xmath46 & 0 & [email protected] & only central region@xmath47 + note . @xmath19cole et al . ( 2007 ) , @xmath22hernndez - martnez et al . ( 2009),@xmath23teyssier et al . ( 2012 ) , @xmath24crowther & bibby ( 2009 ) , @xmath25van eymeren et al . ( 2010 ) , @xmath26marconi et al . ( 2010 ) , @xmath27gil de paz et al . ( 2003 ) , @xmath28lebouteiller et al . ( 2013 ) , @xmath31croxall et al . ( 2009 ) , @xmath33walter et al . ( 2008 ) , @xmath35gonzlez - delgado et al . ( 1994 ) , @xmath38esteban et al . ( 2009 ) , @xmath41bekki & tsujimoto ( 2010 ) , @xmath42mcconnachie et al . ( 2005 ) , @xmath43kniazev et al . ( 2009 ) , @xmath44bresolin et al . ( 2007 ) , @xmath45cannon et al . ( 2005 ) , @xmath46lpez - snchez et al . ( 2007 ) , @xmath48monreal - ibero et al . ( 2012 ) , @xmath49westmoquette et al . ( 2013 ) . in this section , we categorize dirrs in local volume into the ones in which massive agb contribution is basically unseen for the n abundance in the ism and others exhibiting elevated n abundances . as the boundary of this selection , we set @xmath0 ( n / o ) @xmath50 and @xmath51 , respectively , considering a margin of 0.2 dex between the two associated with individual observed errors in n / o ratio . the sample of dirrs with their properties is selected from the list compiled by @xcite . we mainly focus on the data which have the o abundance directly measured using [ o iii ] @xmath524363 line . exceptionally , a well - studied izw 18 beyond local volume is added . the updates for some data are adopted with their source references indicated . the final list is shown in table 1 . the sixth column , @xmath18log(o / h ) , means a deviation from the observed metallicity - luminosity relation within the local volume . here we adopt the relation 12+@xmath0 ( o / h ) = 5.67 - 0.151@xmath17 @xcite . since this relationship can be interpreted as a result of the control of an enrichment achievement by the gravitational potential wells ( e.g. , * ? ? ? * ) , we regard a large deviation as an evidence for a peculiar dark halo such as a massive dark halo ( + @xmath18 ) or the striping of an outer part of dark halo through a tidal interaction(@xmath53 ) . on the other hand , a small deviation yielding a scatter of this relation is possibly regarded as a result of phenomena specific to individual galaxies . for instance , bursts of star formation that have occurred in the recent past may lead to a small + @xmath18 due to a prompt o production . here it should be of note that the direct method of o abundance analysis does not support a level of accuracy for o / h better than 0.2 dex ( e.g. , * ? ? ? * ) . in light of this fact , the values of @xmath18 less than 0.2 may not be taken at face value . figure 1 shows the n / o ratios selected for two groups in table 1 , i.e. , low n / o dirrs ( red circles ) and high n / o dirrs ( blue circles ) , compared with those of extremely metal - poor stars in the galactic halo ( squares ) . we see that the level of n / o for low n / o dirrs is broadly compatible with that of galactic halo stars , which is considered to reflect the nucleosynthesis n / o ratio in massive stars . in addition , for reference , much more data of extragalactic hii regions compiled by @xcite are denoted by small crosses . these data include n / o for bright massive galaxies such as m 31 , m 83 , m 51 , and ngc 4254 which mainly populate at 12 + @xmath0 ( o / h)@xmath549.0 . leo a is one of the lowest luminous gas - rich galaxies . since a deep image provided by hst observation reveals its star formation history , which has continuously lasted at least 8 gyr @xcite , leo a is a good example of a dirr exhibiting a low n / o ratio regardless of a long - term star formation . its low n / o ratio is likely attributed to the galaxy mass which is so small that massive agb ejecta can easily escape from a galaxy potential . ngc 6822 is also found by hst color - magnitude studies to have a long star formation history over @xmath5410 gyr ( e.g. , * ? ? ? * ; * ? ? ? though ngc 6822 is currently isolated from either the milky way or m 31 , its observed distance and velocity in the context of a cosmological simulation suggests that this galaxy belongs to the population which has passed the milky way , together with sextans a and sextans b exhibiting @xmath0 ( n / o ) = @xmath55 , @xmath56 , respectively @xcite . on the other hand , gas dynamics of two magellanic - type galaxies , ic 4662 and ngc 5408 , is well investigated , and expanding superbubbles are detected for both galaxies as well as the feature of twisted velocity field in ic 4662 , likely caused by a recent merger event @xcite . the m 81 group is dynamically unstable , leading to tidal interactions among galaxies that allow the presence of tidal dwarf galaxies ( e.g. , * ? ? ? * ) and tidal streams and debris ( e.g. , * ? ? ? its member galaxies , ho ii , ngc 2366 , ic 2574 , exhibit significantly lower o abundances by @xmath8 0.2 - 0.4 dex than those expected for their corresponding luminosities . that might be a result of stripping of dark halo through tidal interactions . note that all of three galaxies have long - term and broadly constant star formation histories @xcite . likely , a low - luminosity seyfert galaxy ngc 4395 in the canes venatici cloud ( cvn ) group may have a similar stripping history , though its o abundance involves a large uncertainty including the lmc - like high value @xcite . in addition to them , izw 18 is likely to be significantly stripped off a dark halo , which may result in the record as the most metal - deficient galaxy known . this star - forming galaxy is found by hst photometry to be composed of stars of all ages possibly including 13 gyr - old stars @xcite . another sample of dirr beyond the local volume , um 462 , can be added to this category since this galaxy ( @xmath57=14.4 mpc ) is a disrupted system showing sings of interaction in the emission - line map and exhibits @xmath0 ( n / o)[email protected] with 12 + @xmath0 ( o / h)[email protected] @xcite . if we can measure the n / o ratio of a blob of stripped n - rich ejecta in the interacting dirrs , its value must be high , and becomes more direct evidence for stripping of n - rich ejecta from low n / o dirrs . here are two dirrs potentially corresponding to this case . um 448 ( @xmath57=75.9 mpc ) is an interacting system of 2 younger / older bodies with evidence of elevated n / o ratio ( @xmath0 n / o = @xmath59 ) at the zone of interaction @xcite . haro 11(@xmath57=83.6 mpc ) is an interacting lyman - break analogue of 3 bodies , with one knot showing lower o and higher n / o ( @xmath0 n / o = @xmath60 ) than the rest ( e.g. , @xmath61 ) of the system @xcite . pegasus has an extended ( @xmath8 8 kpc ) stellar disk that has no expected distortions caused by a tidal interaction @xcite . the luminosity of this galaxy might be sufficiently low to have low n / o as in the case of leo a. however , pegasus seems to possess a massive dark halo , which is inferred from a higher o abundance than predicted for its luminosity ( see also * ? ? ? in the end , an isolated environment with a massive halo is likely to result in a relatively high n / o . possession of a massive dark halo is also implied for ngc 5264 from a large @xmath62 ( o / h ) of + 0.59 . ic 1613 is a member of local group , but the likelihood of association with the milky way ( passing through the virial radius ) is estimated to be less than 1 percent @xcite . ngc 5253 is well studied about the distribution of o and n abundances inside the galaxy . it has been found by several authors ( e.g. , * ? ? ? * ; * ? ? ? * ) that both o and n abundances are nearly constant except for the central 50 pc region where the n / o ratio is significantly elevated owing to high n abundances . since the constant n abundance yields @xmath0 ( n / o ) @xmath8 -1.5 @xcite , ngc 5253 might be suitable for the category of low n / o dirrs . a more extreme case than ngc 5253 with n - enriched dense core can be seen in mrk 996 , which is an isolated dirr at a distance of 22.3 mpc @xcite . this galaxy exhibits @xmath0 ( n / o)=@xmath63 in the core within a diffuse n - poor envelope ( @xmath0 n / o=@xmath64 ) . incorporating the possible stripping of agb ejecta into the chemical evolution models , we examine the evolutionary paths of n / o for dirrs . since we already construct a successful model for the evolutions of c / o and n / o in extragalactic h ii regions @xcite , we utilize the same model formula as in the previous paper except for sn ia model , together with a new treatment of agb ejecta . we adopt the new formula for sne ia that are composed of prompt and slow sn ia @xcite , though this study is unaffected by sn ia formula owing to a negligible ( no ) contribution of o ( n ) from sne ia to chemical enrichment . here we refer to the adopted yield of n. for the nucleosynthesis n yields of massive stars , we use the empirical yield deduced by @xcite . the derived yield is based on the observed n / o ratios of extremely metal - poor stars @xcite as well as recent nucleosynthesis results for rapidly rotating massive stars giving high n production @xcite . for the agb yield , we assume a metallicity - dependent n yield for stars with the masses of 1 - 8 . as a basic yield , we adopt the results calculated with @xmath65=0.001 , @xmath66=4 , and @xmath67=0.8 by @xcite , and we choose its metallicity dependence so as to follow the observed increasing feature in n / o for extragalactic h ii regions as shown in figure 1 . slow star formation proceeded in dirrs is parameterized by a low sfr coefficient , @xmath68 = 0.08 - 0.25 gyr@xmath69 ( the mass fraction of the gas converted into stars per gyr , and note that @xmath68 = 0.4 for the solar neighborhood ) , adjusting the final o abundance to each observed one in h ii regions . for all galaxies , the galactic age of 13.5 gyr is assumed . among low n / o dirrs , we calculate a few possible paths of n / o evolution for the lmc including the effects of agb stripping . for the initial mass function ( imf ) , we adopt a steep imf ( @xmath70=@xmath6 ) , which is predicted by previous studies on the chemical evolution of the lmc @xcite . first , we take a look at the model result without agb stripping over the whole evolution ( black dashed line in the left panel of fig . 2 ) . owing to a gradual increase in n abundance caused by continuous enrichment by agb stars , the predicted present - day n / o is largely deviated form the observed value . on the other hand , an assumption that massive ( 4 - 8 ) agb stars do not contribute to all chemical enrichment history gives a good agreement with the observation to the result ( blue line ) . more realistic cases for agb stripping are further considered . numerical simulations of agb stripping analogous to the lmc case presented in the next section suggest that during the first @xmath85 - 8 gyr , agb ejecta is kept within an inner halo of the lmc . taking this numerical result into consideration , we exemplify two simple cases that the ejecta of massive agb stars can be recycled for the first 5 ( red solid line ) or 8 gyr ( red dashed line ) while no contribution to chemical enrichment is assumed for the latter evolution . both results are compatible with the observed n / o ratio within an error bar . these results suggest that the present - day low n / o can be explained as long as stripping of massive agb ejecta has occurred at a very high rate during at least last several gyrs . moderate stripping would yield a higher n / o than that expected from the observation . according to our current numerical simulations , a sufficient stripping efficiency while the stripping process is activated is realized only for the model with less massive dark halo ( see 4 ) . since such high agb stripping is essential to achieve the low n / o as in the lmc , its process in the framework of dynamical evolution should be more precisely and in further detail validated by our future simulations with high resolution . here we treat the most interesting case where a high n / o ratio is realized at a very low metallicity . four dirrs , i.e. , pegasus , ic 5152 , ic 1613 , ddo 82 , exhibit @xmath0 ( n / o ) = @xmath71 to @xmath72 at a similar metallicity of 12+@xmath0(o / h)@xmath3 7.9 . these high n / o cases are examined with an assumption that agb contribution is fully included together with a steep imf ( @xmath70=@xmath6 ) as in the same for the lmc . the model result is shown by blue line in the right panel . we see that it predicts a relatively low level of n / o as compared with the observed values among four dirrs . the model calculated with a salpteter imf ( @xmath70=@xmath73 ) lowers a reaching n / o by @xmath8 0.1 dex and thus leads to deviation from the observed point . more enhanced n / o ratio can be obtained by adopting a different form of the imf , that is , the imf with a cutoff at the upper mass end , @xmath74 , at around 20 - 25 , which is implied by high - z metal - poor damped ly@xmath16 ( dla ) systems @xcite as well as by some dsph galaxies such as the fornax @xcite . this is because such a imf lacks very massive stars which synthesize o in large quantities , while the contribution of n from agb stars remains at the same level ( note that a steep imf reduces both contributions ) . as expected , the result calculated with @xmath74=20 and @xmath70=@xmath73 is found to be in better agreement with the observed values by reaching a higher n / o ( red line ) . this predicted path for 12+@xmath0(o / h ) 7.5 coincides with the n / o ratios of high - z dlas @xcite . we , however , do not exclude the possibility that such high n / o ratios are affected to some extent by local contamination from the ejecta of wolf - rayet stars ( e.g. , * ? ? ? * ) , in particular for the extremely high n / o dirrs such as mrk 996 @xcite . except for the unusually high n / o ratios as exhibited by the three galaxies ( ic 5152 , ic 1613 , ddo 82 ) , the observed n / o ratios of dirrs for 12+@xmath0 ( o / h ) 8.5 are basically resided within the triangular - shaped cloud of data points , the vertices of which are ( 12+@xmath0 o / h , @xmath0 n / o)= ( 7.21 , @xmath75 ) , ( 7.93 , @xmath71 ) , ( 8.37 , @xmath76 ) , corresponding to izw 18 , pegasus , and the lmc , respectively ( see fig . 1 ) . since the highest n / o ratio among them is broadly identical to the one of pegasus , which is modeled with an assumption of a complete contribution of agb ejecta to chemical enrichment , a rather small scatter of the n / o ratio in the low - metallicity regime can be explained by the process of agb stripping that can happen at different times and with different intensities . it should be here stressed that the model results in figure 2 can only be compared to the present - day ( i.e. , end - point ) hii region abundances of many other dirrs , while a comparison with n / o ratios prior to this epoch is impossible due to the lack of suitable diagnostics , such as abundances in dim long - lived stars . accordingly , the overall interpretation of the n / o ratios exhibited by dirrs can be summarized as follows ; ( i ) a plateau - like low n / o of @xmath13 is attributed to an efficient stripping of agb ejecta due to a sufficiently small galaxy mass or a strong interaction with other massive galaxy as experienced by the lmc , and ( ii ) a confinement of agb ejecta within a galaxy potential or their moderate stripping eventually lifts a n / o ratio to some extent , @xmath8 0.3 dex at most . a common property of dirrs including unusually high n / o ones in our scenario is their imfs which favor a small number of massive stars . this is an outcome of chemical evolution that demands less enrichment by massive stars to the ism in dirrs , and may be suggestive of the fact that partial removal of sn ii ejecta is ubiquitous in dirrs , as will be discussed in $ 5.1 . we here investigate how much of gas released from agb stars can be stripped from dwarf galaxies orbiting a luminous disk galaxy by using n - body + hydrodynamical simulations . in order to perform numerical simulations on gpu clusters , we have revised our previous simulation code ( grape - sph ; * ? ? ? the present model is a hybrid one in the sense that the two - hold dynamical model ( i.e. , orbital model plus n - body model ) constructed to simulate an interactive evolution between the milky way and the mcs @xcite is combined with a hydrodynamical code including star formation and supernova feedback effects @xcite . details are described in the two papers . we briefly describe the model applying to the present study . we determine the orbital evolution of a dwarf galaxy with respect to the milky way for the last @xmath8 11 gyr , using a backward integration scheme . the milky way influences the orbits of a dwarf galaxy through a fixed gravitational potential of three components : a central bulge , a disk , and an extended dark matter halo . each component is built to give the best model for the magellanic stream formation @xcite . a dwarf galaxy is assumed to have a plummer potential with a different mass of @xmath77 , @xmath78 , and @xmath79 . in addition , dynamical friction is assumed to operate separately on the orbit of a dwarf galaxy as it passes through the milky way s dark halo , following the chandrasekhar formula @xcite . we carry out our simulations in a galactocentric frame in such that the center of the milky way is fixed at ( @xmath80,@xmath81,@xmath65 ) = ( 0,0,0 ) . the assumed initial position vectors and space velocities of a dwarf galaxy are the same as those adopted for the lmc , i.e. , ( @xmath80,@xmath81,@xmath65 ) = ( -0.8 , -41.6 , -27.0 ) kpc and ( @xmath82,@xmath83,@xmath84 ) = ( -50.7 , -226.1 , 229.3 ) km s@xmath69 ( see table 1 in * ? ? ? * ) . in our model , the orbital evolution of a dwarf galaxy depends solely on @xmath85 . figure 3 shows the orbital evolutions with three different @xmath85 . ( blue solid line ) , @xmath86 ( red dotted line ) , and @xmath87 ( green dashed line ) . ] for the determined orbits of dwarf galaxies , we trace the motion of agb ejecta within a modeled dwarf galaxy where star formation proceeds . a dwarf galaxy is represented as a bulgeless disk galaxy embedded in a massive dark matter halo . the mass- and size - ratios of dark matter halo to initial stellar disk are set to be 16.7 and 14 , respectively . the scale length and size @xmath88 of stellar disk are assumed to be 1.5 kpc and 7.5 kpc , which are analogous to the disk structure of the lmc @xcite . thus , the dark halo has a size of 105 kpc . the gas disk is assumed to be exponential and have a mass of @xmath89 with its size of @xmath90 . the total energy of individual sn ii and sn ia is set to be @xmath91 erg , and its 90% is used for thermal feedback while the remaining is used for kinematic feedback . each form of the deposited energy increases the thermal energy ( i.e. , temperature ) and random motions of the surrounding ism , respectively . the total numbers of particles used for dark matter , stellar disk , and gaseous one are 400,000 , 100,000 , and 100,000 , respectively . in the following , @xmath92 represents the elapsed time since a start of simulation . first , we see the detailed results calculated by the model with @xmath86 . the upper panels of figure 4 demonstrate that gas initially in the disk of a dwarf galaxy is stripped and then distributed within the outer halo of the milky way through an orbital evolution . owing to sn feedback effects , some fraction of gas in the disk is ejected into the dwarf s halo , and becomes susceptible to tidal stripping . then real stripping occurs especially when a dwarf galaxy passes its orbital pericenter at @xmath93 and 8 gyr , since a large fraction of its dark matter halo is stripped by the strong tidal field of the milky way . at the first pericenter passage , a small fraction of gas ejected into the dwarf s halo is stripped to form diffuse gas streams , which will not return to the ism in the dwarf s disk . after the second pericenter passage , a dwarf galaxy loses a much larger fraction of gas under the truncated dark halo . the stripped gas finally shows a thin ring - like feature . interestingly , some fraction of its gas is accreted onto the inner region of the milky way . -@xmath94 plane at different four time steps for the model with @xmath86 . the time @xmath92 shown in the upper left corner for each panel is given in units of gyr . the orbit of a dwarf galaxy is indicated by thick green line in the upper left panel for @xmath95 gyr . _ lower panel _ : time evolution of the mass of agb ejecta for the same model as in upper panels . the mass of agb ejecta ( @xmath96 ) within @xmath97 kpc and that of all agb ejecta are shown by red dotted and blue solid lines , respectively . ] the lower panel of figure 4 shows the mass @xmath96 of agb ejecta within 15 kpc of a dwarf galaxy halo , compared with the total mass of gas ejected from agb stars . the difference between the two is seen after the first pericenter distance at @xmath93 gyr , and @xmath96 within a halo clearly drops owing to a tidal stripping of agb ejecta from the halo after the second pericenter passage at @xmath98 gyr . in this model , 28% of all gas ejected from agb stars is finally stripped from the dwarf s halo with no return back to the original gas disk up to the present - day . provided that the stripped ejecta originate in the upper mass range of agb progenitors as implied from the chemical studies , this fraction will meet the condition that n released from massive agb stars is significantly stripped off and thus do not contribute to an enrichment in the ism . since we aim at tracing an overall evolution of a dwarf galaxy under an interactive environment , we are forced to discard accurate resolution to simulate the behavior of individual ejecta released from sne ii , prompt and slow sne ia , and low - mass and massive agb stars . accordingly , our result is limited to show that the ensemble of agb ejecta are lifted up to the halo by feedbacks from the assembly of numerous sne ii and ia ( see also discussion in 5.1 ) . ) within 15 kpc of a dwarf galaxy halo for the isolated ( blue solid line ) and interaction ( red dotted line ) models for three different dwarf masses ; @xmath99 ( top ) , @xmath86 ( middle ) , and @xmath87 ( bottom ) . ] next we present the importance of tidal interaction to strip agb ejecta , introducing the isolated model where the gravitational influence of the milky way on dwarf galaxies is not included . figure 5 shows the comparison of time evolution of @xmath96 within a halo for three different @xmath85 between the models calculated under an orbital evolution presented in figure 1 and the isolated ones . for all of three models , differences in the evolution of @xmath96 within a halo are seen , in particular , a factor of @xmath100 difference in the final mass for the lowest mass model . on the other hand , the highest mass model retains most of agb ejecta in a dwarf halo even after close interactions at @xmath101 1 and 9 gyr . these results are attributed to a different extent of dark halo stripping in the sense that a less massive dwarf galaxy loses more dark halo through tidal interactions . the derived dependence of the removal efficiency on the mass strongly suggests that n stripping should be efficient in low - mass , low - metallicity galaxies and much less efficient in high - mass , high - metallicity galaxies , which is compatible with the observed n / o tendency as shown in figure 1 . in the end , we claim that the chemical evolution of dwarf galaxies is largely influenced by their galaxy surroundings , i , e , satellite galaxies in a luminous galaxy or an isolated environment , together with by their dynamical masses . our theoretical scheme is constructed under a hypothesis that preferential loss of massive agb ejecta together with those of prompt sne ia occurs in dirrs . this view is favoured by hydrodynamical simulations in the starburst model @xcite . in this model , sne ii explode in dense ism where most of energy from sne ii is dissipated away by radiative cooling owing to the high - density environment ( e.g. , 97% ; * ? ? ? * ) , and the rest of energy is used for the formation of expanding shells , which is responsible for the formation of hi hole or very diffuse ism . afterward , sne ia explode and agb ejecta are released in an environment with little ism where these ejecta can be expelled from the disk driven by sne ia s explosion energy . we suppose that an efficient joint mass loss from prompt sne ia and massive agb stars is realized even under a continuous star formation by regarding it as a sequence of small - scale starbursts . in other words , global continuous star formation is an ensemble of localized small bursts . in each local region , stars are intermittently formed at intervals of a few @xmath102 yrs . in localized small bursts of star formation , the ambient ism is first thermalized by sn ii explosions so as to suppress further star formation within @xmath103yrs , then followed by feedback of prompt sne ia , and finally the ism can cool down without a major heat source . such an interval of star formation is supported by the discussion on the turbulent ism following sn explosions . @xcite discuss the mixing timescales of sn ii ejecta and quote a shortest mixing timescale of @xmath104yr for the ambient ism . it might predict an unacceptably large scatter in stellar abundances due to the incomplete chemical homogeneity of the ism unless we adopt the interval of star formation more than a few @xmath102yr . this scheme will be validated by performing highly - resolved hydrodynamical simulations applied to local regions where individual stars and ejecta can be decomposed , extracted from an entire galaxy ( bekki & tsujimoto , in preparation ) . we believe that such future numerical simulations can shed light on the connection between the star formation and the removal efficiencies among the different heavy - element ejectors . observationally , there is evidence for bursting star formation even in the milky way s past over most of cosmic time . @xcite find that the local disk within @xmath105 pc has experienced a complex star formation history , using a chromospheric age distribution of dwarf stars . we will see that the lmc has a much more complex one with the strong sfr bumpiness on account of its small mass if we can have a close look at stellar properties of the individual lmc stars , and thereby deduce the localized , not averaged , star formation history . we do not object to removal of sn ii ejecta from a galaxy potential of dirrs . it is reasonable to anticipate that the ejecta of sne ii is in part removed from dwarf galaxies . as supporting evidence , galactic winds from dwarf starburst galaxy , ngc 1569 , exhibit an enhancement of @xmath16-elements relative to fe which is indicative of a sn ii - like feature @xcite . note that this galaxy shows a marginally low ( n / o ) ratio of @xmath106 with 12+@xmath0 ( o / h ) = 8.19 @xcite . additionally , two dwarf starburst galaxies ( ngc 4449 , he 2 - 10 ) are likely to retain the hot gas with an enhanced @xmath16/fe while a non - enhanced feature is seen for two galaxies ( ngc 3077 , ngc 4214 ) @xcite . as already discussed , chemical evolution models for dirrs including the lmc support a steeper imf . this required reduction of sn ii enrichment can be identical to the outcome of a partial removal triggered by galactic winds associated with sne ii . this angle is supported by the observational studies on the imf of the mcs claiming a salpeter - like imf ( e.g. , * ? ? ? an imf slope of @xmath70=@xmath6 adopted in the present study indeed broadly corresponds to a 50% removal of sn ii ejecta . it turns out that the differences in removal efficiencies among snii , sn ia , and agb ejecta may be quite moderate . recall here that we assume a removal of subset ( not all ) of both agb ( only massive ones ) and sn ia ( only prompt ones ) ejecta . these similar efficiencies are rather favored by previous works @xcite . in any event , the degree of escape of sn ii ejecta is of secondary importance for low n / o dirrs focused in this study : the low n / o ratios for dirrs such as the lmc mimic the effects of sn ii nucleosynthesis , though the underlying cause is the preferential stripping of n - rich agb ejecta . on the other hand , the n / o evolution of dirrs that can confine agb ejecta is largely influenced by a degree of sn ii removal since the n / o track goes up by the combined effects of n enrichment by agb stars and less o contribution by sne ii . the loss of agb ejecta from dirrs is restricted to those originated from massive ( 4 ) agb stars since prompt sne ia play a key role of their thermalization and eventual removal . therefore , in our scenario , @xmath107-process elements such as ba synthesized in agb stars with a mass of 1.5 - 3@xcite are hard to suffer a removal from dirrs . on the contrary , the observed results present that stars belonging to the lmc as well as to some dsph galaxies exhibit unusual high ba abundances @xcite . on the other hand , since the production site of c is inclined to a more massive star , i.e. , a 2 -4 agb star @xcite , a partial removal of c ejected form agb stars is expected . indeed , the mcs show the deficiency of c in h ii regions but to a lesser extent than n ( e.g. , * ? ? ? * ) . the n / o ratio as diagnosis of a galaxy environment as we propose can be applied to discussion on the debatable motion of the lmc . recent measurement of the proper motion of the lmc using hst suggests that the lmc is now in the middle of a first infall to the milky way @xcite . their canonical orbit implies that the lmc first entered within 100 kpc of the milky way only a few @xmath102 yr ago . here we calculate the n / o evolution in the lmc for the case analogous to this first infall model . we assume that n enrichment by massive agb stars ceases from 1.5 gyr ago corresponding the accretion epoch reaching a virial radius ( 300 kpc ) of the milky way . as shown in figure 6 , the recent accretion model does not give the sufficient time to lower the n / o ratio down to the present value ( dashed line ) , while the periodic orbit model which has the orbital period of @xmath108 gyr within a virial radius matches the observation ( solid line ) , as already discussed in $ 4 . these results suggest that the lmc has approached the milky way at least more than a few gyr ago . this approach can be strengthened by acquiring stellar n / o for unevolved stars in the lmc . the n / o ratios for stars with different metallicities will give information on how the n / o ratio changes with metallicity , which acts as an indicator of age . this n / o - age relation thus obtained will enable us to pinpoint an accretion epoch of the lmc , though it is currently a formidable task to measure the n abundance for dim unevolved stars in local group . in future , the same method will be also applied to infer the epoch of accretions of dsph galaxies onto the milky way . in cluster of galaxies , dwarf galaxies may be inclined to exhibit low n / o owing to an efficient stripping of agb ejecta by strong tide as well as ram pressure of hot intra - cluster medium . in the virgo cluster , @xmath870% of observed 21 dirrs exhibit low n / o ratios @xcite . this percentage derived from a small database , in fact , seems not so high since 50 dirrs in the local volume listed by @xcite involves @xmath8 80% low n / o dirrs ( however , @xmath109 shows @xmath0 ( n / o)@xmath110 for 12+@xmath0 ( o / h)@xmath148.2 in extragalactic h ii region data compiled by pettini et al . in addition , among the member dirrs in the virgo , we do not see a clear correlation between a low - n / o cluster member and its location relative to the centre of the cluster ( i.e. , m 87 ) . given the absence of a clear correlation between them , a scenario involving ram stripping of cold galactic gas may lead to an increased n / o ratio ( e.g. , * ? ? ? * ; * ? ? ? * ) , contrary to our main thesis . this ram - stripping process results in a suppression of further star formation and production of o , followed by a delayed ejection of n. it is indeed different from our proposed stripping of halo gas , where star formation should be negligible . more data for the reliable statistical analysis will be surely demanded to validate how the n / o ratio is associated with a galaxy environment . in local group , we predict that n / o ratios in dsph galaxies are essentially low as in the mcs because of their much closer distance to a luminous galaxy , i.e. , the milky way or m 31 than dirrs . in our proposed stripping scenario , severe truncation of recycling of agb ejecta due to tidal stripping of gas from dirrs is responsible for the origin of low n / o . on the other hand , @xcite propose an alternative infall scenario in which the observed unusually low n / o in the young populations of the lmc is due to the infall of gas with very low n / o from the smc in the predicted scheme of a gas - transfer from the smc to the lmc ( e.g. , * ? ? ? it is possible to consider that similar dilution occurred in some dirrs by an infall of the objects analogous to h i high velocity clouds ( hvcs ) that surround the milky way . galactic hvcs are observed to have low [ n / h ] ranging from @xmath111 - @xmath112 , giving @xmath0 ( n / o)@xmath113 @xcite.this infall scenario predicts a low n / o ratio only for young stars and h ii regions . therefore , a way to distinguish between the two scenarios is to measure the n / o ratio for intermediate - age stars in dirrs . we have assessed the origin of low n / o ratios quite commonly seen for h ii regions in dirrs including the mcs , which can be interpreted as lack of enrichment of n by massive agb stars . at various level , we find a signature indicating past tidal interactions for the properties of these dirrs , along with a strong argument that the mcs have closely interacted with the milky way . these findings lead to our claim that the ejecta of massive agb stars are efficiently stripped off from a shallow gravitational potential of dirrs during their dynamical interaction with their larger host galaxies . theoretical basis of this mechanism is that massive agb stars release their ejecta after @xmath80.1 gyr from a stellar birth , which broadly coincides with the delay time of the bursting explosion of prompt sne ia . these ejecta thus dispersed into the halo are eventually stripped off if dirrs suffer tidal interactions . our numerical simulations on the dynamical evolution of dirrs around the milky way have confirmed that the two - stage process ( ejection in halo followed by removal ) for efficient stripping of agb ejecta really occurs in dirrs . our new chemical evolution models predict various paths of n / o evolution for dwarf galaxies according to the possible diversity of agb stripping history that results form past tidal interactions . future measurements of n / o abundance ratios for unevolved turn - off stars as a function of metallicity for nearby dwarf galaxies by the upcoming next - generation telescopes will open a new paradigm where the chemical evolution is discussed in close connection with the effects of galaxy environments . t.t . is grateful for icrar s hospitality during a research visit at which this work was completed . the authors wish to thank the referee for all valuable comments that helped improve the paper . numerical computations were carried out at the university of western australia on a gpu cluster implementing the cuda g5/g6 software package for calculations of gravitational dynamics . this research was supported by the graduate university for advanced studies ( sokendai ) . 99 aloisi a. et al . , 2003 , apj , 595 , 760 asplund m. , grevesse n. , sauval a. j. , scott p. , 2009 , ara&a , 47 , 481 bastian n. , covey k. r. , meyer m. r. , 2010 , ara&a , 48 , 339 bekki k. , 2011 , mnras , 416 , 2359 bekki k. , 2009 , mnras , 399 , 2221 bekki k. , shigeyama t. , tsujimoto t. , 2013 , mnras , 428 , l31 bekki k. , tsujimoto t. , 2012 , apj , 761 , 180 bekki k. , tsujimoto t. , 2010 , apj , 721 , 1515 bekki k. , chiba m. , 2007 , mnras , 381 , l16 binney j. , tremaine s. , 2008 , galactic dynamics ( princeton : princeton univ . press ) bothun g. d. , thompson i. b. 1988 , aj , 96 , 877 bradamante f. , matteucci f. , dercole a. , 1998 , a&a , 337 , 338 bresolin f. , urbaneja m. a. , gieren w. , pietrzyski g. , kudritzki r .- p . , 2007 , apj , 671 , 2028 busso m. , gallino r. , lambert d. l. , travaglio c. , smith v. v. , 2001 , apj , 557 , 802 cannon j. m. et al . , 2012 , apj , 747 , 122 cannon j. m. , skillman e. d. , sembach k. r. , bomans d. j. , 2005 , apj , 618 , 247 cayrel r. et al . 2004 , a&a , 416 , 1117 chiappini c. , romano d. , matteucci f. , 2003 , mnras , 339 , 63 cignoni m. et al . , 2012 , apj , 754 , 130 cole a. a. et al . , 2007 , apj , 659 , l17 collins j. a. , shull j. m. , giroux m. l. , 2007 , apj , 657 , 271 contreras ramos r. et al . , 2011 , apj , 739 , 74 croxall k. v. et al . , 2009 , apj , 705 , 723 crowther p. a. , bibby j. l. , 2009 , a&a , 499 , 455 de avillez m. a. , mac low m .- m . , 2002 , apj , 581 , 1047 diaz j. d. , bekki k. , 2012 , apj , 750 , 36 dufour r. j. , 1975 , apj , 195 , 315 ekstrm s. meynet g. , chiappini c. , hirschi r. , maeder a. , 2008 , a&a , 489 , 685 esteban c. , bresolin f. , peimbert m. , garca - rojas j. , peimbert a. , 2009 , apj , 700 , 654 gil de paz a. , madore b. f. , pevunova o. , 2003 , apjs , 147 , 29 gonzlez - delgado r. m. , prez e. , tenorio - tagle g. , 1994 , apj , 437 , 239 harris j. , zaritzsky d. , 2009 , aj , 138 , 1243 henry r. b. c. , nava a. , prochaska j. x. , 2006 , apj , 647 , 984 henry r. b. c. , edmunds m. g. , kppen j. , 2000 , apj , 541 , 660 hernndez - martnez l. , pea m. , carigi l. , garca - rojas j. , 2009 , a&a , 505 , 1027 hill v. , 2004 , in carnegie observatories astrophysics series , vol . 4 , origin and evolution of the elements , ed . a. mcwilliam & m. rauch ( cambridge : cambridge univ . press ) , 203 hirschi r. , 2007 , a&a , 461 , 571 james b. l. , tsamis y. g. , walsh j. r. , barlow m. j. , westmoquette m. s. , 2013 , mnras , 430 , 2097 james b. l. , tsamis y. g. , barlow m. j. , walsh j. r. , westmoquette m. s. , 2013 , mnras , 428 , 86 james b. l. , tsamis y. g. , barlow m. j. , 2010 , mnras , 401 , 759 james b. l. , tsamis y. g. , barlow m. j. , westmoquette m. s. , walsh j. r. , cuisinier f. , exter k. m. , 2009 , mnras , 398 , 2 kallivayalil n. , van der marel r. p. , besla g. , anderson j. , alcock c. , 2013 , 764 , 161 kallivayalil n. , van der marel r. p. , alcock c. , axelrod t. , cook k. h. , drake a. j. , geha m. , 2006 , apj , 638 , 772 kniazev a. y. , brosch n. , hoffman g. l. , grebel e. k. , zucker d. b. , pustilnik a. , 2009 , mnras , 400 , 2054 kobulnicky h. a. , skillman e. d. , 1998 , apj , 497 , 601 kobulnicky h. a. , skillman e. d. , 1997 , apj , 489 , 636 korn a. j. , keller s. c. , kaufer a. , langer n. , przybilla n. , stahl o. , wolf b. , 2002 , a&a , 385 , 143 lanfranchi g. a. , matteucci f. , 2003 , mnras , 345 , 71 lebouteiller v. , heap s. , hubeny i. , kunth d. , 2013 , a&a , in press lebouteiller v. , kunth d. , thuan t. x. , dsert j. m. , 2009 , a&a , 494 , 915 lee h. , grebel e. k. , hodge p. w. , 2003 , a&a , 401 , 141 letarte b. et al . , 2010 , a&a , 523 , a17 lpez - snchez . r. , esteban c. , 2010 , a&a , 517 , a85 lpez - snchez . r. , esteban c. , garca - rojas j. , peimbert m. , rodrguez m. , 2007 , apj , 656 , 168 makarova l. n. et al . , 2002 , a&a , 396 , 473 mannucci f. , della valle m. , panagia n. , 2006 , mnras , 370 , 773 maoz d. , sharon k. , gal - yam a. , 2010 , apj , 722 , 1879 marcolini a. , brighenti f. , dercole a. , 2004 , mnras , 352 , 363 marcolini a. , brighenti f. , dercole a. , 2003 , mnras , 345 , 1329 marconi m. et al . , 2010 , apj , 713 , 615 martin c. l. , 1998 , apj , 506 , 222 martin c. l. , kobulnicky h. a. , heckman t. m. , 2002 , apj , 574 , 663 mcconnachie a. w. , irwin m. j. , ferguson a. m. n. , ibata r. a. , lewis g. f. , tanvir n. , 2005 , mnras , 356 , 979 moll m. , vlchez j. m. , gaviln m. , daz a. i. , 2006 , mnras , 372 , 1069 monreal - ibero a. , walsh j. r. , vlchez j. m. , 2012 , a&a , 544 , a60 ott j. , walter f. , brinks e. , 2005 , mnras , 358 , 1423 pagel b. e. j. , tautvaiien g. , 1995 , mnras , 276 , 505 pagel b. e. j. , edmunds m. g. , fosbury r. a. e. , webster b. l. , 1978 , mnras , 184 , 569 peimbert m. , torres - peimbert s. , 1974 , apj , 193 , 327 pettini m. , zych b. j. , steidel c. c. , chaffee f. h. , 2008 , mnras , 385 , 2011 pilyugin l. s. , leonid s. , vlchez j. m. , thuan t. x. , 2010 , apj , 720 , 1738 pilyugin l. s. , thuan t. x. , vlchez j. m. , 2003 , a&a , 397 , 487 pompia l. et al . , 2008 , a&a , 480 , 379 recchi s. , matteucci f. , dercole a. , tosi m. , 2004 , a&a , 426 , 37 recchi s. , matteucci f. , dercole a. , tosi m. , 2002 , a&a , 384 , 799 recchi s. , matteucci f. , dercole a. , 2001 , mnras , 322 , 800 rivero gonzlez j. g. , puls j. , najarro f. , brott i. , 2012 , a&a , 537 , a79 rocha - pinto h. j. , scalo j. , maciel w. j. , flynn c. , 2000 , a&a , 358 , 869 rolleston w. r. j. , venn k. , tolstoy e. , dufton p. l. , 2003 , a&a , 400 , 21 russell s. c. , dopita m. a. , 1990 , apjs , 74 , 93 schwartz c. m. , martin c. l. , 2004 , apj , 610 , 201 skillman e. d. , bomans d. j. , kobulnicky h. a. , 1997 , apj , 474 , 205 sbordone l. , bonifacio p. , buonanno r. , marconi g. , monaco l. , zaggia s. , 2007 , a&a , 465 , 815 spite m. et al . , 2005 , a&a , 430 , 655 teyssier m. , johnston k. v. , kuhlen m. , 2012 , mnras , 426 , 1808 tolstoy e. , hill v. , tosi m. , 2009 , ara&a , 47 , 371 totani t. , morokuma t. , oda t. , doi m. , yasuda n. , 2008 , pasj , 60 , 1327 tremonti c. a. et al . , 2004 , apj , 613 , 898 tsamis y. g. , barlow m. j. , liu x .- w . , danziger i. j. , storey p. j. , 2003 , mnras , 338 , 687 tsujimoto t. , 2011 , apj , 736 , 113 tsujimoto t. , shigeyama t. , 2012 , apj , 760 , l38 tsujimoto t. , bekki k. , 2011 , a&a , 530 , a78 van den hoek l. b. , groenewegen m. a. t. , 1997 , a&as , 123 , 305 van eymeren j. , koribalski b. s. , lpez - snchez . r. , dettmar r .- j . , bomans d. j. , 2010 , mnras , 407 , 113 van loon j. th . et al . , 2010 , aj , 139 , 68 van zee l. , haynes m. p. , 2006 , apj , 636 , 214 van zee l. , skillman e. d. , haynes m. p. , 2006 , apj , 637 , 269 van zee l. , salzer j. j. , haynes m. p. , odonoghue a. a. , balonek t. j. , 1998 , aj , 116 , 2805 vila - costas m. b. , edmunds m. g. , 1993 , mnras , 265 , 199 vlchez j. m. , iglesias - pramo j. , 2003 , apjs , 145 , 225 walter f. , brinks e. , de blok w. j. g. , bigiel f. , kennicutt r. c. , thomley m. , leroy a. , 2008 , aj , 136 , 2563 weisz d. r. , dolphin a. e. , skillman e. d. , holtzman j. , dalcanton j. j. , cole a. a. , neary k. , 2013 , mnras , 431 , 364 weisz d. r. , skillman e. d. , cannon j. m. , dolphin a. e. , kennicutt r. c. jr . , walter f. , 2008 , apj , 689 , 160 westmoquette m. s. , james b. , monreal - ibero a. , walsh j. r. , 2013 , a&a , 550 , a88 wyder t. k. , 2003 , aj , 125 , 3097 yoshii y. , tsujimoto t. , nomoto k. , 1996 , apj , 462 , 266 yun m. s. , ho p. t. p. , lo k. y. , 1994 , nature , 372 , 530

we consider an example of a system with two degrees of freedom admitting separation of variables but having a subset of codimension 1 on which the 2-form defining the symplectic structure degenerates . we show how to use separation of variables to calculate the exact topological invariant of non - degenerate singularities and singularities appearing due to the symplectic structure degeneration . new types of non - orientable 3-atoms are found . _ keywords _ : phase topology , separated variables , kowalevski top , double field , loop molecule , non - orientable atoms _ msc 2000 _ : 70e17 , 70g40

1312.5184

the lie co - algebra @xmath0 with coordinate functions @xmath1 has the lie poisson bracket @xmath2 for a given function @xmath3 the system of equations written as @xmath4 is called euler equations on @xmath0 with the hamilton function @xmath5 @xcite . suppose we have a rigid body rotating about a fixed point @xmath6 and relate all vector and tensor objects to a reference frame moving with the body . let @xmath7 be some vectors fixed in the inertial space and @xmath8 the kinetic momentum of the body . the equations of motion have the form with @xmath9 . the constant symmetric matrix @xmath10 is the inertia tensor and @xmath11 is the angular velocity of the body . the function @xmath12 is treated as the potential energy . in what follows we use the coordinates @xmath13 of @xmath14 in the moving frame instead of @xmath15 for convenience . in the generic case @xmath16 , common levels @xmath17 of the casimir functions @xmath18 are 6-dimensional symplectic leaves of the lie poisson bracket . system restricted to @xmath17 becomes a hamiltonian system with three degrees of freedom . the hamilton function @xmath19 defines an integrable generalization of the kowalevski top @xcite to a double force field . additional integrals in involution are @xcite : @xmath20\\ \qquad -\alpha_1 \boldsymbol{\beta}^2-\beta_2 \boldsymbol{\alpha}^2+(\alpha_2+\beta_1)(\boldsymbol{\alpha}{\cdot}\boldsymbol{\beta } ) . \end{array}\ ] ] it is known that for a wide class of hamilton functions including without loss of generality one can choose the following constants of the casimir functions @xcite @xmath21 for the function the cases @xmath22 and @xmath23 correspond to the classical kowalevski case @xcite and the case of yehia @xcite . both of these cases have symmetries , globally reduce to systems with two degrees of freedom and are not considered here . for irreducible cases the system , admits a lax representation given by a.g . reyman and m.a . semenov - tian - shansky @xcite but has not been yet reduced to quadratures . let us call this system _ the rs - system_. the study of irreducible integrable 3d - systems begins with detecting the so - called critical subsystems . these are even - dimensional invariant submanifolds of the phase space with the induced hamiltonian systems having less than three degrees of freedom . all critical subsystems for the rs - system were found in @xcite . separation of variables is known for two of them @xcite . consider a critical subsystem on a four - dimensional invariant submanifold . the 2-form defining the hamiltonian type of the induced dynamics is obtained as the restriction of the symplectic structure of @xmath17 . it appeared that , in the rs - system , all critical subsystems with two degrees of freedom have 3-dimensional subsets on which this 2-form degenerates . such systems are now called _ almost _ hamiltonian . for the first subsystem @xcite , as shown in @xcite , the fomenko zieschang invariant @xcite can be applied . here we study the second subsystem found in @xcite . it is denoted by @xmath24 . this notation stands for the dynamical system and therefore includes both the phase space and the induced dynamics . for the sake of brevity we also call @xmath24 the phase space of the subsystem meaning the corresponding subset of @xmath17 . the rough phase topology of the system @xmath24 was described in @xcite . nevertheless , some properties of @xmath24 has not been completely established and the character of some `` strange '' bifurcations has not been explained . in this paper we fulfil the complete topological investigation of the system @xmath24 in terms of the topological invariants @xcite calculated using the global separation of variables . new topological effects are revealed due to non - orientable bifurcations , which are possible in almost hamiltonian systems . we use the definition of @xmath24 given in @xcite . consider the first integral of the rs - system @xmath25 where @xmath26 , @xmath27 @xmath28 . as shown in @xcite , the equation @xmath29 applied to the integral constants gives a leaf of the bifurcation diagram of the global integral map @xmath30 . therefore , we define a non - empty invariant set as follows @xmath31 the invariant submanifold @xmath32 was first found in @xcite in terms of invariant relations on @xmath17 having singularities on the set @xmath33 , where @xmath34 the lie poisson bracket @xmath35 of these relations is defined as @xmath36 m\right\ } , \quad m=\dfrac{1}{r^4}(2g - p^2 h).\ ] ] it is proved in @xcite that in the domain @xmath37 the set @xmath24 is a smooth imensional manifold , @xmath35 is a partial integral on @xmath24 and the restriction to @xmath24 of the symplectic structure is non - degenerate everywhere except for the subset @xmath38 . this subset is non - empty . moreover , it does not contain critical points of @xmath35 and therefore is a 3-dimensional submanifold in @xmath24 . thus , the induced system on @xmath24 is almost hamiltonian , i.e. , the 2-form defining the hamiltonian field degenerates on a subset of codimension 1 . below we give some simple explanation of the fact that @xmath24 is everywhere a smooth 4-dimensional manifold . moreover , it appears to be non - orientable . in the sequel , we call @xmath24 a manifold without further comments on the fact . from now on having a first integral denoted by an upper case letter we denote its arbitrary constant by the corresponding lower case letter . on @xmath24 , the following identities hold @xcite @xmath39 these relations show that it is convenient to take @xmath40 for the functionally independent pair of the first integrals on @xmath24 , while the pairs @xmath41 and @xmath42 are not in one - to - one correspondence with the triple @xmath43 . thus , we define the integral map @xmath44 and study the bifurcations of the integral manifolds @xmath45 for almost all integral manifolds @xmath46 we can take two values of @xmath47 with different signs according to . introduce the new variables @xmath48 as @xmath49 it readily follows from that @xmath50 consider the two - valued ( algebraic ) radicals @xmath51 [ prop02 ] the differential equations induced on @xmath24 by the rs - system separate in variables @xmath48 , @xmath52 and on the integral manifolds @xmath46 the phase variables have the following expressions in terms of @xmath48 @xmath53 , \\ \displaystyle \alpha _ 2 = \phantom{-}{2a } { \theta^{2 } } [ ( a u_1-b u_2 ) \pr_1 \pr_2 - b \qr_1 \qr_2 \psi ] , \\ \displaystyle \beta _ 1 = \phantom{-}{2b } \,{\theta^{2}}[a \qr_1 \qr_2 \psi - ( b u_1 - a u_2 ) \pr_1 \pr_2 ] , \\ \displaystyle \beta _ 2 = \phantom{-}{2b } \,{\theta^{2}}[a \qr_1 \qr_2 \pr_1 \pr_2- ( b u_1-au_2)\psi],\\ \displaystyle \omega _ 1 = { r ( \ell u_1 - 2 a m ) } { \theta^{}}\pr_2,\quad \omega _ 2 = { r ( 2b m u_2-\ell ) } { \theta^{}}\pr_1,\\ \displaystyle \alpha _ 3 = { a r } { \theta^{}}\qr_1 , \quad \beta _ 3 = { b r } { \theta^{}}u_1 \qr_2 , \quad \omega _ 3 = { 2}{\theta^{}}(b \qr_2 \pr_1 - a \qr_1 \pr_2 ) . \end{array}\ ] ] the result easily follows from @xcite . the variables of separation @xmath54 found in @xcite can oscillate on infinite segments . therefore here we use dimensionless variables @xmath55 , always restricted to the segment @xmath56 $ ] . let us note that , according to and , @xmath57 everywhere , so the factor @xmath58 in is well defined and always positive . fixing the values @xmath59 and choosing @xmath47 according to , let us treat as a map @xmath60 then the integral manifold @xmath46 is the @xmath61-image of the direct product of two curves @xmath62 ( @xmath63 ) @xmath64 where @xmath65 introduce the enhanced space @xmath66 , and define @xmath67 by equations and . let @xmath68 be the map given by . [ lem1 ] the set @xmath69 is a connected smooth 4-dimensional manifold in @xmath70 . it is easy to check that the system of five equations , always has rank 5 . therefore , @xmath69 is a smooth 4-dimensional manifold . let us call a point @xmath71 admissible if the corresponding set @xmath72 is not empty . for a given @xmath73 , we call the projection of it onto the @xmath74-plane a region of possible motions . if not empty , this region is a rectangle in the square cut out by the system of inequalities @xmath75 . this yields that the image of the integral map @xmath76 in the @xmath77-plane is @xmath78 , \ ; 2p^2 m^2 + 2 h m + 1 \geqslant 0\}. \end{array}\ ] ] the set @xmath79 of all admissible points is a two - sheet covering of @xmath80 . the sheets with opposite signs of @xmath47 are glued together along the curve @xmath81 and @xmath79 is connected . take any two points in @xmath69 and connect their images in @xmath79 by a path @xmath82 with @xmath83 . at the points of @xmath84 the curves @xmath85 never degenerate to one point ( we omit technical details , since this fact follows from the rough topological analysis given below ) . therefore the initial two points in @xmath69 can also be connected by a path . [ lem3 ] consider the involution @xmath86 @xmath87 @xmath88 the manifold @xmath69 and the map @xmath89 are @xmath90-invariant and @xmath90 changes the orientation on @xmath69 . indeed , for @xmath91 let @xmath92 denote the jacobi matrix of the system , . we readily obtain that @xmath93 , where @xmath94 is the matrix of @xmath90 . so @xmath90 changes the orientation of @xmath70 but preserves the orientation of the space normal to @xmath69 . therefore , it changes the orientation of @xmath69 . let us emphasize that the regions of possible motions for @xmath95 and @xmath96 are centrally symmetric to each other and do not coincide except for @xmath97 . in the latter case @xmath90 becomes a @xmath98-symmetry of @xmath73 . in particular , @xmath99 is a diffeomorphism and @xmath100 . summarizing the above statements we come to the following theorem . [ theo1 ] the set @xmath24 is a smooth connected 4-dimensional non - orientable submanifold in @xmath17 . for any point @xmath101 the integral manifold @xmath46 is diffeomorphic to @xmath102 . on @xmath103 we have @xmath104 . as we can see from , the set @xmath85 in @xmath105 is an intersection of the round cylinder generated by the unit circle in the @xmath106-plane and a cylinder with elliptical directrix in the @xmath107-plane . the form of @xmath85 depends on the position of the roots of @xmath108 with respect to @xmath109 . the standard transformations which can happen to @xmath85 are shown in fig . [ figgamma ] : @xmath110 @xmath111 ; @xmath112 @xmath113 ; @xmath114 @xmath115 . here @xmath116 and @xmath117 stand respectively for the eight - curve @xmath118 with transversal self - intersection and for the curve @xmath119 formed by two circles transversally intersecting at two points . the difference between cases @xmath120 and @xmath121 is that the two circles of @xmath120 differ by the sign of @xmath122 , and the two circles of @xmath121 differ by the sign of @xmath123 . in the first case there is an exit to @xmath124 when the common root of @xmath125 and @xmath126 turns out to be an isolated point . in the second case both exits would be of hyperbolic type because @xmath126 never has a multiple root in this system . to accomplish the _ rough _ topological analysis of an integrable system one has to do the following : \1 ) find the admissible region ( the image of the integral map ) ; \2 ) for each regular value of the integral map find the number of liouville tori in the pre - image of this value , i.e. , establish the topology of regular integral manifolds ; \4 ) for each critical value of the integral map find the topological type of its pre - image , i.e. , establish the topology of irregular integral manifolds usually called ( critical ) integral surfaces ; \5 ) show a collection of paths in the integral constants space with complete description of the topological type of bifurcations taking place along this path ; this set of paths should be sufficient to find out the character of any bifurcation occurring in the phase space . for the classical integrable systems in the rigid body dynamics , this program was fulfilled in @xcite . in this section , we find out the rough topology of the system @xmath24 , thus describing its rough liouville equivalence class . further , to describe the _ exact _ topology of the system , i.e. , to establish its liouville equivalence class ( for detailed definitions see @xcite ) , we need to use general classifications of critical points and their neighborhoods @xcite and _ exact _ topological invariants . by this term we mean the invariants that completely define liouville foliations on some sufficient collection of 3-dimensional integral manifolds , e.g. fomenko zieschang invariants @xcite or marked loop molecules @xcite . this will be done in the following sections . according to lemma [ lem3 ] , the integral manifolds undergo topological transformations only in two cases , namely , when @xmath77 crosses the discriminant set of one of the product polynomials @xmath127 ( @xmath63 ) or when @xmath77 reaches the curve @xmath103 . in the first case , for @xmath128 , the map @xmath90 identifies in @xmath24 two sets @xmath129 and @xmath130 different in @xmath69 . therefore , in the sequel we by default suppose that @xmath131 . in the second case the set @xmath132 is factorized by the action on @xmath102 diagonal with respect to the induced actions on @xmath105 @xmath133 in what follows if @xmath134 stands for some set and @xmath135 is a positive whole number , then @xmath136 denotes @xmath135 isolated copies of @xmath134 . [ theo2 ] the bifurcation diagram @xmath137 of the system @xmath24 consists of the half - lines @xmath138 and of the curve @xmath103 defined by . the set @xmath137 divides the admissible region in the @xmath77-plane into six chambers @xmath139 shown in fig . the corresponding regular integral manifolds are listed in the last column of table @xmath120 . the integral surfaces in the pre - image of the @xmath141 segments forming the set @xmath137 are listed in the last column of table @xmath121 . ' '' '' chamber & position of @xmath142 & @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath46 + ' '' '' @xmath147&@xmath148 & & & & & + ' '' '' @xmath149&@xmath150 & & & & & + ' '' '' @xmath151&@xmath152 & & & & & + ' '' '' @xmath153&@xmath154 & & & & & + ' '' '' @xmath155&@xmath156 & @xmath157 $ ] & @xmath158 $ ] & @xmath159&@xmath160 & @xmath161 + ' '' '' @xmath162&@xmath163 & @xmath164 $ ] & @xmath165 $ ] & @xmath159&@xmath159 & @xmath166 + ' '' '' @xmath167&@xmath168 & @xmath169 $ ] & @xmath170 $ ] & @xmath159&@xmath159 & @xmath166 + ' '' '' @xmath171&@xmath172 & & & & & + ' '' '' @xmath173&@xmath174 & & & & & + the proof almost obviously follows from lemma [ lem3 ] . let us make only some remarks . the half - lines @xmath175 and @xmath176 correspond to the cases of a multiple root in one of the polynomials @xmath127 ( @xmath63 ) . it happens if @xmath177 or @xmath178 . the corresponding critical motions are pendulum type motions @xmath179 they include closed orbits of rank 1 , singular points of rank 0 at the intersections of the half - lines and separatrices of rank 1 of unstable singular points . from here we readily obtain the inequalities for @xmath180 on the half - lines . note that the mutual position of the values @xmath181 is completely defined by the position of @xmath182 with respect to @xmath183 . in table 1 , we present the inequalities for @xmath142 and , consequently , define the accessible regions ( segments of oscillation ) for @xmath48 , the topology of @xmath85 and the resulting type of regular integral manifolds @xmath46 . here the @xmath184 sign attached to the notation of a chamber means the sign of @xmath185 in the corresponding part of this chamber . it affects the values @xmath142 but does not change the rest of information . the number of tori in @xmath46 for @xmath128 equals @xmath186 where @xmath187 is the number of those radicals among @xmath188 which have constant sign on @xmath102 . [ theagre1 ] suppose that the radical @xmath123 or @xmath122 does not vanish on the connected component of a regular integral manifold or a critical integral surface . then we respectively denote @xmath189 ' '' '' seg . & @xmath143 & @xmath144 & @xmath145 & @xmath146 & @xmath46 + ' '' '' @xmath120 & @xmath190 & @xmath191 $ ] & @xmath192 & @xmath160 & @xmath160 + ' '' '' @xmath121 & @xmath193 $ ] & @xmath194 & @xmath160 & @xmath192 & @xmath160 + ' '' '' @xmath195 & @xmath190 & @xmath164 $ ] & @xmath192 & @xmath159 & @xmath159 + ' '' '' @xmath196 & @xmath169 $ ] & @xmath194 & @xmath159 & @xmath192 & @xmath159 + ' '' '' @xmath197 & @xmath198 $ ] & @xmath199 $ ] & @xmath160 & @xmath116 & @xmath200 + ' '' '' @xmath201 & @xmath202 $ ] & @xmath203 $ ] & @xmath116 & @xmath160 & @xmath200 + ' '' '' @xmath204 & @xmath202 $ ] & @xmath205 $ ] & @xmath116 & @xmath159 & @xmath206 + ' '' '' @xmath207 & @xmath169 $ ] & @xmath208 $ ] & @xmath159 & @xmath116 & @xmath206 + ' '' '' @xmath209 & @xmath169 $ ] & @xmath210 $ ] & @xmath159 & @xmath116 & @xmath206 + ' '' '' @xmath211 & @xmath212 $ ] & @xmath205 $ ] & @xmath116 & @xmath159 & @xmath206 + ' '' '' @xmath213 & @xmath169 $ ] & @xmath214 $ ] & @xmath159 & @xmath159 & @xmath161 + ' '' '' @xmath215 & @xmath169 $ ] & @xmath205 $ ] & @xmath159 & @xmath159 & @xmath161 + ' '' '' @xmath141 & @xmath216 $ ] & @xmath205 $ ] & @xmath159 & @xmath159 & @xmath166 + in table 2 , we collect the information on the critical cases . in the first column we give the notation of the smooth segments of @xmath137 according to fig . [ figbifset ] . in the corresponding accessible regions stand for the double root of @xmath218 . more analysis is needed for the segments on the curve @xmath103 . here the roots of @xmath219 become centrally symmetric and are denoted by @xmath220 , @xmath221 for @xmath63 . the curve @xmath103 is tangent to @xmath222 and @xmath223 respectively at the points @xmath224 @xmath225 let us formulate the result in a separate statement . [ prop03 ] the integral manifolds in the pre - image of the curve @xmath103 are as follows : @xmath161 on segments @xmath226 ; @xmath166 on segment @xmath141 ; @xmath227 at the point @xmath228 ; @xmath206 at the point @xmath229 . we see from table 2 that the four components of the set @xmath230 differ by the signs of @xmath231 on segment @xmath213 , by the signs of @xmath232 on segment @xmath215 , and by the signs of @xmath233 on segment @xmath141 . at the same time according to the @xmath234-symmetry @xmath90 of @xmath230 simultaneously changes the signs of @xmath235 and @xmath236 . this means that @xmath90 identifies the components having the same product @xmath237 on segment @xmath213 as in fig . [ figglu0],@xmath238 . here the arrows show the connected components of the direct product @xmath102 , the numbers stand for the resulting connected components in the quotient set . on segment @xmath215 ( fig . [ figglu0],@xmath239 ) @xmath90 identifies the components with the opposite sign @xmath240 but the same sign @xmath241 . therefore , the result is @xmath242 . on segment @xmath141 the symmetry preserves all four components of @xmath230 . the result is @xmath243 ( fig . [ figglu0],@xmath244 ) . let us take the points @xmath224 . the critical points in the pre - image are of rank 1 and form closed orbits . for the accessible regions we have @xmath245 , & \tau > 1 , & u_2 \in [ -1_*,1 _ * ] , & \sigma = 1;\\ q_2 : \quad u_1 \in [ -1_*,1 _ * ] , & \tau = 1 , & u_2 \in [ -1,1 ] , & \sigma > 1 . \end{array}\ ] ] the sets @xmath246 for @xmath228 and @xmath229 are shown in fig . [ figpointsq1q2],@xmath238 and @xmath239 . again , @xmath90 acts as simultaneous central symmetry in @xmath247- and @xmath248-planes . at @xmath228 , it glues together the components of @xmath145 and preserves @xmath146 . the result is @xmath227 . at @xmath229 , @xmath90 preserves the components of @xmath146 and therefore factorizes @xmath145 . the result is @xmath206 . this proves the statement . now we describe all bifurcations in terms of the atoms according to the contemporary notation @xcite . let us recall some terminology from @xcite . consider a 2-surface ( two - dimensional compact manifold without boundary ) and a morse function @xmath249 on it having a critical value @xmath250 . a 2-atom is a neighborhood of a connected component of the set @xmath251 for sufficiently small @xmath252 , foliated into level lines of @xmath249 and considered up to the fiber equivalence . a 2-atom is supposed to have only one connected singular fiber @xmath253 . if an atom @xmath254 is given , its singular fiber is denoted by @xmath255 . in fig . [ figatoms1 ] , the following atoms are shown : @xmath256 ( @xmath257 , @xmath258 ) ; @xmath259 ( @xmath260 , @xmath261 ) ; @xmath262 ( @xmath263 , @xmath264 ) ; @xmath265 ( @xmath266 , @xmath267 ) . for integrable hamiltonian systems with two degrees of freedom , bifurcations of the liouville tori are described in terms of 3-atoms . let @xmath268 be the integral map . fixing a regular value @xmath269 of @xmath270 we obtain a 3-dimensional iso-@xmath270 manifold @xmath271 . let us suppose for simplicity that @xmath272 is connected , otherwise we take one connected component of @xmath272 . let @xmath273 be a critical value of @xmath274 on @xmath272 and @xmath275 be the connected component of @xmath276 . then @xmath275 is called a singular leaf of the liouville foliation on @xmath272 . a 3-atom is a small enough connected neighborhood of @xmath275 in @xmath272 containing no other singular leaves and invariant under the phase flow . in fact , 3-atoms as foliated manifolds are considered up to the fiber equivalence as defined in @xcite . in the same way as before , for a given 3-atom @xmath254 we denote its singular leaf by @xmath255 . we easily obtain 3-atoms from the above mentioned 2-atoms by considering their direct products with a circle . then , traditionally @xcite , we keep for them the same notation . thus , @xmath277 , @xmath278 , @xmath279 and the bifurcations of tori when crossing the singular leaves are @xmath280 note that the 3-atom @xmath265 was predicted @xcite but never has been met in real systems . of course , bifurcations with non - symmetric atoms can be written the other way round depending on the direction in which we cross the bifurcation diagram . in the case of minimal or maximal integral surfaces , symmetric atoms can be folded twice , so that the bifurcation @xmath281 turns into @xmath282 . let us use the notation @xmath283 for the atom of a minimal ( maximal ) torus . here @xmath272 is a regular level of @xmath270 and on it @xmath274 has the form @xmath284 with regular function @xmath285 . the 2-atom @xmath283 ( already not associated with any morse function ) is just an annulus foliated into circles , and the 3-atom @xmath283 is the direct product of an annulus and a circle foliated into 2-tori . considering the torus @xmath286 as the singular leaf we have the following bifurcation @xmath287 . the atoms in correspond to the critical points which are called simple @xcite or non - degenerate @xcite . for all points forming the motions including those of rank 0 in the preimage of @xmath288 ( see fig . [ figbifset ] ) the non - degeneracy is proved in @xcite . the only exceptions are the motions in the pre - image of the points @xmath289 , which are degenerate as critical points of rank 1 in the rs - system ( with three degrees of freedom ) . let us also mention that all tori in the pre - image of @xmath103 are degenerate as critical points of rank 2 in @xmath17 @xcite . it is now easy to describe all non - degenerate bifurcations . we have the following atoms : \1 ) @xmath256 on segments @xmath290 ; \2 ) @xmath291 on segments @xmath292 ; \3 ) @xmath259 on segments @xmath293 ; \4 ) @xmath294 on segments @xmath295 . the points @xmath296 @xmath297 have @xmath180-coordinate equal to @xmath298 and are enumerated along the @xmath180-axis . the points @xmath299 of rank 0 in the pre - image of @xmath296 are also non - degenerate . for such points the local phase topology is described in terms of the almost direct products of 2-atoms @xcite . here we have only direct products . this fact follows immediately from the general classification of the cases with one singular point on a singular leaf @xcite and the atoms on the adjacent segments . thus , small enough invariant under the phase flow neighborhoods @xmath300 of the points @xmath299 ( called extended or saturated neighborhoods ) are @xmath301 to finish the description of the rough topology we need to point out the bifurcations occurring on the pre - image of the curve @xmath103 . two of them are obvious . as shown in the proof of proposition [ prop03 ] , the symmetry @xmath90 on segments @xmath226 glues together pairwise the four tori of chambers @xmath162 and @xmath167 respectively . therefore we have here bifurcations with two atoms @xmath283 . m0.75 m0.18 & let us consider a path reaching segment @xmath141 from chamber @xmath302 . each of the sets @xmath303 along this path has four components corresponding to the pair of signs @xmath304 . consider a continuous set of these components @xmath305 marked by @xmath306 . the component of @xmath307 defined by @xmath304 is well projected onto the plane @xmath308 and the corresponding component of @xmath309 is well projected onto the plane @xmath310 as shown in fig . [ figchambervi ] . we then see that both @xmath311 are almost central symmetries on these planes and become real central symmetries on @xmath312 . finally for a point on segment @xmath141 we obtain the 3-atom @xmath313 , where @xmath314 is the mbius band foliated into circles in the natural way with one singular central circle ( the band s axis ) twice shorter than all close ones . the product of the axis with a circle stands for the torus in the pre - image of @xmath103 . the mbius band itself foliated this way gives a non - orientable 2-atom , which we denote by @xmath315 . the same notation we use for the corresponding 3-atom @xmath316 . its bifurcation in the direction from the border into the chamber is @xmath317 , but unlike the atom @xmath256 having a circle as a singular leaf , here the singular leaf is a torus covered twice by the close regular torus as @xmath318 . this 3-atom is impossible if we deal with a hamiltonian system without degenerations of the symplectic structure . to describe the topology in small enough neighborhoods of @xmath289 we first consider the situation arising in the covering manifold @xmath69 , where all transformations are easily seen from the sets @xmath319 . consider a neighborhood of @xmath228 and unfold the picture from the integral constants space onto the plane @xmath320 . we obviously obtain the cross formed by the lines @xmath321 and @xmath322 . on each of the lines the bifurcation is @xmath294 , along the horizontal line the bifurcation is @xmath323 and along the vertical line the bifurcation is @xmath324 . in fact , this picture reflects two connected components of the neighborhood of the pre - image of @xmath228 in @xmath69 . on each component the bifurcation diagram together with the atoms is shown in fig . [ figbifq1],@xmath238 . the vertical iso-@xmath325 graphs change as in fig . [ figbifq1],@xmath239 , and the horizontal iso-@xmath35 graphs change as in fig . [ figbifq1],@xmath244 . this phenomenon can be called the splitting of the atom @xmath262 . the possibility of topologically unstable systems is discussed in @xcite . the case we obtain here is described as a possible transformation of iso - energy fomenko graphs in @xcite . let us consider a path reaching the point @xmath228 from chamber @xmath155 . all the way the curve @xmath145 has two components with different signs of @xmath235 while the closed curve @xmath326 covers the whole set @xmath117 as @xmath318 ( fig . [ figchamberiii ] ) . in @xmath69 , we unfold this path to a small vertical path @xmath327 and this process forms two atoms @xmath328 , where @xmath265 denotes the 2-atom . representing the sets @xmath329 and @xmath330 on the plane as in fig . [ figatomc1onplane ] , we see that in this representation both @xmath331 and @xmath332 act as central symmetries , so @xmath331 identifies the components of @xmath329 and @xmath332 acts as a @xmath234-symmetry on @xmath333 . factorizing by the diagonal action , we can write , admitting some inexactness , @xmath334 . therefore the pre - image of the chosen path in @xmath24 gives one atom @xmath328 , i.e. , one 3-atom @xmath265 . on the diagram @xmath137 of the system @xmath24 , crossing the point @xmath228 from the border into chamber @xmath155 we have the bifurcation @xmath265 written in the form @xmath335 . let us consider a neighborhood of @xmath229 and , in the same way as in the previous case , unfold the picture from the integral constants space onto the plane @xmath320 . we obtain the cross formed by the lines @xmath336 and @xmath337 . the topological picture in @xmath69 again gives two copies of the bifurcation shown in fig . [ figbifq1 ] . the essential difference appears after applying the factorization with respect to @xmath90 . consider a small vertical path @xmath338 covering the path reaching @xmath229 transversally from chamber @xmath339 . the set @xmath340 does not transform in the whole neighborhood of @xmath229 . along the chosen path the union @xmath341 fills the 2-atom @xmath265 ( see fig.[figchamberii ] ) . in this case we can show these sets in the plane as in fig . [ figatomt1onplane ] , where @xmath331 and @xmath332 act as central symmetries . factorizing by the diagonal action , we can write @xmath342 and get the pre - image of the chosen path in @xmath24 as a union of two atoms of the new type , which we denote by @xmath343 . here the 2-atom @xmath344 is shown in fig . [ figatomt1 ] . again , we keep the same notation @xmath345 for the corresponding 3-atom @xmath343 . on the diagram @xmath137 of the system @xmath24 , crossing the point @xmath229 from the border into chamber @xmath339 we have two simultaneous bifurcations @xmath346 . the 3-manifold with boundary @xmath345 is non - orientable and is impossible in a system with non - degenerate symplectic structure . finally , we have obtained the topological description of the integral manifolds , critical integral surfaces and the bifurcations in @xmath24 along any chosen path in the plane of the integral constants . this completes the rough topological analysis of the system @xmath24 . the exact topological analysis is a way to establish the phase topology of the considered system up to liouville equivalence @xcite . in order to calculate main topological invariants of such equivalence , we need to describe the families of regular liouville tori and to find the exact rules by which these families are glued to the boundaries of the bifurcation atoms . let us recall some notions from the general theory @xcite . given a liouville integrable hamiltonian ( or almost hamiltonian ) system on a imensional manifold @xmath347 , let us remove from the phase space all connected components of the integral surfaces containing critical points of the integral map @xmath348 , i.e. , all singular leaves of the liouville foliation . in our case , these leaves also include the whole level @xmath349 . the connected component of the remaining set is called a _ family _ of liouville tori . consider a path @xmath350 \to \bbr^2 $ ] which is either closed @xmath351 or has its ends beyond the admissible region @xmath352 . the pre - image @xmath353))$ ] is called a loop manifold . under some simple transversality conditions it is indeed a smooth 3-dimensional manifold without boundary @xcite . if the path @xmath82 is a fixed level line of some first integral @xmath354 , we call @xmath355 an iso-@xmath354 manifold . frequently , the role of @xmath354 is played by the hamiltonian @xmath5 . then @xmath355 is called an iso - energy manifold . identifying the points that belong to the same leaf of the liouville foliation on @xmath355 we obtain the rough fomenko graph @xmath356 with edges representing the families of regular tori and vertices corresponding to singular leaves of the foliation . consider an edge of this graph bounded by two vertices . on the boundary tori of the atoms pointed by these vertices some pairs of coordinate cycles ( bases of cycles ) are defined called admissible coordinate systems ( or admissible bases ) . shift these bases to one regular torus corresponding to an inner point of the graph s edge . two obtained bases are connected ( in the one - dimensional homology group ) with the so - called gluing matrix . it is an integer - valued matrix whose determinant is equal to @xmath109 . in the orientable case without minimal or maximal tori the bases are chosen in such a way that the determinant is always equal to @xmath357 . endowing each edge of the graph with the gluing matrix we obtain one of the forms of the exact topological invariant of the liouville foliation on @xmath355 . usually , since gluing matrices are defined up to the changes of admissible coordinate systems , they are replaced by some sets of numerical invariants called marks . the resulting topological invariant is called the marked molecule and is denoted by @xmath358 ( see @xcite for complete details ) . the goal of the exact topological analysis of an integrable system is to find the existing marked molecules and the corresponding loop manifolds for a reasonably full set of paths in the integral constants plane . we return to the system @xmath24 and use the advantages of the separation of variables to describe formally the families of regular tori and introduce , in some universal way , the coordinate system ( the pair of independent cycles ) on each family . in the general case of an integrable system with two degrees of freedom , each family is parameterized by the value of the integral map and the image of a family is an open connected set in @xmath359 . the image of one family can cover several chambers and the walls between them if there are walls on which some families do not bifurcate . for the system @xmath24 this is not the case . indeed , let us collect all information about the existing atoms and the number of regular tori in the chambers in fig . [ figbifwithatoms ] . the arrows show the atoms and the direction in which the number of tori increases , the number of tori itself is given in squares . we see that on each wall all tori of the adjacent chambers are involved in the corresponding bifurcation . indeed , in each chamber the accessible region for the separated variables in this system consists of one rectangle , therefore , all the tori projecting onto that rectangle bifurcate simultaneously . fix some chamber and consider the corresponding accessible regions @xmath360\qquad ( i=1,2).\ ] ] here @xmath361 are the roots of @xmath362 . theorem [ theo1 ] states that regular tori are the connected components of @xmath102 , where @xmath85 are defined by . let us introduce the angular variables @xmath363 in such a way that @xmath364 \sqrt{\eta_i - u_i}= \sqrt{\eta_i-\xi_i}\cos(\varphi_i ) , \qquad \sqrt{u_i-\xi_i}= \sqrt{\eta_i-\xi_i}\sin(\varphi_i ) . \end{array}\ ] ] square roots of constant values are always supposed non - negative . substituting into we get the equations of @xmath85 containing @xmath365 and , maybe , some radicals which have constant sign on each connected component of the integral manifold . according to agreement [ theagre1 ] this sign is either @xmath366 or @xmath367 and , for the chosen connected component of @xmath46 , it does not change inside the corresponding chamber . finally we obtain that the family of tori is defined by the set of those signs out of @xmath368 which remain in the final expressions for @xmath188 on the connected components of @xmath85 . [ theagre2 ] having expressed @xmath188 on the connected component of @xmath85 in terms of @xmath369 we will always consider this component oriented by the direction of increasing of the angle @xmath370 . in fact , the separation of variables allows us to assign the universal orientation to the tori of each family . indeed , the signs of the radicals in equations , though arbitrary , are strictly consistent with the signs in equations . substituting into , we obtain @xmath371 where @xmath372 and @xmath373 can equal @xmath366 , @xmath367 or @xmath109 . it means that the orientation of the cycle @xmath374 on all tori is given by the phase flow of the system @xmath24 . [ theagre3 ] we call the orientation of a regular torus positive if it is defined by the pair of cycles @xmath375 , where @xmath376 and the orientation of @xmath374 is induced by the phase flow . since @xmath377 in are the same for the whole family , by the described universal algorithm we fix a positive orientation of the family . consider an arbitrary atom @xmath254 together with some direction in which it is crossed along a chosen path @xmath82 in @xmath378 . the boundary @xmath379 consists of regular tori divided in a natural way along @xmath82 into incoming and outgoing ones . the admissible bases @xmath380 on all boundary tori can be composed from the curves @xmath85 in one of the following ways @xmath381 or @xmath382 depending on the type of @xmath254 . some of the atoms may not have incoming or outgoing tori . [ theagre4 ] choosing admissible bases on the boundary tori of an atom we always suppose that the orientation of these bases is positive ( in the sense of agreement [ theagre3 ] ) for all outgoing tori and negative for all incoming ones . let us give the explicit formulas for the parameterized cycles @xmath85 and the positive bases on the families . here we use the information given in table 1 . for the curve @xmath145 we have three different cases . in chambers @xmath383 the variable @xmath384 oscillates in @xmath385 $ ] and therefore we put @xmath386 } \sin\varphi_1,\\ \sgn \dot \varphi _ 1 = + 1 . } \right.\ ] ] let us once give remarks to the sign of @xmath387 applied in all similar representations of @xmath85 . making the formal substitution of the introduced expressions for @xmath388 into the first equation we get @xmath389}.\ ] ] in all such cases the remaining non - constant square roots will have constant sign along the corresponding trajectory . we suppose them to be positive . here there are two of them . to change both signs , it is enough to substitute @xmath390 without changing the sign of @xmath391 . to change only one of the signs we may substitute @xmath392 or @xmath393 this will change both the default orientation of @xmath145 and the sign of @xmath391 all over the chamber and , of course , will not affect the class of equivalent gluing matrices on the corresponding families . using the same method of the formal substitution with square roots considered positive , we obtain for chambers @xmath394 @xmath395[u_1(\varphi_1)-\tau_2]},\\ \sgn \dot \varphi _ 1 = e_1 , \end{array } \right.\ ] ] and for chamber @xmath302 @xmath396 analogously , for the curve @xmath146 we obtain in chambers @xmath397 @xmath398 } \cos\varphi_2,\\ \sgn \dot \varphi_2 = + 1 , \end{array } \right.\ ] ] in chambers @xmath399 @xmath400[u_2(\varphi_2)-\sigma_2]},\\ \sgn \dot \varphi_2 = e_2 , \end{array } \right.\ ] ] and in chamber @xmath162 @xmath401 finally we can say that for a given chamber the number of families is equal to the number of different combinations of @xmath402 and on each family the positive orientation is defined by the basis @xmath403 . the complete information on the families and the corresponding bases is given in table 3 . ' '' '' chamber & @xmath404 & @xmath339 & @xmath155 & @xmath162 & @xmath167 & @xmath302 + & @xmath120 & @xmath121 & @xmath121 & @xmath196 & @xmath196 & @xmath196 + ' '' '' bases & @xmath405 & @xmath406 & @xmath407 & @xmath408 & @xmath409 & @xmath410 + let us denote by @xmath411 the atoms arising in the pre - image of segments @xmath412 of the bifurcation diagram , where @xmath413 is the number assigned to the corresponding segment . if there are two atoms in the pre - image of a point on segment @xmath413 , we denote them by @xmath414 using for @xmath415 the sign built out of the fixed signs of the radicals as defined above . the only exception is @xmath416 having four components . to each of them we assign the pair of sings . obviously , in this notation we have the following atoms @xmath417 for a non - degenerate 3-atom , we can say that its sign is defined by the sign of the connected component of the curve out of @xmath418 that does not bifurcate at this moment . the signs of the atoms in the pre - image of a point on the curve @xmath103 ( segments @xmath213 , @xmath215 , and @xmath141 ) separate those tori families which are not identified with each other upon reaching @xmath103 . to establish for each atom ( excluding for a while the points of @xmath103 ) the uniquely defined admissible cycles ( @xmath419-cycles for @xmath256 and @xmath420-cycles for hyperbolic atoms ) , we use table 2 and equations giving the orientation on @xmath85 induced by the phase flow . the general rule of choosing this cycle is as follows . first , we take the curve @xmath85 for which the accessible region does not contain a double root ( i.e. , @xmath217 ) . then we multiply it by the sign of @xmath421 from the adjacent chamber . this sign , of course , coincides with the sign of the corresponding cycle in the basis taken for the positive orientation of the family in this chamber in table 3 . for example , on segment @xmath196 with two atoms @xmath256 we take @xmath422 from table 2 and choose @xmath240 for the sign from the basis of chamber @xmath155 in table 3 . on segment @xmath211 with two atoms @xmath259 we take @xmath423 from table 2 and then choose @xmath241 for the sign from the basis of any of chambers @xmath339 or @xmath302 in table 3 . the second cycle in the pair @xmath380 defining an admissible coordinate system is chosen according to agreements [ theagre3 ] and [ theagre4 ] . it is convenient to collect in one table all admissible coordinate systems for the atoms on the segments of the bifurcation diagram written out for some globally fixed direction of crossing these atoms . let us take the direction of the increasing @xmath180-coordinate and denote the corresponding pairs of cycles for an atom @xmath411 by @xmath424 and @xmath425 . while constructing the loop molecules we may meet with the necessity to cross some atoms in the inverse direction . then we denote such admissible bases for an atom @xmath411 by @xmath426 and @xmath427 . the connection between these bases is obvious @xmath428 = 0.35em ' '' '' seg . & 1 & 2 & 3 & 4 & 5 & 6 & 7 + ' '' '' @xmath429 & @xmath430 & @xmath430 & @xmath430 & @xmath430 & @xmath431 & @xmath432 & @xmath433 + ' '' '' @xmath434 & @xmath405 & @xmath435 & @xmath436 & @xmath437 & @xmath438 & @xmath437 & @xmath439 + ' '' '' seg . & 8 & 9 & 10 & 11 & @xmath440 & @xmath441 & 13 + ' '' '' @xmath429 & @xmath442 & @xmath442 & @xmath433 & @xmath443 & @xmath444 & @xmath445 & @xmath446 + ' '' '' @xmath434 & @xmath447 & @xmath448 & @xmath449 & @xmath430 & @xmath430 & @xmath430 & @xmath430 + the result is given in table 4 . for the curve @xmath103 the general rule does not work since the tori in the pre - image are in fact regular manifolds . for the new atoms @xmath283 , @xmath315 , @xmath265 and @xmath345 we need to establish some other rules . for the atoms @xmath283 we take admissible bases with the @xmath420-cycle coming from the adjacent atom @xmath259 in the molecule . thus , the notation @xmath440 stands for the case of the edge coming from segment @xmath207 ( the part of the loop molecule of the point @xmath228 ) and @xmath441 is used for the bases obtained along the edge coming from segment @xmath204 ( the part of the loop molecule of the point @xmath229 ) . consider the 3-atom @xmath450 on segment @xmath141 . obviously , @xmath160 here stands for the global @xmath420-cycle @xmath451 which came from segment @xmath211 . so let it be the first cycle of the admissible basis on @xmath452 . for the @xmath419-cycle let us take the circle in the mbius band that covers twice the middle line of the band oriented according to agreement [ theagre4 ] . on the family in chamber @xmath302 corresponding to the signs @xmath453 , the basis @xmath454 is negatively oriented . then from table 3 , its orientation coincides with that of @xmath455 , which is the same as of @xmath456 . let us write out equations and on the curve @xmath103 ( @xmath457 ) : @xmath458},\\ \sgn \dot \varphi_2 = e_2 . \end{array } \right.\ ] ] it is easy to see that @xmath90 acts on @xmath102 as the simultaneous shift @xmath459 and @xmath460 . therefore the circle folding twice is @xmath461 and we come to the basis for segment @xmath141 as in table 4 . since the 3-atom @xmath345 is non - orientable and @xmath462 consists of only one torus , the solution here is standard , i.e. , for an outgoing torus we take the positive orientation of the family , but for an incoming one we take the negative orientation of the family . on the contrary , the 3-atom @xmath265 is orientable and the choice of its orientation defines the orientations on two boundary tori . in our case at the point @xmath228 both of them are incoming ( supposing that one of the coordinates @xmath180 or @xmath185 increases while another is fixed ) . two families come to the point @xmath228 from chamber @xmath339 . choosing the negative orientation from the families as given by the pair @xmath463 ( see table 3 ) , we have one of them different from the atom s orientation . we mark this situation by the notation @xmath464 . note that we may choose the orientation on @xmath465 as @xmath466 . then it is consistent with some orientation of @xmath265 . obviously , in this case one of the gluing matrices on the incoming edges will have the determinant equal to @xmath467 . if the topology of the molecule makes it possible to change the direction of this edge and the following ones without general contradiction , then we can change orientations of the corresponding families of tori and obtain an orientable molecule . but if this edge is a part of a loop , such a change may be impossible . then the molecule defines a non - orientable loop manifold . as it was mentioned above , in the system @xmath24 we have exactly four critical points of rank 0 , all four are non - degenerate and their neighborhoods have representations . therefore the corresponding loop molecules are well - known and completely classified by the bolsinov theorem @xcite . nevertheless , in this section we demonstrate the use of separated variables for the process of constructing these loop molecules . let us take small closed paths around the points @xmath468 in the clockwise direction . then from tables 3 and 4 we have the following chains : @xmath469 substituting the bases from table 4 and we readily obtain not only the gluing matrices but also the rules by which families choose their bounding atoms . indeed , let us take for example the edge connecting @xmath470 and @xmath471 . we have @xmath472 we see that the two families started from the @xmath259-atom @xmath473 ( @xmath474 ) differ by the sign @xmath241 and therefore come to different @xmath259-atoms on segment @xmath204 . for the sake of brevity we denote some @xmath475-matrices needed below @xmath476 as usual , @xmath477 stands for the identity matrix . the loop molecules of non - degenerate singularities @xmath299 ( @xmath478 ) generated by the chains are shown in fig . [ figmols1234 ] : ( @xmath238 ) for @xmath479 , ( @xmath239 ) for @xmath480 , and ( @xmath244 ) for @xmath481 . of course , our method of global choice of orientation affects the gluing matrices , but since the possible changes ( e.g. changing the order of the separation variables ) are applied to all tori simultaneously , the gluing matrices of the molecules considered in this section simultaneously change their signs ( @xmath482 ) . this only leads to another representatives of the same exact topological invariants . in the system @xmath24 the set of degenerate closed orbits consists of the motions in the pre - image of the points @xmath289 . [ theo3 ] the loop molecules of the points @xmath289 can be represented in the form shown in fig . [ figlmolq1q2 ] . the first one is connected , while the second consists of two equivalent connected components differing by the sign @xmath241 . consider a closed path surrounding the point @xmath228 clockwise . recall that the atoms @xmath259 both in @xmath483 and @xmath484 differ by the sign @xmath240 . at the same time on @xmath103 we have to identify the families from chamber @xmath162 having the same product @xmath485 and from chamber @xmath167 having the same sign @xmath241 . then we obtain the picture of the molecular edges printed over a piece of the bifurcation diagram as shown in fig . [ figtomolq1xn ] . in the whole neighborhood of @xmath228 all @xmath420-cycles on the regular tori are induced by the periodic critical trajectories of the atoms @xmath486 . then from table 2 we readily obtain that both @xmath487 and @xmath488 ( no matter , incoming or outgoing ) are defined as @xmath489 . taking the chains marked with the points @xmath490 we get @xmath491 from table 4 , the gluing matrix is @xmath492 at both points @xmath493 . let us find the gluing matrices on the arcs connecting the atoms @xmath494 . to be definite , let us take the points @xmath495 on the arc with @xmath496 and shift the corresponding bases @xmath497 to the torus @xmath498 . we get two bases @xmath466 and @xmath499 . on segment @xmath213 with @xmath500 we obtain from and @xmath501 where @xmath502}>0 $ ] . the obvious substitution @xmath503 turns @xmath504 into @xmath505 . a similar reasoning leads to the equality @xmath506 on @xmath498 . finally we have that the map @xmath90 identifies two incoming bases at @xmath507 , so they give the same basis on the image torus in @xmath24 . therefore the gluing matrix on the arc is the identity matrix @xmath477 . another arc @xmath508 is considered analogously and gives the same result . now let us turn to the arcs in chamber @xmath167 . shifting the bases @xmath509 from the points @xmath510 to the point @xmath511 ( i.e. , to the torus @xmath512 ) we get two bases @xmath513 and @xmath499 . in the same way as above , yields @xmath514 , but from we obviously have @xmath515 . so again @xmath90 identifies two bases at @xmath511 and they give the same basis on the image torus in @xmath24 . the gluing matrix is @xmath477 . this proves the statement on the loop molecule of @xmath228 . consider a closed path surrounding the point @xmath229 clockwise . recall that the atoms @xmath259 both in @xmath516 and @xmath517 differ by the sign @xmath241 . at the same time on @xmath103 we have to identify the families from chamber @xmath167 having the same sign @xmath241 . the families in chamber @xmath302 are not identified with each other and end with the atoms @xmath315 . this leads to two rough invariants homeomorphic to that shown in fig . [ figlmolq1q2 ] . the picture of the molecular edges together with a piece of the bifurcation diagram is shown in fig . [ figtomolq2xn ] . in this case the globally defined @xmath420-cycles are @xmath451 . since the connected components of the loop molecule differ exactly by the sign @xmath241 , this cycle is fixed for the whole component and is not affected by any identifications in the pre - image of @xmath103 . the edges in chamber @xmath339 give @xmath518 . the gluing matrix is @xmath492 . in chamber @xmath167 let us fix @xmath241 and shift the bases @xmath519 to the point on @xmath103 . for @xmath520 , we get two bases @xmath521 and @xmath522 . previously we stated that , in chamber @xmath167 , @xmath514 and @xmath515 , therefore the obtained bases at the point of @xmath103 give the same basis on the image torus in @xmath24 . thus , the gluing matrix on the loops is the identity matrix . finally , comparing the bases for segments @xmath211 and @xmath141 from table 4 we see that the gluing matrix is @xmath523 where @xmath524 . the theorem is proved . note that the structure of the molecule of @xmath228 in fig . [ figlmolq1q2 ] allows us to pass the whole molecule in one direction . instead of going globally right - to - left , let us change the direction in the upper half . then we have to change the admissible bases on the boundary tori of the atoms @xmath525 . the matrix @xmath492 remains unchanged but all the identity matrices turn into @xmath526 . so , in this case we can avoid gluing matrices with the determinant equal to @xmath467 and the loop manifold is orientable . obviously , this can not be done for the molecules of the point @xmath229 and for some similar iso - integral molecules having loops rather than arcs . let us denote by @xmath527 the result of gluing @xmath187 handles and @xmath528 mbius bands to a closed 2-surface @xmath529 ( of course , it supposes cutting out small discs first ) . from theorem [ theo3 ] we find the topology of the loop manifolds of the degenerate points . [ thecorol ] the loop manifold of the point @xmath228 is homeomorphic to the direct product @xmath530 , i.e. , to the loop manifold of the non - degenerate tom @xmath262 . the loop manifold of the point @xmath229 is homeomorphic to two copies of the direct product @xmath531 . indeed , all @xmath532-marks in the molecules of @xmath289 are equal to @xmath533 and the loop manifolds are easily restored from the topology of the graph . this fact corresponds to the mentioned above property of globally defined @xmath420-cycles ( at the point @xmath228 two @xmath420-cycles @xmath489 are identified by the @xmath234-symmetry @xmath90 in the pre - image of @xmath103 ) . in this section , we present the result of constructing all possible iso-@xmath325 and iso - energy marked molecules which occur in the system @xmath24 , except for the critical values of the integrals at four singular points @xmath534 , @xmath535 and , for the restriction of @xmath5 to @xmath24 , the minimal value @xmath536 of the @xmath180-coordinate on the curve @xmath103 . in table 5 , the notation @xmath537 stands for orientable molecules , @xmath538 for non - orientable ones , @xmath539 for splitting ones . we also include the non - compact iso-@xmath35 molecule @xmath540 for the level @xmath541 where the symplectic structure degenerates . for non - negative values of @xmath185 the iso-@xmath325 manifolds are non - compact . thus , crossing the zero value , these manifolds bifurcate without any critical points of @xmath325 appearing on the zero level . three corresponding non - compact molecules are denoted by @xmath542 . note that the molecule @xmath543 was found in @xcite as describing the zero level of the bogoyavlensky integral . this level , due to , coincides with the manifold @xmath544 . the molecules with the gluing matrices are representatives of their liouville equivalence classes obtained by our way of choosing the coordinate systems and orientations on the families of tori . in @xcite , one can find the description of possible changes that occur due to the changes of directions and orientations . finally , in table 6 we show the correspondence between the values of @xmath185 and @xmath180 and the molecules found . @xmath185-value & code & @xmath180-value & code + + @xmath185-value & code & @xmath180-value & code + @xmath545 & @xmath546 & @xmath547 & @xmath548 + @xmath549 & @xmath546 & @xmath550 & @xmath551 + @xmath552 & @xmath553 & @xmath554 & @xmath555 + @xmath556 & @xmath557 & @xmath558 & @xmath559 + @xmath560 & @xmath561 & @xmath562 & @xmath563 + @xmath564 & @xmath553 & @xmath565 & @xmath566 + @xmath567 & @xmath568 & @xmath569 & @xmath570 + @xmath571 & @xmath572 & @xmath573 & @xmath574 + @xmath575 & @xmath576 & @xmath577 & @xmath570 + @xmath578 & @xmath543 & & + @xmath579 & @xmath580 & & + @xmath581 & @xmath582 & & + this completes the exact topological analysis of the system . the author is grateful to a.v.bolsinov for extremely valuable discussions and advices , to v.n.roubtsov , the university of angers and the organizers of fdis-2013 for hospitality and support . m. p. kharlamov , bifurcation diagrams of the kowalevski top in two constant fields , regular and chaotic dynamics 10 ( 4 ) ( 2005 ) 381398 . [ ] , http://dx.doi.org/10.1070/rd2005v010n04abeh000321 [ ] . m. p. kharlamov , one class of solutions with two invariant relations of the problem of motion of the kowalevski top in double constant field , mekh . tela 32 ( 2002 ) 3238 . http://arxiv.org/abs/0803.1028 [ ] . m. p. kharlamov , a. y. savushkin , separation of variables and integral manifolds in one problem of motion of generalized kowalevski top , ukrainian mathematical bulletin 1 ( 4 ) ( 2004 ) 569586 . http://arxiv.org/abs/0803.0882 [ ] . m. p. kharlamov , separation of variables in the generalized 4th appelrot class , regular and chaotic dynamics 12 ( 3 ) ( 2007 ) 267280 . http://arxiv.org/abs/0803.1024 [ ] , http://dx.doi.org/10.1134/s1560354707030021 [ ] . t. i. pogosian , m. p. kharlamov , bifurcation set and the integral manifolds in the problem of motion of the rigid body in linear force field , journal of applied mathematics and mechanics 43 ( 3 ) ( 1979 ) 452462 . http://dx.doi.org/10.1016/0021-8928(79)90093-5 [ ] . m. p. kharlamov , bifurcation of common levels of first integrals of the kovalevskaya problem , journal of applied mathematics and mechanics 47 ( 6 ) ( 1983 ) 737743 . http://dx.doi.org/10.1016/0021-8928(83)90109-0 [ ] . l. m. lerman , y. l. umanski , classification of four - dimensional integrable systems and the poisson action of @xmath583 in extended neighborhoods of simple singular points . i. , sbornik : mathematics 183 ( 12 ) ( 1992 ) 141176 . e. ryabov , m. p. kharlamov , classification of singularities in the problem of motion of the kovalevskaya top in a double force field , sbornik : mathematics 203 ( 2 ) ( 2012 ) 257287 . http://dx.doi.org/10.1070/sm2012v203n02abeh004222 [ ] .

the chiral two - dimensional electron gas in landau levels of a bernal - stacked graphene bilayer has a valley - pseudospin ising quantum hall ferromagnetic behavior at odd filling factors of these fourfold degenerate states . at zero interlayer electrical bias , the ground state at these fillings is spin polarized and electrons occupy one valley or the other while a finite electrical bias produces a series of valley pseudospin - flip transitions . in this work , we extend the ising behavior to chirally - stacked multilayer graphene and discuss the hysteretic behavior of the ising quantum hall ferromagnets . we compute the transport gap due to different excitations : bulk electron - hole pairs , electron - hole pairs confined to the coherent region of a valley - pseudospin domain wall , and spin or valley - pseudospin skyrmion - antiskyrmion pairs . we determine which of these excitations has the lowest energy at a given value of the zeeman coupling , bias , and magnetic field .

1410.4232

the chiral two - dimensional electron ( c2deg ) gas in bilayer graphene ( blg ) in a quantizing magnetic field exhibits a rich variety of broken - symmetry states in mean - field theory . these states have been extensively studiedbarlasreview in the zeroth - energy landau level @xmath2 which is eightfold degenerate when valley @xmath3 and spin @xmath4 degrees of freedom are counted because of the presence of an extra orbital degree of freedom @xmath5@xcite@xmath6 these broken symmetry states are best described using the pseudospin language where a valley pseudospin up(down ) is associated with valley @xmath7 and an orbital pseudospin up(down ) is associated with the orbital @xmath8 in this language , the broken - symmetry states are spin , valley , or orbital quantum hall ferromagnets ( qhfs ) . they are gapped , have charged quasiparticle excitations , and support an integer quantum hall effect@xcite . the effect of the coulomb interaction on the higher landau levels @xmath9 has so far received less attention . in a recent paper@xcite , we have shown that the c2deg in these levels behaves as a valley ising qhf when the filling factor of these levels ( which are fourfold degenerate ) is odd , i.e. when @xmath10 the c2deg at these fillings is spin polarized and the two valley states with valley pseudospin @xmath11 are degenerate at zero and at some finite values of the bias @xmath12 ( the electrical potential difference between the two layers ) . for @xmath13 transitions occur between states with total spin and total valley pseudospin @xmath14 changing the bias can induce first - order transitions between the different spin and valley - pseudospin ground states . the phase diagram for different landau levels shows a marked difference between the positive @xmath15 and negative @xmath16 landau levels . the ising behavior of the higher landau levels contrasts with that of landau level @xmath2 where valley , spin , or orbital qhf can all have a @xmath17 symmetry@xcite . ising qhfs have been studied previously both theoretically@xcite and experimentally@xcite in a variety of semiconductor quantum wells or bilayer systems . a complete classification of the qhf states , including isotropic , easy - axis , and easy - plane qhfs has been derived for a model of two nearby , infinitely narrow , two - dimensional layers@xcite . a well - known case of ising qhf occurs when two levels with different electrical subbands or landau level indices are brought to degeneracy by tuning the electrical bias or the zeeman coupling at even filling factor . the _ valley_-pseudospin ising transitions in the higher landau levels , however , occur at odd filling factors and are of a slightly different nature : they do not occur near landau level crossings . as in other ising qhfs , however , the exchange part of the coulomb interaction energy plays a major role as it competes with the noninteracting energy . what is special with blg is that an electronic state is described by a four - component spinor and so consists of a superposition of four states with different orbital wave functions . as the bias is varied , the coefficients of this superposition are modified and so is the coulomb exchange interaction which is orbital - dependent . in a way , the ising transitions in blg occur in part because of a change in the internal structure of the electron . the discontinuous transitions between spin - polarized states at @xmath18 that occur when the bias is varied cause abrupt changes in the transport gap that may explain some recent experimental results@xcite on the transport properties of the c2deg in bilayer graphene . in the present work , we study the behavior of the transport gap of the c2deg with bias at odd fillings of the higher landau levels . we first show that the ising behavior found previously@xcite should also occur in chirally - stacked multilayer graphene , at least at zero bias . we then compute the energy of different types of charged excitations of the ising ground states . we consider bulk quasiparticles ( quasi - electrons and quasi - holes ) and topological quasiparticles ( skyrmions and antiskyrmions ) . in each case , we evaluate the transport gap due to the excitation of a particle - antiparticle pair using a microscopic hartree - fock calculation . our approach allows us to compute the excitation gap at finite zeeman and/or bias and takes into account the possibility of intertwined spin and valley - pseudospin textures . when appropriate ( at zero bias or zeeman coupling ) , we also compute the transport gap using a long - wavelength approximation of the hartree - fock energy functional . when the zeeman coupling is taken into account , we find that valley skyrmions are the lowest - energy excitations in bilayer graphene , at least for @xmath19 . this contrasts with previously studied ising systems where skyrmions are not the lowest - energy excitations@xcite . the skyrmions that we find have either a spin or a valley - pseudospin texture , but not both@xcite and , as in monolayer graphene@xcite , they exist only in a small range of bias or zeeman coupling . magnetic ising ferromagnets show an hysteretic behavior of the magnetization as the magnetic field is swept upward or downward . in the absence of disorder and at @xmath20 the external magnetic field must exceed a certain coercive field in order to produce the spin reversal . we find a similar behavior with bias for the valley pseudospins and so we compute the coercive bias for landau levels @xmath21 owing to the fact that more valley - flip transitions are possible in levels @xmath22 than in @xmath23 we find that the hysteretic behavior for @xmath22 is also more complex , with a very large coercive field . at finite temperature or in a disordered system , domains with particular valley - pseudospin orientations are formed in the c2deg . regions of different valley pseudospins are separated by domain walls which can have complicated patterns . when domain walls are present , the electronic band structure and so the electron - hole excitation gap is modulated in space . in the coherent region of a domain wall , the valley pseudospins rotate from one polarization to the other . for a linear domain wall , the coherent region forms a linear channel of a few magnetic lengths in width . at the center of these regions , the band gap is minimal . it should be possible , then , to excite electron - holes pairs that are confined to a coherent region of a wall and that have smaller excitation energy than the bulk electron - hole or the skyrmion - antiskyrmion pairs . these mobile particles may be responsiblejungwirth for the increase in dissipation and breakdown of the quantum hall effect that is observed in some ising qhfs@xcite . we compute the energy of such excitations and find that , in some range of bias , they can be the lowest - energy excitations . this paper is organized in the following way . sections ii and iii introduce the tight - binding model and the electronic states in bilayer graphene as well as most of the hartree - fock formalism necessary to derive the energy gap for the particle - antiparticle pairs . section iv discusses the hysteretic behavior of the ising qhfs and extend this behavior to other , chirally - stacked , multilayer graphene structures . the transport gaps for the electron - hole pairs , confined electron - hole pairs and skyrmion - antiskyrmion pairs are computed in secs . v , vi , and vii respectively . section viii is the conclusion . the appendix provides a derivation of the projected representation of a skyrmion that is used in sec . bilayer graphene consists of two graphene layers separated by a distance @xmath24 . in each layer , the crystal structure is a honeycomb lattice that can be described as a triangular bravais lattice with a basis of two carbon atoms @xmath25 where @xmath26 is the layer index . the in - plane lattice constant is @xmath27 . in the bernal stacking , the upper @xmath28 sublattice is directly on top of the lower @xmath29 sublattice , while the upper @xmath30 sublattice is above the center of a hexagonal plaquette of the lower layer . in the continuum approximation , the hamiltonian is expanded around the two non - equivalent valley points @xmath31 of the brillouin zone@xcite . in a transverse magnetic field @xmath32 , the tight - binding hamiltonian for an electron with the valley index @xmath33 is given , in the basis @xmath34 by@xmath35where @xmath36 with @xmath37 the magnetic length and @xmath38 the hopping parameters . a recent calculation@xcite gives @xmath39 ev for the nearest - neighbor hopping , @xmath40 ev for the interlayer hopping between carbon atoms that are immediately above one another ( i.e. @xmath41 ) and @xmath42 ev for the interlayer next nearest - neighbor hopping between carbons atoms in the same sublattice ( i.e. @xmath43 and @xmath44 ) . the magnetic field considered in the present work is assumed sufficiently large for the warping term @xmath45 to be negligible@xcite . the parameter @xmath46 ev represents the difference in the crystal field between sites @xmath47 and @xmath48 . the ladder operators @xmath49 and @xmath50 are defined in the landau gauge @xmath51 in the absence of a magnetic field , the electronic dispersion has four bands which we denote by @xmath52 with @xmath53 the negative and @xmath54 the positive energy bands . the two middle bands @xmath55 touch each other at the six valley points @xmath56 where they have zero energy in the absence of bias . the high - energy bands @xmath57 are separated from the middle two energy bands by a gap of order @xmath58 in a quantizing magnetic field , each band is split into landau levels . we shall be concerned only with bands @xmath59 in this work . the eigenstates of @xmath60 for the levels of these bands have energies @xmath61 with @xmath62 and the corresponding eigenvectors are given by @xmath63 and@xmath64where @xmath65 is the guiding - center index . the functions @xmath66 where @xmath67 is the position of the graphene layer @xmath68 by definition , @xmath69 if @xmath70 otherwise@xmath71where @xmath72 is the level index and @xmath73 the sample dimension in the @xmath74 direction . the functions @xmath75 are the eigenstates of the one - dimensional harmonic oscillator . in a minimal tight - binding model where only @xmath76 are kept , levels @xmath5 are degenerate and have zero energy . when spin and valley degrees of freedom are counted and the zeeman term is neglected , there are in total eight degenerate levels with zero energy . this octet of states defines the landau level @xmath77 the higher landau levels have @xmath78 with @xmath79 the positive energy landau levels @xmath80 have band index @xmath81 while the negative energy ones have @xmath82 with @xmath83 this notation is illustrated in fig . 1 of ref . . contrary to @xmath84 the higher - energy landau levels @xmath9 are fourfold degenerate . each level with quantum numbers @xmath85 has the macroscopic degeneracy @xmath86 where @xmath87 is the c2deg area , associated with the guiding - center index @xmath65 . in this work , we assume that landau level mixing can be neglected . as shown in ref . , this limits the electric bias to @xmath12 @xmath88 ev for @xmath89 t and @xmath12 @xmath90 ev for @xmath91 t. we project the hamiltonian into level @xmath92 so that the electron field operator is given by@xmath93where @xmath94 annihilates an electron is state @xmath95 and the super - index @xmath96 corresponds to the four states@xmath97where the second index in the parenthesis is for the spin . hereafter , we use the notation @xmath98 and @xmath99 the various energy gaps are computed in this work in the hartree - fock approximation ( hfa ) . in a recent work on the ising qhfs@xcite , the hfa was used to compute the energy of ground states with uniform spin and/or valley polarizations . in this section , the hfa is extended to the study of excitations with spin and/or pseudospin textures . to simplify the notation , the subscript @xmath100 is omitted whenever possible , since the hilbert space is restricted to level @xmath92 only . the hartree - fock hamiltonian for electrons in level @xmath92 is given by @xmath101 \notag \\ & & + \sum_{\mu , \mu ^{\prime } } \overline{\sum_{\mathbf{q}}}h^{\left ( \mu _ { \xi } , \mu _ { \xi } ^{\prime } \right ) } \left ( \mathbf{q}\right ) \left\langle \rho _ { \mu , \mu } \left ( -\mathbf{q}\right ) \right\rangle \rho _ { \mu ^{\prime } , \mu ^{\prime } } \left ( \mathbf{q}\right ) \notag \\ & & -\sum_{\mu , \mu ^{\prime } } \sum_{\mathbf{q}}x^{\left ( \mu _ { \xi } , \mu _ { \xi } ^{\prime } \right ) } \left ( \mathbf{q}\right ) \left\langle \rho _ { \mu , \mu ^{\prime } } \left ( -\mathbf{q}\right ) \right\rangle \rho _ { \mu ^{\prime } , \mu } \left ( \mathbf{q}\right ) , \notag\end{aligned}\ ] ] where @xmath102 with @xmath103 and @xmath104 the bohr magneton , is the zeeman coupling . the bar over the summation in the hartree term means that the @xmath105 is omitted from the sum because a positive background is considered in order to make the system neutral . we have defined the constant @xmath106and @xmath107 is the filling factor of level @xmath92 with @xmath108 the number of electrons in this level . in @xmath109 @xmath110 is the dielectric constant of the substrate . in all the calculations of this paper , @xmath111 the ground state average value of the operators @xmath112can be considered as the order parameters of a specific phase of the electron gas . the operators @xmath113with @xmath114 and the projectors@xmath115and@xmath116are defined such that @xmath117 is the filling factor of _ layers _ @xmath114 . it is important to remember that , unlike level @xmath84 layer and valley indices are not equivalent in the higher landau levels . the hartree and fock interactions that enter @xmath118 are defined by ( here , @xmath119)@xmath120with @xmath121 and ( @xmath122 for @xmath114 ) and @xmath123 , \notag \\ v^{\left ( \xi , \overline{\xi } \right ) } \left ( x\right ) & = & g_{1}^{\left ( \xi \right ) } g_{2}^{\left ( \overline{\xi } \right ) } + g_{2}^{\left ( \xi \right ) } g_{1}^{\left ( \overline{\xi } \right ) } \\ & & + e^{-qd}\left [ g_{1}^{\left ( \xi \right ) } g_{1}^{\left ( \overline{\xi } % \right ) } + g_{2}^{\left ( \xi \right ) } g_{2}^{\left ( \overline{\xi } \right ) } % \right ] , \notag\end{aligned}\]]with the functions@xmath124 \notag \\ g_{2}^{\left ( \xi \right ) } \left ( x\right ) & = & e^{-x^{2}/4}\left [ \left\vert b_{3,\xi } \right\vert ^{2}l_{n-2}\left ( \frac{x^{2}}{2}\right ) \right . \label{yy3 } \\ & & + \left . \left\vert b_{4,\xi } \right\vert ^{2}l_{n-1}\left ( \frac{x^{2}}{2}% \right ) \right ] , \notag\end{aligned}\]]where @xmath125 is a laguerre polynomial . the hartree - fock energy per electron is @xmath126 ^{2 } \label{ener1 } \\ & & + \frac{1}{\nu _ { n}}\sum_{\mu } \left ( e_{\mu _ { \xi } } ^{\left ( 0\right ) } -% \frac{\mu _ { \sigma } \delta _ { z}}{2}\right ) \left\langle \rho _ { \mu , \mu } \left ( 0\right ) \right\rangle \notag \\ & & + \frac{1}{2\nu _ { n}}\sum_{\mu , \mu ^{\prime } } \overline{\sum_{\mathbf{q}}}% h^{\left ( \xi , \zeta \right ) } \left ( \mathbf{q}\right ) \left\langle \rho _ { \mu , \mu } \left ( -\mathbf{q}\right ) \right\rangle \left\langle \rho _ { \mu ^{\prime } , \mu ^{\prime } } \left ( \mathbf{q}\right ) \right\rangle \notag \\ & & -\frac{1}{2\nu _ { n}}\sum_{\mu , \mu ^{\prime } } \sum_{\mathbf{q}}x^{\left ( \xi , \zeta \right ) } \left ( \mathbf{q}\right ) \left\langle \rho _ { \mu , \mu ^{\prime } } \left ( -\mathbf{q}\right ) \right\rangle \left\langle \rho _ { \mu ^{\prime } , \mu } \left ( \mathbf{q}\right ) \right\rangle . \notag\end{aligned}\ ] ] to compute the order parameters , we define the matrix of matsubara green s functions @xmath127where @xmath128 is the time - ordering operator . its fourier transform is defined by @xmath129so that@xmath130 the matrix of green s functions is computed from its hartree - fock equation of motion @xmath131 g_{\mu , \mu ^{\prime } } \left ( \mathbf{q},\omega _ { n}\right ) = \hslash \delta _ { \mathbf{q},0}\delta _ { \mu , \mu ^{\prime } } \label{eqmotion } \\ & & + \sum_{\mu ^{\prime \prime } } \overline{\sum_{\mathbf{q}^{\prime } \neq \mathbf{q}}}h^{\left ( \mu _ { \xi } ^{\prime \prime } , \mu _ { \xi } \right ) } \left ( \mathbf{q}^{\prime } -\mathbf{q}\right ) \left\langle \rho _ { \mu ^{\prime \prime } , \mu ^{\prime \prime } } \left ( \mathbf{q - q}^{\prime } \right ) \right\rangle \notag \\ & & \times e^{-i\left ( \mathbf{q}\times \mathbf{q}^{\prime } \right ) \cdot \widehat{\mathbf{z}}\ell ^{2}/2}g_{\mu , \mu ^{\prime } } \left ( \mathbf{q}% ^{\prime } , \omega _ { n}\right ) \notag \\ & & -\sum_{\mu ^{\prime \prime } } \sum_{\mathbf{q}^{\prime } } x^{\left ( \mu _ { \xi } ^{\prime \prime } , \mu _ { \xi } \right ) } \left ( \mathbf{q}^{\prime } -% \mathbf{q}\right ) \left\langle \rho _ { \mu ^{\prime \prime } , \mu } \left ( \mathbf{q - q}^{\prime } \right ) \right\rangle \notag \\ & & \times e^{-i\left ( \mathbf{q}\times \mathbf{q}^{\prime } \right ) \cdot \widehat{\mathbf{z}}\ell ^{2}/2}g_{\mu ^{\prime \prime } , \mu ^{\prime } } \left ( \mathbf{q}^{\prime } , \omega _ { n}\right ) , \notag\end{aligned}\]]where @xmath132 is the chemical potential and the renormalized energies are given by @xmath133where@xmath134 it is instructive to rewrite the hartree - fock energy in a pseudospin language . since the skyrmions found in the hfa do not contain intertwined spin and valley - pseudospin textures , it is sufficient to define pure spin and valley - pseudospin fields . for the total spin field , including both valleys , @xmath135for the total valley pseudospin , including both spin states , @xmath136finally , for the total electronic density@xmath137the fourier transform of these fields is defined by@xmath138where @xmath139 is one of the fields and @xmath140 is defined in term of the order parameters @xmath141 it does not include the form factors that reflect the character of the different orbitals present in the electronic spinor ( see eqs . ( [ spinor1 ] ) and ( [ spinor2]))@xmath6 we say that @xmath142 is given in the guiding - center representation ( gcr ) . we define the field @xmath143 as the part of the total spin field which is in valley @xmath144 while @xmath145 is the part of the total in - plane valley - pseudospin field with spin @xmath146 . the fields are coupled by the constraint@xmath147 \notag \\ & = & \nu _ { n } , \notag\end{aligned}\]]where @xmath148 etc . in the gcr , the hartree - fock energy per electron @xmath149 becomes @xmath150 the interactions are functions of @xmath151 only and are given by@xmath152 , \\ \lambda _ { \rho , z } & = & \sum_{\xi } \xi \left [ h^{\left ( \xi , \xi \right ) } -% \frac{1}{2}x^{\left ( \xi , \xi \right ) } \right ] , \\ \lambda _ { z , z } & = & \sum_{\xi } \left [ h^{\left ( \xi , \xi \right ) } -h^{\left ( \xi , \overline{\xi } \right ) } -\frac{1}{2}x^{\left ( \xi , \xi \right ) } \right ] .\end{aligned}\ ] ] in the limit of zero zeeman or bias coupling , skyrmions usually become very large and their energies can not be computed by the microscopic hfa of eq . ( [ eqmotion ] ) since the matrix of green s functions becomes very large . in this limit , it is more useful to make a gradient approximation of the total energy in eq . ( [ totalener ] ) , taking advantage of the fact that the spin or pseudospin texture varies slowly in space@xcite so that only the first few terms in the expansion can be kept . in levels @xmath0 of bilayer graphene , the gradient approximation works well for spin skyrmion but , as we will show below , can not be applied to valley - pseudospin skyrmions since these later topological excitations have a small finite size at zero bias . when all electrons occupy the valley @xmath153 at @xmath154 , there is no intervalley coherence and eq . ( [ totalener ] ) at zero bias and zero zeeman coupling simplifies to @xmath155 with@xmath156note that , at zero bias,@xmath157so that @xmath158 is independent of the valley index . for a spin - textured excitation , all electrons remain in the same valley but some of them flip their spin . if we neglect the charge density term ( i.e. the third term on the right - hand side of eq . ( [ espin ] ) ) , then a gradient approximation gives for the excitation energy with respect to the ground state @xmath159where @xmath160 and the field @xmath161 ( where @xmath87 is the 2deg area ) has unit modulus@xcite . the spin stiffness is defined by@xmath162eq . ( [ efunspin ] ) , with the condition that @xmath163 is the nonlinear @xmath146 model ( nl@xmath146 m ) . note that , @xmath164 is the energy of an uncharged spin texture . no electron is added or removed to the system . to compute the energy of a charged spin texture , it is necessary to consider the chemical potential@xcite . there is no need to worry about this problem in the present work since we are interested only in the transport gap which is the energy required to excite a skyrmion - antiskyrmion pair . this energy can be expressed solely@xcite in terms of @xmath165 ( see eq . ( [ eskask ] ) below ) . a finite zeeman coupling favors small spin textures while the charge - density term favors large textures . at finite zeeman coupling , the size of the spin texture is a compromise between these two terms and it can be small . the gradient approximation is then no longer valid . a microscopic hartree - fock calculation must be done to compute the energy of small spin textures . when all electrons occupy the spin state @xmath166 at @xmath154 , there is no spin coherence and the total hartree - fock energy at zero bias is @xmath167 where@xmath168 in a valley - pseudospin texture , no spin are flipped . neglecting again the charge - density term , the gradient approximation gives for the excitation energy with respect to an ising ground state with @xmath169 * * * * @xmath170 \label{evallee } \\ & & + \frac{1}{2}\sum_{\nu } \int d\mathbf{r}\left [ \rho _ { z}\left ( \partial _ { \nu } p_{z}\left ( \mathbf{r}\right ) \right ) ^{2}+\rho _ { \bot } \left ( \partial _ { \nu } p_{\bot } \left ( \mathbf{r}\right ) \right ) ^{2}\right ] \notag \\ & & + n_{\varphi } \alpha _ { d}\left ( n_{1,+}-n_{1,-}\right ) ^{2}\left [ \left ( p_{z}(q=0)\right ) ^{2}-\frac{1}{4}\right ] , \notag\end{aligned}\]]where @xmath171 is a unit field and the pseudospin stiffnesses are given by@xmath172and@xmath173while the easy - axis anisotropy term is given by @xmath174 the last term in eq . ( [ evallee ] ) is related to the capacitive energy . if @xmath175 it can be rewritten as@xmath176where @xmath177 is the number of electrons in valley @xmath178 even though the bias is zero , there is a capacitive energy because in each state @xmath179 there is a charge imbalance between the two layers which is given by @xmath180 numerically , @xmath181 and @xmath182 are all positive quantities . as an example , for @xmath183 t@xmath184 their value is @xmath185 mev@xmath186 mev , @xmath187 mev/@xmath188 and @xmath189 mev@xmath6 the easy - axis anisotropy and capacitive energy are comparable in size . in this section , we extend the ising behavior of bilayer graphene@xcite to chirally - stacked multilayer graphene and discuss the hysteretic behavior associated with the valley - pseudospin flip transitions at integer filling @xmath190 the ground state at @xmath154 is spin polarized and so , for a uniform state , the energy functional of eq . ( [ totalener ] ) simplifies to ( with @xmath191)@xmath192where @xmath193 is a constant and @xmath194 which depend implicitly on the bias @xmath195 are given by @xmath196and@xmath197with all interactions evaluated at @xmath198 the form of the energy functional @xmath199 ( with @xmath193 neglected ) is shown in fig . [ figure1](a ) for a finite bias . there are two extrema at @xmath200 and @xmath201 and another one at @xmath202 with respective energies@xmath203the metastable minimum ( chosen to be @xmath201 in the figure ) disappears when the curvature of @xmath199 changes sign or , equivalently , when @xmath204 or @xmath205 we show in fig . [ figure1](b ) and ( c ) the behavior of these three energies for @xmath206 at @xmath89 t and @xmath207 at @xmath208 t. when @xmath209 and @xmath210 or equivalently,@xmath211the two states @xmath212 are degenerate and the c2deg has an ising behavior@xmath6 this situation occurs at zero bias . numerically , the coefficient @xmath213 at all bias but @xmath214 . [ figure1 ] shows , @xmath215 so that the ising behavior persists to finite bias although the two minima at @xmath216 then have unequal energies . one minimum is a metastable state and the other a stable state . a hysteretic behavior is expected since the system can be trapped into the metastable state if the thermal energy is not sufficient to overcome the energy barrier between the two minima . a similar behavior was discussed in semiconductor ising qhfsjungwirthprb , but the hysteretic behavior found in bilayer graphene for valley pseudospins seems more complex . for @xmath217 the coefficient @xmath209 only at zero bias where there is a transition from @xmath218 ( @xmath219 ) to @xmath220 ( @xmath221 ) . in the ground state , all electrons are in the valley which has the _ lowest _ non - interacting energy @xmath222 . this is the situation depicted in fig . [ figure1](b ) . for positive bias , the metastable state at @xmath200 disappears at the bias @xmath223 mev where @xmath224 . in analogy with a magnetic system , @xmath225 can be considered as a coercive bias . for negative bias , @xmath201 is the metastable state and a similar scenario occurs . hereafter , we consider positive bias only since the transitions for negative bias are the opposite of those for positive bias . all landau levels @xmath226 behave in the same fashion . the smallness of @xmath227 for @xmath206 reflects the fact that @xmath228 increases very rapidly with bias . for @xmath229 the ground state has @xmath218 at finite bias ( the electrons are in the valley state with the _ biggest _ non - interacting energy ) until a critical bias @xmath230 mev indicated in fig . [ figure1](c ) where there is a jump from @xmath231 to @xmath220 . unlike @xmath217 the two minima in @xmath232 coexist up to a very large coercive bias @xmath233 mev that is well outside the limit of validity of our model that assumes no landau level mixing . in order to find the correct coercive bias for @xmath229 it is necessary to include this mixing . we discuss the hysteretic loop for @xmath19 below ( see fig . [ figure3 ] ( c ) ) , after we obtain the gap for the bulk and confined electron - hole excitations . with the two minima at @xmath234 and the maximum at @xmath235 energy per electron at the extremal values of @xmath236 for ( b ) @xmath237 ( c ) @xmath238 t@xmath6 ] the ising behavior can be extended to chirally - stacked @xmath239layer graphene ( @xmath240 ) , at least at zero bias . in the simplest tight - binding model where only @xmath241 and @xmath242 are kept and in the two - component modelmccann , the hamiltonian in the basis @xmath243 is @xmath244where @xmath245 the eigenvectors are of the form @xmath246with @xmath247 ( landau level @xmath248 corresponds to @xmath249 ) this model can be captured by our approach if only two - coefficients in the spinors of eqs . ( [ spinor1],spinor2 ) are kept . this implies , from eqs . ( [ yy0]-[yy5 ] ) , that @xmath250 the ising criteria in eq . ( [ ising ] ) , becomes@xmath251or @xmath252with@xmath253where @xmath254 is the separation between two adjacent layers . this inequality is satisfied for @xmath255 in chirally - stacked multilayer graphene . we now study the charged excitations of the ising qhfs at @xmath154 in landau levels @xmath21 we start in this section with the hartree - fock bulk quasiparticles ( electron and hole ) and consider in the next two sections the hartree - fock electron and hole quasiparticles confined to the coherent region of a valley - pseudospin domain wall and the spin and/or valley - pseudospin skyrmions and antiskyrmions . in each case , we are mostly interested in computing the energy to create a quasiparticle - antiquasiparticle pair with infinite separation which is simply the sum of the quasiparticle and antiquasiparticle energies . the minimal such energy defines the transport gap of the ising qhf which is measurable in transport experiment . in the uniform ising phases , the equation of motion for the matrix of green s functions ( eq . ( [ eqmotion ] ) ) gives four energy levels . when the ground state is @xmath256 ( i.e. @xmath257)@xmath184 the only nonzero order parameter is @xmath258 while when @xmath259 ( i.e. @xmath260 ) the only parameter is @xmath261 the first energy level is fully occupied and amongst the three possible bulk electron - hole excitations , two are always lower in energy : a spin flip and a valley - pseudospin flip . for a ground state in valley @xmath262 the spin and valley flips have the energy @xmath263 these energies are plotted as a function of the bias in fig . [ figure2](a ) for @xmath183 t and fig . [ figure2](b ) for @xmath238 t . for @xmath217 the lowest - energy bulk excitation ( i.e. the transport gap ) is a valley - pseudospin flip at small bias and it changes to a spin flip above @xmath264 mev . there is a discontinuity in the slope of the gap at @xmath265 for @xmath229 the lowest - energy excitation is also a valley - pseudospin flip until the critical bias @xmath266 mev . at this bias , a ground state transition causes a discontinuity in the gap : it abruptly changes from a valley - pseudospin flip to a spin flip . this is indicated by a the downward arrow in fig . [ figure2 ] . and @xmath267 ) and confined ( @xmath268 ) electron - hole excitations as a function of bias for : ( a ) @xmath269 t and ( b ) @xmath238 t. ] in the presence of disorder or at finite temperature , domain walls ( dw ) are created in an ising ferromagnet@xcite . these walls can have different shapes , even closing on themselves ( i.e. domain wall loops ) . in this section , we find a straight domain wall solution to the microscopic hartree - fock equation and compute the minimal energy required to excite an electron - hole pair in its coherent region . at filling factor @xmath270 the ground state is spin polarized and all electrons must be in valley @xmath271 or @xmath272 for a state modulated in only one direction , say along @xmath273 , we can take @xmath274 in the definition of the fourier transform of the green s function in eq . ( [ gx ] ) and so @xmath275the order parameters are now given by@xmath276the equation of motion ( eq . ( [ eqmotion ] ) ) simplifies to@xmath277 g_{\xi , \xi ^{\prime } } \left ( x,\omega _ { n}\right ) = \hslash \delta _ { \xi , \xi ^{\prime } } \label{dwequation } \\ & & + h^{\xi } \left ( x\right ) g_{\xi , \xi ^{\prime } } \left ( x,\omega _ { n}\right ) -\sum_{\xi ^{\prime \prime } } x^{\xi , \xi ^{\prime \prime } } \left ( x\right ) g_{\xi ^{\prime \prime } , \xi ^{\prime } } \left ( x,\omega _ { n}\right ) , \notag\end{aligned}\]]with the renormalized energies@xmath278where @xmath279and the interactions @xmath280the hartree - fock energy per electron becomes @xmath281 the formal solution of the green s function equation gives the band structure ( we leave implicit the dependence on @xmath65 of the variables here)@xmath282 , \]]where@xmath283and@xmath284 the order parameters are obtained from the self - consistent equations @xmath285 these equations are constrained by the condition @xmath286 so that there is no modulation of the _ total _ guiding - center density when the c2deg is modulated in one direction only . the electron - hole gap for the confined excitations is given by the minimal separation between the two bands i.e. by @xmath287 .\ ] ] we look for a dw solution by imposing the boudary conditions : @xmath288 and @xmath289 we start the iteration process with the simple seed@xmath290where @xmath291 is the length of the system in the direction of the modulation and @xmath292 the number of values of @xmath293 we take both parameters as large as possible in the numerical calculation . this seed produces a nel wall . multiplying the right - hand side of eqs . ( [ s1 ] ) and ( [ s2 ] ) by @xmath294 produces a bloch wall . both walls have the same energy . to represent the solution , we use a pseudospin representation where@xmath295so that @xmath296 is a unit field in @xmath65-space . figure [ figure3](a ) shows the profile of the angle @xmath297 between the pseudospin and the @xmath298 axis ( i.e. @xmath299 $ ] ) in the dw solution at zero bias for @xmath217 and @xmath89 t@xmath6 the corresponding band structure is shown in fig . [ figure3](b ) where the confined quasiparticle gap is indicated by a double arrow . the intervalley coherence @xmath300 is very strong near @xmath301 so that the gap for the confined electron - hole pairs is smaller than the electron - hole bulk gap by a small amount only . bulk and confined gaps as a function of bias are compared in fig . [ figure2 ] . since the spin is not considered in our dw solution , @xmath302 should be compared with the valley - pseudospin gap only in this figure . as the bias is increased positively , the coherence region of the dw for @xmath206 in fig . [ figure3](a ) moves to the left , indicating that the proportion of the stable @xmath303 phase in the dw solution increases . at the same time , the metastable minimum increases in energy . for @xmath304 mev , only the @xmath305 phase survives and there is no dw solution anymore . the dw solution disappears before the coercive bias @xmath306 mev is reached . for @xmath229 there is a dw solution up to @xmath307 mev , a value much lower than @xmath308 mev . already at @xmath309 mev , the proportion of the stable phase @xmath271 in the dw has reached @xmath310 no dw solution is found at larger bias until near the critical bias @xmath230 mev . at this bias , the two ising states have again the same energy ( i.e. @xmath209 ) . our numerical analysis shows that this second dw solution exists in the range @xmath311 $ ] mev . like the case @xmath217 the dw solution is lost before the coercive bias is reached . the critical bias @xmath312 is above the limit of validity of our model but it is possible to modify either the magnetic field or the dielectric constant to find a smaller @xmath313 figure [ figure3](c ) shows the hysteretic loop expected around zero bias for @xmath206 . the dotted lines indicate the position of the coercive bias . if the confined quasiparticles are not taken into account , then the arrows indicate the progression of the gap for a downward or upward sweep in bias . for a downward sweep , the gap decreases and then jumps suddenly , at @xmath314 to its value in the phase where the electrons occupy the valley @xmath315 just the opposite occurs for an upward sweep . if skyrmions are considered , this behavior is not changed substantially since the energy of a skyrmion - antiskyrmion pair is slightly different from that of the bulk quasiparticles only in a small range of bias ( see sec . if confined quasiparticles are considered however ( i.e. when domain walls are present ) , then the gap jumps discontinuously at @xmath316 and its value remains almost constant for @xmath317 the large value of the coercive field for @xmath229 if it survives landau - level - mixing corrections , should lead to a large hysteretic loop . in the absence of interaction , the transport gap is zero at zero bias and the qhe is lost if @xmath318 the exchange interaction , however makes the gap finite . in this case , the qhe is lost then when the gap is smaller than the disorder broadening of the landau levels . and hysteretic behavior . ( a ) comparison of the sine - gordon and hartree - fock solutions . ( b ) band structure , @xmath319 in the domain wall solution . the gap for a confined electron - hole excitation is indicated by @xmath320 ( c ) hysteretic behavior of the transport gap around zero bias . the plain ( dashed ) arrows indicate the progression of the gap for a downward ( upward ) sweep in bias . ] it is interesting to compare the hartree - fock solution with the dw solution that can be obtained by solving the euler - lagrange equation@xcite derived from the energy functional of eq . ( [ evallee ] ) . neglecting the capacitive term and at zero bias , the excitation energy is @xmath321 \left ( \partial _ { x}\theta \right ) ^{2 } \\ & & + kl_{y}\int dx\cos ^{2}\theta . \notag\end{aligned}\]]considering a nel wall and neglecting the small difference between the stiffnesses @xmath322 and @xmath323 at zero bias , the euler - lagrange equation simplifies to the sine - gordon equation@xmath324where @xmath325 and @xmath326imposing the boundary conditions @xmath327 and @xmath328 a solution that minimizes the energy is the sine - gordon soliton@xcite @xmath329where @xmath330 is the domain wall width ( i.e. the coherence region ) . note that , in the sine - gordon solution , @xmath331 and the capacitive term can effectively be neglected . in fig . [ figure3](a ) , the hartree - fock and sine - gordon dw solutions for @xmath332 and @xmath89 t are compared . the sine - gordon solution is actually very close to the hartree - fock result . the domain wall width extracted from eq . ( [ gordon ] ) is @xmath333 . this is much larger than the domain wall width found in alas quantum wells@xcite . to compute the skyrmion excitations , it is more convenient to use the symmetric gauge @xmath334 with @xmath335 instead of the landau gauge . the landau gauge wave functions @xmath336 in the spinors of eqs . ( [ spinor1])-([spinor2 ] ) are then be replaced by the functions @xmath337 where the quantum number @xmath338 is related to the angular momentum @xmath339 by @xmath340 the functions @xmath341 are given by@xcite ( by definition , @xmath342 if @xmath343 or @xmath344)@xmath345where @xmath346 is the angle between the vector @xmath347 and the @xmath273 axis , @xmath348 is a generalized laguerre polynomial and the normalization constant is given by@xmath349with@xcite @xmath350 if @xmath351 and @xmath352 if @xmath353 . in the hartree - fock approximation , the hamiltonian in the symmetric gauge is given by @xmath354where the matrix elements of the coulomb interaction are defined by@xmath355with@xmath356and @xmath357where @xmath358 or @xmath359 when @xmath360 or @xmath361 the hamiltonian @xmath118 includes an interaction with a positive background where the positive charges are assumed to occupy the same state , @xmath362 as the electrons in the ground state . at @xmath270 the ground state has @xmath256 or @xmath259 and can be written as @xmath363if , for example , @xmath364 then a general antiskyrmion ( one electron removed from the ground state ) that allows for the possibility of an intertwined spin and valley - pseudospin textures can be written as @xmath365where the operator@xmath366for a skyrmion ( one electron added to the ground state ) , three types of excitations are possible@xcite depending on the level the extra electron is added to i.e. : @xmath367with @xmath368 and@xmath369@xmath370@xmath371the normalization condition imposes @xmath372 for each @xmath373 a similar scheme is applied if the ground state has @xmath374 when the coefficients @xmath375 for @xmath376 the antiskyrmion or skyrmion excitations are reduced to the bulk hole @xmath377 or bulk electron excitations @xmath378 respectively . if @xmath379 the excitation is a pure valley - pseudospin skyrmion , @xmath380 or valley - pseudospin antiskyrmion , @xmath381 with spin @xmath382 if @xmath383 ( or @xmath384 ) , the excitation is a pure spin - skyrmion , @xmath385 or spin - antiskyrmion , @xmath386 in valley @xmath387 ( or @xmath388 ) . to compute the coefficients @xmath389 for the antiskyrmion excitation , we define the @xmath390 matrix of green s functions @xmath391 where @xmath392 and write down its hartree - fock equation of motion using the hamiltonian of eq . ( [ hhf ] ) . the procedure is described in detail in ref . and is equivalent to the canonical transformation method used to compute spin - skyrmion in ref . . the resulting self - consistent equation for the matrix green s function is solved numerically in an iterative way until a converging solution is found . we proceed in a similar way for the three skyrmion excitations . the occupation of each quantum state @xmath393 is given by @xmath394 and the coherence factors by @xmath395 ( @xmath396 ) . these coherences are responsible for the spin and valley - pseudospin textures . they are obtained from the off - diagonal elements of the matrix @xmath397 . in the numerical calculation , the maximum value of @xmath393 is set to @xmath398 so that skyrmions up to @xmath399 in size can be obtained . for a pure spin skyrmion , the number of flipped spins per skyrmion when the ground state has electrons with spin @xmath166 in valley @xmath400 or @xmath401 is given by@xmath402while the number of flipped valley pseudospins per skyrmion is @xmath403 we choose @xmath256 for the ground state at zero bias . figure [ figure4](a ) shows the excitation energies of the antiskyrmion and bulk hole as a function of the zeeman coupling for landau level @xmath206 . note that , with a maximum number of angular momenta @xmath398 , we are limited to the range @xmath404 mev * * * * otherwise the skyrmion is too big . * * * * figure [ figure4 ] shows that the antiskyrmion has lower energy than the bulk hole in the small zeeman range @xmath405 $ ] mev . there is no intertwined spin - valley pseudospin texture in this range . indeed , for @xmath406 mev , the excitation is a pure spin antiskyrmion , @xmath407 while above this value it is a pure valley - pseudospin antiskyrmion,@xmath408 . the corresponding number of spin flips @xmath409 or valley - pseudospin flips @xmath410 is shown in fig . [ figure5](a ) as a function of the zeeman coupling . naturally , @xmath411 is independent of the zeeman coupling while @xmath412 increases with decreasing @xmath413 and magnetic field @xmath414 t. ] figure [ figure4](b ) shows the excitation energy for the skyrmion and bulk electron states for @xmath206 and zero bias . the state @xmath415 has the lowest energy amongst the bulk electron states since the two other bulk states @xmath416 and @xmath417 ( not shown in the figure ) are degenerate and involve a spin flip . the lowest - energy skyrmion is @xmath418 for @xmath419 mev and @xmath420 above this value . the transition between these two skyrmion types is discontinuous . again , no intertwined texture is found . the excitation @xmath421 and @xmath422 are respectively a spin and a valley - pseudospin skyrmion . apart from a global zeeman shift of their energy due to added or removed electron , the energy of the valley - pseudospin skyrmion and antiskyrmion is independent of the zeeman coupling since these excitations do contain spin flip . moreover , the coefficients of the eigenspinors in eqs . ( [ spinor1]-[spinor2 ] ) do not depend on the zeeman coupling either . the number of spin flips for @xmath423 is identical to that for @xmath424 . the sum of the skyrmion and antiskyrmion energies gives the energy needed to create a skyrmion - antiskyrmion pair with an infinite separation . at zero zeeman coupling , the spin skyrmion - antiskyrmion pair has lower energy than the bulk electron - hole pair in the range @xmath425 $ ] mev . in graphene however , @xmath426 mev if the magnetic field is not tilted . at this value , the relevant excitations for the transport gap ( i.e. the pair excitations with the lowest energy ) are the valley - pseudospin skyrmion - antiskyrmion pairs . ) or valley pseudospin flips ( @xmath411 ) as a function of the zeeman coupling for the lowest - energy excitations in fig . 4 . ] to study the valley - pseudospin skyrmions as a function of bias , we take @xmath427 mev , i.e. a value large enough to kill the spin texture . the ground state for @xmath16 has @xmath259 for @xmath221 and @xmath256 for @xmath219 . we find a pure valley - pseudospin skyrmion in this case : @xmath428 for @xmath221 and @xmath420 for @xmath429 the excitation energy of valley - pseudospin antiskyrmion and skyrmion is plotted in fig . [ figure6 ] ( a ) and ( b ) together with the energy of the corresponding bulk electron and hole . the valley - pseudospin texture gradually disappears as @xmath430 increases . the range of bias where the topological excitations win over the bulk ones is small : @xmath431 mev and @xmath432 mev respectively . the number of valley - pseudospin flips , @xmath411 is shown in fig . [ figure7 ] . it is maximal at zero bias and decreases rapidly with increasing bias . at zero bias , @xmath433 and so these skyrmions are very small in space unlike the spin skyrmions at zero zeeman coupling . @xmath89 t and @xmath434 the excitations are @xmath435 and @xmath436 for positive bias and @xmath437 and @xmath438 for negative bias . ] the induced density profiles @xmath439 of the spin and valley - pseudospin skyrmions or antiskyrmions are similar to that of spin skyrmions in a semiconductor 2deg@xcite . the density @xmath440 is maximal and positive at the origin for a skyrmion ( or minimal and negative for an antiskyrmion ) . it is isotropic and decreases(increases ) to zero at infinity for skyrmion(antiskyrmion ) . the integrated density @xmath441 contains one more(less ) electron for a skyrmion(antiskyrmion ) . a plot of the valley - pseudospin texture in the projected representation , which is defined in appendix a , gives the pattern represented in fig . [ figure8 ] . the pairing with @xmath442 in eqs . ( [ l1])-([l4 ] ) leads to a spin or valley - pseudospin vortex around the origin where the pseudospins rotate by @xmath443 . and magnetic field @xmath89 t. ] and magnetic field @xmath89 t in the projected representation . ] figure [ figure9 ] shows the energy of the different pair excitations in landau level @xmath206 with @xmath89 t and in level @xmath207 with @xmath208 t with the zeeman coupling @xmath444 in both cases . spin skyrmion - antiskyrmion pairs are not relevant at these zeeman couplings and so only the bulk and confined electron - hole pairs and the valley - pseudospin skyrmion - antiskyrmion pairs are considered . from this figure , it is clear that , if domain walls are present , then the confined pairs are the relevant excitations at small bias . otherwise , skyrmion - antiskyrmion pairs win but their energy is only slightly lower than that of the bulk quasiparticles . t ; ( b ) @xmath238 t. ] we were not able to obtain spin or valley - pseudospin skyrmions in landau levels @xmath445 . for the spin skyrmions , it was found@xcite that in graphene ( monolayer ) , the range of zeeman coupling where skyrmions are the lowest - energy excitations decreases very rapidly with increasing @xmath446 we assume that the same behavior applies to bilayer graphene and that , if skyrmions exist ( according to the nonlinear @xmath146 model discussed in the next section , they do for some higher values of @xmath447 ) , then they do so in a zeeman range not accessible to our numerical approach . for valley - pseudospin skyrmions , the number of pseudospin flips shown in fig . [ figure8 ] indicates that they are very small . increasing the landau level index will make them even smaller , indistinguishable from the bulk charged excitations . it may seem surprising that the excitation energy of the spin skyrmion is lower than that of the valley - pseudospin skyrmion at small zeeman coupling and zero bias ( see fig . [ figure4 ] ) since , in the non - interacting picture , there is no cost for a valley pseudospin flip and the exchange - energy cost seems similar in the two excitations . however , as shown in sec . iii , the energy functional in the gradient approximation is very different for spin and valley - pseudospin textures . for valley pseudospins , the exchange interaction is anisotropic and there is an easy - axis anisotropy term and an easy - plane term ( the capacitive energy ) . in the limit @xmath448 skyrmions have a slowly - varying spin texture and their excitation energy can be computed from the nonlinear @xmath449 model ( nl@xmath146 m ) energy functional in eq . ( [ efunspin ] ) . the energy needed to create a spin skyrmion - antiskyrmion pair with infinite separation is@xcite@xmath450where the spin stiffness is defined in eq . ( [ stifspin ] ) . this value provides a lower bound for the hartree - fock calculations shown in fig . figure4 since the limit @xmath451 can not be reached numerically . according to eq . ( [ eskask ] ) , a spin skyrmion - antiskyrmion pair at zero zeeman and zero bias has an energy lower than a bulk electron - hole pair for @xmath452 $ ] and @xmath453 $ ] at @xmath89 t. table 1 lists the value of these two gap energies for different landau levels . .energy of a bulk electron - hole pair and a spin skyrmion - antiskyrmion pair in the nl@xmath454 m calculated as a function the landau level index @xmath92 at zero bias and zero zeeman coupling and at @xmath89 t. the results are the same in the two valleys . in each entry , the upper(lower ) value is for the positive(negative ) values of @xmath92 . [ cols="<,<,<",options="header " , ] if the easy - axis anisotropy and capacitive terms could be neglected in the anisotropic nonlinear @xmath146 model ( anl@xmath146 m ) energy functional of eq . ( [ evallee ] ) , then the valley - pseudospin skyrmion - antiskyrmion excitation ( or bimeron - antibimeron ) energy would be given by @xcite@xmath455and the number of valley - pseudospin flips would diverge at zero bias . as our microscopic calculations shows , this does not happen . valley - pseudospin skyrmions have a finite small size at zero bias and so eq . ( [ bimeron ] ) can not be used to compute their energies . it is necessary to include the capacitive and easy - axis term as well as the hartree energy due to the density modulation induced by the skyrmions . all these terms are included in our microscopic hartree - fock calculation of the skyrmion energy . inter - landau - level spin - skyrmions have been studied before in semiconductor quantum wells@xcite but it was shown that are never the lowest - lying charged excitations . by contrast , we find here , in bilayer graphene , that skyrmions where different orbitals ( but of the same landau level ) are mixed can be the lowest - lying charged excitations . we have computed the energy of different types of pair excitations in the ising quantum hall ferromagnetic states of bilayer graphene in landau levels @xmath456 . these excitations include bulk electron - hole pairs , electron - hole pairs confined to the coherent region of a domain wall , and skyrmions - antiskyrmion pairs . for a disorder - free systems where the coulomb interaction is treated in the hartree - fock approximation , valley - pseudospin skyrmion - antiskyrmion pairs have a lower excitation energy than bulk electron - hole pairs . at finite temperature or in a disordered system where domain walls are created , confined electron - hole pairs are the lowest - energy excitations but only in a small range of bias . when confined electron - hole pairs are excited , it should be possible , according to the theory developed in ref . to observe a breakdown of the quantum hall effect . in previously studied ising systems in semiconductor quantum wells , this breakdown lead to the apparition of magnetoresistance spikes near the ising transitions . these spikes were observed experimentally@xcite . according to ref . , when the chemical potential is pinned in the bulk quasiparticle energy , the confined electron - hole pairs can be easily excited . if the domain wall loops are dense enough to overlap , these confined quasiparticles can cross the sample by scattering from one loop to an adjacent one and then increase the dissipation in the sample , creating magnetoresistance spikes that are detected near integer fillings close to the transition between the ordered and the paramagnetic phases at finite temperature . as far as we know , there has been to date no experimental study of such effect in the higher landau levels of bilayer graphene . the domain wall solution discussed in this work comes from the coulomb interaction . it is different from the domain walls studied before in bilayer graphene@xcite which were artificially created in the middle of a double - gated bilayer graphene sample by placing it in an electric gate where the potential profile changes sign across the center@xcite . domain walls have also recently been studied in bilayer graphene but at zero magnetic field@xcite . another type of excitation should be considered in addition to those studied in this work . it consists of confined textured excitations in the coherent regions of the domain walls i.e. confined solitons and antisolitons of the valley pseudospin in the one - dimensional coherent region of a wall . it is possible to extract an effective pseudospin hamiltonian for such excitations by combining the long - wavelength energy functional presented in this paper with the microscopic hartree - fock calculation . this has been done previously in ising qhfs@xcite and in the coherent stripe phase in a semiconductor double quantum - well system@xcite . we leave this problem to future work . r. ct was supported by a grant from the natural sciences and engineering research council of canada ( nserc ) . computer time was provided by calcul qubec and compute canada . the total density in landau level @xmath92 is given by@xmath457 in the absence of valley coherence , the spin density field in real space is given by@xmath458and@xmath459similarly , in the absence of spin coherence , the valley - pseudospin field given in real space by@xmath460and@xmath461 these density and fields contain the character of the different landau level orbitals @xmath72 in @xmath341 that enter the definition of the spinors in eq . ( [ spinor1],[spinor2 ] ) with the wave functions in the symmetric gauge . since these spinors contain a mixture of the orbitals @xmath462 and @xmath463 the resulting field pattern of a skyrmion in real space is complex . to get a simpler visualization of the spin and pseudospin fields , it is useful to use a projected representation that we define in the following way . we write@xmath464where , for example,@xmath465and the other components are easily obtained from the spinors in eqs . ( spinor1,[spinor2 ] ) . we define the fourier transform of @xmath466 as @xmath467 in the symmetric gauge , the matrix elements that enter the fourier transform are given by@xmath468 ^{\ast } f_{m_{1},m_{2}}\left ( -\mathbf{q}\right ) , \notag\end{aligned}\]]with @xmath469where the upper sign is for @xmath470 and the lower sign for @xmath471 and @xmath472 is a generalized laguerre polynomial@xmath6 it follows that @xmath473where @xmath474 does not involve the quantum number @xmath393 and is given by @xmath475 ^{\ast } \\ & & -b_{+,2}^{\ast } b_{-,2}\left [ f_{n-2,n}\left ( \mathbf{q}\right ) \right ] ^{\ast } \notag \\ & & -b_{+,3}^{\ast } b_{-,3}\left [ f_{n , n-2}\left ( \mathbf{q}\right ) \right ] ^{\ast } \notag \\ & & + b_{+,4}^{\ast } b_{-,4}\left [ f_{n-1,n-1}\left ( \mathbf{q}\right ) \right ] ^{\ast } . \notag\end{aligned}\]]we thus get@xmath476 for a review of some of the properties of graphene and bilayer graphene , see for example : a. h. castro neto , f. guinea , n. m. r. peres , k. s. novoselov and a. k. geim , rev . phys . * 81 * , 109 ( 2009 ) ; d. s. l. abergel , v. apalkov , j. berashevich , k. ziegler and tapash chakraborty , advances in physics * 59 * , 261 ( 2010 ) ; m. o. goerbig , rev . phys . * 83 * , 1193 ( 2011);edward mccann and mikito koshino , rep . prog . phys . * 76 * , 056503 ( 2013 ) . the value of the hopping parameters that we use are those recently computed by jeil jung and allan h. macdonald , phys . rev . b * 89 * , 035405 ( 2014 ) . our definition of the parameters @xmath477 differs by a minus sign from that used by these authors . chia - wei huang , efrat shimshoni , and h. a. fertig , phys . b * 85 * , 205114 ( 2012 ) ; victoria mazo , chia - wei huang , efrat shimshoni , sam t. carr , and h. a. fertig , phys . rev . b * 89 * , 121411 ( r ) ( 2014 ) .